THEORETICAL AND EXPERIMENTAL

INVESTIGATION OF THE NON-NORMAL NATURE

OF THERMOACOUSTIC INTERACTIONS

A THESIS

submitted by

SATHESH MARIAPPAN

for the award of the degree

of

DOCTOR OF PHILOSOPHY

DEPARTMENT OF AEROSPACE ENGINEERING

INDIAN INSTITUTE OF TECHNOLOGY MADRAS
MARCH 2012
 THESIS CERTIFICATE

This is to certify that the thesis titled THEORETICAL AND EXPERIMENTAL IN-
VESTIGATION OF THE NON-NORMAL NATURE OF THERMOACOUSTIC
INTERACTIONS, submitted by Sathesh Mariappan, to the Indian Institute of Tech-
nology, Madras, for the award of the degree of Doctor of Philosophy, is a bona fide
record of the research work done by him under my supervision. The contents of this
thesis, in full or in parts, have not been submitted to any other Institute or University
for the award of any degree or diploma.

Dr. R. I. Sujith
Research Guide
Professor
Department of Aerospace Engineering
Indian Institute of Technology Madras, 600 036

Place: Chennai
Date : 22nd March 2012
 ACKNOWLEDGEMENTS

I would like to thank and acknowledge the help rendered by all the people during my
past four years of research. It was a wonderful experience for me to perform my Ph.
D., in this pleasant ambiance. During the years, several people have influenced my
academic career at various point of time.

First and foremost, I would like to express my gratitude to my supervisor, Prof. R.

I. Sujith for his exceptional guidance and support. I am always amazed by his enthusi-
asm and extraordinary efforts that he puts in learning new things, which is a source of
motivation. In fact, without his motivation and effort, I would not have joined the Ph.
D. programme. He has shaped my career and taught me lot of things, both academic
and non-academic, which helped and will help me to continue my career in the future.
Further, I owe him for the enormous freedom that he has given and the faith he had in
me.

I would like to express my gratitude to the members of my Doctoral Committee,

Prof. S. Narayanan, Prof. K. Balasubramaniam, Prof. S. R. Chakravarthy, Dr. S. Sarkar
and Dr. N. K. Sinha for their valuable comments and suggestions during the meetings. I
would like to thank our HODs, Prof. J. Kurian, Prof. P. Sriram and Prof. V. Bhaskar for
their support and encouragement during my stay. My sincere gratitude to my teachers,
Prof. S. Santhakumar, Prof. M. Ramakrishna, Dr. Velmurugan, Dr. N. R. Pancha-
pakesan, Dr. Amit Kumar, Dr. T. Muruganandam (Aerospace Engineering), Prof. N.
Gupte, Prof. S. Govindarajan and Prof. V. Balakrishnan (Physics) for their exceptional
lectures, which helped to understand the concepts clearer and better. I would like to
acknowledge the excellent suggestions from Prof. V. Jayashankar (Electrical Engineer-
ing) in designing the high current heater that I use in my experimental setup. Special
thanks to Dr. A. Tangirala and Dr. N. Kaisare (Chemical Engineering) for their interests
and fruitful discussions during the course of my study.

Throughout my stay, I have been supported constantly by the administrative staff,

i
Mrs. Mekala, Mr. Sundar, Mr. Dhanapal, Mr. Stephen, Mr. Siva, Mr. Manikandan,
Ms. Rajalakshmi, Ms. Pavitra and Mrs. Saraswathi (institute administration) during my
paper work. I would like to acknowledge the great help provided by the staff members in
the work shop, Mr. Shankarakumarasamy, Mr. Kennedy, Mr. Divakaran, Mrs. Yamuna
and Mr. Dayalan. Special thanks to Mr. John George (Central workshop incharge)
for his suggestions regarding safety in my experimental setup. I am thankful to the
Engineering Unit of IITM, for laying the three phase outlet for my experiment. Many
thanks to Mr. Ranganathan and Mr. Chandru for their excellent fabrication of most of
the components in my experimental setup.

I am indebted to Prof. P. J. Schmid (LadHyX, Ecole Polytechnique) for answering

patiently to my never ending questions regarding non-normality and warm hospitality
during my internship in LadHyX. I wish to thank Prof. W. Polifke (Technical University,
Munich) for his generous support and critical comments during his visits, which made
me to think further and understand the problem better. Further, his great hospitality dur-
ing my visits to Munich gave me an opportunity to collaborate with other researchers. I
am indebted to Dr. P. Wahi, Prof. M. K. Verma and Mr. A. Saha (IIT Kanpur)for intro-
ducing me to the technique of continuation to analyse nonlinear dynamical systems and
their support during my stay in IIT Kanpur. Special thanks to Prof. Rama Govindarajan
(JNCASR) and Prof. Joseph Mathew (IISc) for their enthusiastic support and interest in
my work. My understanding of thermoacoustic engines and refrigerators have greatly
improved during the discussions with Dr. P. Bala Subrahmanyam (CEERI, Chennai)

During my Ph. D., I have greatly benefited by attending many workshops. To indi-
cate a few, SERC school on nonlinear dynamics at IISc, AIM meetings in IIT Madras
and JNCASR, organised by Dr. M. Juniper (University of Cambridge) and Prof. Sujith
and FLOW Summer School Aeroacoustics in Low Mach Number Confined Flows, or-
ganised by Dr. Gunilla Efraimsson (Royal Institute of Technology, Stockholm). These
workshops gave me a good exposure to the tools of nonlinear dynamics and aeroacous-
tic/hydrodynamic stability theory.

I would like to acknowledge the financial assistance from the institute and alumni
association to present some of my papers in the international conferences, which gave
me a good exposure of the outside research world. I am also grateful to the Depart-

ii
ment of Science and Technology (DST) and Ministry of Human Resource Development
(MHRD), India for funding my research, stay and travel. Further, the travel support
from Deutsche Forschungsgemeinschaft (DFG), Institute for Advanced Studies (IAS),
Munich, Germany, Ecole Polytechnique, Paris, France and Royal Institute of Technol-
ogy (KTH), Stockholm, Sweden are greatly acknowledged.

This acknowledgement would never be complete, if I do not mention my lab and

hostel friends. I am very lucky to got some of best friends of my life over the years in
the lab. Everybody in the lab have helped in some way or other during my stay. I am
greatly indebted to all of them, particularly, Priya for her endless hospitality especially
during my initial period in the lab, Lipika for her kindness in the fabrication of the ex-
perimental setup, which gave me a quick start for my Ph. D., Rahul for his soft and
friendly company. My sincere thanks to Balaji for teaching me SIMPLE algorithm and
I am always amazed by his patience, which was a motivation during my hard times. I
enjoyed the company of Ramgopal & Vivek, with whom I have spent a lot of time in
the later part of my stay and will cherish the moments in the future. I thank my senior
friends Kushal, Sharath, Bharath, Koushik, Guru, O. J. Sreenivasan, Jayaraman, Jaya-
surya and Sreerekha for their support and suggestions in handling numerical codes and
hardware instruments. My understanding have greatly improved during my discussions
with Rajini, Devendra, Irfan, Rajarishi Das, Rajeev, Sonu, Srinath, Mayank, Joseph,
Thomas Fiala, David, Ramesh (Linux Guru), Ganesh, Avishek, Vikrant, Vineeth, Vinu,
Tharun, Girish sir, Rana sir, Ralf, Santhosh, Trinath, Rajesh, Vishnu, Vishnu Unni,
Roopa, Pushkarini, Harshini, Gopalakrishnan and Meenakshi.

In the hostel, I am happy to have got the company of Rajasekar and I use to stay in
his room whenever I get bored. I have shared lots of happy and memorable moments
with my friends, Somasundaram (taught me to play tennis), Raju, Mahendran, Shyam,
Shankar (tennis mate and helped me with his electrical knowledge in my experimental
instrumentations), Paramasivam, Ravikumar, Gnanasekhar, Mayil, Venkatesh, Mahesh,
Joseph. Further thanks for the help rendered by the hostel administrative staff, Mr.
Rajamani, Mr. Balaji (Swamy).

Finally, I should acknowledge that all my success that I had and will have in my life
are due to my parents. I always thank them for their careful attention towards me and

iii
dedicate the present thesis to them. I thank God for blessing me with good health and
giving me everything that I enjoy in my life.

iv
 ABSTRACT

KEYWORDS: thermoacoustic instability; non-normality; transient growth;

global-acceleration; adjoint optimisation; dynamic mode decom-
position

Thermoacoustic instability is a challenging problem in solid and liquid rockets, ram-

jets, aircraft and industrial gas turbines etc. During instability, acoustic oscillations are
sustained by the unsteady heat release rate from the heat source (flame) in the com-
bustion chamber, which causes detrimental effects to the operation and leads to prema-
ture failure of aero engines. Thermoacoustic instability occurs when acoustic pressure
oscillations in the combustion chamber are amplified by the positive feedback of the
unsteady heat release rate. As a first step in the theoretical analysis of stability of ther-
moacoustic interactions, the governing equations are linearized around the steady state.
The location of the eigenvalues in the complex plane determines the linear stability of
the system. The predictions of stability thus obtained are valid only in the asymptotic
time limit; i.e., t .

Recently, it was shown that thermoacoustic systems are non-normal and that classi-
cal linear stability analysis is not sufficient to completely determine the linear stability
of the system. Due to non-normal nature of the thermoacoustic interaction, the eigen-
modes of the system are non-orthogonal. In such systems, a transient growth in the
amplitude of the initial perturbation is observed, even for a system stable according to
classical stability analysis. The transient growth thus obtained can be of several or-
ders of magnitudes, which along with the nonlinearities present in the system plays an
important role in the regime of subcritical transition to instability.

In the present thesis, two thermoacoustic systems are analysed: 1. solid rocket
motor (SRM) and 2. Rijke tube. To begin with, the case of a SRM is particularly
chosen, as orthogonality of the eigenmodes (normal system) was assumed explicitly

v
in the earlier investigations. An analytical framework is developed to understand and
predict thermoacoustic instability in SRM, taking into account the non-orthogonality
of the eigenmodes of the unsteady coupled system. The coupled system comprises the
dynamics of the acoustic field and the propellant burn rate. The terms contributing to
the non-normality in the acoustic field and unsteady burn rate equations are identified.
These terms, which were neglected in the earlier analyses, are incorporated in this anal-
ysis. Furthermore, the short-term dynamics are analysed using a system of differential
equations that couples the acoustic field and the burn rate, rather than using ad hoc
response functions which were used in earlier analyses. In this thesis, a solid rocket
motor with homogeneous propellant grain has been analysed. Modeling the evolution
of the unsteady burn rate using a differential equation increases the degrees of freedom
of the thermoacoustic system. Hence, a new generalized disturbance energy (analogous
to Chus energy) is defined which measures the growth and decay of the oscillations.
This disturbance energy includes both acoustic energy and unsteady energy in the pro-
pellant and is used to quantify the transient growth in the system. Nonlinearities in
the system are incorporated by including second-order acoustics and a physics-based
nonlinear unsteady burn rate model. Nonlinear instabilities are analysed with special
attention given to pulsed instability. Pulsed instability is shown to occur with pressure
coupling for burn rate response. Transient growth is shown to play an important role in
pulsed instability.

As experimental measurements in SRM are difficult to perform in consideration of

the harsh environment encountered, a simpler thermoacoustic system, the horizontal
Rijke tube, is chosen for further analysis. This constitutes the rest of the thesis. In
the past, apart from the non-normal nature, the coupling between the acoustic field and
the heat release rate was ad hoc. It is necessary to establish the above coupling with
mathematical rigor, as it forms the basis for non-modal stability analysis.

An analysis of thermoacoustic instability is performed for a horizontal Rijke tube

with an electrical resistance heater as the heat source. The governing equations for this
fluid flow become stiff and are difficult to solve using computational fluid dynamics
(CFD) technique, as the Mach number of the steady state flow and the thickness of the
heat source (compared to the acoustic wavelength) are small. Therefore, an asymptotic

vi
analysis is performed in the limit of small Mach number of the steady flow and com-
pact heat source to eliminate the above stiffness problem. The unknown variables are
expanded in powers of steady flow Mach number. Two systems of governing equa-
tions are obtained: one for the acoustic field and the other for the unsteady flow field
in the hydrodynamic zone, which is around the heater. In this analysis, the coupling
between the acoustic field and the unsteady heat release rate from the heater appears
from the asymptotic analysis. Furthermore, a non-trivial additional term, referred to as
the global-acceleration term, appears in the momentum equation of the hydrodynamic
zone, which has serious consequences for the stability of the system. This term can be
interpreted as a pressure gradient applied from the acoustic zone onto the hydrodynamic
zone. The asymptotic stability of the system with the variation of system parameters
is presented using the bifurcation diagram. Numerical simulations are performed using
the Galerkin technique for the acoustic zone and CFD techniques for the hydrodynamic
zone. The results confirm the importance of the global-acceleration term. Bifurcation
diagrams obtained from the simulations including and excluding the above term are dif-
ferent. Acoustic streaming is shown to occur during the limit cycle and its effect on the
unsteady heat release rate is discussed.

A theoretical framework has been developed to understand the non-normal nature

of thermoacoustic interaction in the above Rijke tube system. The amount of transient
growth is measured in a norm, which in general physically represents the energy in
the disturbance. There has been a significant discussion in the past on choosing the
correct energy in order to describe the non-normal nature of the system. The present
framework allows one to obtain a norm systematically from the definition of disturbance
energy (Myers energy). As the degrees of freedom encountered is large ( 104 ), ad-
joint optimisation technique is chosen over the traditional singular value decomposition
to obtain the optimum initial condition. As before in the case of SRM, the optimum ini-
tial condition thus obtained has significant contributions from the variables governing
the dynamics of the heat source. Some interesting flow structures are observed in the
optimum initial condition near the heat source.

As a final step, an experimental investigation of the non-normal nature of thermoa-

coustic interactions in the same Rijke tube system is performed. Since non-normality

vii
and the associated transient growth are linear phenomena, experiments have to be con-
fined to the linear regime. The bifurcation diagram for the subcritical Hopf bifurcation
into a limit cycle behavior has been determined, after which the amplitude levels, for
which the system acts linearly, have been identified for different power inputs to the
heater. There are two main objectives for this experimental investigation. The first one
deals with the extraction of the linear eigenmodes associated with the acoustic pressure
from experimental data. This is accomplished by the Dynamic Mode Decomposition
(DMD) technique applied in the linear regime. The non-orthogonality between the
eigenmodes is determined for various values of heater power. The second objective is
to identify evidence of transient perturbation growth. The total acoustic energy in the
duct has been monitored as the thermoacoustic system has been initialized by linear
combinations of the two dominant eigenmodes. Transient growth in acoustic energy, on
the order of previous theoretical studies, has been found, and its parameter dependence
on amplitude ratio and phase angle of the initial eigenmode components has been de-
termined. This study represents the first experimental confirmation of non-normality in
thermoacoustic systems.

viii
 TABLE OF CONTENTS

ACKNOWLEDGEMENTS i

ABSTRACT v

LIST OF TABLES xv

LIST OF FIGURES xxviii

ABBREVIATIONS xxix

NOTATION xxxi

1 Introduction 1
1.1 Classical linear stability analysis . . . . . . . . . . . . . . . . . . . . 4
1.2 Non-normality and transient growth . . . . . . . . . . . . . . . . . . 5
1.3 Non-Normality in thermoacoustic system . . . . . . . . . . . . . . . 8
1.4 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.1 Thermoacoustic instabilities in solid rocket motors . . . . . . 10
1.4.2 Thermoacoustic instabilities in Rijke tube . . . . . . . . . . . 11
1.4.3 Non-modal stability theory . . . . . . . . . . . . . . . . . . . 12
1.4.4 Non-normality of thermoacoustic systems . . . . . . . . . . . 14
1.4.5 Mathematical modelling of thermoacoustic systems . . . . . . 18
1.5 Existing understanding and outstanding issues . . . . . . . . . . . . . 18
1.6 Objective of the present thesis . . . . . . . . . . . . . . . . . . . . . 22
1.7 Tools to be used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.8 Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2 Thermoacoustic instability in a solid rocket motor 27

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

ix
 2.2 Role of non-normality . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.1 Chamber acoustics . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.2 Solution procedure . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.3 Damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.4 Unsteady burn rate . . . . . . . . . . . . . . . . . . . . . . . 38
2.4 Short term dynamics and transient growth . . . . . . . . . . . . . . . 42
2.5 Linear analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.6 Generalised disturbance energy . . . . . . . . . . . . . . . . . . . . . 46
2.7 Result and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.7.1 Linearly stable and unstable system . . . . . . . . . . . . . . 50
2.7.2 Pseudospectra and transient growth . . . . . . . . . . . . . . 54
2.7.3 Effects of the internal degrees of freedom . . . . . . . . . . . 55
2.7.4 Pulsed instability . . . . . . . . . . . . . . . . . . . . . . . . 58
2.7.5 Bootstrapping . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.8 Interim summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.9 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3 Modelling thermoacoustic interactions in a Rijke tube 69

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 Governing equations . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.3 Asymptotic analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3.1 Continuity equation: acoustic zone O(M ) . . . . . . . . . . . 79
3.3.2 Momentum equation: acoustic zone O(M ) . . . . . . . . . . 79
3.3.3 Energy equation: acoustic zone O(M ) . . . . . . . . . . . . . 80
3.3.4 Continuity equation: hydrodynamic zone O(1) . . . . . . . . 81
3.3.5 Momentum equation: hydrodynamic zone O(1) . . . . . . . . 82
3.3.6 Energy equation: hydrodynamic zone O(1) . . . . . . . . . . 83
3.4 Solution technique . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.4.1 Equations governing the acoustic zone : one-dimensional form 85
3.4.2 Equations governing the hydrodynamic zone . . . . . . . . . 86

x
 3.5 Results and discussions . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.5.1 Stability regimes . . . . . . . . . . . . . . . . . . . . . . . . 88
3.5.2 Effect of global-acceleration on the stability of the system . . 91
3.5.3 Unsteady flow field in the hydrodynamic zone . . . . . . . . . 93
3.6 Interim summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4 Non-modal stability analysis in a Rijke tube system 99

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.2 Direct equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.3 Myers energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.3.1 Acoustic zone . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.3.2 Hydrodynamic zone . . . . . . . . . . . . . . . . . . . . . . 110
4.4 Definition of the cost functional . . . . . . . . . . . . . . . . . . . . 113
4.5 Adjoint system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.5.1 Adjoint equations . . . . . . . . . . . . . . . . . . . . . . . . 115
4.5.2 Adjoint boundary conditions . . . . . . . . . . . . . . . . . . 117
4.5.3 Mapping between direct and adjoint variables . . . . . . . . . 118
4.6 Solution procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.7 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.7.1 Evolution of disturbance energy . . . . . . . . . . . . . . . . 122
4.7.2 Convergence studies . . . . . . . . . . . . . . . . . . . . . . 124
4.7.3 Optimum initial condition . . . . . . . . . . . . . . . . . . . 124
4.7.4 Comparison with Chus energy . . . . . . . . . . . . . . . . . 125
4.7.5 Evolution of the optimum initial condition . . . . . . . . . . . 127
4.7.6 Parametric study . . . . . . . . . . . . . . . . . . . . . . . . 130
4.8 Interim summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

5 Experimental investigations of non-normality 133

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.2 Experimental configuration of the Rijke tube . . . . . . . . . . . . . . 134
5.3 Experimental procedures and theoretical background . . . . . . . . . 136
5.3.1 Bifurcation diagram . . . . . . . . . . . . . . . . . . . . . . 138

xi
 5.3.2 Regime of linear behavior . . . . . . . . . . . . . . . . . . . 139
5.3.3 Dynamic Mode Decomposition . . . . . . . . . . . . . . . . 139
5.3.4 Theoretical investigation of the eigenmodes . . . . . . . . . . 141
5.3.5 Extraction of dynamic modes . . . . . . . . . . . . . . . . . 143
5.3.6 Transient growth experiment . . . . . . . . . . . . . . . . . . 149
5.4 Results and discussions . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.4.1 Bifurcation diagram . . . . . . . . . . . . . . . . . . . . . . 150
5.4.2 Triggering amplitudes . . . . . . . . . . . . . . . . . . . . . 151
5.4.3 Regime of linearity . . . . . . . . . . . . . . . . . . . . . . . 154
5.4.4 Non-orthogonality of the eigenmodes . . . . . . . . . . . . . 155
5.4.5 Evidence of transient growth . . . . . . . . . . . . . . . . . . 157
5.4.6 Optimal initial condition . . . . . . . . . . . . . . . . . . . . 162
5.4.7 Triggering energy . . . . . . . . . . . . . . . . . . . . . . . . 165
5.5 Interim summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

6 Comparison of experimental and theoretical results 170

6.1 Comparison of theoretical and experimental bifurcation diagram . . . 170
6.1.1 Nature of the bifurcation . . . . . . . . . . . . . . . . . . . . 171
6.1.2 Amplitude levels . . . . . . . . . . . . . . . . . . . . . . . . 172

7 Conclusions and outlook 174

7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.1.1 Non-modal stability analysis in a solid rocket motor . . . . . 174
7.1.2 Asymptotic formulation of thermoacoustic interaction . . . . 175
7.1.3 Transient growth and nonlinear instabilities . . . . . . . . . . 176
7.1.4 Experimental investigation of non-normality . . . . . . . . . 176
7.2 Scope for future work . . . . . . . . . . . . . . . . . . . . . . . . . . 177
7.2.1 Theoretical investigations . . . . . . . . . . . . . . . . . . . 177
7.2.2 Experimental investigations . . . . . . . . . . . . . . . . . . 179

A Coupling terms in Eqns. (2.9) and (2.10) 181

B Linearised equations governing thermoacoustic interactions in SRM 183

xii
C Governing equations and boundary conditions involved in the hydrody-
namic zone: 187

D Steps involved in obtaining adjoint equations: 193

D.1 Acoustic zone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
D.1.1 Algebraic term . . . . . . . . . . . . . . . . . . . . . . . . . 193
D.2 Hydrodynamic zone . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
D.2.1 Laplacian term . . . . . . . . . . . . . . . . . . . . . . . . . 195

E Detailed drawing of the Rijke tube experimental setup 197

F Calibration of microphones 203

G Measurement of acoustic damping 207

H Experimental determination of the linear regime 209

I Dynamic mode decomposition 213

I.1 Determination of the eigenmodes . . . . . . . . . . . . . . . . . . . . 214

J Application of DMD in the presence of noise 217

K Determination of acoustic velocity using two microphone technique 221

 LIST OF TABLES

2.1 SRM parameter values and operating conditions. . . . . . . . . . . . 50

3.1 Physical parameters in acoustic zone, hydrodynamic zone and numeri-

cal parameters used for CFD simulations . . . . . . . . . . . . . . . . 88

4.1 Physical parameters in acoustic zone, hydrodynamic zone and conver-

gence parameters in for solving the direct and adjoint equations . . . . 121
4.2 Relative comparison of the optimum initial computed using Chus en-
ergy with the results obtained by using Myers energy as the reference. 126

5.1 The locations of the microphones, heater, thermocouple and loudspeak-

ers, measured from the upstream edge of the Rijke tube. . . . . . . . . 138

C.1 Comparison of various steady state flow properties between the present
simulation and experiments. Numerical values in parentheses indicate
values obtained from experiments. Experimental results for the recircu-
lation zone length and the separation angle are obtained from Coutanceau
and Bouard (1977), the Nusselt number, defined as N ud =
whereas 
2T0u R2  Tp 
2(T Tu )
r d is obtained from Collis and Williams
w 0 r=1
0
(1959). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

E.1 Materials used to make the major components of the Rijke tube setup. 197

F.1 Calibration data, Acalib = Am /Aref for all the microphones used in the
present investigation. All the microphones used belongs to PCB SN
series. The model number along with the location in the Rijke tube are
also shown. The sensitivity of the reference microphone, 103B02 SN
5236 is 223.65 mV /kP a with 1% uncertainty (Piezotronics, 2008). 206

xv
 LIST OF FIGURES

1.1 A typical combustion chamber, showing a flame as the heat source. The
figure is adapted with permission from Chong et al. (2011). . . . . . . 1
1.2 Schematic representation of the coupling between acoustic field and
unsteady heat release rate in the combustion chamber. . . . . . . . . . 2
1.3 A typical experimental data obtained during thermoacoustic instability
of an electrically heated horizontal Rijke tube. (a) Evolution of acoustic
pressure. (b) Zoomed in view of the oscillations during limit cycle.
This corresponds to the region represented by the rectangle in (a). (c)
Spectral content in the oscillations of the limit cycle. Two discrete tones
of frequencies 171 and 342 Hz are observed. Power supplied to the
heater is 800 W , mass flow rate through the system is 2.19 g/s, the
heater is located at a distance of 1/8th from the upstream end of the
tube and the acoustic pressure is measured at a location of 0.525 m
from the upstream end. . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 a & c) Spectrum for a linearly stable and unstable system. b & d) Evo-
lution of acoustic pressure for the same system. . . . . . . . . . . . . 5
1.5 a) Evolution of an initial condition at various time instants t1 < t2 <
t3 for non-normal and normal systems. b) Evolution of energy in the
disturbance (||(t)||2 ) for the above initial conditions. . . . . . . . . . 6
1.6 a) Evolution of the energy in the disturbance for various initial condi-
tions. The initial conditions are marked. b) Initial condition along only
one eigenmode. c) Initial condition along more than one eigenmode. d)
Distribution of optimum initial condition along various eigenmodes. . 7
1.7 Region of bistability obtained for the bifurcation between non-dimensional
heater power (K) and time lag ( ) between the response of the heater
to acoustic velocity fluctuations. This figure is adapted with permission
from Subramanian et al. (2010) . . . . . . . . . . . . . . . . . . . . . 9

xvii
1.8 Outstanding issues and the current objective of the present thesis. Rep-
resentative existing literature on the stability analysis of thermoacoustic
systems are shown in a qualitative manner. The horizontal axis rep-
resents the mathematical rigor with which the coupling between the
acoustic field and heat source is performed. The vertical axis represents
the sophistication in the stability analysis. The axis out of the paper
represents the increasing experimental investigations. Planes A and B
represent the numerical and experimental investigations respectively. In
plane A, the various levels indicates the following, I - classical stability
analysis, II - basic non-modal stability analysis and III - advanced sta-
bility analysis. Similarly, in plane B, I & II represent the experiments
associated with classical stability and non-modal stability analysis re-
spectively. Area shaded in grey represents the investigations performed
in the present thesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.9 Schematic representation of the structure of the thesis. . . . . . . . . 24

2.1 Schematic diagram of the combustion chamber geometry of the SRM

considered. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2 Geometry of the pressure coupled propellant response model. . . . . . 40
2.3 Schematic representation of performing SVD to determine the optimum
initial condition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.4 The evolution of acoustic pressure at x = 0.25 for a linearly stable sys-
tem. U1 (0) = 3, P1 (0) = 3, Pm6=1 (0) = Um6=1 (0) = 0, M Tp (, 0) =
0, C1 = 0.05, C2 = 0.001. . . . . . . . . . . . . . . . . . . . . . . . 51
2.5 The phase portrait of the acoustic pressure and the unsteady burn rate
at x = 0.25. U1 (0) = 3, P1 (0) = 3, Pm6=1 (0) = Um6=1 (0) = 0,
M Tp (, 0) = 0, C1 = 0.05, C2 = 0.001. . . . . . . . . . . . . . . . 51
2.6 The evolution of acoustic pressure at x = 0.25 for a linearly unsta-
ble system. U1 (0) = 0.3, P1 (0) = 0.3, Pm6=1 (0) = Um6=1 (0) =
0, M Tp (, 0) = 0, C1 = 3 104 , C2 = 1 104 . . . . . . . . . . 53
2.7 The phase portrait of the acoustic pressure and unsteady burn rate at
x = 0.25. U1 (0) = 0.3, P1 (0) = 0.3, Pm6=1 (0) = Um6=1 (0) = 0,
M Tp (, 0) = 0, C1 = 3 104 , C2 = 1 104 . . . . . . . . . . . . 53
2.8 a) Pseudospectra of the non-normal linear operator L. b) Zoomed in
view near the origin and the calculation of the lower bound for the max-
imum transient growth. The contour value represents log10 . C1 =
0.03, C2 = 0.02, kLk = 1.88 104 , max = 1 102.5 , max /||L|| 
1. <(z) and =(z) indicates the real and imaginary part of z respectively. 55

2.9 a) Comparison of the evolution of eLt with differential equation for
unsteady burn rate and response function (YR ) calculations, C1 = 0.03,
C2 = 0.02, (0) = Vopt . b) The magnitude of the response functions
(|YR |)of the propellant for various values of H. . . . . . . . . . . . . 56

xviii
2.10 Relative amplitude of the optimum initial condition direction. Inset
shows the relative amplitude of the acoustic modes, C1 = 0.03, C2 =
0.02, (0) = Vopt , N = 5, Mg = 150. . . . . . . . . . . . . . . . . 57
2.11 Evolution of acoustic pressure at x = 0.25 from the linear simulation.
U2 (0) = 3, P2 (0) = 3, Pm6=2 (0) = Um6=2 (0) = 0, M Tp ( = 1, 0) =
0.03, M Tp ( 6= 1, 0) = 0, C1 = 0.02, C2 = 0.02. . . . . . . . . . . 59
2.12 Evolution of acoustic pressure at x = 0.25 from the nonlinear simulation.
U2 (0) = 3, P2 (0) = 3, Pm6=2 (0) = Um6=2 (0) = 0, M Tp ( = 1, 0) =
0.03, M Tp ( 6= 1, 0) = 0, C1 = 0.02, C2 = 0.02. . . . . . . . . . . 59
2.13 The phase portrait of the acoustic pressure and unsteady burn rate at x
= 0.25 from the linear simulation. U2 (0) = 3, P2 (0) = 3, Pm6=2 (0) =
Um6=2 (0) = 0, M Tp ( = 1, 0) = 0.03, M Tp ( 6= 1, 0) = 0, C1 =
0.02, C2 = 0.02. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.14 The phase portrait of the acoustic pressure and unsteady burn rate at
x = 0.25 (near the limit cycle) from the nonlinear simulation U2 (0) =
3, P2 (0) = 3, Pm6=2 (0) = 0, Um6=2 (0) = 0, M Tp ( = 1, 0) = 0.03,
M Tp ( 6= 1, 0) = 0, C1 = 0.02, C2 = 0.02. . . . . . . . . . . . . . 61
 PN 
2 2
2.15 The evolution of acoustic energy Eac (t) = m=1 (Um + Pm ) from
linear and nonlinear simulations. U2 (0) = 3, P2 (0) = 3, Pm6=2 (0) =
Um6=2 (0) = 0, M Tp ( = 1, 0) = 0.03, M Tp ( 6= 1, 0) = 0, C1 =
0.02, C2 = 0.02. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
 PN 
2 2
2.16 a) The evolution of acoustic energy Eac (t) = m=1 (Um + Pm ) with
optimum initial condition (0) = Vopt , C1 = 0.03, C2 = 0.02. b)
Convergence
qP study for the previous figure with N. Note that, N =
I
i=1 ((N (ti ) N 1 (ti )) /N (ti )) 100 is the measure used for
2

studying the convergence of the simulations. Moreover, N represents

any one of the variables P (x = 1/4), R(x = 1/4) & E. The summa-
tion index i represents the value of the variables at the ith time step
in the numerical simulation. The threshold of N is chosen as 1% for
convergence of the solution, which corresponds to N=5 in the present
case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.17 The evolution of Galerkin pressure modes a) first mode, b) second
mode, c) third mode, d) fourth mode, e) fifth mode, f ) acoustic pres-
sure at x = 0.25, U2 = 3.0, P2 = 3.0, Pm6=2 (0) = Um6=2 (0) = 0,
M Tp (, 0) = 0, C1 = 0.02, C2 = 0.02. . . . . . . . . . . . . . . . . 64
2.18 The Fourier transform of acoustic pressure at x = 0.25 during differ-
ent time intervals, U2 = 3.0, P2 = 3.0, Pm6=2 (0) = Um6=2 (0) =
0, M Tp (, 0) = 0, C1 = 0.02, C2 = 0.02. . . . . . . . . . . . . . . 64

3.1 Configuration of the Rijke tube showing the acoustic and hydrodynamic
zones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

xix
3.2 Linearly stable system: (a) evolution of acoustic velocity (uf ) at the
heater location (xf ), (b) phase portrait between uf and unsteady heat
release rate q, xf = 0.25, K = 0.10, U1 (t = 0) = 0.05, Um6=1 (t =
0) = 0, Pm (t = 0) = 0. . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.3 Linearly unstable system: (a) evolution of acoustic velocity (uf ) at the
heater location (xf ), (b) phase portrait between uf and unsteady heat
release rate q (only limit cycle is shown, transients are not shown for
clarity), (c) evolution of uf during a period of the limit cycle, (d) dis-
tribution of the acoustic velocity ua in the Rijke tube during a period of
the limit cycle. xf = 0.25, K = 0.1785, U1 (t = 0) = 0.5, Um6=1 (t =
0) = 0, Pm (t = 0) = 0. . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.4 Bifurcation diagram with the non-dimensional heater power K as the
control parameter. The other parameters are, xf = 0.25. The two
regimes are R1 - linearly stable regime and R2 - linearly unstable. . . 90
3.5 (a) Comparison of the evolution of acoustic velocity (uf ) at the heater
location (xf ) with and without the global-acceleration term in Eqn.
(3.15), (b) Evolution of uf for a longer period of time with the global-
acceleration term, zoomed out view of (a). xf = 0.25, K = 0.1785,
U1 (t = 0) = 0.5, Um6=1 (t = 0) = 0 and Pm (t = 0) = 0. . . . . . . . . 91
3.6 Bifurcation diagram with non-dimensional heater power K as the con-
trol parameter without the global-acceleration term in Eqn. (3.15). The
parameters chosen are the same as for the simulation shown in Fig. 3.4. 92
3.7 Streamlines (a-e) in the hydrodynamic zone (corresponding to up ) at
various instants during one period of the limit cycle, (f ) evolution of ua
at xf during the limit cycle: K = 0.1785, xf = 0.25, lw = 10 m. . . 94
3.8 Flow streaming in the hydrodynamic zone, (a) streamlines averaged
over one period of the limit cycle, (b) streamlines of the steady base
flow, (c) streamlines of the velocity difference between (a) and (b),
(d) evolution of the non-dimensional unsteady heat release rate (q),
showing a significant mean shift: K = 0.1785, xf = 0.25, lw =
10 m, U1 (t = 0) = 0.5, Um6=1 (t = 0) = 0 and Pm (t = 0) = 0. . . . . 95

4.1 Flow domain and boundary conditions in the hydrodynamic zone. . . 101
4.2 Schematic of adjoint looping to determine the optimum initial condi-
tion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.3 Evolution of the disturbance energy (Ed ) obtained from nonlinear sim-
ulations. a) Globally stable system, C1 = 0.27, C2 = 0.03, xf =
0.25, K = 0.1, 1 = 1.9 103 , 2 = 3 = 3.7 103 , = 5.28
103 , Lcu = 35, u0a (xa ) = 1.5cos (xa ) , p0a (xa ) = u0 = v 0 = T 0 = 0.
b) Linearly unstable system, C1 = 0.27, C2 = 0.03, xf = 0.25, K =
0.1785, 1 = 3.3103 , 2 = 3 = 6.6103 , = 9.4104 , Lcu =
35, u0a (xa ) = 0.5cos (xa ) , p0a (xa ) = u0 = v 0 = T 0 = 0. . . . . . . 121

xx
4.4 Convergence of the variables during each adjoint looping for a) state
, Pi , u, v & and b) cost functional =. Residue is
space variables, Ui
PNgrid  Nadj   Nadj1   Nadj 
defined as i=1 i  i  / i  to measure the con-
vergence towards the optimum initial condition. i represents any one
of the state space variables Ui , Pi , u, v & . The summation index
i represents the value of the variables at the ith spatial location. Nadj
represents the adjoint loop number. The threshold of residue is cho-
sen as 0.1% for convergence, which corresponds to Nadj = 23. The
area shaded in grey indicates converged solution. The parameters are
Red = 10, C1 = 0.27, C2 = 0.03, xf = 0.25, K = 0.2856, 1 =
1.06 102 , 2 = 3 = 2.11 102 , = 2.49 103 , Lcu = 35. . . 122
4.5 Streamlines of u & v (top half) and contour of temperature T (bottom
half) corresponding to the optimum initial condition for maximum tran-
sient growth, a) Lcu = 15, b) Lcu = 20, c) Lcu = 25, d) Lcu = 30, e)
Lcu = 35, f ) Lcu = 40. g) Variation of = with Lcu . Other parameters
are same as in Fig. 4.4. . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.6 Distribution of optimum initial condition in the state space variables,
(a) streamlines and magnitude of velocity corresponding to u & v, (b)
temperature field T in the hydrodynamic zone, (c) acoustic field in the
acoustic zone. The parameters are same as that in Fig. 4.4 e. . . . . . 125
4.7 Comparisons of the optimum initial condition, computed using (a) My-
ers and (b) Chus energy in the definition of the cost functional =. In
the case of Chus energy, 2 = 3 = 0 in Eqn. 4.24. =max using My-
ers and Chus energy are 1843 and 1861 respectively. The legends for
the plots are as follows. Top: streamlines corresponding to u & v &
contours of temperature T , bottom: distribution of acoustic velocity
and pressure. Other parameters are same as that in Fig. 4.4 (e). . . . . 126
4.8 Evolution of the optimum initial condition. a) Evolution of the amplifi-
cation of disturbance energy (Edt /Ed0 ), b) zoomed in view. The symbol
indicate the selected time instants, for which the flow fields are shown
in subsequent subfigures. Optimum initial condition in the acoustic
zone, c) acoustic velocity & d) acoustic pressure. Evolution of the
acoustic field, e) acoustic velocity & f ) acoustic pressure. The parame-
ters are Red = 35, C1 = 0.31, C2 = 0.1, xf = 0.25, K = 0.06, 1 =
2.12 104 , 2 = 3 = 4.24 104 , = 7.41 102 , Lcu = 35.
Other parameters are same as that in Fig. 4.4 (e). . . . . . . . . . . . 128
4.9 Evolution of the optimum initial condition in the hydrodynamic zone,
a) t = 0, b) t = 0.25, c) t = 0.5 & d) t = 1.0. In each subfigure, top
half: velocity field corresponding to u & v, bottom half: temperature
field T . The parameters are are same as that in Fig. 4.8. . . . . . . . . 129

xxi
4.10 Variation of the optimum initial condition with the time scale ratio . a
& c)  = 0.68, b & d)  = 1.10. =max for the above  are 1843 and 1253
respectively. Top: streamlines corresponding to u & v & contours of
temperature T , bottom: distribution of acoustic velocity and pressure.
Other parameters are same as that in Fig. 4.4 (e). . . . . . . . . . . . 130

5.1 Schematic diagram of the Rijke tube experiment. All dimensions are
given in mm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.2 Configuration of the Rijke tube showing the heater, the microphone
ports and the acoustic drivers. The locations of the microphone ports
are numbered, and their distances from the upstream edge of the tube
are listed in table 5.1. All dimensions are given in mm. . . . . . . . . 135
5.3 Top: electrical heater along with the ceramic holder. All dimensions are
in mm. Bottom: operating at 1 kW . . . . . . . . . . . . . . . . . 137
5.4 Comparison of the first two eigenmodes obtained analytically (see Eqn.
5.4) from the n model (left column) with the eigenmodes obtained
by performing DMD on the data obtained from the same model (right
column). The value of the non-dimensionalised heater power n and the
absolute value of the inner product hp1 , p2 i are indicated in the title of
each subfigures. The eigenmodes are displayed following the scheme:
real part of first eigenmode, real part of second eigen-
mode, imaginary part of first eigenmode, imaginary
of second eigenmode. The governing parameters are xf = 0.125 and
= 0.2 (prescribed by Lighthill 1954). . . . . . . . . . . . . . . . . . 144
5.5 Eigenvalues obtained by applying DMD to data produced from (Eqn.
5.1). The decomposition is applied separately to data from the upstream
and downstream side of the heater; the corresponding eigenvalues are
displayed by symbols, and , respectively. Using Eqn. (5.4), the
initial condition consists of a first theoretical eigenmode in (a) and a
second theoretical modes in (b). A magnified view of the dominant
eigenvalues is given in the right column. Finally, a comparison of the
eigenvalues from theoretical investigations in -symbols and from the
DMD in -symbols is presented in (e). The governing parameters have
been chosen as n = 0.5, = 0.2 and xf = 0.125. . . . . . . . . . . . 145

xxii
5.6 Validation of the extraction of eigenmodes via the Dynamic Mode De-
composition. The subfigures on the left (right) hand side are asso-
ciated with initial conditions along the first (second) mode. The ini-
tial condition is (a) p(x, 0) = 0.6p1 (x) 0.6p1 (x), and (b) p(x, 0) =
0.3p2 (x)0.3p2 (x). The relative contribution (root-mean-square value)
of the dynamic modes in the data sequence from the (c,d) upstream and
(e,f ) downstream side of the heater. The first two dynamic modes in
each subfigure correspond to the actual eigenmodes. (g,h) The per-
centage difference (xj ) = (||p(xj , t) pr (xj , t)||) /||p(xj , t)|| 100
between the input data and the reconstructed data sequence using the
eigenmodes of the system at all microphone locations. The parameters
are the same as in Fig. 5.5. . . . . . . . . . . . . . . . . . . . . . . . 147
5.7 Comparison of the non-orthogonality measure |hp1 , p2 i| obtained from
theoretical calculations (solid line) and from processing data by the
DMD technique (symbols). . . . . . . . . . . . . . . . . . . . . . . . 148
5.8 Bifurcation diagrams displaying the rms values of the acoustic pressure
at x = 0.15 versus the power supplied to the heater. The mass flow rate
is (a) m
= 2.34 g/s, (b) m
= 2.19 g/s, (c) m
= 2.03 g/s. . . . . . . . 150
5.9 Determination of triggering and the corresponding limit cycle ampli-
tudes. (a & b) Evolution of the system that is just triggered and (c) just
decayed. K = 782W , m = 2.34 g/s. . . . . . . . . . . . . . . . . . . 152
5.10 Bifurcation diagram showing hysteresis behavior along with the trigger-
ing (GH) and limit cycle (IJ) amplitudes. The experiments have been
repeated four times to ensure repeatability of the results. A mass flow
rate of m
= 2.34 g/s has been chosen. . . . . . . . . . . . . . . . . . 153
5.11 Steady state temperature recorded downstream of the heater during ex-
perimental run 1. The steady-state temperature during increasing and
decreasing values of K are shown. Vertical dashed lines indicate the
power level associated with the region IJ shown in Fig. 5.10. . . . . . 154
5.12 Bifurcation diagram showing averaged triggering amplitudes (GH) and
averaged limit cycle amplitudes (IJ). The limit of the linearity (LN)
defines the linear regime (in gray); all experiments investigating the
non-normal nature of the thermoacoustic system are performed in this
regime. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.13 (a-c) First two dominant eigenmodes obtained by applying DMD to
the experimental acoustic pressure data. The respective values of the
heater power K and the absolute value of the inner product hp1 , p2 i are
indicated in the title of each subfigure. The use is symbols is identical to
the one in Fig. 5.4. (d) Dependence of |hp1 , p2 i| on the power supplied
to the heater K, together with the associated error bars. . . . . . . . . 156

xxiii
5.14 Experimental acoustic pressure data at (a) location x = 0.1 and (b)
location x = 0.8 which is used in the DMD-analysis. The initial con-
dition is closely aligned in the direction of the second eigenmode of
the system. The relative contributions (root-mean-square value) of the
dynamic modes to the full data sequence for the (c) upstream and (d)
downstream side of the heater. The first two dynamic modes, along with
their complex conjugates, represent the second eigenmode of the sys-
tem. (e) The spatial structure of the first dynamic mode with
signifying the real part and the imaginary part. (f ) The rel-
ative error (in percent) at all locations of the microphones, (xj ) =
(||p(xj , t) pr (xj , t)||) /||p(xj , t)|| 100, between the input data p and
a reconstructed data sequence pr using the first two dynamic modes. . 158
5.15 Three kinds of initial perturbation imposed on the system through the
acoustic driver. Sinusoidal input of the form (a) 0.54sin(2f1 t), (b)
0.54sin(2f2 t), and (c) 0.3sin(2f1 t) + 0.3sin(2f2 t), where f1 and
f2 are the frequencies of the first and second eigenmodes of the system.
The vertical black line indicates the instant in time when the acoustic
driver is switched off. . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.16 Evolution of non-dimensional acoustic energy, which is defined as E(t) =
RL
0
(u2 + p2 /(pu )) dx/ (u u2u ) . For (a), the heater is switched off; for
(c), the heater is switched on (K = 747 W ). Evolution of the normal-
ized amplification of acoustic energy E(t)/E(0) with heater switched
off (b) and switched on (d). The colors indicate the three kinds of ini-
tial condition shown in Fig. 5.15; the time interval during which the
acoustic driver is active is indicated in gray. . . . . . . . . . . . . . . 160
5.17 Evolution of the acoustic pressure at x = 0.3 in the linear regime. The
two horizontal dash-dotted lines in all subfigures represent the ampli-
tudes delimiting the linear regime. The amplitude of the initial perturba-
tion is scaled and the subsequent evolution of the system is recorded for
the following: (a) 0.1 sin(2f1 t)+0.1 sin(2f2 t), (b) 0.075 sin(2f1 t)+
0.075 sin(2f2 t), (c) 0.05 sin(2f1 t)+0.05 sin(2f2 t). In (d) the acous-
tic pressure signals from subfigures (a)-(c) are rescaled and superim-
posed. Circle and cross symbols are associated with data from subfig-
ures (b) and (c), respectively. The duration of the systems evolution
is taken as 0.2s, which corresponds to a non-dimensional time span
t/(L/a) of 69. The supplied heater power is K = 747W. . . . . . . . 161

xxiv
5.18 Variation of the maximum energy amplification Emax with (a) ampli-
tude ratio A2 /A1 for fixed phase angle = 0o and (b) with phase
angle for fixed amplitude ratio A2 /A1 = 1. The symbols () indi-
cate the experimental data point; the continuous line represents a Gaus-
sian fit. (c) Maximum transient growth over initial conditions of the
form A1 sin(2f1 t) + A2 sin(2f2 t + ). The optimal initial condition
for maximum transient growth is identified for the parameter combina-
tion A2 /A1 = 0.93 and = 90o . The corresponding maximum tran-
sient growth is 2.3. The remaining parameters are f1 = 173Hz, f2 =
346Hz, m = 2.34g/s and K = 747W. . . . . . . . . . . . . . . . . . 163
5.19 Evolution of (a) acoustic pressure P (x = 0.525) and (b) acoustic en-
ergy E(t) for the triggering of an instability. The inset (c) shows the
initial evolution of E(t). Zones A, B, C and D represent the time in-
tervals where the system is excited by the acoustic driver (A), near the
triggering amplitude (B), in the nonlinearly unstable state (C) and in
the final limit cycle (D). The gray area represents the region when the
acoustic driver is switched on. The governing parameters are A2 /A1 =
= 0o , K = 782W and m = 2.34g/s. . . . . . . . . . . . . . . . . . 165
5.20 (a) Acoustic energy Etrig required for triggering an instability as a func-
tion of amplitude ratio A2 /A1 . (b) Magnified version of (a) for a value
of A2 /A1 between 0.1 and 3.0. This parameter interval is shown in
gray in (a). The remaining parameters are = 0o , K = 747W and
m = 2.34g/s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

6.1 Bifurcation diagram with the non-dimensional heater power K(Tw

T0u )/T0u as the control parameter, a) with out corrections and b) with
corrections for the heat transfer losses from Matveev (2003a). Solid
lines and triangles indicate the results from numerical simulations and
experiments respectively. The parameters for the experiments are xf =
0.125, Reef f = 9.13, N ud = 1.84 (1.83 from Collis and Williams
1959), kth = 2.4 102 w/m K (Kreith, 2000), T0u = 295 K, M =
5 104 , lc = 1.4 104 m, Sc = 0.01 m2 . . . . . . . . . . . . . . . 171

7.1 An outlook and possible future extension of the present thesis. The plot
is similar to the one shown in Fig. 1.8 described in the introduction
chapter. The axis start from the current investigations and recommen-
dations for future work are arranged according to the expected difficulty
level. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

C.1 Schematic representation of the heat source, (a) rack of the heater wire
filament (b) boundary conditions for the two dimensional flow over a
heated circular cylinder. . . . . . . . . . . . . . . . . . . . . . . . . . 188

xxv
C.2 Grid generated (121101) for flow in the hydrodynamic zone (a) Phys-
ical domain with grid clustering near the cylinder surface with k = .
Flow domain is shown only up to r = 25 so that the presence of the
cylinder can easily be visible in the figure. Numerical simulations are
performed for the domain size r = 50. Convergence tests are performed
and it is found that there is less than 5% change in the results with the
domain size for r = 50, (b) Computation domain with uniform grids. . 188
C.3 Staggered grid arrangement, indicating density, pressure, temperature
and velocity nodes. (a) Physical domain, (b) Computational domain. . 189
C.4 (a-b) r direction velocity component discretisation stencil. (c-d)
direction velocity component discretisation stencil. (a & c) Physical
domain, (b & d) Computation domain. . . . . . . . . . . . . . . . . . 189
C.5 Response of the unsteady heat transfer from the heated cylinder for the
forcing of the freestream velocity, up (r , /2 < < ) = 1 +
0.1 sin(ta /2.43). The parameters are, Red = 10, Tu = 295 K, Tw =
700 K. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

E.1 Detailed drawing for the Rijke tube along with provisions for instru-
mentations; a) isometric, b) Orthographic view. Microphones and ther-
mocouples are mounted in the small holes, while acoustic drivers are
mounted in the large holes. All dimensions are in mm. . . . . . . . . 198
E.2 Detailed drawing for the decoupler (orthographic projections). All di-
mensions are in mm. . . . . . . . . . . . . . . . . . . . . . . . . . . 199
E.3 Detailed drawing for the heater stand; a) isometric, b) Orthographic
view. All dimensions are in mm. . . . . . . . . . . . . . . . . . . . . 200
E.4 Detailed drawing for the heater mesh brazed with the copper rod. All
dimensions are in mm. . . . . . . . . . . . . . . . . . . . . . . . . . 201

F.1 (a) Schematic representation of the microphone calibration setup. (b)

Microphone arrangements for calibration. All dimensions shown are in
mm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
F.2 Image showing two models of microphones, 103B02, 377B10 used for
the measurement of acoustic pressure. . . . . . . . . . . . . . . . . . 205
F.3 Calibration data for a microphone, 103B02 SN 5235 in the frequency
range 50500 Hz. Calibration is performed twice to ensure repeatabil-
ity and are shown as cross and hollow circular symbols. (a) Amplitude
ratio Acalib = Am /Aref , Am & Aref represent the amplitudes measure
from the above microphone and the reference. (b) Phase difference be-
tween the microphone and the reference. . . . . . . . . . . . . . . . . 205

xxvi
G.1 Estimation of acoustic damping in the system. (a) Acoustic pressure
measured at x = 0.525, which is used for damping measurement. Dark
grey region indicates the period during which the acoustic driver is
switched on at a frequency of 156 Hz. Data used for the determination
of acoustic damping is shown in light grey region. (b) Determination of
the decay rate of the system by plotting the evolution of the logarithm
of the instantaneous amplitude ratio (log (|H(P (t))|/|H(P (tst ))|)). The
exponential decay rate () is obtained by the slope of the above curve. 208

H.1 Characteristics of the acoustic driver unit, obtained for three different
frequencies. The rms-value of the voltage supplied to the acoustic driver
unit is indicated on horizontal axis. The symbols represent the experi-
mental data, while the continuous line presents a linear fit of the data. 210
H.2 Identification of the linear regime. The acoustic forcing is conducted at
300 Hz with K = 764 W and m = 2.34 g/s. A power law is used to
fit the response of the system. The resulting equation for the fit reads
28.36Vp0.67 14.24. The shaded area indicates the linear regime. . . . 210

I.1 Schematic illustration of the application of DMD technique. . . . . . 213

J.1 Estimation of noise levels encountered in the experiments. (a) and (c)
noise levels measured at locations x = 0.10 and x = 0.80. (b) and (d)
the input waveform measured at the same locations, used to perform
DMD. The horizontal line indicates the rms-level of the corresponding
signal. The relative noise level is estimated as the ratio of the rms-level
of the noise to the rms-level of the signal at the microphone locations.
Results are shown for determining the (e) second and (f ) first eigen-
mode of the system. The remaining parameters are m = 2.34g/s and
K = 747W. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
J.2 Illustration of the robustness of the DMD algorithm in the presence of
noise. (a) Histogram plot based on 20 bins for the value of the inner
product | < pi , pj > | from 200 realizations with random noise added
to the experimental data. The added noise level is equivalent to the one
shown in Fig. J.1 (e,f ). (b) Probability density function (PDF) for the
above histogram. The cross symbols indicate the histogram data, and
the continuous line indicates a Gaussian distribution fit. The equation
of the fit is also indicated. The mean value xm of | < pi , pj > | is 0.127,
and the standard deviation is 3.124 103 . The percentage error
associated with | < pi , pj > | is evaluated as /xm 100 = 2.46. This
value is used as the amplitude of the error bar (Holman, 1994) indicated
in Fig. 5.13d. The governing parameters are the same as in Fig. J.1. . 219

xxvii
K.1 Evolution of (a) acoustic pressure P (x = 0.525) for the case, when
the system is triggered by external forcing. Evolution of (b & d) non-
dimensional acoustic pressure, P (x = 0.525)/(M Pu ) and (d & e)
acoustic velocity during triggering (initial phase of zone B) and limit
cycle (zone D) respectively. The governing parameters are A2 /A1 =
= 0o , K = 782 W , m = 2.34 g/s, uu = 0.24 m/s, Tu = 298 K,
M = 6.9 104 and Pu = 1.01235 105 P a. . . . . . . . . . . . . 222

xxviii
 ABBREVIATIONS

BC Boundary Condition
BTCS Backward Time Central Space
CFD Computational Fluid Dynamics
DMD Dynamic Mode Decomposition
FFT Fast Fourier Transform
IC Initial Condition
ODE Ordinary Differential Equation
PDE Partial Differential Equation
POD Proper Orthogonal Decomposition
RK4 Runge-Kutta Fourth Order
SIMPLE Semi-Implicit Method for Pressure Linked Equations
SRM Solid Rocket Motor
SVD Singular Value Decomposition

xxix
 NOTATION

Upper case variables

Cn Constants defining the initial condition
C1 First damping coefficient in the acoustic zone
C2 Second damping coefficient in the acoustic zone
F ratio of timescales of chamber acoustics and transient heat conduction
in the propellant
H ratio of the heat release at the propellant surface to the thermal
capacity of the propellant
K Non-dimensional heater power
Kc Keulegan-Carpenter number
L Linear operator matrix
M Mach number of steady state flow
N Number of Galerkin modes used for the simulation
Pe Peclet number
Pelec Electrical power supplied to the heater, W
Qs Overall heat release per unit mass at the propellant surface, J/kg
R Propellant regression rate, m/s
Re Reynolds number
Sc Port/cross section area, m2
Sl Port circumference, m
Sp Specific heat capacity of the propellant, J/kgK
T Temperature, K
Vn nth eigenmode of the system

Lower case variables

a Average speed of sound, m/s
er Unit vector along the radial direction in the hydrodynamic zone
h1 , h2 , h3 Empirical constants in the correlation between electrical power to heat release rate
l Length of the propellant grain, m
la Length of the Rijke tube, m
lc Radius of the electrical heater wire, m
lw Effective length of the heater wire participating in heat transfer, m
m Mass influx rate from the propellant per unit volume at any axial
location, kg/m3 s
m Mass flow rate through the Rijke tube system, Kg/s
mp Pyrolysis coefficient of the propellant
n Burn rate index of the propellant
p Acoustic pressure
q Non-dimensional unsteady heat release rate from the heater

xxxi
r, Co-ordinates in the hydrodynamic zone
tac Time scale in the acoustic zone, s
tcc Time scale in the hydrodynamic zone, s
u Acoustic velocity
uf Non-dimensional acoustic velocity at the heater location
x Non-dimensional co-ordinate in the axial direction of the motor from
the head end
xa Non-dimensional co-ordinates in the acoustic zone
xf Non-dimensional axial location of the heat source
y Non-dimensional distance from the propellant surface

Greek variables
Thermal diffusivity of the propellant, m2 /s
n Non-dimensional growth rate of nth eigenvalue
n Non-dimensional angular frequency of nth eigenvalue
Ratio of specific heat capacities
Length scale ratios (lc /la )
 Time scale ratios (tac = tcc )
Density of the fluid flow
State space vector
n Frequency of the nth Galerkin mode
< Specific gas constant, J/KgK
<n Real part of the eigenvalue
=n Imaginary part of the eigenvalue
n Projection along the direction of nth eigenmode

Subscript
0 Steady state variables
a Variables corresponding to acoustic zone
c Variables corresponding to hydrodynamic zone
p Transformed variables in the hydrodynamic zone
pp Peak to peak value of the variable in the asymptotic time
s Steady state variables in non-dimensional form

Superscript
d Downstream variables
u Upstream variables

Symbols
Reference variables for non-dimensionalisation
Dimensional variables
_ Total dimensional variables
Adjoint operation
Complex conjugate
|| || L2 norm of the state space vector
hi Inner product
|| Modulus of a complex number
Of the order

xxxii
 Approximately equal

xxxiii
 CHAPTER 1

Introduction

Thermoacoustic instability is a challenging problem in solid and liquid rockets, ramjets,

aircraft and industrial gas turbines etc. (Mcmanus et al., 1993). A typical combustion
chamber, showing a flame as the heat source is indicated in Fig. 1.1. Thermoacoustic
instability occurs when the acoustic oscillations in a duct (combustion chamber) are
amplified by the positive feed back of the unsteady heat release rate from the heat source
(Fig. 1.2).

The mechanism of instability is as follows. In combustion chambers, disturbances

are always present due to turbulence of the incoming flow, change in the operating con-
ditions etc. The velocity fluctuations generated due to the above phenomenon perturb
the heat release rate. For example in a premixed flame, an upstream velocity perturba-
tions causes wrinkles on the flame surface, which eventually leads to heat release rate
fluctuations (Schuller et al., 2003; Subramanian and Sujith, 2011). In turn, the heat re-
lease rate fluctuations cause the surrounding fluid to expand and contract. This motion
near the heat source is like a monopole source of sound and creates acoustic oscillations
in the duct. The acoustic velocity thus generated again perturbs the heat source and the
feed back loop is completed. If the driving from the acoustic - heat source coupling
overcomes the acoustic damping (viscous and end losses) in the duct, the amplitude of

Figure 1.1: A typical combustion chamber, showing a flame as the heat source. The
figure is adapted with permission from Chong et al. (2011).
Figure 1.2: Schematic representation of the coupling between acoustic field and un-
steady heat release rate in the combustion chamber.

Figure 1.3: A typical experimental data obtained during thermoacoustic instability of

an electrically heated horizontal Rijke tube. (a) Evolution of acoustic pres-
sure. (b) Zoomed in view of the oscillations during limit cycle. This
corresponds to the region represented by the rectangle in (a). (c) Spectral
content in the oscillations of the limit cycle. Two discrete tones of frequen-
cies 171 and 342 Hz are observed. Power supplied to the heater is 800 W ,
mass flow rate through the system is 2.19 g/s, the heater is located at a dis-
tance of 1/8th from the upstream end of the tube and the acoustic pressure
is measured at a location of 0.525 m from the upstream end.

2
the oscillations increases and eventually saturates to a limit cycle due to the nonlinear-
ity present in the system. The limit cycle thus obtained has well defined frequencies
associated with it.

Figure 1.3 (a) represents the evolution of acoustic pressure measured experimen-
tally during thermoacoustic instability in an electrically heated horizontal Rijke tube.
Initially, the amplitude of the oscillations are small and can be described by linear the-
ory. The oscillations grow exponentially and as the amplitude increases, the nonlinearity
saturates the amplitude of the oscillations to reach a limit cycle. The region represented
by the rectangle is zoomed up and is shown in Fig. 1.3 (b). The wave form during limit
cycle shows well defined periodic oscillation. Further, the spectral content of the acous-
tic pressure oscillations during limit cycle indicates two well defined peaks as shown in
Fig. 1.3 (c). The above peaks are associated with the first and second eigenmodes of
the system.

The above self sustained acoustic oscillations cause excessive turbine blade vibra-
tion in gas turbines and damage them eventually. Further, in industrial gas turbines,
stringent emission norms require the combustor to operate in lean equivalence ratios
where thermoacoustic instabilities are more prone to occur (Annaswamy et al., 1997).
In rockets the problem is more severe, as catastrophic mission failures occur (Culick,
2006; Blomshield et al., 1997b). The problems arising due to thermoacoustic instabil-
ity are identified only in the later stages of the development programme. The complex
interaction between the chamber acoustic field, the unsteady fluid flow and the combus-
tion makes the problem challenging. Thus, identifying instabilities in the early phase
of the design stage is a formidable task. Thermoacoustic instability occurs, when the
acoustic pressure oscillations in the combustion chamber are amplified by the positive
feedback of the unsteady heat release rate from the heat source in the chamber (Mc-
manus et al., 1993). Hence, understanding the coupling between the chamber acoustic
field and the heat source is vital for the prediction and control of these instabilities. A
theoretical analysis uses mathematical tools and aims at the prediction of the onset of
instability.

3
1.1 Classical linear stability analysis

In the analysis of thermoacoustic instability, the system is initially modelled using math-
ematical equations. These are in general nonlinear partial differential equations (PDEs)
(Poinsot and Veynante, 2005). As the first step in the investigation of the stability of
the system, the governing equations are linearised around the steady state. Further, the
linearised equations are valid, when the amplitude of the perturbations are small. Even
for a system which is unstable, like the one shown in Fig. 1.3 (a), the amplitude during
the initial growth phase of the disturbance is small. Hence the initial evolution can be
described by linear theory. The obtained linearised PDEs can be converted to ordinary
differential equations (ODEs) in time by suitable spatial discretisation (Culick, 2006).
The ODEs thus obtained can be cast in the following standard form (Strogatz, 2000):

d
= L (1.1)
dt

where, represents the column vector, its elements are the state space variables defining
the system. For example, can contain the evolution of acoustic velocity (u) and pres-
sure (p) at various discretised locations of the system. L is a matrix, which determines
the linear evolution of the system.

In classical stability analysis, the eigenvalues of L determines the linear stability

of the system (Strogatz, 2000). To describe the analysis, a representative spectrum
and the corresponding linear evolution of the perturbation from the investigation of
thermoacoustic instability in a Rijke tube system (Balasubramanian and Sujith, 2008b)
are shown in Fig. 1.4. If all the eigenvalues are in the left half of the imaginary axis,
as shown in Fig. 1.4 (a), the system is stable. On the other hand, even if one of the
eigenvalue is to the right of the imaginary axis (Fig. 1.4 c), the system is unstable
(Strogatz, 2000). A typical evolution of the acoustic pressure oscillation for a linearly
stable and unstable system are shown in Fig. 1.4 (b & d) respectively.

The stability predicted by this theory is valid only at asymptotic time limit (t )
(Strogatz, 2000). No predictions are made about the evolution of the disturbance, in the
initial period (just after the initial perturbation). The non-normal nature of the system
plays an important role in this initial period.

4
Figure 1.4: a & c) Spectrum for a linearly stable and unstable system. b & d) Evolution
of acoustic pressure for the same system.

1.2 Non-normality and transient growth

A system is said to be non-normal if the linear operator L governing the evolution of the


system does not commute with its adjoint L LL 6= L L, indicates adjoint operator
(Golub and VanLoan, 1989). For such systems, the eigenmodes (eigenvectors) are non-
orthogonal. The solution of Eqn. (1.1) can be written in terms of eigenmodes as follows
(Strogatz, 2000):
X
N
(t) = Cn e(n +in )t Vn (1.2)
n=1

where, n = n + in is the nth eigenvalue, Vn is the corresponding eigenmode and

Cn s are the constants, which are determined by the initial condition. The evolution of
the system is represented by a linear combination of the projection along the eigen-
modes. In order to visualise transient growth, a system with two degrees of freedom is
considered and is shown in Fig. 1.5. Then only two eigenmodes are present and can
be represented easily in a figure. Let n = e(n +in )t Vn represent the projection of the
disturbance along the direction of eigenmodes. An initial condition (t = t1 ), which
is represented by an arrow is chosen such that, it has non-zero projections on both the
eigenmodes. For a linearly stable system, n s decay (as n < 0) according to their
decay rate monotonically to zero. Two cases are considered: Non-normal and normal

5
Figure 1.5: a) Evolution of an initial condition at various time instants t1 < t2 < t3 for
non-normal and normal systems. b) Evolution of energy in the disturbance
(||(t)||2 ) for the above initial conditions.

system. Non-normal system is first analysed and is represented by 1 & 2 drawn non-
orthogonally. As time evolves, the amplitude of the disturbance grows for a short time,
which can be seen from the left hand side of Fig. 1.5 (a). The disturbance amplitude
decays eventually to zero, as the individual n also decays to zero, thus confirming
the predictions of linear stability. On the other hand, in a normal system (1 & 2 are
orthogonal), the same initial condition decays monotonically.

The evolution of the amplitude of the disturbance (||(t)||2 ) are shown in Fig. 1.5
(b). The amplitude of the disturbance is defined as the L2 norm of the state space
vector . For non-normal system, ||(t)||2 increases transiently and eventually decays,
whereas it decreases monotonically for a normal system. One more important point
to be noted is the following. If an initial condition has components along only one

6
Figure 1.6: a) Evolution of the energy in the disturbance for various initial conditions.
The initial conditions are marked. b) Initial condition along only one eigen-
mode. c) Initial condition along more than one eigenmode. d) Distribution
of optimum initial condition along various eigenmodes.

eigenmode (along n ), then there is no transient growth as all n s decay monotonically.

Thus the amount of transient growth depends critically on the initial condition specified
to the system. This is illustrated in Fig. 1.6, where the analysis have been performed in
a non-normal system (Balasubramanian and Sujith, 2008b) with 20 degrees of freedom.

Figure 1.6 (a) shows the evolution of energy in the disturbance for various initial
conditions. The evolutions are marked according to the subfigure labels in Fig. 1.6
(b-d), where the initial conditions are shown. When an initial condition, which has
projection along only one eigenmode (Fig. 1.6 b) is specified, the disturbance energy
decays monotonically. Disturbance energy in the present case is same as ||(t)||2 . How-
ever, when the initial condition, which has projections on more than one eigenmodes
(Fig. 1.6 c) is chosen, transient growth in the disturbance energy is observed. Moreover,
the optimum initial condition which gives the maximum transient growth can also be
obtained and is shown in Fig. 1.6 (d). A detailed technique for obtaining the same is
given in Section 2.4.

To summarise, for a linearised system, stable under the classical linear stability,

7
all eigenmodes are decaying monotonically in time. However, in the case of the non-
normal system, the vectorial sum of the eigenmodes which gives the state of the system
at any time t can increase (for a suitable initial condition) for a short time and eventually
decays after a long time. The transient growth occurring might be of several orders
of magnitude and the system might reach a different dynamical behaviour, e.g. limit
cycles, if the nonlinearities become significant as the amplitude increases during this
growth (Balasubramanian and Sujith, 2008a).

1.3 Non-Normality in thermoacoustic system

Recently, thermoacoustic systems are shown to be non-normal, which leads to the non-
orthogonality of the eigenmodes (Balasubramanian and Sujith, 2008b). Consider the
following acoustic momentum and energy equation for low Mach number flow in non-
dimensional form:
u p
+ =0 (1.3a)
t x
p u
+ = Q(x) (1.3b)
t x
where, u and p represent the non-dimensional acoustic velocity and pressure in
the duct respectively, Q represents the non-dimensional unsteady heat release rate from
the heat source . In general, the above heat release rate can be written as Q(x) =
R(x)u(x, t) + S(x)p(x, t) (Balasubramanian and Sujith, 2008b). R(x) & S(x) repre-
sent the strength of the coupling between the acoustic variables and the heat source.
Equation (1.3) can be cast in the standard form (Eqn. 1.1) as follows:

u 0 x

u
= (1.4)
t p x

+R S p
| {z }
L

When R = 0, the linear operator L in the above equation is self-adjoint. As the

value of R increases due to the presence of heat source, L becomes non-normal. In
1)
The non-dimensionalisation is as follows. u = u u, p = p/(M P ), Q = Q(
/ l/(P u
), x =
x
/l & t = t/(l/a)

8
Figure 1.7: Region of bistability obtained for the bifurcation between non-dimensional
heater power (K) and time lag ( ) between the response of the heater to
acoustic velocity fluctuations. This figure is adapted with permission from
Subramanian et al. (2010)

most of the combustion system, fluctuations in heat release rate from the heat source
by acoustic velocity fluctuations is prominent (Candel, 2002) and hence R(x) is non-
zero in most of the combustion chamber. Hence thermoacoustic interactions are in
R
general non-normal. Transient growth occurs in the evolution of (u2 + p2 ) dv and
domain
is identified as non-dimensional acoustic energy of the system (Nagaraja et al., 2009).

A typical stability map for a thermoacoustic system is shown in Fig. 1.7 for the
variation of two parameters, namely non-dimensional heater power (K) and time lag
( ) between the response of the heater to acoustic velocity fluctuations (Subramanian
et al., 2010). White region indicates globally stable region i.e., system is stable for all
amplitude perturbations. Globally unstable region, where the system is unstable to any
amplitude perturbations is marked in light grey colour. The transition from globally
stable to unstable behaviour occurs via a bistable region (marked in dark grey colour).
In this region, the system is stable for small amplitude perturbations, while it is unstable
for large amplitude perturbations. From dynamical systems point of view, this region
is known as subcritical transition zone (Strogatz, 2000). In the bistable region, transient
growth arising due to the non-normal nature of thermoacoustic interaction plays an
important role in the asymptotic stability of the system (Balasubramanian and Sujith,
2008a,b).

9
1.4 Literature review

Thermoacoustic oscillations were first observed by Rijke (1859). He used a vertical

tube with a coiled electrical heating filament as the heat source. Self-sustained ther-
moacoustic oscillations were observed when the heater was positioned at some axial
location of the tube and beyond some threshold power level. For example, the acoustic
oscillations were observed only, when the heater was positioned at the lower half of the
duct. The tube was name after him: Rijke tube. Rijke gave an explanation based on
the pressure pulse generated due to volumetric expansion of the fluid near the heater
zone. However, this argument did not explain the fact that instabilities were observed
only for some selected range of heater locations. The condition for the occurrence of
thermoacoustic instability was given by Rayleigh (1878). Thermoacoustic oscillations
are amplified, when the acoustic pressure oscillations are in phase with the unsteady
heater release rate from the heat source. In aero engines, thermoacoustic oscillations
were first observed in solid rocket motors (SRM).

1.4.1 Thermoacoustic instabilities in solid rocket motors

Thermoacoustic instabilities in SRMs are known to exist since 1930 (Culick, 2006).
Since then, many investigations were conducted to understand the mechanisms behind
them and to arrive at measures to control the same. Furthermore, combustion instability
alone is not the only source of instability in an SRM. Although other instabilities, such
as chuffing or L instability (Sutton and Biblarz, 2001), acoustic instability due to vortex
shedding in segmented motors (Vuillot, 1995; Kourta, 1997; Anthoine et al., 2002) and
instability of shear waves at the propellant surface (Flandro and Majdalani, 1996) are
important, acoustic oscillations due to thermoacoustic instabilities play a dominant role.
In the SRM, the driving mechanism for combustion instability to occur is the response
of the unsteady burn rate of the propellant to chamber acoustics. This leads to unsteady
mass addition to the combustion chamber and ultimately causes an unsteady heat release
rate.

Initial attempts to analyse thermoacoustic instabilities theoretically were by Culick

10
(1963), and Friedly and Petersen (1966), where the linearised equations were inves-
tigated. They obtained an explicit expression for the complex frequency (eigenvalue)
of the system. The real part of the frequency gave the growth rate of the oscillations.
When the real part is positive, the system becomes linearly unstable (Section 1.1) and
the amplitude of the oscillations grow exponentially, eventually reaching a limit cycle.
Culick (1976a) was the first to derive an analytical condition for the existence of sta-
ble and unique limit cycle behaviour using a two mode Galerkin approximation with
second order acoustics as the only nonlinear process. Nonlinearity in the combustion
response was also included in the analysis and the dynamical behaviour was analysed
(Levine and Baum, 1983; Flandro et al., 2007).

Another interesting dynamical behaviour observed experimentally was that SRMs

stable for small amplitude disturbances were seen to become unstable for larger ones
(Blomshield et al., 1997a). This was called pulsed instability or triggering. Many
mechanisms were proposed for explaining triggering in the past. Second order gas dy-
namics alone were proved to be insufficient to cause triggering because of the absence
of self coupling terms (Culick, 1994). Higher order gas dynamics (third order acous-
tics) also proved the same (Yang et al., 1990). Hence, a nonlinear combustion response
was thought of as an alternate candidate for triggering. Culick et al. (1995) used an
ad hoc nonlinear velocity coupling model for burn rate response to show triggering nu-
merically. Wicker et al. (1996) analysed various forms of nonlinear coupling between
unsteady burn rate and acoustic variables. By suitably adjusting the parameters and
the form of the coupling terms, triggering was demonstrated. Anathakrishnan et al.
(2005) further explained that velocity coupled models are the only possible candidates
for causing triggering in realistic operating regimes of SRMs.

1.4.2 Thermoacoustic instabilities in Rijke tube

Even though the details and characteristics of thermoacoustic instabilities depend on

the particularities of the physical and geometric setup, a great deal of insight into the
fundamental mechanisms underlying these instabilities can be gained from a study of
a simple model problem: the Rijke tube an elementary thermoacoustic device that

11
contains much of the essential physics of thermoacoustic interactions.

Due to its simplicity and representative character, the study of thermoacoustic in-
stabilities in a Rijke tube has a long history, going back to the original investigation
of Rijke (1859) who observed acoustic amplifications in a vertical tube as a coiled elec-
trical heating filament is turned on. Over the past decades, this observation has given
rise to a significant body of literature exploring this phenomenon by numerical simu-
lations (see, e.g., Kwon and Lee, 1985; Hantschk and Vortmeyer, 1999; Subramanian
et al., 2010) or by experimental efforts (see, e.g., Matveev, 2003b; Song et al., 2006).
The goal of these studies was a better understanding of the basic mechanisms linking
the acoustic field in the chamber to the unsteady heat release rates. A detailed survey of
the existing literature, followed by the motivation to perform further analysis are dealt
in Chapter 3.

The stability analysis described in the above two subsections focussed on the be-
haviour of individual eigenvalues and the associated eigenmodes. Hence, the analysis
were termed as modal stability analysis. During the early 90s, a new stability analy-
sis gained popularity in the investigation for stability of fluid flow problems (Schmid,
2007). In this case, the stability of the system is defined based on the dynamics of the
energy associated with the perturbations. A mechanism for triggering instabilities (as
introduced in the previous paragraph) is described by the above stability analysis. This
analysis is termed as non-modal stability analysis, as it involves the dynamics associated
with more than one eigenvalue and the corresponding eigenmode.

1.4.3 Non-modal stability theory

The governing linear operator of most of the physical systems are non-normal (Tre-
fethen et al., 1993). Non-modal stability theory was first applied in the analysis of
stability of fluid flow problems. Transient growth was first observed by Gustavsson
(1991) from the numerical simulations of a plane channel flow. The paper concentrated
on the evolution of a three dimensional disturbance. They used Orr-Sommerfeld equa-
tions (Orr, 1907) to describe the evolution of wall normal components of velocity and
vorticity. Non-zero initial conditions are prescribed only in the wall normal velocity

12
component and a transient growth in the kinetic energy is observed even for a damped
mode.

Butler and Farrell (1992) then performed non-modal stability analysis in the same
plane channel flow. The operator governing the linear evolution of the system was
shown to be non-normal. As described in Section 1.2, the amount of transient growth
in the energy of the perturbations depends on the initial condition. Optimum initial
condition for maximum transient growth was obtained using a variational approach.
They observed that three-dimensional perturbations are optimal in transiently increasing
the amplitude of the perturbations. Streamwise vortices are obtained as the optimal
perturbations, which further evolve to streamwise streaks. The ratio of the energy in
the streamwise streaks to the streamwise vortices were of the order 1000 for a Reynolds
number of 1000. The higher transient growth thus obtained might trigger the system to
become nonlinearly unstable, eventually reach turbulence. This was augmented by the
fact that the formation of three-dimensional structures during the onset of turbulence
was observed much earlier experimentally in boundary layer flows (Klebanoff et al.,
1962).

A framework for the non-modal stability analysis was developed by Farrell and
Ioannou (1996a,b) to analyse autonomous and non-autonomous systems respectively.
They applied the theory to analyse the stability of atmospheric flows. The idea of ob-
taining the maximum transient growth and the associated optimum initial condition by
singular value decomposition was also introduced. Later numerical simulations in sev-
eral fluid mechanic problems like, evolution of a pair of oblique waves in a plane chan-
nel flow (Schmid and Henningson, 1992), stability analysis of a falling liquid curtain
(Schmid and Henningson, 2002), transient growth in a swept Hiemenz flow (Guegan
et al., 2008), etc. were performed and the importance of non-normal nature of the
system was identified. Since the nonlinear terms in the Navier-Stokes equation for in-
compressible flow conserve kinetic energy (Henningson et al., 1993), non-normality
was shown to be a necessary condition for the subcritical transition to turbulence (Hen-
ningson and Reddy, 1994).

There were a few experimental evidence of transient growth in fluid flow problems.
The earliest experimental results were reported by Mayer and Reshotko (1997), who

13
observed transient growth of the disturbance amplitude in a pipe flow experiment. The
formation of streaky structures (regions of high and low velocity field), which appear
due to non-normal nature of the system was observed in experiments in boundary layer
(Matsubara and Alfredsson, 2001). In the same experiment, stable laminar streaks were
generated and a transient growth in the amplitude of disturbance was observed in the
stream wise direction (Fransson et al., 2004). Further, a passive control of transition to
turbulence in boundary layer was performed by exploiting the non-normal nature of the
system (Fransson et al., 2006).

To summarise, in fluid flows, the interplay between non-normality and nonlinear-

ity is shown to be one of the routes for the sub-critical transition to turbulence (Geb-
hardt and Grossmann, 1994; Baggett et al., 1995; Barkley and Tuckerman, 1999; Crim-
inale and Drazin, 1999). Non-normality and transient growth are observed in many
areas of research including magnetohydrodynamics (Krasnov et al., 2004), aeroacous-
tics (Chagelishvili et al., 1997; George and Sujith, 2009), astrophysics (Mukhopadhyay
et al., 2006) and atmospheric flows (Farrell and Ioannou, 1996a). Motivated by the
role of non-normality in the stability of fluid flow problems, similar phenomenon were
explored in thermoacoustic instability.

1.4.4 Non-normality of thermoacoustic systems

Recently, it was shown that thermoacoustic systems are in general non-normal (Bal-
asubramanian and Sujith, 2008b). They performed stability analysis on electrically
heated horizontal Rijke tube. The acoustic field in the duct was described by the acous-
tic momentum and energy equations. The response of the heat source (electrical heater)
to acoustic velocity fluctuations is modelled using a time lagged quasi-steady correla-
tion given by Heckl (1990). A transient growth of 4 5 times in acoustic energy was
observed, even for a linearly stable system.

A similar analysis was performed to analyse the stability of ducted diffusion flame
by Balasubramanian and Sujith (2008a). Burke-Schumann equations, along with the
infinite rate chemistry assumption was used to model the dynamics of the heat release
rate. Transient growth in the energy of the order of 105 was observed, which was much

14
higher than that obtained in the earlier analysis (Balasubramanian and Sujith, 2008b).
The consequences of non-normality in the subcritical transition regime were discussed
in both the papers.

Around the same period, it was shown analytically by Nicoud et al. (2007) that
the eigenmodes associated with acoustic pressure of a thermoacoustic system are non-
normal in the presence of heat source or complex impedance in the boundary.

The dynamics of the system was shown to change dramatically during the initial
time because of the non-orthogonality of the eigenmodes (Kedia et al., 2008). While in
the same paper, the frequency shifts are shown to be small when the terms contributing
to the non-normality are neglected. Furthermore, Nagaraja et al. (2009) discussed the
use of singular values, which described the transient growth of acoustic energy during
instabilities.

The effects of internal flame dynamics on the non-normal nature of a ducted pre-
mixed flame interaction was analysed by Subramanian and Sujith (2011). The dynam-
ics of the premixed flame was modelled using G equation, where the flame front was
tracked in a kinematic approach. Further, it was observed that the contribution from the
degrees of freedom associated with the internal flame dynamics is more compared to
that of the acoustic variables in the optimum initial condition.

Tulsyan et al. (2009) have analysed combustion instability involving vortex shed-
ding in the light of non-normal nature of thermoacoustic interaction. The effect of
dynamics of the vortex shedding on the acoustic field is modelled as a kicked oscilla-
tor. Transient growth was observed from period to period defined by the kicks. The
above problem was considered as discrete time system and non-modal stability analysis
developed for the same (Schmid, 2007) is employed.

Wieczorek et al. (2010) have performed non-modal stability analysis for flame mod-
elled as one-dimensional, which was located in a tube fitted with a nozzle. Apart from
the acoustic oscillations, entropy fluctuations were present and convected by the non-
zero base flow. Eigenmodes associated with acoustic and entropy fluctuations were
obtained and shown to be non-orthogonal to each other. Further, transient growth based
on acoustic energy led to spurious values. It was concluded that energy associated with

15
the entropy fluctuations should be included in the definition of disturbance energy of
the system.

Linear system identification (SI) technique was developed to perform non-modal

stability analysis by Selimefendigil et al. (2011). They performed the investigation on a
Rijke tube system. The response of the electrical heater to acoustic velocity fluctuations
were determined numerically using computational fluid dynamic techniques. From the
numerical data, a reduced order model from SI was obtained for the response of the heat
source, which is then coupled with the acoustic field. The SI technique described in the
paper captures the transient dynamics of the heater to acoustic velocity fluctuations,
making it suitable to analyse the no-normal nature of the thermoacoustic system.

Recently, Mangesius and Polifke (2011) have developed a discrete-time, state-space

approach to analyse non-normal effects in thermoacoustic systems. They applied this
framework to investigate Rijke tube system . The advantage of this approach is that
the response of the heat source obtained from SI technique can be easily included.
Furthermore, complex impedance boundary conditions can be implemented, which is a
source of non-normality.

Optimum initial conditions obtained in the earlier analysis were from the linearised
equations. Juniper (2011b) used the technique of nonlinear adjoint optimisation, in
order obtain both the linear and nonlinear optimal initial conditions for maximum tran-
sient growth in a Rijke tube system described in Balasubramanian and Sujith (2008b).
The role of the above optimal initial conditions in the asymptotic stability of the system
is also investigated. In the above model, there are two kinds of nonlinear optimal states;
the first one has strong energy growth during the initial period and the other one has a
weaker growth, but retains the energy to render the system unstable (Juniper, 2011a).
Furthermore, the effect of noise on the stability of thermoacoustic oscillations in a Rijke
tube has been explored numerically by Waugh et al. (2011). They observed that pink
noise (higher noise amplitudes in the lower frequency range) is effective in causing
the system to become unstable and postulated that this is because the initial state that
causes triggering from lowest energy (the most dangerous initial state) also has higher
amplitudes in the low frequency range. During this excitation, the system is observed
to reach a limit cycle via an unstable periodic solution. The stability limits in the sub-

16
critical transition zone are found to be influenced by the strength of noise (Waugh and
Juniper, 2011).

The role of non-normality in active control of thermoacoustic instability was inves-

tigated by Kulkarni et al. (2011). They showed that traditional controllers, which were
designed based on eigenvalue analysis will fail when the system is non-normal. They
observed secondary peaks in the linearised system and the nonlinearity in the system
sustains the secondary peaks leading to nonlinear instability. A controller was designed
so that the amount of transient growth is minimum, ensuring a monotonic decay in the
evolution of the acoustic energy.

Later, Agharkar et al. (2011a) have extended to include the dynamics of the heat
source in the design of controllers. They performed investigation to control thermoa-
coustic instability in a ducted premixed flame. The dynamics of the flame is modelled
using G equation, similar to Subramanian and Sujith (2011). A reduced order model
based on balanced truncation method is obtained and is shown to be superior to that
obtained from traditional proper orthogonal decomposition method in capturing the dy-
namics of the premixed flame. Further, in Agharkar et al. (2011b), a linear quadratic
Gaussian framework is formulated to this lower order model and control of instabilities
were achieved. The favourable locations of the sensors (microphones) and actuators
(acoustic drivers) are determined from the balanced truncation modes.

In the non-modal stability analysis of thermoacoustic systems described earlier, the

coupling between the acoustic field and the heat source was phenomenological. Fur-
thermore, there were two systems of equations involved in the problem; one for the
acoustic length scale and the other for the length scale of the heat source. In most of
the problems of thermoacoustics, the above two length scales are disparate (Poinsot
and Veynante, 2005). Moreover, the governing equations for this fluid flow become
stiff and are difficult to solve by computational fluid dynamics (CFD) technique, as
the Mach number of the steady flow and the thickness of the heat source (compared to
the acoustic wavelength) are small. Therefore asymptotic analysis is performed in the
limit of small Mach number and compact heat source to eliminate the above stiffness
problem.

17
1.4.5 Mathematical modelling of thermoacoustic systems

Wu et al. (2003) have pioneered the application of asymptotic expansions to analyse

combustion instabilities. They analysed the amplification of sound waves, when a flame
propagates in a gravity field. Separate systems of equations for acoustic field and flame
zones were derived, by performing asymptotic analysis on the conservation equations.
They have performed linear stability analysis for the acoustic-flame coupling, followed
by a weakly nonlinear theory for Darrieus-Landau mode of instability of the flame and
the acoustic field. The paper explained the experimental observation of the transition
from curved to flat flame during instability.

Moeck et al. (2007) have also performed asymptotic analysis to investigate ther-
moacoustic instability in a Rijke tube with flame as the heat source. They obtained
with mathematical rigor, the correct systems of equations and the coupling between the
acoustic field and the heat source. The presence of an additional global-acceleration
term in the momentum equation of the hydrodynamic zone was also observed. Further,
they concluded that since one dimensional equations were used in the hydrodynamic
zone for their analysis, the above term vanishes leaving the conventional momentum
equation intact in the hydrodynamic zone. The unsteady heat release rate from the
flame may be due to the equivalence ratio fluctuations at the inlet of the hydrodynamic
zone, which in turn may be caused by the acoustic velocity at the location of the flame
(Moeck et al., 2007). Recently, Wu and Moin (2010) investigated the generation of
acoustic waves from premixed flame due to freestream enthalpy fluctuations, using
asymptotic analysis. A vigorous subharmonic parametric instability was observed at
moderate levels of enthalpy fluctuations.

1.5 Existing understanding and outstanding issues

A summary of the literature review and some of the outstanding issues that have been
identified are shown schematically in Fig. 1.8. The existing investigations are ordered
in a way explained as follows. The horizontal axis represents the mathematical rigor
with which the model for a particular thermoacoustic system is developed. The verti-

18
cal axis represents the sophistication in the stability analysis for the above developed
model. The axis out of the plane of the paper represents the experimental investigations
performed in the same thermoacoustic system. The scales are qualitative. Planes A
and B represent the numerical and experimental investigations respectively. In plane
A, the various levels indicates the following, I - classical stability analysis, II - basic
non-modal stability analysis and III - advanced stability analysis. Similarly, in plane B,
I & II represent the experiments associated with classical stability and non-modal sta-
bility analysis respectively. The papers indicated in the figure are only representatives
of the kind of analysis or experimental investigations. Area shaded in grey represents
the investigations performed in the present thesis. Since the present thesis is focussed
on the stability analysis in SRM and Rijke tube, references are made that are associated
with the above thermoacoustic systems.

Numerical analysis (plane A) is considered first. In level I, Culick and his coworkers
pioneered and developed a framework to perform classical stability analysis in SRMs.
Their mathematical modelling is based on asymptotic expansion in terms of the steady
state Mach number. In this manner, the governing equations for the acoustic field is
obtained. However, the response of the propellant to acoustic fluctuations are modelled
using response functions. Further, those response functions are ad hoc and lack phys-
ical interpretations. On the other hand, Wu et al. (2003) pioneered the application of
asymptotic analysis to investigate the stability of flames to various inlet fluctuations.
Moeck et al. (2007) performed similar analysis in a Rijke tube with flame as the heat
source. The governing equations at low Mach numbers ( 103 ) of the steady state flow
and compact size of the heat source compared to the acoustic field become stiff (Ander-
son, 2001). Solving the same by CFD directly is difficult. Hence the other approach of
asymptotic analysis is preferred in multiscale problems (Ting et al., 2007). In perform-
ing asymptotic analysis, Moeck et al. (2007) observed an additional global-acceleration
term in the momentum equation governing the dynamics of the heat source. However,
they did not consider the above term, as they performed one dimensional calculations
for the dynamics of the heat source. Hence it is important to include the effects of
global-acceleration term and understand the effect of the same in thermoacoustic sys-
tem.

19
Figure 1.8: Outstanding issues and the current objective of the present thesis. Repre-
sentative existing literature on the stability analysis of thermoacoustic sys-
tems are shown in a qualitative manner. The horizontal axis represents the
mathematical rigor with which the coupling between the acoustic field and
heat source is performed. The vertical axis represents the sophistication in
the stability analysis. The axis out of the paper represents the increasing
experimental investigations. Planes A and B represent the numerical and
experimental investigations respectively. In plane A, the various levels in-
dicates the following, I - classical stability analysis, II - basic non-modal
stability analysis and III - advanced stability analysis. Similarly, in plane
B, I & II represent the experiments associated with classical stability and
non-modal stability analysis respectively. Area shaded in grey represents
the investigations performed in the present thesis.

20
 As one moves on to the next level of stability analysis, Balasubramanian and Sujith
(2008a,b) showed that thermoacoustic interactions are non-normal and non-modal sta-
bility analysis have to be performed in order to gain a more complete understanding.
Similar analysis was performed by Wieczorek et al. (2010), which included entropy
fluctuations along with acoustic fluctuations. Further, system identification technique
that can be used to perform non-modal stability analysis was developed by Selime-
fendigil et al. (2011). In plane III, investigations using advanced non-modal analysis
is shown. Juniper (2011b) applied the technique of adjoint optimisation to obtain the
optimum initial condition for maximum transient growth for both linear and nonlinear
systems.

The contributions of the degrees of freedom associated with the flame dynamics on
the optimum initial condition is examined in Subramanian and Sujith (2011). By in-
cluding the dynamics of the heat source, the total number of degrees of freedom of the
system is increased. Hence, it is important to include the contribution from the state
variables associated with the dynamics of the heat source to the disturbance energy.
This brings us to the question of the proper choice of physically relevant norm. There
has been a significant discussion in the past on choosing the correct norm in order to
describe the non-normal nature of the system (Hanifi et al., 1996; Sameen and Govin-
darajan, 2007; George and Sujith, 2011a). Although in hydrodynamic stability theory,
kinetic energy of the perturbations serves as the relevant norm, a generalisation of the
same to thermoacoustic systems is yet to be developed. George and Sujith (2011a) has
discussed various definitions of disturbance energy and their relevance in the usage in
non-modal stability analysis. However, this paper is restricted to non-reacting flows
and extension to combustion systems remains open. Hence, a rational method has to
be adopted in this thesis in order to choose the correct norm for non-modal stability
analysis.

From the experimental perspective (plane B), Blomshield et al. (1997a,b) performed
experiments in SRM. The experiments were focussed to obtain the linear and nonlinear
stability behaviour. The growth rates during nonlinear instability were also obtained.
The instrumentation in the experiments with SRMs were performed in harsh environ-
ments and hence only a fewer quantities can be measured. In order to gain a fundamen-

21
tal understanding of thermoacoustic instability, experiments in a simpler thermoacoustic
device; Rijke tube is performed. Matveev (2003b); Song et al. (2006) performed con-
trolled experiments in an electrically heated Rijke tube. They obtained the bifurcation
diagram and stability regimes for the system. The above experiments were focussed
from the point of view of classical stability analysis and are designated in level I. As
an extension from Matveev (2003b), experimental investigation in order to capture the
effects of non-normality and its consequence have to be performed.

1.6 Objective of the present thesis

The last section dealt with the existing understanding and some of the outstanding issues
in the modeling and analysis of stability in thermoacoustic systems. On the basis of the
above conclusions, following objectives are identified and are described as follows.

1. The first objective deals with the investigation of non-normal nature of thermoa-

coustic interaction in a solid rocket motor with homogeneous propellant. The

reason for choosing the above configuration is the following. Earlier analysis

(Culick, 2006), assumed the orthogonality of the eigenmodes in the investigation

of the same. The consequence of relaxing the above assumption will indicate the

effects and role of non-normality in the stability of the system.

2. The experiments associated with investigation of non-normality and transient

growth are difficult to perform in a solid rocket motor. Moreover, the coupling
between the acoustic field and the heat source was performed in a phenomeno-
logical way (Poinsot and Veynante, 2005) and was not obtained from rigorous
mathematical derivations. To address the above two issues, analysis is performed
on a simpler thermoacoustic system: Rijke tube. A mathematical framework will
be developed to derive the coupling between the acoustic field and a heat source.

3. Modeling of the Rijke tube system will now include the dynamics of the heat
source. The issues related to the internal dynamics of the heat source in regard to
non-modal stability analysis will be discussed. A discussion on the proper choice

22
 of norm in thermoacoustic systems will be made.

4. In order to obtain a more complete view, experimental investigation of non-

normal nature of Rijke tube system will be performed. From experimental data,

thermoacoustic interaction will be shown to be non-normal. Further, a search will

be performed to obtain an evidence of transient growth. Ultimately, the role of

transient growth on the asymptotic stability of the system will be examined.

1.7 Tools to be used

In order to achieve the objectives, following tools will be used. A brief description of the
same is itemized as follows. The tools are ordered in correspondence to the objectives.

1. Basic tools on non-modal stability analysis will be used (Schmid, 2007). Opti-

mum initial condition for maximum transient growth will be obtained by singular

value decomposition (Golub and VanLoan, 1989).

2. The coupling between the acoustic field and the dynamics of the heat source will

be obtained by performing asymptotic analysis (Ting et al., 2007) on the govern-

ing equations of fluid flow. Separate systems of equations governing the dynamics
of the two are obtained. The obtained equations are solved numerically.

3. The degrees of freedom encountered in the previous case will be large ( 104 ),
as the dynamics of the heat source will also be taken into account. Hence, an
advanced tool, adjoint optimisation (Juniper, 2011b) will be preferred to SVD

in obtaining the optimum initial condition and the associated maximum transient
growth.

4. In the experiments performed in Rijke tube, a simultaneous measurement of

acoustic pressure along the length of the duct will be made. Later, a data pro-
cessing technique, dynamic mode decomposition (DMD, Schmid 2010) will be
used to extract the eigenmodes of the system. The non-orthogonality of the ob-

23
 Figure 1.9: Schematic representation of the structure of the thesis.

tained eigenmodes in the presence of the heater will be determined. Further, two
microphone technique (Bellows, 2006) will be used to obtain the acoustic veloc-

ity data and ultimately the acoustic energy of the system. An evidence of transient

growth will be identified from the evolution of the obtained acoustic energy

The details of the individual tools and their techniques of application will be dis-
cussed in the places, where they are applied.

1.8 Structure of the thesis

As stated, the present investigation is focussed on four main objectives. Each objective
is presented in individual chapters. A schematic representation of the structure of the
thesis is shown in Fig. 1.9. The first objective of the investigation of non-normal nature
of thermoacoustic interaction in a solid rocket motor is discussed in Chapter 2. In
particular, the role of transient growth in the regard to the pulsed instability is focussed.

24
 The third chapter deals with the formulation of the framework to obtain the coupling
between the dynamics of the acoustic field and the heat source. The effect of global
acceleration on stability of the thermoacoustic oscillations is focussed as the main issue.

Non-normal nature of the governing equations obtained in Chapter 3 are exam-

ined. The optimum initial condition for maximum transient growth and the effect of
non-acoustic initial conditions on the stability of the system are analysed in Chapter 4.
Experimental investigations related to the identification of the non-normal nature of Ri-
jke tube system are performed in Chapter 5. The non-orthogonality of the eigenmodes
and its consequence, transient growth are captured. Chapter 6 presents the comparison
between the predictions from the numerical simulations and experimental results.

The last chapter deals with the conclusion and future outlook that could be used
in order to gain more understanding of the non-normal nature of the thermoacoustic
system.

25
 CHAPTER 2

Thermoacoustic instability in a solid rocket motor

2.1 Introduction

Solid rocket motors (SRMs) are often prone to combustion instability. The prediction
of combustion instability in the early stage of the design is a formidable task due to the
complex unsteady flow field existing in the combustion chamber. Combustion instabil-
ity occurs when the unsteady burn rate from the propellant (in SRMs) is amplified by
the positive feed back of the acoustic oscillations in the chamber. Combustion insta-
bility causes excessive pressure oscillations, which might resonate with the structural
modes of the rocket, leading to excessive vibrations, causing structural failure or dam-
age of the payload. Furthermore, during the occurrence of the combustion instability,
the heat transfer to the combustion chamber walls is increased, eventually melting them
(Sutton and Biblarz, 2001). As described in the previous chapter, the driving mecha-
nism for combustion instability to happen in SRM is the response of the unsteady burn
rate of the propellant to chamber acoustics. This leads to unsteady mass addition to the
combustion chamber and ultimately causes an unsteady heat release rate.

Thermoacoustic instabilities in SRMs are attributed to the time lag between the un-
steady burn rate and the chamber acoustics. The phase delay for a particular frequency
can be identified in the time domain as a time lag. The physical origin of this time
lag can be attributed to various dynamical processes involved in the burn rate dynamics
of the propellant. Instability occurs if the time lag is in some suitable range such that
fluctuating energy is added to the system. The above idea gave rise to the n model
(n is the interaction index that gives the coupling strength between the acoustic veloc-
ity and the unsteady combustion process and is the time lag) developed by Crocco
(1956) for liquid rocket engines. The model was simple; n and vary with frequency
in realistic situations. However, it gave a basic understanding of the physical origin of
instability.
 Initial attempts to tackle the instabilities theoretically were by Culick (1963), and
Friedly and Petersen (1966), where the linearised equations were investigated. They
obtained an explicit expression for the complex frequency of the system. The real part
of the frequency gave the growth or decay of the oscillations. Because the unsteady heat
release rate was the main driving source for the acoustic oscillations, a relation between
the two was developed in order to characterize the system dynamics. This coupling be-
tween the unsteady burn rate in response to acoustic oscillation was captured by the fre-
quency dependent admittance function (Y = Mb [(m0 /m)
/ (p0 / p) (0 /
) / (p0 / p)])
described by Culick (1968), where Mb denotes the Mach number at the burning surface,
m is the mass addition rate, p is the pressure and is the density. The prime (0 ) denotes
fluctuating quantities, while the over bar () denotes mean quantities. The admittance
function calculated was used to determine the growth of the acoustic oscillations. An-
alytical expressions for the admittance function in the linear regime were derived by
Williams (1962); Deluca et al. (1995). The governing equations are nonlinear partial
differential equations for the burn rate, which makes the problem difficult to solve for
composite propellants. Some semi-empirical theories were proposed to handle these
difficulties (Brewster and Son, 1995). The form of the admittance function was de-
rived and its exact parameters were found experimentally (Price, 1984). One of the
successful experimental techniques used to calculate the parameter values was the T-
burner technique (from the admittance function), which was demonstrated by Lin and
Wang (1995). Experimental and theoretical analysis were used in tandem to predict
instabilities.

Theoretical analysis starts with linearising the governing equations and analyzing
their stability. This leads to finding the eigenvalues (complex frequency) and eigen-
modes of the system. In a classical linear stability analysis, a system is said to be
linearly stable if the oscillations decay to zero in the asymptotic time limit, reaching
finally the steady state (stable fixed point). The system is linearly unstable if the oscil-
lations grow exponentially. Both the definitions are for infinitesimally small distur-
bances with respect to the corresponding mean quantities. The stability of the system
is determined by the real part of the complex frequency as described earlier. A linearly
unstable system grows exponentially, and after some time the oscillation amplitudes are
not small. The nonlinear effects start playing a major role in the time evolution of the

28
system, causing the evolution to reach a limit cycle (oscillations with constant ampli-
tude). This nonlinearity is attributed to the nonlinear chamber acoustics and nonlinear
combustion response of the propellant (Culick, 2006).

Culick (1976a) was the first to derive an analytical condition for the existence of
stable and unique limit cycle behaviour using a two mode Galerkin approximation with
second order acoustics as the only nonlinear process. Nonlinearity in the combustion
response was also included in the analysis and the dynamical behaviour was analysed
(Levine and Baum, 1983; Flandro et al., 2007). Apart from these, the effect of partic-
ulate damping (Culick, 1976b), vorticity (Flandro, 1995a) and flow turning (Flandro,
1995b) on SRM stability was discussed. A Computational fluid dynamics (CFD) analy-
sis was performed by Shimada et al. (2007) using the admittance function for burn rate
acoustic coupling and quasi-steady flame model in the gas phase. Growth rates at low
Mach numbers were predicted more accurately by a second order finite volume method
than the Galerkin technique usually used for such analysis (Culick, 2006).

Another interesting dynamical behaviour observed experimentally was that SRMs

stable for small amplitude disturbances were seen to become unstable for larger ones
(Culick, 2006). This was called pulsed instability or triggering (triggering occurs in
other thermoacoustic systems as well). Many mechanisms were proposed for explaining
triggering in the past. Second order gas dynamics alone were proved to be insufficient to
cause triggering because of the absence of self coupling terms (Culick, 1994). Higher
order gas dynamics (third order acoustics) also proved the same (Yang et al., 1990).
Hence, a nonlinear combustion response was thought of as an alternate candidate for
triggering. Culick et al. (1995) used an ad hoc nonlinear velocity coupling model for
burn rate response to show triggering numerically. Wicker et al. (1996) analysed various
forms of nonlinear coupling between unsteady burn rate and acoustic variables. By
suitably adjusting the parameters and the form of the coupling terms, triggering was
demonstrated. Anathakrishnan et al. (2005) further explained that velocity coupled
models are the only possible candidates for causing triggering in realistic operating
regimes of SRMs.

In the above analyses, there are some common assumptions and procedures that
were adopted to solve the thermoacoustic instability problem in the SRM. They are

29
(i) the orthogonality of the eigenmodes, (ii) the use of admittance (response) functions
or ad hoc models for burn rate-acoustic coupling and (iii) the use of classical linear
stability theory for all time t. Culick (1997) showed that the frequency shift due to
the non-orthogonality of the eigenmodes is second order in a mean flow Mach number.
However, the system dynamics change dramatically during the initial time because of
the non-orthogonality of the eigenmodes (Kedia et al., 2008). The use of admittance
(response) functions or ad hoc coupling models for burn rate acoustic coupling does
not account for the initial transients in the burn rate. These can lead to erroneous predic-
tion of the system dynamics, both qualitatively and quantitatively. If the evolution of the
unsteady propellant burn rate is modelled using a differential equation in time, then the
degrees of freedom for the system are increased. This implies that the thermoacoustic
system variables include not only acoustic pressure and velocity but also other variables
related to the unsteady burn rate. Therefore, the initial condition on the problem is re-
stricted not only to acoustic variables but also to unsteady burn rate, which might show
an interesting behaviour. Indeed, it is found that the effects of non-orthogonality of
the eigenmodes play an important role in the initial conditions related to the unsteady
burn rate. The last assumption, i.e. the validity of the classical linear stability theory, is
applicable only in the asymptotic time limit (Strogatz, 2000). This assumption is valid
during the initial period, only if the eigenmodes are orthogonal to each other. There-
fore, a new generalized stability criterion (Farrell and Ioannou, 1996a) has to be used to
account for the short term dynamics. This chapter relaxes the above three assumptions
and offers a more complete analysis of the thermoacoustic instability in the SRM. A
homogeneous propellant is considered for the present analysis.

2.2 Role of non-normality

In the past, linear stability analyses of SRMs were performed using the classical linear
stability theory (Culick, 2006). Energy in the eigenmodes can be transferred from one
eigenmode to another, either due to nonlinear coupling (Culick, 1976a) or due to non-
orthogonality of the eigenmodes (Balasubramanian and Sujith, 2008b). In the former
case, the coupling comes from the nonlinear time evolution equation for the eigen-

30
modes. This is a direct interaction. This requires some threshold amplitude (system
dependent) to be reached for it to have significant effects in the dynamical evolution of
the system. In the latter case, the energy transfer occurs even with small disturbances.
In this limit, the nonlinearity present in the system (will have a mild effect at small
amplitudes) transfers a small amount of energy from one eigenmode to another. If the
energy transfer leads to a distribution of energy among the modes in such a way that
transient growth happens, there is a net energy transfer from the base flow to the eigen-
modes. For a linearly stable case, the energy given in one eigenmode is fed to the base
flow and from it, the energy is fed back again to other eigenmodes. Thus energy transfer
happens through the base flow. It is an indirect interaction.

It is evident that the stability analysis of the SRM has to be modified including taking
the non-orthogonality of eigenmodes into consideration. The phenomenon of pulsed
instability observed in SRMs has been demonstrated theoretically only with ad hoc
burn rate-acoustic coupling models. This chapter concentrates on the four main issues.
The first issue is the inclusion of the non-orthogonality of eigenmodes, which plays an
important role in the short term dynamics of the system. The second is to use a physics
based model for the burn rate response of a homogeneous propellant to demonstrate
the various observed phenomena, especially pulsed instability. The model also captures
the transients (important for non-normal systems), which facilitates the prediction of
system dynamics during the initial time. Therefore, a differential equation for the time
evolution of the state vector is derived for the unsteady burn rate acoustic coupling and
is solved simultaneously along with the equations for the acoustic field. The dynamical
system now considered comprises not only the acoustic variables with a source from the
unsteady propellant burn rate, but also an extended one, also comprising the unsteady
burn rate variables. The third issue is to include all the nonlinear processes involved,
so that the energy transfer and large amplitude oscillations (limit cycles) are accurately
predicted. The fourth issue is that experiments performed by Blomshield et al. (1997a)
indicate that the triggering pressure amplitude required for pulsed instability is very
small (see Fig. 13 and table 5 of Blomshield et al. 1997a) compared to the steady
state and mean pressure during the limit cycle (triggering pressure amplitude is 4%
of the steady state pressure). Furthermore, the triggering pressure amplitude in their
study is much smaller than the limit cycle amplitude. This chapter investigates if pulsed

31
Figure 2.1: Schematic diagram of the combustion chamber geometry of the SRM con-
sidered.

instability can be obtained from a small amplitude initial pulse (compared with the limit
cycle amplitude) as observed by Blomshield et al. (1997a).

2.3 Formulation

The SRM considered here has a prismatic cylindrical propellant grain of length l, port
circumference Sl and a constant port area Sc . A schematic of the geometry considered
with the coordinate system used is shown in Fig. 2.1. A cylindrical geometry is studied
so as to make the analysis simple.

2.3.1 Chamber acoustics

In the SRM, the amplitude of the limit cycle pressure oscillations is often about 20% of
the mean chamber pressure (Culick, 1976a; Flandro, 1996). In this range of pressure
oscillations, nonlinear acoustics plays an important role in the nonlinear dynamical evo-
lution of the system. This is in contrast to gas turbines, where the limit cycle pressure
oscillations are 3-4% of the mean chamber pressure (Lee and Santavicca, 2005) and the
nonlinearity in combustion response alone can be assumed to play an important role.
Hence the present analysis includes a second order nonlinearity in acoustics (Culick,
1976a,b; Yang et al., 1990) and a physics based nonlinear model for combustion re-

32
sponse (Krier et al., 1968). The acoustic oscillations are assumed to be isentropic,
which is valid up to the inclusion of second order nonlinearities (Culick, 1997). Reddy
and Trefethen (1994) have shown that convection differential operator causes non-
 

normality in the system (d/dx) = d/dx, indicates adjoint operator . There-
fore, the stronger the convection, the higher is the level of non-normality. In com-
bustion chambers of SRM, the mean flow Mach number is around 0.1, which creates
strong convective effects. Hence, the distribution of mean axial velocity plays a key
role in characterizing the non-normality of the system. The above picture can be looked
as an asymmetry in the flow field, which eventually leads to non-normality. A similar
kind of asymmetry is present in the earlier analysis with diffusion flame (Balasubra-
manian and Sujith, 2008a) and Rijke tube (Balasubramanian and Sujith, 2008b) due
to the presence of a localized heat source. In contrast to these, SRM has distributed
heat sources; however, the strong convection leaves the system non-normal. In general,
non-normality in a convection-diffusion problem occurs from the convection term that
creates asymmetry in the distribution of any flow variable because of the directionality
of the base flow, whereas diffusion is a gradual process which does not create any asym-
metry (Balasubramanian and Sujith, 2008a). In order to keep the analysis tractable, we
assume a constant mean chamber pressure p, density and temperature T, which are
valid assumptions for SRM (Culick, 1997). The effects of flow turning (stabilizing)
and pumping (destabilizing) cancel each other exactly for cylindrical propellant geom-
etry (Flandro, 1995b); hence, they do not appear in this analysis. The one-dimensional
steady continuity equation is

(
u) p Sl
=m
& m
= R (2.1)
x Sc

where, m
is the mean mass influx rate from the propellant per unit volume at any axial
is the mean propellant regression rate and p is
location, u is the base flow velocity, R
the propellant density. Integrating the equation with head end velocity as zero gives

m

u = x (2.2)

The above equation gives the axial variation of mean velocity, which is used for further
calculations. At low Mach numbers M, the isentropic relation for fluctuations in density

33
can be used up to second order acoustics (Culick, 1976a). Hence, the continuity equa-
tion is decoupled from momentum and energy equations. The unsteady momentum and
energy equations are given by
 _ _  _
u _u p
= mu
_ __
+u (2.3a)
t x x

 _ _ _ 
p _p _u
_
+u + p = ( 1) Q (2.3b)
t x x
_
. _
_ _
where m = Rp Sl Sc , Q =mh, h is the heat of reaction of the propellant at a
constant pressure per unit mass and is the ratio of specific heat capacities at constant
__
pressure and volume. The source term mu in Eqn. (2.3) is due to the reduction of
the momentum of the fluid in the chamber by the low velocity inflow of the burning
_
propellant. On the other hand, the source term Q in Eqn. (2.3) is due to the energy
released by the propellant to the fluid in the chamber. Decomposing the flow variables
_ _
as, u = u + u0 , p = p + p0 , m = m 0, Q = Q 0 and R = R
+Q 0 , followed
+R
_ _ _
+m
by substituting the above in Eqn. (2.3), one obtains the governing equations for the
perturbations. The acoustic momentum and energy equations thus obtained are (with
nonlinear terms given within curly {})
   
u0 u0 0 d
u 0 u0 0 p0 p0
+ u + u + u = [m 0 u + m
u0 + { m
0 u0 }]
t x d
x x x x
    (2.4)
p0 p0 u0 du p
0
u
0
+ u + p + p0 + u0 + p0 = ( 1)Q 0 (2.5)
t x x dx x x
0 = m
where Q 0 h = R
0 p h (Sl /Sc ), the tilde () denotes dimensional quantities, u0 is

the acoustic velocity, p0 is the acoustic pressure, m

0 is the fluctuations in mass influx rate
0 is the fluctuating heat release rate per unit volume by the propellant
per unit volume, Q
0 is the fluctuating burn rate.
combustion and R

Non-dimensionalising the above equations as follows: p = p0 /

p, u = u0 /um , U =
.
0 R,
u/um , R = R x = x/l, M = um /a, where um = (ml) / (2 ) is the average
base flow velocity, U = 2x is the non-dimensional base flow velocity, a is the sonic
speed and M is the average base flow Mach number. The non-dimensionalised acoustic

34
momentum and energy equations are
   
u u dU 1 p u p p
+ M U + u + = km [RU + u] + km Ru M u +
t x dx M x x M x
    (2.6)
p u dU p p u
+ M + p + M U = ke R M u + M p (2.7)
t x dx x x x
where km = (mu pM ), ke = (( 1) lmh)
m l) / ( / (
pa).

2.3.2 Solution procedure

The Galerkin technique (Zinn and Lores, 1971; Padmanabhan, 1975) is used to solve
Eqns. (2.6) and (2.7). The dependent variable is expanded as a linear combination
of basis functions, which are chosen to satisfy the boundary conditions. The basis
functions are chosen for a duct, which is acoustically closed at both ends, in spite of
the non-zero admittance at the nozzle entry. The actual eigenmode shape is shown
to deviate from the above, which is of the order of average mean flow Mach number
(M 0.1) (Culick, 1976a). Although non trivial boundary conditions lead to the
non-normality of the system (Nicoud et al., 2007), our investigation mainly focuses on
the non-normal nature of the system, arising purely from the interaction of chamber
acoustics and unsteady burn rate. This can be regarded as the first step in analysing the
non-normal nature of the thermoacoustic interaction in the SRM. Hereafter, the term
mode specifies only the Galerkin mode unless specified. The spatial distribution of
c s
the unsteady burn rate is expanded on the above basis. The coefficients Rm and Rm are
obtained from the unsteady burn rate equation discussed in Section 2.3.4. The variables
are expanded as follows:

X
N X
N
u (x, t) = Um (t) sin(m x), p (x, t) = M Pm (t) cos(m x)
m=1 m=1
(2.8)
X
N
c s
R (x, t) = [Rm (t) cos(m x) + Rm (t) sin(m x)] , m = m
m=1

where N is the number of Galerkin modes used in the above expansion, The above ex-
pressions are substituted in Eqns. (2.6) and (2.7). Then the evolution of the coefficients
c
(Um , Pm , Rm and Rsm ) is obtained by projecting the obtained equation onto the basis

35
function (Galerkin mode) used for expansion, utilizing the orthogonality of the basis
function. The following evolution equations are finally obtained:

X
N


Un +2 1
Um In,m 2
+ Pm In,m c 3
+ Rm In,m + Rm In,m = 2 km Nn1 M Nn2 Nn3
s 4

m=1
! (2.9)
2 X
N
  2 
Pn + 5
Um In,m 6
+ Pm In,m c 7
+ Rm In,m = M 2 Nn4 (M )2 Nn5
M m=1
M
(2.10)
The coupling terms are given in Appendix A.

The set of 2N coupled first order ordinary differential equations is solved by using
1 3
fourth-order Runge-Kutta scheme (Riley et al., 2006). The integrals In,m , In,m and
4
In,m are the major contributors for the non-normality of the system. The terms contain
the convective terms U = 2x, which is a linear function of x giving a non-vanishing
coupling integral. Thus the coupling among the Galerkin modes is formed and termed
as apparent linear coupling. These terms lead to the initial transient growth and its
relation to eigenmodes is discussed by Kedia et al. (2008). Note that the mean flow
velocity U = 2x creates asymmetry in the flow field as discussed in Section 2.3.1.

2.3.3 Damping

Nozzle damping and viscous dissipation are the sources of damping of acoustic oscil-
lations in SRMs. The former contributes much to the damping of acoustic waves. A
part of the incident wave at the choked nozzle is carried away by the mean flow (the
remaining is reflected at the choked throat). Thus, some part of the acoustic energy is
carried away from the system and hence leads to loss. The loss coefficient N O evalu-
ated for short nozzles (nozzle length is small compared to acoustic wavelength) is given
by (Zinn, 1972)
   
+1 +1 2 ml

N O = MN 1+ MN , MN = , (2.11)
2 2 a

where, MN is the Mach number at the nozzle entrance plane. Even though a closed-
closed boundary condition is assumed for the rocket combustion chamber, some part

36
of the acoustic energy is carried away by the mean flow. The effect of which has to be
accounted for in the governing equations. It is a standard practice (Culick, 2006) to use
the natural duct modes as the basis for projecting the equations and at the same time
include the effect of nozzle as a damping term in the governing equation. An acoustic
boundary layer develops because of the viscous effect and no slip boundary condition
at the propellant surface. The effect is modelled as a volumetric sink term with the
damping coefficient m given by Matveev (2003b).
 r 
m 1
m = C1 + C2 (2.12)
1 m

where m is the frequency of the mth Galerkin mode, C1 and C2 are constants that deter-
mine m . For the rocket motor considered in table 2.1, l = 8 m (a medium sized motor)
and the frequency of oscillations encountered during the limit cycle oscillation is 66.25
Hz (Section 2.7.4), which is close to the fundamental mode of the rocket configuration.
Hence, the frequency of oscillation is not very small and viscous damping is expected to
play a role. The higher acoustic modes will have higher frequencies and are affected by
viscous damping and hence viscous damping has to be included. Moreover the viscous
damping coefficients chosen for the present simulations are small (C1 , C2 O (102 ))
compared with the nozzle damping (N O O (101 )). The decay of mth Galerkin
pressure mode due to damping is given by

Pm
= (m + N O ) Pm (2.13)
t

The right hand side of Eqn. (2.13) is added to the right hand side of Eqn. (2.10) to
account for the losses. A similar analysis is performed by (Matveev, 2003b). The final
acoustic momentum and energy equations are as follows:

X
N


Un +2 1
Um In,m 2
+ Pm In,m c 3
+ Rm In,m + Rm In,m = 2 km Nn1 M Nn2 Nn3
s 4

m=1
! (2.14)
2 X
N
 
Pn (n + N O ) Pn + 5
Um In,m 6
+ Pm In,m c 7
+ Rm In,m
M m=1 (2.15)
2 
= M 2 Nn4 (M )2 Nn5
M

37
2.3.4 Unsteady burn rate

The unsteady burning of the propellant in response to the acoustic oscillations is shown
to be the main driving source of acoustic instabilities in SRMs (Williams, 1962; Kuo
and Summerfield, 1984). Gusachenko and Zarko (2008) give an excellent review of
the unsteady solid propellant burn rate models. The burn rate fluctuates in response
to acoustic pressure and velocity (parallel to the propellant surface) oscillations in the
chamber (Culick, 1968). Much less is known about the velocity coupling models and
only ad hoc response functions are used for the SRM stability analysis (Levine and
Baum, 1983; Baum and Levine, 1986). The acoustic velocity leads to the convective
heat transfer at the propellant surface leading to unsteady burn rates. Sometimes flow
reversal takes place which further increases the nonlinearity of the response. In this
investigation, acoustic pressure - burn rate coupling alone is investigated, as the acoustic
velocity burn rate coupling needs a more involved treatment. The models available to
study acoustic velocity burn rate coupling are few and are insufficient to represent the
dynamics involved in sufficient detail.

On the other hand, a number of models exist for unsteady burn rate - acoustic pres-
sure coupling. When a propellant burns, there exist a condensed phase and a gas phase
above the propellant surface. The flame is present in the gas phase and most of the
heat release is from that phase. There are three basic time scales involved in the prob-
lem: (i) the reaction time scale of the flame, (ii) the flow time scale in the gas and
condensed phase, (iii) the conduction timescale in the solid phase. A homogeneous
propellant is analysed in the present case, which results in a premixed flame in the gas
phase (Williams, 1985). Premixed flame reaction time scales are very small compared
with the acoustic time scales of the chamber. Furthermore, assuming a quasi-steady gas
phase and a small condensed phase (solid and pyrolysed gaseous propellant) leads to
analyzing the dynamic response only in the solid phase (Williams, 1962; Krier et al.,
1968). Later, the restriction of small condensed phase was relaxed (Romanov, 1999).
Thermal inertia in the gas phase has been studied by Kumar and Lakshmisha (2000).
Culick and Isella (2000) incorporated the dynamics in the condensed and gas phases.
Apart from these, the burn rate response with phase transitions in the condensed phase
(Cozzi et al., 1999), propellant heterogeneity (Cohen and Strand, 1985) and some spe-

38
cific class of propellants (Ward et al., 1998) have been investigated. An asymptotic
analysis has also been used to obtain the burn rate response of the propellant (Margolis
and Armstrong, 1986, 1988). However, the above analysis is mathematically complex
and analytical solutions are very much limited.

The unsteady burn rate model used in this paper is from Krier et al. (1968). The
model used here is simple but it captures the essential physics of the problem. The time
lag between the acoustic pressure and burn rate is due to the finite speed of thermal
 2

wave propagation in the solid phase and the time scale th = R associated is

comparable with the chamber acoustic time scale (a = l/a). The ratio F = a /th for
the SRM parameters shown in table 2.1 equals 1.37. A differential equation in time
for non-dimensional temperature (T) inside the propellant grain is derived from the
energy equation. The propellant burn rate is then related to the surface temperature by
a power law (Krier et al., 1968). The dynamical boundary condition is written, relating
the acoustic pressure and heat transfer at the propellant surface. The derivation of the
equation is given in Krier et al. (1968) and the final nonlinear equation is as follows.

At each x location,

T T 2T
(1 + R) = 0, 0 y < , 0 <
y y 2 (2.16)


m 2 (a) = F
R = TS p 1, TS ( ) = T (y = 0, ), /t = lR

Boundary Condition (BC):


T  (1 + p)2n ((1 + p)n/mp H)
= H(1 + R) (2.17a)
y y=0 1+R

T (y , ) = 0 (2.17b)

Initial condition (IC):

T (y, 0) = Tst (y) + Tp0 (y) (2.18)
 
  2
.  
where, y = y R , = t(l/a) R , H = QS Sp TS,0 T , T =
 . 
T T TS,0 T ,Tst = ey , T is the temperature of the propellant at y
, TS,0 is the surface temperature of the propellant, Tst is the non-dimensional steady
state temperature, Tp0 is the non-dimensional temperature fluctuation at t = 0, n is the

39
 Figure 2.2: Geometry of the pressure coupled propellant response model.

burn rate index, mp is the pyrolysis coefficient, is the thermal diffusivity of the pro-
pellant, QS is the overall heat release per unit mass at the propellant surface, Sp is the
specific heat capacity of the propellant, y is the non-dimensional distance from the pro-
pellant surface, H is the ratio of the heat release at the propellant surface to its thermal
capacity and F is the ratio of timescales of the chamber acoustics (a ) and transient heat
conduction in the propellant (th ).

The co-ordinate system is fixed to the propellant surface which regresses according
to the burn rate. The geometry is shown in Fig. 2.2. A Dirichlet type boundary con-
dition far from the propellant surface and a Neumann type at the surface Eqn. (2.17a),
which comes from the balance between the amount of heat transfer from the flame
in the gas phase to the propellant surface, are applied. The steady state temperature
profile (Tst = ey ) obtained as the solution of the corresponding steady state problem
T /y + 2 T /y 2 = 0, is exponentially decaying in y. The problem is formulated
in one dimension as the response function predicted is shown to be accurate by Baum
and Levine (1986); Culick and Isella (2000). Note that (1+R) in Eqn. (2.16) appears
as a convection term, as the coordinate system is fixed to the propellant surface, which
is regressing with the burn rate at that time (see Fig. 2.2). As described in Section
2.3.1, this term contributes to the non-normality in the burn rate response. The other
term in Eqn. (2.16), i.e. the diffusion term has no preferred direction associated with
 

it (d2 /dx2 ) = d2 /dx2 and hence does not contribute to the non-normality of the
burn rate response. The above advection diffusion equation is shown to produce high
transient growth by Reddy and Trefethen (1994). The presence of the non-normal be-
haviour in both combustion and acoustics leads to high transient growth when coupled

40
together. Hence, the short-term dynamics obtained from both the unsteady burn rate
and acoustic equations will be very different from those predicted by the classical lin-
ear stability theory for asymptotic time. It is now important to understand the various
physical processes that contribute to the dynamics of the unsteady burn rate.

The mechanism of the burn rate - acoustic pressure coupling is as follows. The
reaction rate of a premixed flame is dependent on the pressure and, to a weaker extent,
on the temperature. During the compression part of the acoustic cycle, the flame speed
increases, and the flame comes closer to the solid thereby causing more heat transfer
to the solid. The pyrolysis of the propellant is assumed to obey the Arrhenius law,
which under practical values of activation energy leads to a power law dependence
on surface temperature (Krier et al., 1968). Higher heat transfer to the solid phase
increases the solid phase surface temperature and hence the burn rate. Thus the acoustic
forcing comes through the boundary condition stated in Eqn. (2.17a), which is a crucial
difference from the earlier analysis of Balasubramanian and Sujith (2008a), where the
forcing explicitly appears through the convective term in the equation.

From the physics of the problem, the temperature fluctuations are expected to be
high near the propellant surface and decrease towards the chamber casing (Dirichlet
boundary condition). Thus, it is necessary to cluster more grid points near the surface,
which will yield accurate results with fewer grid points and hence lesser computational
time (Anderson, 2001). The transformation = eky is performed on Eqn. (2.16) and
k controls the amount of grid clustered near the propellant surface.

Equation (2.16) with boundary (Eqn. 2.17) and initial (Eqn. 2.18) conditions are
transformed into

2
T T 2 T
+ ((1 + R)k k )2
(k) = 0, 0 10



2



mp

R = TS 1, TS ( ) = T ( = 1, )

  

T  1 (1 + p) ((1 + p) p H)
2n n/m
(2.19)
BC : = H(1 + R) ,
=1 k (1 + R)






T ( 0, ) = 0





1/k
IC : T (, 0) = + Tp ()

41
Equation (2.19) is solved by a semi-implicit backward time central space (BTCS) scheme
similar to that used by Junye (2000). The burn rate R(x) at each axial location is ob-
tained for the corresponding pressure at that location. Now, in order to use the source
terms R(x) in Eqns. (2.14) and (2.15), R(x) is projected onto the Galerkin modes
c s
andRm , Rm are obtained as:
Z 1 Z 1
c s
Rm (t) = 2 R(x, t) cos(m x)dx, Rm (t) = 2 R(x, t) sin(m x)dx (2.20)
0 0

To track the evolution of the system, the nonlinear equations (2.14) and (2.15) are
integrated using the fourth order RungeKutta (RK4) method with Eqn. (2.19) updated
at each sub-step of RK4 using the semi-implicit BTCS scheme. General conclusions
about non-normality of the system can be made with the linearised equations. To quan-
tify the effect of non-normality, the equations are cast in a standard linearised form,
which can be analysed using an existing frame work (Schmid and Henningson, 2001).

2.4 Short term dynamics and transient growth

To analyse the non-normal system, a state space vector formulation is used. Farrell and
Ioannou (1996a) developed a generalized stability theory for non-normal linear opera-
tors, in the context of atmospheric sciences. In the state space vector representations, a
general N dimensional linear dynamical system is given by (same as Eqn. 1.1)

d (t)
= L (t) (2.21)
dt

where L is an N dimensional square matrix. Balasubramanian and Sujith (2008a,b)

have shown that L is non-normal (LL 6= L L) for a thermoacoustic system. The
general solution is given by (t) = eLt (0), where (0) is the initial condition. An
amplification factor is defined as 2 = h(t) | (t)i/h(0) | (0)i (h | i denotes in-


ner product in the linear vector space, |(t)i = (t), h(t)| = (t) as a measure of
growth or decay of fluctuations in the system during its time evolution. Note that 2
can be related to some physical quantity such as the disturbance energy of the system
(Schmid and Henningson, 2001; Nagaraja et al., 2009). Moreover, 2 will depend on

42
the choice of the initial condition. The maximum value of 2 for all possible initial con-
2
ditions is given by G(t) = M ax (h(t) | (t)i/h(0) | (0)i) = eLt (Golub and
(0)
VanLoan, 1989). Here, kAk denotes the 2-norm of the matrix A. The optimum initial
condition (Vopt ) to attain the maximum transient growth is obtained as follows. The
formal solution of Eqn. (2.21) is given by:

(t) = eLt (0) (2.22)

where, (0) is the initial condition of the state space variables. The matrix eLt is called
as the propagator matrix, which advances the system from (0) to (t). Now a singular
value decomposition (Golub and VanLoan, 1989) of eLt is performed as below

eLt = U DV (2.23)

where U & V are the left and right singular vectors, D is the matrix, whose diagonal
elements form the singular values. Further, U & V are orthogonal matrices, which
implies U U = V V = I. Multiplying Eqn. (2.23) throughout by V , the following
equation is obtained
eLt V = U D (2.24)

The above matrix equation can be interpreted schematically as in Fig. 2.3. Now, com-
paring the left hand side of Eqn. (2.24) with the right hand side of Eqn. (2.22), one can
interpret V containing the initial condition (0). The rows of V , i.e., v1 , v2 , ... forms
the initial conditions. Since the matrix D is diagonal, the action of eLt on v1 , v2 , ...
gives the output u1 , u2 , ... amplified by the numerical value of d1 , d2 , .... As the sin-
gular values are arranged in the descending order, maximum amplification is obtained
for v1 and is termed as the optimum initial condition. u1 is the corresponding optimum
output and the associated amplification is given by d1 . Further, the numerical value of
d1 is the L2 norm of the matrix eLt as discussed in the previous paragraph. Now, from
the evolution of G (t), maximum amplification over all possible initial conditions and
time can be obtained. For a system that is unstable according to classical linear stability
theory, Gmax = max(G(t)) . For a system, that is stable according to the classical
linear stability and is highly non-normal, Gmax >> 1. This shows high initial transient

43
Figure 2.3: Schematic representation of performing SVD to determine the optimum ini-
tial condition.

growth. For a linearly stable normal system Gmax = 1.

2.5 Linear analysis

To analyse the generalized stability of the system as discussed in the previous section,
Eqns. (2.14), (2.15) and (2.19) are linearised to give the following:

X
N


1 2 c 3 s 4
Un +2 Um In,m + Pm In,m + Rm In,m + Rm In,m = 0 (2.25)
m=1

!
2 X
N
 
Pn (n + N O ) Pn + 5
Um In,m 6
+ Pm In,m c 7
+ Rm In,m =0 (2.26)
M m=1

2
Tp Tp 2 Tp 1/
+ (k k )
2
(k) k
+ mp Tps = 0

2




2
R = mp Tps , Tps (t) = Tp ( = 1, t), /t = lR a = F (2.27)
  

Tp  ATps Bp



BC :  = , T p ( 0, ) = 0
=1 k
 .
where A = (2H 1) /mp , B = 2H 1/mp 2 n, Tp is the fluctuating tem-
1
perature in the propellant given by Tp = T Tst = T /k . Equation (2.27) has
to be applied at all axial locations. Because R is a linear function of Tps , Tp can be

44
decomposed as follows:

X
N
Tp (x, t) = [Tkc (t) cos(k x) + Tks (t) sin(k x)] ,
k=1 (2.28)
Rkc = Tkc /mp , Rks = Tks /mp

The above expression, when substituted in Eqn. (2.27) and then projected onto the
Galerkin basis as described earlier, leads to the following equations:

Tnc T c 2 Tnc
+ (k k 2 ) n (k)2 + mp 1/k Tsn
c
=0

2



c
Tsn (t) = Tnc ( = 1, t) (2.29)
  

Tnc  c
BM Pn

ATsn c


BC :  = , T n ( 0, ) = 0
=1 k

Tns Tns 2 s
+ (k k )
2
(k) 2 Tn
+ mp Tsn = 0
1/k s


2


s s
Tsn (t) = Tn ( = 1, t) (2.30)
  

Tns  s

ATsn s


BC :  = , T n ( 0, ) = 0
=1 k

An important observation in the linear regime is that the Galerkin pressure mode is
absent in the boundary condition for Eqn. (2.30). Hence, Rns is not affected (in the
linearised equations) by the acoustic fluctuations, and it evolves depending only on its
initial condition. However, it affects the acoustic momentum equation Eqn. (2.25)
s 4
through the Rm In,m term. Equations (2.25), (2.26), (2.29) and (2.30) can be cast in the
form of (2.21). Equations (2.29) and (2.30) are discretised using a second-order central
difference at Mg equally spaced points in 0 0 . The linearised equations are

d
= L (2.31)
dt
 
= c1 c2 . . cN s1 s2 . . . sN
12N Mg

 T
= U1 P1 . . . UN PN T represents matrix transpose
12N
 T
cn = c
1 Tn(1) c
2 Tn(2) c
. . . . Mg 1 Tn(M g 1) 1(Mg 1)

45
  T
sn = s
1 Tn(1) s
2 Tn(2) . . . . s
Mg 1 Tn(M g 1) 1(Mg 1)

where the subscripts n(1), n(2), n(3)....n(Mg 1) represents the fluctuating tempera-
ture at 1st , 2nd , 3rd .... (Mg 1)th points from the propellant surface for the nth mode.
The homogeneous boundary condition (Tns ( 0, ) = 0) at the last point, leave cn
and sn with (Mg 1) discrete points. The linear operator matrix is expanded in Ap-
pendix B. Now, as mentioned in Section 2.2, Eqn. (2.27) indicates the extra degree of
freedom for the system apart from acoustics and it appears as extra variables cn , sn in
Eqn. (2.31). This implies that not only does the system comprise acoustic variables, but
it is also an extended one with variables from the burn rate response. Hence, acoustic
energy (Rienstra and Hirschberg, 2008), which defines the growth or decay of acous-
tic oscillations is inadequate. A new generalized disturbance energy is defined which
accounts for the perturbations in burn rate variables (cn , sn ). A formal derivation of
fluctuating thermal energy in the propellant is performed. The thermal energy obtained
is then added to the acoustic energy with appropriate weight factors (arrived at from
the consideration of entropy generation at the propellant surface) to get the generalized
disturbance energy. The factor i in Eqn. (2.31) is present for the reason, that the 2
norm or L2 norm of (t) represents the disturbance energy. The following section deals
with the disturbance energy and its relation to h(t) | (t)i.

2.6 Generalised disturbance energy

As mentioned in Section 2.4, 2 , which is a relative measure of the 2-norm of (t)

 
k(t)k = (h(t) | (t)i)1/2 , can be related to the energy in the disturbance calculated
from the state space variables in Eqn. (2.31). From our analysis, this energy has two
components. The first component is from the chamber acoustic field and the second is
from the unsteady thermal energy of the propellant. Energy in the acoustic field can be
characterized by the familiar acoustic energy (Rienstra and Hirschberg, 2008) given by
ZZZ h
1 
2 i
Eac (t) = u0 (x, t))2 + p0 (x, t) a2
( dV (2.32)
2 chambervolume

46
Non-dimensionalising the above by u2m Sc l/2 and assuming 1-D variation, leads to the
following:

Z 1"  2 #
1X 2

N
Eac (t) 2 p(x, t)
Eac (t) = 1 2 = (u(x, t)) + = Un + Pn2 (2.33)
2
um Sc l 0 M 2 n=1

Now a similar expression for the fluctuating energy stored in the solid phase of the
propellant has to be calculated to get the total disturbance energy. Equation (2.16) is
written in its linearised and dimensional form for the fluctuating temperature T0 as
!
T0 0
T R

0 Tst 2 T0
p Sp R = p (2.34)
t y y y2

where p is the thermal conductivity of the solid phase of the propellant. Multiplying
.
0
the above equation by T T and rearranging, we obtain

! !2
p Sp T02p T0 T 0 02
= T0 + p Sp R T + p Sp Tst R
0 T0
2T t T y y y 2T y 2T y
(2.35)
Integrating Eqn. (2.35) over the entire solid phase propellant volume and applying
Gauss divergence theorem, we obtain
RRR !
p Sp T02 dV 1
ZZ 0    

V1
= p T 0 T
dS +
p Sp R 02
T  T  02
2T
t T St y 2 Su Sl

ZZZ ZZZ ! 2 (2.36)

p Sp Tst 0 0 p T0
+ R T dV dV
2T V1 y T V1 y

where V1 is the entire propellant volume, St is the entire propellant surface, Su is pro-
pellant upper surface and Sl is the propellant lower surface (casing). The left-hand side
of Eqn. (2.36) is a positive definite quantity, which is the fluctuating energy due to
temperature fluctuations in the propellant. The first term in the right-hand side is the
corresponding energy flux term. The second term is due to the contributions from the
unsteady burn rate and the last term is from the loss due to thermal conduction (the term
is always negative). Now the equation is the conservation equation for the fluctuating

47
energy Ep present in the solid propellant. The fluctuating energy Ep is
ZZZ Z Z 1
p Sp 02 p Sp  2
Ep = T dV = TS,0 T lSl Tp2 dxdy (2.37)
2T V1 2T R y=o x=0

Non-dimensionalising the above by u2m Sc l/2, substituting for Tp from Eqn. (2.28) and
substituting = eky , we obtain

Z  c 2  s 2 !
1 X
N
Ep T T
Ep = 1 2 = n + n d (2.38)
2
um Sc l 2 =o n=1

  2  

where = p Sp Sl TS,0 T Tu2 Sc k . Discretising in the domain
R m

leaves Eqn. (2.38) as

N Mg 1  2  s 2 !
Ep X X Tnc T
Ep = 1 2 = + n (2.39)
2
um Sc l 2 n=1 i=1 i i

Now, the weightage to energies Eac and Ep in forming the total disturbance energy ET
is fixed by considering the energy from entropy fluctuations. Chu (1965) has derived an
 
expression for the disturbance energy Echu and is given as follows:

ZZZ "    0 2 #
1 p02 (x, t) ( 1) P s
Echu (t) = u0 (x, t))2 +
( + dV
2 chamber volume a2 <
(2.40)
where s0 is the entropy fluctuation and < is the characteristic gas constant of the gas
in the combustion chamber. Flame is the source of entropy fluctuations in the motor.
The entropy fluctuations are computed from propellant surface temperature and acoustic
fluctuations, as derived by Krier et al. (1968), as

s0 h  i
= n
2nH 2n /mp 1 p + (2mp + 1 2mp H) Tps (2.41)
< 1
 . 
where = TS,0 T Tf . Note that Tf is the steady flame temperature. Now
2
the coefficient proportional of Tps in the entropy part of Echu after substituting Eqn.
(2.41) in Echu is [ (2mp + 1 2mp H)]2 (P ) / ( 1). This gives the weight factor

48
Wf for Ep which has to be added to Eac . Hence, the total disturbance energy is

" #2
2 P T (2mp + 1 2mp H)
ET = Eac + Wf Ep , Wf = (2.42)
( 1) p Sp TS,0 T

Non-dimensionalising as before, we get the total non-dimensional generalized distur-

bance energy ET . This energy incorporates the contributions from the entropy fluctua-
tions released by the flame as well as the temperature fluctuations inside the propellant.

ET
ET = 1 = Eac + Wf Ep (2.43)
2
u2m Sc l

Now, using Eqns. (2.33) and (2.39), we get

" Mg 1
#
1X
X 2

N
ET
ET (t) = 1 2 = Eac + Wf Ep = Un2 + Pn +
2
(i Tnc )2 + (i Tns )
2
um Sc l 2 n=1 i=1
1
= k (t)k2
2
(2.44)

p
where i = Wf /i . Thus, the 2-norm of (t) is related to the physical gener-
alized disturbance energy and 2 gives the amplification of the same during the system
evolution. It should be noted that Chus energy is derived in the limit of zero mean
flow Mach number. However, in the present case the mean flow Mach number is not
zero (M 0.1). If one includes the energy contribution due to mean flow, the distur-
bance energy (ET ) can not be represented by the L2 norm. The reason for choosing ET
as the L2 norm is as follows. In a dynamical system, L2 norm can be calculated eas-
ily using singular value decomposition (SVD). The optimum initial condition and the
maximum transient growth can then be obtained directly from the SVD. The choice of
norms other than the L2 norm brings the complication of defining new inner products.
The inner products thus defined can be mathematically inconsistent, and some special
techniques (such as adjoint optimisation) other than SVD should be used. To keep the
analysis simple, Chus energy is used, so that ET coincides with the L2 norm. The L2
norm can then be computed using SVD. However, in Chapter 4, a more general norm
(based on Myers energy) other than L2 norm is used. As indicated earlier, adjoint op-

49
 Gas properties Propellant properties Combustion chamber dimensions

p=60 bar h=7 MJ/kg l=8 m

T=2900 K =1.6106 m2 /s Sl =1.59 m
=4.82 kg/m3 mp =6 Sc =0.2 m2
=1.35 n=0.4 Numerical parameters
<=287 J/kg k
R=0.017 m/s N=5
H=0.76 p =1800 kg/m3 Mg =150
= 0.19 Sp = 3542 J/KgK Time step t = 0.005

Table 2.1: SRM parameter values and operating conditions.

timisation technique is used to obtain the maximum transient growth and the associated
optimum initial condition.

2.7 Result and discussion

Simulations are performed for a rocket motor, whose system parameters are given in
table 2.1. Increasing N, Mg and decreasing t beyond the above values leads to a dif-
ference of less than 1 %. Hence, the above values are chosen for all simulations shown
below. The damping coefficients C1 and C2 are varied to get a different dynamical
behaviour for the simulation.

2.7.1 Linearly stable and unstable system

The linearised equation (2.21) is analysed for the stability of the system to small ampli-
tude disturbances. Eigenvalues of the discretised linear operator L determine the linear
stability of the system in the asymptotic time limit. If all the eigenvalues lie in the left
half of the complex plane, the system is asymptotically stable to small disturbances.
This above predictions are given by classical linear stability. However, the short term
behaviour is different due to the non-normal nature of the linearised operator. Using
Farrell and Ioannou (1996a) terminology, a system with Gmax = 1 is called a lin-

50
Figure 2.4: The evolution of acoustic pressure at x = 0.25 for a linearly stable system.
U1 (0) = 3, P1 (0) = 3, Pm6=1 (0) = Um6=1 (0) = 0, M Tp (, 0) = 0, C1 =
0.05, C2 = 0.001.

Figure 2.5: The phase portrait of the acoustic pressure and the unsteady burn rate at x
= 0.25. U1 (0) = 3, P1 (0) = 3, Pm6=1 (0) = Um6=1 (0) = 0, M Tp (, 0) =
0, C1 = 0.05, C2 = 0.001.

51
early stable (generalised linear stability) and the finite amplitude small disturbances
die down monotonically. The time evolution of unsteady oscillations is shown at one
fourth of the motor length (l/4) from the head end as a representative position. Acous-
tic pressure oscillation is shown in Fig. 2.4 for a system, stable according to classical
linear stability. The amplitude decays in the asymptotic time limit, eventually reaching
a stable fixed point. The phase space plot between p (x=1/4) and R (x=1/4) (Fig. 2.5)
shows a spiral trajectory, eventually collapsing to a single point (0, 0) corresponding to
the mean flow (arrows indicate the direction of the time evolution). This type of fixed
point is called stable focus. The actual dimensions of the phase space in the modal
and discretised form is 2N +2N (Mg -1)=2N (Mg ), which corresponds to the total num-
ber of state space variable in Eqn. (2.31). The plot in Fig. 2.5 is just the projection of
2N (Mg ) space onto a two dimensional space. Hence we observe apparent intersections
of the phase trajectories, which are actually evolving without intersection in a higher
dimension space.

On the other hand, if the real part of one of the eigenvalues is positive, the system is
linearly unstable. Figure 2.6 shows the acoustic pressure evolution from a small initial
disturbance for a linearly unstable system. Initially, the oscillations grow exponentially
as predicted by linear stability theory, reaching amplitudes where nonlinear terms start
dominating. The nonlinear terms in Eqns. (2.14) and (2.15) start dominating, balanc-
ing the driving terms resulting in the formation of oscillations of constant amplitude
called limit cycle. The corresponding phase plot between p (x=1/4) and R (x=1/4) af-
ter removing the transients leaves a closed curve (Fig. 2.7). This corresponds to limit
cycles, where the trajectories close itself as t . The presence of apparent multiple
intersections in Fig. 2.7 shows the presence of more dominant frequencies, which is
a characteristic of limit cycle oscillations in SRMs. The existence of limit cycle was
explained by Culick (1976a) with second-order nonlinear acoustics. However, another
important phenomenon is the occurrence of pulsed instabilities (Blomshield et al.,
1997a) in SRMs. This is a type of instability leading to unpredicted damage of the mo-
tors and was not explained to date with a physics-based acoustic burn rate coupling
model. This issue is addressed in Section 2.7.4. Before examining the nonlinear regime
of oscillations, it is important to discuss some more interesting results pertaining to the
non-normal nature of the linear operator L in the following section.

52
Figure 2.6: The evolution of acoustic pressure at x = 0.25 for a linearly unstable sys-
tem. U1 (0) = 0.3, P1 (0) = 0.3, Pm6=1 (0) = Um6=1 (0) = 0, M Tp (, 0) =
0, C1 = 3 104 , C2 = 1 104 .

Figure 2.7: The phase portrait of the acoustic pressure and unsteady burn rate at x =
0.25. U1 (0) = 0.3, P1 (0) = 0.3, Pm6=1 (0) = Um6=1 (0) = 0, M Tp (, 0) =
0, C1 = 3 104 , C2 = 1 104 .

53
2.7.2 Pseudospectra and transient growth

The linear operator (Eqn. 2.31) in its discretised form (L matrix) is used for pseudospec-
tra computation. For normal operators, resonance (maximum amplification) happens at
the eigenvalues; at other points, the amplification is inversely proportional to the dis-
tance of the forcing frequency from the nearest eigenvalue. However, for non-normal
operators, resonant amplification of many orders occurs far from the eigenvalues and is
called pseudoresonance (Trefethen and Embree, 2005). The pseudospectra plot is
used to analyse non-normal operators and z is called an pseudoeigenvalue of the oper-

ator L, if it satisfies (zI L)1 1 (Trefethen and Embree, 2005). The perturba-
tions given are very small compared with the size of the linear operator ( << kLk).
For normal operators, pseudospectrum consists of concentric circles, confirming the
inversely proportional relationship between the amplification and the distance between
the excitation frequency and the nearest eigenfrequency. However, for non-normal op-
erators the contours are distorted.

Figure 2.8 (a) shows that the pseudospectra of the L matrix are highly distorted near
the imaginary axis. The system considered is stable according to the classical linear
stability theory. All the eigenvalues of the system lie on the left half of the complex
plane. The perturbation in the linear operator L is depicted in its pseudospectra. The
relation between transient growth and the geometry of the pseudospectra is described in
Trefethen and Embree (2005). The contours spill over to the right half of the complex
plane, which is an indication of transient growth. The zoomed-in contour of Fig. 2.8
(a) near the origin is shown in Fig. 2.8 (b). For example, a perturbation of = 101
(/ kLk = 5.33 104 ) leads to the spillage of the pseudospectra to the right by z =
72 units from the imaginary axis. From this, the transient growth is estimated to be
(z/)2 = 7.22 = 51.84. This is just one point in the contour. Maximizing this over
all contours results in Kreiss constant (), which gives the lower bound for the max-
imum transient growth (Gmax ). For a normal system, contours move proportionally
outwards with z and hence = 1 with no transient growth. Thus, qualitative informa-
tion can be obtained from the contours of the pseudospectra. As is shown in Section 2.6
the square of the 2-norm of the state space vector equals the total disturbance energy
in the system. Hence, the transient growth obtained now directly gives the disturbance

54
Figure 2.8: a) Pseudospectra of the non-normal linear operator L. b) Zoomed in view
near the origin and the calculation of the lower bound for the maximum
transient growth. The contour value represents log10 . C1 = 0.03, C2 =
0.02, kLk = 1.88 104 , max = 1 102.5 , max /||L||  1. <(z) and
=(z) indicates the real and imaginary part of z respectively.

energy amplification. We also note that a very small perturbation (0.1%) leads to an
energy rise of 2 orders of magnitude. All these are obtained from investigating the
geometry of the pseudospectra.

2.7.3 Effects of the internal degrees of freedom

2
The exact calculation of maximum transient growth is to evolve G(t) = e(Lt) and
find its maximum as discussed in Section 2.4. Figure 2.9 (a) shows the evolution of
(Lt)
e and the maximum transient growth is found to be 128.92 = 1.66 104 , which is

higher than the previous estimate based on the lower bound of Gmax from pseudospec-
tra. Also, the same figure shows that the use of response function for modelling the
acoustic - burn rate coupling gives rise to very small transient growth. The response
function is calculated as follows. The equation for unsteady burn rate (Eqn. 2.27) is
solved for a forced pressure oscillation p = p0 sin (t) to get the unsteady burn rate

55

Figure 2.9: a) Comparison of the evolution of eLt with differential equation for
unsteady burn rate and response function (YR ) calculations, C1 = 0.03,
C2 = 0.02, (0) = Vopt . b) The magnitude of the response functions
(|YR |)of the propellant for various values of H.

in the form R = R0 sin (t + ). Here R0 is obtained from the Fourier transform

of the signal R at the frequency /2. The associated phase () is determined from
R . qR qR 
2 2
cos = 0 p(t)R(t)dt 0
p (t)dt 0
R (t)dt . The response function (YR )

is then calculated as YR = R/p = R0 ei p0 . The magnitude of YR for various exci-
tation frequencies is shown in Fig. 2.9 (b). It is also observed that magnitude of YR
reaches maximum around 1, which corresponds to the timescale for unsteady conduc-
tion inside the unburnt propellant.

Next for the extended system, the optimum initial condition (Vopt ) for the maximum
transient growth is calculated. The state space vector (t) has N pairs of acoustic vari-
ables called acoustic modes. The remaining 2N (Mg -1) variables describe the unsteady
propellant burn rate response called burn rate modes. Neither the acoustic modes, nor
the burn rate modes are the eigenmodes of the system. They are only reference modes
(basis functions) satisfying the boundary condition and the variables are just projected
along these modes.

The relative amplitude of various modes in the initial condition (Vopt ) is shown
in Fig. 2.10 with the acoustic modes shown in the inset. The important observation is
that, for obtaining maximum transient growth, one should also excite in the burn rate
modes. Although, the appropriate response function (YR ) is used instead of solving

56
Figure 2.10: Relative amplitude of the optimum initial condition direction. Inset
shows the relative amplitude of the acoustic modes, C1 = 0.03, C2 =
0.02, (0) = Vopt , N = 5, Mg = 150.

simultaneously the unsteady burn rate equations along with the acoustic equations, there
is a huge difference in the transient growth between the two curves in Fig. 2.9 (a). By
using a response function, the dynamics involved in the unsteady burn rate are implicitly
not taken into account. Now, if the optimum initial condition for the maximum transient
growth is distributed more in the burn rate modes, then it is natural to expect a small
transient growth, if the dynamics in the burn rate modes are not taken into account.
Figure 2.10 shows indeed that the optimum initial condition is distributed among the
burn rate modes, and hence there is a large difference in transient growth of both curves
in Fig. 2.9 (a).

Therefore, for a stable system, according to classical linear stability, a very small
local change in propellant burn rate (might be due to inhomogeneity in the propellant)
can give rise to an initial perturbation in the burn rate mode. Also, as the motor is fired
initially, the temperature distribution is uniform in the propellant. As the SRM operates,
and as the port configuration changes as time evolves, after some time, there might be
fluctuations in the temperature at the surface of the propellant. This serves as an initial
condition, where non-normality of the system plays its role and transient growth will
occur. This can cause transient growth and the amplitude increases, eventually reaching
a limit cycle (in the presence of nonlinearities). This important observation cannot be
made, if one uses the propellant response for modelling the burn rate, which neglects
the transient dynamics of the burn rate response. Higher modes in both acoustics and

57
burn rate do not contribute to the initial condition showing the modal independence of
the discretisation with increasing number of modes. The role of transient growth is
shown to play an important role in pulsed instability.

2.7.4 Pulsed instability

In SRMs, experiments indicate that rockets which are stable to small-amplitude dis-
turbances become unstable for larger ones (Blomshield et al., 1997a). They then ex-
hibit limit cycle oscillations or the rocket motor may be damaged. This phenomenon is
known as pulsed instability or triggering (Culick, 2006). From a dynamical systems
point of view, this kind of phenomenon is termed as sub-critical transition to instabil-
ity. The system is linearly stable, but nonlinearly unstable. In the previous studies,
only ad hoc models for burn rate acoustic velocity coupling were used to simulate the
experimental results (Wicker et al., 1996; Anathakrishnan et al., 2005; Flandro et al.,
2007). Wicker et al. (1996) tried different forms of nonlinear propellant response func-
tions YR to demonstrate triggering. Flandro et al. (2007) have given a comprehensive
compilation of his earlier work and new formulations to predict the nonlinear stability
of SRMs. However, his model also assumes an ad hoc propellant response function,
which is not derived from the physics of the problem. Moreover, the coefficients in the
forms of the ad hoc function are obtained from experiments (i.e. like matching limit
cycle waveforms) and there is no rigorous theoretical reasoning behind them.

This chapter solves both acoustic and propellant response equations simultaneously,
without any ad hoc assumptions on burn rate dependence on the acoustic field being
made in the formulation. The present numerical simulations show pulsed instability
in some parameter range. Pulsed instability can possibly occur in two ways. The
first is when the initial disturbance amplitude is large enough for the nonlinear terms
to be dominant compared with the linear terms right from the start of the evolution.
The linearised equations (2.25), (2.26) and (2.27) are solved numerically and Fig. 2.11
shows decaying acoustic pressure oscillations. This means that the system is linearly
stable. Linearised equations scale with initial conditions and the dynamical evolution
will look similar for all scaled amplitudes. Now, for the same parameters and initial

58
Figure 2.11: Evolution of acoustic pressure at x = 0.25 from the linear simulation.
U2 (0) = 3, P2 (0) = 3, Pm6=2 (0) = Um6=2 (0) = 0, M Tp ( = 1, 0) =
0.03, M Tp ( 6= 1, 0) = 0, C1 = 0.02, C2 = 0.02.

Figure 2.12: Evolution of acoustic pressure at x = 0.25 from the nonlinear simulation.
U2 (0) = 3, P2 (0) = 3, Pm6=2 (0) = Um6=2 (0) = 0, M Tp ( = 1, 0) =
0.03, M Tp ( 6= 1, 0) = 0, C1 = 0.02, C2 = 0.02.

59
Figure 2.13: The phase portrait of the acoustic pressure and unsteady burn rate at x
= 0.25 from the linear simulation. U2 (0) = 3, P2 (0) = 3, Pm6=2 (0) =
Um6=2 (0) = 0, M Tp ( = 1, 0) = 0.03, M Tp ( 6= 1, 0) = 0, C1 =
0.02, C2 = 0.02.

condition, the nonlinear terms are included and equations (2.14), (2.15) and (2.19) are
solved. Figure 2.12 shows that the acoustic pressure initially decays, and after sometime
it starts growing with the amplitude eventually reaching a limit cycle. The initial high
amplitude disturbance leads to the modal energy transfer from one mode to another
by direct interaction as explained in Section 2.2. The energy transfer sustains the
oscillations by keeping the disturbance energy among the modes, while in the linearised
case, the energy can only get transferred to the base flow leading to the eventual decay
of the acoustic oscillations. The same picture is shown in the phase space plot in Fig.
2.13. The trajectory from the linear evolution ends in a stable focus showing classical
linear stability. On the other hand, in Fig. 2.14, which is obtained from nonlinear
simulation, the trajectory ends in a limit cycle eventually. A plot of acoustic energy
(Fig. 2.15) shows that the linear and nonlinear simulations initially show a similar
qualitative behaviour and after some time they both diverge, leading to a qualitatively
different dynamical behaviour.

The second route is by non-normal transient growth. Here even if one starts with a
finite small amplitude suitable initial condition, transient growth due to the non-normal
nature of the system makes the oscillation grow even for a system stable according to
linear stability theory. The transient growth leads to large amplitude oscillations, which
cause the nonlinear terms to play dominant roles and direct interaction of eigenmodes

60
Figure 2.14: The phase portrait of the acoustic pressure and unsteady burn rate at
x = 0.25 (near the limit cycle) from the nonlinear simulation U2 (0) =
3, P2 (0) = 3, Pm6=2 (0) = 0, Um6=2 (0) = 0, M Tp ( = 1, 0) = 0.03,
M Tp ( 6= 1, 0) = 0, C1 = 0.02, C2 = 0.02.

 P 
Figure 2.15: The evolution of acoustic energy Eac (t) = N
m=1 (Um
2
+ P 2
m ) from lin-
ear and nonlinear simulations. U2 (0) = 3, P2 (0) = 3, Pm6=2 (0) =
Um6=2 (0) = 0, M Tp ( = 1, 0) = 0.03, M Tp ( 6= 1, 0) = 0, C1 =
0.02, C2 = 0.02.

61
  PN 
2
Figure 2.16: a) The evolution of acoustic energy Eac (t) = m=1 (Um + Pm2 ) with
optimum initial condition (0) = Vopt , C1 = 0.03, C2 = 0.02. b)
Convergence
qP study for the previous figure with N. Note that, N =
i=1 ((N (ti ) N 1 (ti )) /N (ti )) 100 is the measure used for
I 2

studying the convergence of the simulations. Moreover, N represents

any one of the variables P (x = 1/4), R(x = 1/4) & E. The summation
index i represents the value of the variables at the ith time step in the nu-
merical simulation. The threshold of N is chosen as 1% for convergence
of the solution, which corresponds to N=5 in the present case.

occur. Figure 2.16 (a) shows the comparison of acoustic energy evolution with linear
and nonlinear simulations. The transient growth in the linear simulation decays eventu-
ally, while the nonlinear simulation leads to a limit cycle. The higher transient growth in
the linear simulation than that in the nonlinear one is due to the damping effect from the
nonlinear terms. Here, the initial condition is chosen to be the optimum initial condition
for the maximum transient growth ( (0) = Vopt ) to show the importance of this route
to triggering. Also note that the initial acoustic energy (Eac (t = 0) = 6.4 104 ) is
very small compared with that in Fig. 2.15 (Eac (t = 0) = 8.43).

A convergence study is performed to evaluate the number of Galerkin modes (N )

used for the above simulations. Figure 2.16 (b) shows the plot between percentage
change (N ) in the results of P (x = 1/4) , R (x = 1/4) and E for various values
of N. It is found that for a threshold of 1% change in the solution variables, N=5 is
sufficient. The non-dimensionalised dominant Fourier frequency of the acoustic pres-
sure during limit cycle corresponding to Fig. 2.16 (a) is 0.48, which is very close to

62
the fundamental frequency for the corresponding to a pipe, closed at both the ends.
The non-dimensional frequency in Fig. 2.16 (a) corresponds to 66.25 Hz for the SRM
configuration given in table 2.1.

Transient growth is quantified in a particular measure (Section 2.4). In this chapter,

generalized disturbance energy ET (t) (defined to include the contribution of energy in
the disturbance from the degrees of freedom associated with the unsteady burn rate, see
Section 2.6) is used as a measure to quantify non-normality of the system. Transient
growth from the present simulation can be observed and compared with the experi-
ments, when ET (t) is measured. It is very difficult to devise an experiment to measure
ET (t) in an SRM. The existing experiments with SRM (Blomshield et al., 1997a,b;
Harris and Champlain, 1998) measured only the acoustic pressure at the head end of
the motor. Therefore, it is difficult to have one-to-one comparison of transient growth
of thermoacoustic oscillations in SRM from the available experimental acoustic pres-
sure data with the present simulation.

2.7.5 Bootstrapping

Bootstrapping is a phenomenon, where the dominant frequency of a system changes

during the dynamical evolution of the system. This phenomenon is observed in SRMs
(Yoon et al., 2001). Yoon et al. (2001) attributed this phenomenon to the nonlinear-
ity alone. This is due to the transfer of energy among the modes by either nonlinear
coupling or non-orthogonality of eigenmodes. The phenomenon of bootstrapping is
discussed in the context of turbulence (Gebhardt and Grossmann, 1994) and thermoa-
coustic system (Yoon et al., 2001; Balasubramanian and Sujith, 2008a,b). An ad hoc
acoustic velocity combustion coupling model is used to simulate the observed be-
haviour in Rijke tube (Yoon et al., 2001; Balasubramanian and Sujith, 2008b), whereas
in Balasubramanian and Sujith (2008a), it comes by actually solving the unsteady equa-
tions for a ducted diffusion flame.

The initial condition (t = 0) is chosen as given in Fig. 2.17. The system, which is
stable according to the classical linear stability theory, is excited in the second Galerkin
mode and the evolution of the other Galerkin modes is tracked. Figure 2.17 (a-e) shows

63
Figure 2.17: The evolution of Galerkin pressure modes a) first mode, b) second mode,
c) third mode, d) fourth mode, e) fifth mode, f ) acoustic pressure at x =
0.25, U2 = 3.0, P2 = 3.0, Pm6=2 (0) = Um6=2 (0) = 0, M Tp (, 0) =
0, C1 = 0.02, C2 = 0.02.

Figure 2.18: The Fourier transform of acoustic pressure at x = 0.25 during different time
intervals, U2 = 3.0, P2 = 3.0, Pm6=2 (0) = Um6=2 (0) = 0, M Tp (, 0) =
0, C1 = 0.02, C2 = 0.02.

64
the evolution of the individual Galerkin pressure modes. Initially, the projection on the
second mode decays and transfers the energy to the first mode (Fig. 2.17 a & b). After
some time, the first mode grows to sufficient extent, transferring the energy back to the
second mode causing it also to grow.

The time evolution of acoustic pressure at x = 0.25 plotted in Fig. 2.17 (f ) shows
that the amplitude of the pressure oscillation decreases initially and then, after some
time, it increases by the modal energy transfer. The plot of the fast Fourier transform
(FFT) of the acoustic pressure (Fig. 2.18) illustrates this phenomenon. For 0<t<15, the
second and fourth eigenmodes are the dominant ones, which decay as time evolves. In
15<t<45, the first eigenmode grows because of the energy transfer from the second and
fourth modes. Then, the first eigenmode transfers energy back to the second and fourth
eigenmodes (45<t<62.5) causing them to grow again. In the end (62.5<t<120), there
are higher harmonics due to the energy transfer to higher eigenmodes. The dominant
frequencies present in the system in the limit cycle are close to the natural acoustic
frequency of a closed-closed duct. A crucial difference is that the past analyses have
demonstrated bootstrapping in acoustic velocity combustion coupling systems (Yoon
et al., 2001; Balasubramanian and Sujith, 2008a,b), whereas the present analysis used
the acoustic pressure combustion coupling model.

2.8 Interim summary

A thermoacoustic stability analysis of a solid rocket motor is performed with emphasis

on the following. First, the non-orthogonality of the eigenmodes is accounted for in-
corporating the mean flow (convection) effects in the acoustic equations. The classical
linear stability theory predicts stability, which is valid only in the asymptotic time limit.
For non-normal systems, the short term behaviour can be completely different from the
prediction by the classical linear stability theory for some initial conditions. The tran-
sient dynamics of the unsteady propellant burn rate are included instead of the ad hoc
response models used in the earlier analysis. The acoustic and burn rate equations are
solved simultaneously.

65
 In the present case, burn rate equations are solved for a homogeneous propellant.
The inclusion of a differential equation in time for the unsteady burn rate leads to an in-
crease in the degrees of freedom of the system. Therefore, the growth or decay of oscil-
lations is quantified by a generalized disturbance energy, which includes the acoustic
energy and the unsteady energy in the propellant. The same energy is related to the 2-
norm of the state space vector that shows a transient growth because of the non-normal
nature of the system. The optimum initial condition for maximum transient growth in-
dicates large contribution from the unsteady burn rate modes. The use of a burn rate
response function would have not captured this observation, because it neglects tran-
sients in burn rate modes. Using this model, exponential and pulsed instabilities are
simulated.

In the past, pulsed instability is simulated only by using ad hoc response functions,
which are not physics based. Moreover, in the present case, pulsed instabilities are
simulated with burn rate pressure coupling as against the burn rate velocity coupling
used in the earlier analysis. Pulsed instabilities can occur in two ways. First, is by
introducing a large pulse into the system where nonlinearities are important, leading to
a limit cycle. Second, is through a small initial condition in the appropriate direction
that causes transient growth. As the amplitude of the oscillation increases, nonlinear
terms can then contribute, leading to limit cycle. Finally, other observed phenomena
such as dominant frequency switching during the dynamical evolution of the system
and bootstrapping are also demonstrated. In summary it is essential to include all the
dynamics in the propellant response and the non-orthogonality of eigenmodes to predict
instabilities more accurately in solid rocket motors. The system is no longer purely
acoustic, but an extended one that degrees of freedom in the combustion dynamics.

2.9 Outlook

There are three issues that need to be addressed in the above analysis. They are listed
as follows. The first and most important is performing experiments in SRM in order to
verify the above conclusions regarding transient growth. A time resolved measurement
for the state space variables () is difficult. Further, the coupling between the acous-

66
tic field (Eqn. 2.3) and the heat source (burn rate equation, Eqn. 2.16) is performed
phenomenologically. This forms the second issue. The third is that the generalised
disturbance energy (ET ) derived in Section 2.6 is also phenomenological, as the cou-
pling is not obtained with mathematical rigor. The amount of transient growth obtained
depends on the weight factors associated with the acoustic and burn rate variables in
the disturbance energy. In order to resolve the above mentioned issues, analysis is per-
formed in a simpler thermoacoustic system: Rijke tube. The next chapter deals with
the description of a theoretical framework to analyse thermoacoustic interaction in the
Rijke tube system. The equations governing the acoustic field and the heat source are
obtained from the conservation equations of fluid flow with mathematical rigor.

67
 CHAPTER 3

Modelling thermoacoustic interactions in a Rijke tube

3.1 Introduction

Rijke tube is a simple thermoacoustic device, but has much of the essential physics of
thermoacoustic interaction. Rijke tube is an acoustic resonator tube, with a heat source
(in the present case, an electrical heater). The heat source is positioned at some axial
location as shown by the configuration of a horizontal Rijke tube in Fig. 3.1.

Thermoacoustic instability of the Rijke tube has been studied for a long time. Rijke
tube oscillations were first observed by Rijke (1859). He used a vertical tube with
a coiled electrical heating filament as the heat source. Self-sustained thermoacoustic
oscillations were observed when the heater was positioned at some axial location of the
tube and beyond some threshold power level. Rijke gave an explanation based on the
pressure pulse generated due to volumetric expansion of the fluid near the heater zone.
However, this argument did not explain the fact that instabilities were observed only for
some selected range of heater locations.

Earlier theoretical investigations in Rijke tube was started by Neuringer and Hudson
(1952). They performed their analysis by identifying the zeroth and first order fluctu-
ations as hydrodynamic and acoustic fluctuations. Along with the fact that more heat
transfer is caused due to higher velocity, they were able to identify the possibility of ex-
citing the fundamental mode only when the heater is located in the first half (in the flow
direction) of the duct. Furthermore, the growth of sound was found to be maximum at
a heater location of quarter duct length.

Later, modelling thermoacoustic interactions in Rijke tube using the concept of

phase lag between flow rate and heat transfer was performed by Putnam and Dennis
(1954). In this investigation, experiments dealing with acoustic oscillations in a cham-
ber with flame burning as the heat source were performed. They were able to predict
Figure 3.1: Configuration of the Rijke tube showing the acoustic and hydrodynamic
zones

the regimes of instability using a phase lag model for the heat source, with various pa-
rameters, which compares with their experiments. This confirmed the usage of phase
lag as a convenient representation of the coupling between the velocity fluctuations and
heat source.

Carvalho et al. (1989) employed a linear model for the unsteady heat release rate to
calculate the stability of the modes. They applied Rayleigh criteria (Rayleigh, 1878) and
predicted the axial locations of the heater for which the thermoacoustic oscillations are
unstable. A transfer function for the response of the heat source to velocity fluctuations
in the limit of small Mach number is obtained by Nicoli and Pelce (1989). This is
further used to determine the linear stability of the system.

After the advent of computers and codes, numerical techniques gained importance
in modelling thermoacoustic systems. Raun and Beckstead (1993) numerically deter-
mined the eigenvalues of the Rijke tube system. Acoustic equations are used along with
the one dimensional equations governing the dynamics of the heat source. Boundary
conditions to the acoustic field on the upstream and downstream side of the heat source
are therefore emphasized from hydrodynamic equations. Frequencies and growth rates
were observed to agree well with that from experiments. Further, they concluded that
the model developed by Mclntosh (1987) for response of various configurations/types
of heat source also nearly match with the numerical simulations.

In order to identify and explain the coupling between acoustic fluctuations and heat

70
release rate, a computational fluid dynamics (CFD) simulation was performed in a Rijke
tube like pulse combustor by Entezam et al. (1997). They identified that the strength of
the coupling is proportional to the product of the acoustic pressure and velocity. The
same group continued further and determined a similarity parameter obtained by the
combination of acoustic velocity, pressure and square of a characteristic length (Maj-
dalani et al., 2001). Using the inputs from the above numerical, experimental and scal-
ing considerations, they were able to develop an analytical model for the heat source.
The results, using the model are in agreement with the Rayleigh criteria for the presence
of maximum driving at one-quarter length of the tube.

An excellent review of Rijke tube, along with the associated devices was given by
Raun et al. (1993). Their review indicated some kind of commonality was associated
with the thermoacoustic interactions in these devices. For example, the specific forms of
the response depended on the configuration/type of the heat source. While the governing
equations for the acoustics and the coupling mechanisms remained the same.

The use of transfer function for heat source in the prediction of thermoacoustic in-
stability was first applied by McIntosh and Rylands (1996). They used velocity transfer
models to demonstrate the applicability of this technique. Further, in the same paper,
transfer functions for more realistic burner configurations were determined and sophis-
ticated acoustic models were used. In the end, a fair comparison on the prediction of
the stability regimes, frequency of oscillations in the parameter space with that of the
experimental results have been highlighted.

In all the above papers, models for the heat source are valid only for small acoustic
velocity fluctuations. The stability thus predicted is the linear stability of the system.
For a linearly unstable system, the oscillations grow exponentially as predicted by the
linear stability theory for small amplitudes of oscillations and eventually reach a limit
cycle due to nonlinearities in the heat release rate response of the heater. Nonlinear-
ities in the acoustic field do not contribute to the dynamical evolution of the system
as the Mach number (M 103 ) of the steady state flow is very low (Culick, 2006).
Linear stability theory was further applied to analyse thermoacoustic instability in con-
figurations with multiple heat sources and complex geometries (Bittanti et al., 2002).
Recently, Heckl and Howe (2007) used Greens function technique to determine the

71
occurrence of thermoacoustic oscillations in a ducted premixed flame.

Estimation of the amplitude of acoustic oscillations during limit cycle is important

from the design point of view for gas turbines (Zinn and Lieuwen, 2006). In order to
achieve this, the nonlinearity in the heat release rate response of the heater has to be
accounted for in the model. The nonlinear response of the heater can be determined
by solving the governing equations for the fluid flow over the electrical heater, using
CFD. On the other hand for low dimensional modeling, the nonlinear response of the
heat source can be obtained from a correlation of the heat transfer between the heater
and the local flow velocity (Heckl, 1985, 1990). The response of the heat source thus
obtained can be considered as a source of acoustic energy, thus coupling the chamber
acoustic field with the heat source (Poinsot and Veynante, 2005).

A CFD based analysis was also used to study Rijke tube oscillations. The earliest
one was performed by Kwon and Lee (1985). They determined the stability curve
for various mean flow rates. Hantschk and Vortmeyer (1999) numerically investigated
thermoacoustic oscillations by considering the heat source to be a heated flat plate. They
obtained the amplitude and frequency of the thermoacoustic oscillations during limit
cycle. Kopitz and Polifke (2008) used Nyquist criterion along with CFD to determine
the stability of the system. Nyquist criterion technique has the advantage of being
applicable to complex geometries and non-compact heat sources. Recently, Moeck
et al. (2009) have numerically investigated thermoacoustic instabilities with the heating
source being a flat flame and compared the numerical results with those of experiments.

The technique of numerical continuation, a tool from nonlinear dynamics was used
to determine the bifurcation diagram of a Rijke tube system by Subramanian et al.
(2010). They used the model from Balasubramanian and Sujith (2008b) and determined
the linear and nonlinear stability regimes. From the simulations, a period doubling route
to chaos (Strogatz, 2000) was observed in the oscillations. Further, it was shown that
the system becomes unstable only via subcritical Hopf bifurcation due to the presence
of square root nonlinearity in the response of the heat source. This effort is continued
to determine analytical estimates on the subcritical transition regime, nature of bifurca-
tions using the method of multiple scales (Subramanian et al., 2011).

72
 Recently, thermoacoustic systems are shown to be non-normal, which leads to the
non-orthogonality of the eigenmodes. This non-orthogonality was first demonstrated by
(Balasubramanian and Sujith, 2008b) in an electrically heated Rijke tube system. The
response of the heat source to velocity fluctuations were modelled from the correlation
given by Heckl (1990). A detailed discussion on the effects of non-normality in the
stability of SRM is performed in the previous chapter.

These extensive analytical and numerical analysis for thermoacoustic instability of

a Rijke tube are supplemented by experiments. Matveev and Culick (2003b) and Song
et al. (2006) performed experiments in a horizontal Rijke tube with a mesh-type electri-
cal heating element. The acoustic pressure oscillations were monitored and limit cycle
amplitudes were obtained. For low values of the heater power, the system was stable.
As the heater power was increased, the system became linearly unstable and eventually
reached a limit cycle. Chaotic oscillations were observed in the experiments with mul-
tiple flame (premixed) burner as the heat source (Kabiraj et al., 2011). They performed
nonlinear time series analysis determined the presence of Ruelle-Takens route to chaos
from a limit cycle.

All the investigations in the analysis of thermoacoustic instability in a Rijke tube de-
scribed above except Moeck et al. (2009) have used either a response function or solved
the Navier-Stokes and energy equations for fluid flow by CFD to obtain the dynamics
of the heat source. They assumed a compact heat source (the size of the heat source
along the length of the tube is small compared to the acoustic wavelength) and coupled
the unsteady heat release rate to the acoustic energy equation as a source. In the above
method, the coupling between the chamber acoustic field and the unsteady heat release
rate was not obtained with mathematical rigor. Furthermore, there were two systems of
equations involved in the problem; one for the acoustic length scale and the other for the
length scale of the heat source. They were written and used without mathematical jus-
tification. In a rigorous analysis, the above two systems of equations have to be derived
from the conservation equations of fluid flow by performing asymptotic analysis.

Asymptotic analysis is a perturbation technique used ubiquitously in a wide branch

of fluid mechanics. In this analysis, the solution is expanded in powers of a small
quantity over its base (steady state) value. The small quantity is of a parameter, that is

73
relevant to the problem. An excellent presentation of the theory and application of per-
turbation technique in fluid flow problems is given in Dyke (1975); Kevorkian and Cole
(1981). In particular, multiple length and time scale problems are dealt in Kevorkian and
Cole (1996); Nayfeh (1998). In this chapter, matched asymptotic expansion technique
is used and Lagree (2011) serves as an excellent reference. The mathematical details
of this technique are given in Hoppensteadt (1975). Perturbation method was first em-
ployed to develop boundary layer theory by Prandtl (1928). Soon, this technique was
applied to a variety of fluid flow problems; boundary layers (Bush and Fendell, 1972;
Mellor, 1972), free shear flows (Pier, 2002; Stephane and Fabre, 2007) etc.

Wu et al. (2003) have pioneered the application of asymptotic expansions to analyse

combustion instabilities. They analysed the amplification of sound waves, when a flame
propagates in a gravity field. Separate systems of equations for acoustic field and flame
zones were derived, by performing the asymptotic analysis on the conservation equa-
tions. They have performed linear stability analysis for the acoustic-flame coupling,
followed by a weakly nonlinear theory for the Darrieus-Landau mode of instability of
the flame and the acoustic field. The paper explained the experimental observation of
the transition from curved to flat flame during instability. Moeck et al. (2007) have
also performed the asymptotic analysis to investigate thermoacoustic instability in a
Rijke tube with flame as the heat source. They obtained with mathematical rigor the
correct systems of equations and the coupling between the acoustic field and the heat
source. The presence of an additional global-acceleration term in the momentum equa-
tion of the hydrodynamic zone was also observed. Furthermore, they concluded that
since one dimensional (1D) equations are used in the hydrodynamic zone for their anal-
ysis, the above term vanishes, leaving the conventional momentum equation intact in
the hydrodynamic zone. The unsteady heat release rate from the flame may be due
to the equivalence ratio fluctuations at the inlet of the hydrodynamic zone, which in
turn may be caused by the acoustic velocity at the location of the flame (Moeck et al.,
2007). Recently, Wu and Moin (2010) investigated the generation of acoustic waves
from premixed flame due to freestream enthalpy fluctuations, using asymptotic analy-
sis. A vigorous subharmonic parametric instability was observed at moderate levels of
enthalpy fluctuations.

74
 In this thesis, an investigation of the thermoacoustic instability in an electrically
heated Rijke tube is performed starting from the governing equations of fluid flow. An
asymptotic analysis (Zeytounian, 2002; Ting et al., 2007) is then performed in the limit
of a compact heat source and zero Mach number of the steady flow to obtain two sys-
tems of equations: one governing the acoustic field and the other governing the unsteady
flow and heat transfer near the heat source. The separation of equations for the acoustic
field and heater (hydrodynamic zone, see Fig. 3.1) occurs. The coupling between the
above two systems of equations is obtained. Also, the additional global-acceleration
term, as obtained by Moeck et al. (2007, 2009), appears in the momentum equation for
the hydrodynamic zone. It is also found in the present investigation that the presence
of global-acceleration term has serious consequences for the bifurcation diagram. The
nonlinear evolution equations obtained from the asymptotic analysis in both acoustic
and hydrodynamic zones are solved simultaneously. The limit cycle amplitudes are
obtained, which are required to estimate the tolerance limit of the realistic combustors
(Zinn and Lieuwen, 2006) during instability.

3.2 Governing equations

The Rijke tube configuration considered here has a length la with the heater positioned
at an axial location xf (Fig. 3.1). A mean flow is maintained in the tube at a desired
flow rate using a blower. The electrical resistance heater, which acts as a heat source is
made up of a thin wire of radius lc strung around the heater frame. The effective length
of the wire filament, which participates in the heat transfer to the fluid flow, is lw . The
typical length of the duct (la ) is around 1m and the dimension of the heater along the
axial direction of the tube (thickness, lc ) is around 1mm. The thickness of the heater is
very small compared to the length scale of the acoustic field. Hence, the heater can be
assumed to be compact compared to the acoustic field in the tube. The zone around the
heat source is termed as the hydrodynamic zone. Acoustic and hydrodynamic zones are
schematically shown in Fig. 3.1.

The length of the hydrodynamic zone in the axial direction is of the order of the
thickness of the heater. Hence the hydrodynamic zone can also be assumed to be com-

75
pact compared to the acoustic field. The heater heats the flow and creates a temperature
rise across the heater. Since the heater is compact, piecewise constant steady flow prop-
erties can be assumed on either side of the heater (Kaufmann et al., 2002). The Mach
number of the flow is O(103 ), which leads to a negligible steady state pressure loss.
Hence the steady state pressure is assumed to be constant along the duct. All upstream
steady state variables are known and specifying any one downstream steady state vari-
able, such as density (obtained by solving the steady state version of the equations
governing the hydrodynamic zone, which are described in Section 3.4.2), is enough to
compute the other steady state variables from the following ideal gas and steady state
continuity equations:
u0 T0u d u0 uu0
d0 = , u0 = d (3.1)
T0d 0
where superscripts u and d represent upstream and downstream variables, subscript
0 represents steady variables and indicates dimensional variables. The governing
equations are

+ (
u) = 0 (3.2a)
t
   
1  
+ u u +
p = +
2
u (3.2b)
t 3
 
1 ( 1) 2
+ u p + p u =
k T (3.2c)
t
Non-dimensionalising the above equations using the following scales = / , p =
. 
p/ p, u = u/u, T = T T, x = x/la , ta = t (la /c0 ), where p = pu0 = pd0 , =
p
), u = uu0 , c0 = <T, < is the specific gas constant and c0 is the
u0 , T = p/(<
local speed of sound, which leads to

+ M a (u) = 0 (3.3a)
ta
   
1 M 1
+ M u a u + a p = a + a (a ) u
2
(3.3b)
ta M Rea 3
 
1 M 2
+ M u a p + M pa u = T (3.3c)
ta P ea a
where Rea = ula /, P ea = ula Cp /k, M = u/c0 and subscript a indicates that
non-dimensionalisation is performed with the acoustic length scale a .

76
 An analysis of thermoacoustic instability in a Rijke tube involves the study of a
coupled system which consists of the acoustic field in the tube and the unsteady heat
transfer from the heat source (hydrodynamic zone). Therefore, it is important to track
variations on the length scale of the tube (acoustic scale, la 1m) and the length
scale of the radius of the heater wire filament (lc 1mm) in the hydrodynamic zone.
Furthermore, the acoustic time scale tac = la /c0 and the wire heat transfer time scale
tcc = lc /
u are of the same order for typical values mentioned above. This leads to an
effective coupling of the dynamics of the acoustic field and the unsteady heat release
rate.

The length and time scale ratios are defined as = lc /la , = tac /tcc = M /
1. The system has two length scales separated by a large factor (1/ ) and a
single time scale. Since the flow is at very low Mach number (M 0), the system of
equations (Eqn. 3.3) becomes ill-conditioned (Anderson, 2001). Moreover, a smaller
grid size near the heater will restrict the maximum timestep that can be allowed for
the numerical scheme. All these make Eqn. (3.3) stiff. As a consequence, solving the
problem using CFD is a difficult task. An alternative technique available to solve such a
two length-scale problem is asymptotic analysis (Zeytounian, 2002; Ting et al., 2007),
which is used in the present chapter.

3.3 Asymptotic analysis

In the present investigation, asymptotic analysis is performed in the limit: M 0,

1, 0. The flow variables are expanded in powers of Mach number (M ). The
following ansatz for the flow variables is used:

= s + c + M a , u = u s + ua + u c ,
(3.4)
p = 1 + M pa + M 2 pc , T = Ts + Tc + M Ta

where subscript s stands for steady state variables, a for fluctuations due to the
acoustic field and c for fluctuations in the hydrodynamic zone. Here, the acoustic
fluctuations exist all along the tube, while the fluctuations due to the heater are con-

77
fined to a zone around the heater (hydrodynamic zone), which is small compared to the
acoustic length scale. Hence the variables with subscript a exist over the length of
the tube (acoustic zone, see Fig. 3.1), while the variables with subscript c exists only
in the region around the heater (hydrodynamic zone, see Fig. 3.1) and vanishes as one
moves away from it. It is important to note that the form of the power series used in
ansatz (3.4) are different for various flow variables and the reason is as following.

The acoustic fluctuations in u are zeroth order in M , whereas the fluctuations in

, p, T are first order in M . This is in agreement with the experimental observations
and theoretical formulations (Rienstra and Hirschberg, 2004). On the other hand, tem-
perature fluctuations near the heater (hydrodynamic zone) are comparable to the steady
state temperature Ts and hence Tc appears as a zeroth order fluctuation (Fu and Tong,
2002). The mode of heat transfer from the heater to the gas is by convection. Hence
the zeroth order fluctuation Tc in temperature (T ) is caused by a zeroth order fluctu-
ation uc in velocity (u). Because of zeroth order temperature fluctuations (Tc ) and a
constant leading order pressure (Eqn. 3.4), a zeroth order fluctuation of density (c ) in
the hydrodynamic zone is present . Due to the occurence of above flow fluctuations in
a prescribed form, pressure fluctuations in the hydrodynamic zone should be of second
order in M to obtain final consistent equations. It should be emphasized that choosing a
wrong ansatz will eventually lead to inconsistent equations or wrong physical interpre-
tations. Furthermore, the fluid properties are assumed to be independent of temperature.
Now the ansatz (Eqn. 3.4) is substituted in Eqn. (3.3) to get the following:

(s + c + M a ) + M a ((s + c + M a ) (us + ua + uc )) = 0
ta
  (3.5a)

(s + c + M a ) + M (us + ua + uc ) a (us + ua + uc )
ta
 
1
M 1
+ a 1 + M p a + M pc =
2
a + a (a ) (us + ua + uc )
2
M Rea 3
(3.5b)

The authors are grateful to an anonymous reviewer of their article published in Journal of Fluid
Mechanics, who pointed out the inadequacy of the constant density assumption adopted in an earlier
version of the analysis. The asymptotic analysis is reformulated with the variable density formulation as
per the suggestions of the reviewer.

78
  
1

+ M (us + ua + uc ) a 1 + M pa + M 2 pc
ta

M 2
+M 1 + M pa + M 2 pc a (us + ua + uc ) = (Ts + Tc + M Ta )
P ea a
(3.5c)

Initially equations which are of zeroth order in M are obtained. Then, first order
equations in M are obtained using the solutions from the zeroth order equations. This
process is repeated until governing equations for all the variables in the ansatz (Eqn. 3.4)
are obtained. Furthermore, the system of equations for various orders of M is written
both for the acoustic and hydrodynamic zones. In the following analysis, equations
governing the acoustic zone are first derived and the same exercise is repeated for the
hydrodynamic zone.

3.3.1 Continuity equation: acoustic zone O(M )

The zeroth order continuity equation in M reduces to the steady state equation, which is
already used in the analysis (Eqn. 3.1). Collecting terms of first order in M , Eqn. (3.6)
is obtained. Since the non-dimensionalisation is performed with respect to the acoustic
length scale for Eqn. (3.2), Eqn. (3.6) represents continuity equation in the acoustic
zone.
a
+ a ((s + c ) (us + ua + uc )) = 0 (3.6)
ta
Furthermore, using the continuity equation for steady state in the acoustic zone, a
(s us ) = 0, c = 0 and uc = 0 as described in the ansatz, Eqn. (3.6) becomes

a
+ a (s ua ) = 0 (3.7)
ta

3.3.2 Momentum equation: acoustic zone O(M )

In Eqn. (3.5b), as M is in the denominator of the pressure term, the entire equation is
multiplied by M to obtain,

79
  



M (s + c ) + M a2
+ M (us + ua + uc ) a (us + ua + uc )
ta
  (3.8)
1
M2 1
+ a 1 + M pa + M p c =
2
a + a (a ) (us + ua + uc )
2
Rea 3

The zeroth order equation in M gives a zero spatial gradient for the steady state pressure
in the system, i.e. a (1) = 0, where 1 appears due to non-dimensionalisation of p
with p. The condition for constant p along the duct is already included in the analysis
(Eqn. 3.1). Gathering O(M ) terms from Eqn. (3.8) in the limit Rea , leads to the
momentum equation in the acoustic zone.

ua 1
s + a pa = 0 (3.9)
ta

3.3.3 Energy equation: acoustic zone O(M )

The zeroth order terms from Eqn. (3.5c) represent the steady state energy equation.
Since the upstream (T0u ) steady state temperature is known, steady state equations gov-
erning the heat transfer from the heater are used to obtain the corresponding downstream
value (T0d ) thus making the collection of O(1) terms from Eqn. (3.5c) redundant. Now
the O(M ) terms are gathered to obtain the acoustic energy equation:

1 pa 1
+ a ua = 2 T c (3.10)
ta P ec c

where P ec = ulc Cp k. The unsteady heat release rate from the heater is delivered
into the acoustic zone by local thermal conduction. Hence, the last term in Eqn. (3.10)
represents the coupling term from the hydrodynamic to the acoustic zone. Assuming
one dimensional acoustic field in the axial direction xa , Eqn. (3.10) is integrated over
the cross-sectional area (Sc ) of the tube. In order to convert from 3D to 1D space, the
terms with gradients (a ) in the acoustic length scale are replaced by /xa in Eqn.
(3.10). The acoustic energy equation takes the following form.

80

  ZK
ZZ
1 pa ua 1 2c Tc dVc (
_
Sc + = x xf ) (3.11)
ta xa P ec
Vc

_
where Vc is the volume of the hydrodynamic zone and (x) is the Dirac-delta func-
tion, which is used to indicate the compactness of the hydrodynamic zone in the equa-
tions. The volume integral is converted into surface integral using Gausss divergence
theorem. In order to apply the above theorem, the last term in Eqn. (3.11) is dimen-
sionalised, followed by the application of Gauss divergence theorem and again non-
dimensionalised back to get the following acoustic energy equation,
2
Z
1 pa ua lw lc (c Tc ) d (x xf )
_
+ = er (3.12)
ta xa P ec Sc
0

where lw is the length of the wire which contributes to the heat transferred to the fluid,
er is the unit vector along the radial direction from the cylinder surface, (c Tc )er rep-
resents the component of gradient of Tc along er . The term on the right hand side of
Eqn. (3.12) is identified as the coupling term from hydrodynamic zone to the acoustic
zone, which drives the acoustic oscillations in a Rijke tube. The unsteady heat release
R 
2
rate from the hydrodynamic zone is given by q = (lw lc /P ec Sc ) 0 (c Tc )er d
with the integral evaluated over surface of the cylinder. The equations governing the
acoustic zone are thus obtained and the coupling of the acoustic field with the unsteady
heat release rate of the heater appears from the asymptotic analysis. The equations for
the hydrodynamic zone are derived in the following subsections.

3.3.4 Continuity equation: hydrodynamic zone O(1)

In order to obtain the equations with respect to the hydrodynamic zone, the spatial
derivatives in the acoustic length scale (la ) have to be converted to the length scale (lc )
of the heater, only for variables with subscript c. The transformation a = c / is
applied to Eqn. (3.5). The subscript c in the operator represents the derivatives that
are non-dimensionalised with lc . During the scale change from la to lc , terms which are
second order in M , for e.g. M 2 a , become first order in M , M c . The inclusion of

81
such terms leads to the continuity equation in the hydrodynamic zone as follows.

(s + c ) + c (s + c ) us + uc + ua |xf = 0 (3.13)
tc

where, ua |xf represents ua at the non-dimensionalised heater location xf (xf = xf /la )

and /tc = (1/)/ta . In equation (3.13), s + c and us + uc + ua |xf appear
effectively as one variable. Hence the following change of variables p = s + c and
up = us + uc + ua |xf is applied in Eqn. (3.13) to obtain

p
+ c (p up ) = 0 (3.14)
tc

which is the conventional continuity equation for flows of variable density fluid.

3.3.5 Momentum equation: hydrodynamic zone O(1)

The zeroth order momentum equation in the hydrodynamic zone is as follows.

   

(s +c ) + ua |xf + us + uc c (ua |xf + us + uc )
tc
   
1 1 1 1 pa 
+ c p c = c + c (c ) (ua |xf + us + uc ) a
2
Rec 3 xf

Boundary condition (BC) : uc 0, as xc

(3.15)


where Rec = ulc and a (pa /)|xf represents a (pa /) evaluated at xf . The
equations are then written in terms of p and up and the acoustic momentum equation
(3.9) is used to replace the last term in Eqn. (3.15) to obtain,
    
1 1 1 ua 
p + up c up + c p c = c + c (c ) up + s
2
tc Rec 3 tc xf
(3.16)
BC : up us + ua |xf , as xc ,

where ua /tc |xf represents ua /tc evaluated at the non-dimensional heater location
xf . The important point to be noted is that, Eqn. (3.16) is the momentum equation
for unsteady variable density flow over the heater, with an additional term ua /tc |xf .

82
The above additional term is referred to as the global-acceleration term (Moeck et al.,
2009), which can be identified as a coupling term for the momentum equation in the hy-
drodynamic zone from the acoustic zone, apart from that due to the boundary condition
associated with Eqn. (3.16).

The global-acceleration term occurs in two length scale problems (Klein, 1995;
Klein et al., 2001). This term would not have been identified had the asymptotic anal-
ysis not been performed. In most thermoacoustic systems, there are at least two length
scales; the length scale of the acoustic field and the length scale of the heat source.
Therefore, the above term is expected to be present in the analysis of thermoacoustic
systems. The same term can be interpreted as the pressure gradient imposed by the
acoustic field on the hydrodynamic zone (Ting et al., 2007). An important point to
be emphasized here is the following. If one performs response function calculations
numerically (for example Preetham and Lieuwen 2008) for the unsteady heat release
rate from the heater, where no acoustic field is imposed, then ua /tc |xf will not be
anticipated and hence will not be included. This leads to solving an incorrect system of
equations and prediction of the dynamics of thermoacoustic interaction in a Rijke tube
system. Numerical simulations are performed with and without the global-acceleration
term. It is observed that the error due to neglecting the above term is large and pre-
dictions of the dynamical evolution of the system are modified to a large extent. This
observation has been emphasized in Section 3.5.2.

3.3.6 Energy equation: hydrodynamic zone O(1)

Since the temperature fluctuations due to the heater are zeroth order in M as explained
in the ansatz (Eqn. 3.4), O(1) terms are gathered from Eqn. (3.5c) to obtain

  1 2
c us + ua |xf + uc = (Ts + Tc ) (3.17)
P ec c

A change of variable, Tp = Ts + Tc is performed in Eqn. (3.17) leading to,

1 2
c up = Tp
P ec c (3.18)
BC : c Tp 0, as xc , & Tp = Tw /T, xc = cylinder surface,

83
where, Tw represents the surface temperature of the heater wire. The energy equation
(Eqn. 3.3c) simplifies to an algebraic constraint (Eqn. 3.18) on the velocity field up .
The amount of local dilatation of the fluid is governed by the Eqn. (3.18). Integration of
the same equation over the hydrodynamic zone gives the net dilatation in the volume of
the fluid as it passes through the hydrodynamic zone. The presence of the heat source
which leads to the dilatation in the volume of the fluid manifests as the acoustic velocity
jump (ua |xf ) across the heat source. This in turn drives the acoustic field in the tube.
The above argument is consistent with the derivation of the acoustic energy equation
described in Section 3.3.3, where the Dirac-delta function in Eqn. (3.12) causes the
acoustic velocity jump across xf . Thus the acoustic velocity jump is used as the match-
ing condition across the hydrodynamic zone. The above conclusion is independent of
the type of the heat source and the interpretation is valid as long as the heat source can
be considered as compact with respect to the acoustic field.

In multiple scale asymptotics, averaging the flow variables over the small scale - hy-
drodynamic zone, is performed to obtain the flow variables in the large scale - acoustic
zone (Klein et al., 2001). Accordingly, the term ua /tc |xf in Eqn. (3.16) is evalu-
ated as the average of the values at the upstream and downstream locations of the heat
source, due to acoustic velocity jump. Averaging the momentum equation of the hy-
drodynamic zone (Eqn. 3.16) leads to acoustic momentum equation (Eqn. 3.9) defined
at the location of the heat source, with the acoustic velocity obtained as the average of
ua upstream and downstream of the heat source. The same averaging procedure is per-
formed for the global acceleration term as well, which is used in equation (3.16). The
system of equations for the hydrodynamic zone (3.14, 3.16 & 3.18) are not closed. The
ideal gas equation is used to obtain the relation between p & Tp in the hydrodynamic
zone as follows.

(1 + M pa + M 2 pc ) = (s + c + M a )(Ts + Tc + M Ta ) (3.19)

Equating the zeroth order terms gives the following.

1
p = (3.20)
Tp

84
Solving Eqn. (3.18), simultaneously with Eqns. (3.14, 3.16 & 3.20) gives the tem-
perature field Tp . The unsteady heat release rate from the hydrodynamic zone is then
R 
2
determined as q = (lw lc /P ec Sc ) 0 c (Tp Ts )er d from Tp . This q serves as
the source term for Eqn. (3.12).

3.4 Solution technique

The governing equations in the acoustic and hydrodynamic zones are solved using dif-
ferent solution techniques. First, the solution technique used in the acoustic zone is
discussed.

3.4.1 Equations governing the acoustic zone : one-dimensional form

The acoustic continuity (Eqn. 3.7) and momentum (Eqn. 3.9) equations are converted
from 3D to 1D space as described earlier in Section 3.3.3. The governing equations
thus obtained for the acoustic zone are as follows.

ua 1 pa
s + =0 (3.21a)
ta xa

1 pa ua _
+ = q (x xf ) (3.21b)
ta xa
The above partial differential equations (Eqn. 3.21) are converted to ordinary differen-
tial equations (ODEs) by the Galerkin technique (Zinn and Lores, 1971; Padmanabhan,
1975). In the Galerkin technique, the unknown variables are expanded using basis
functions, which satisfy the boundary conditions. In the present chapter, the Rijke tube
considered is open at both ends. The basis functions are chosen accordingly to satisfy
the acoustic boundary conditions. The acoustic variables ua and pa are expanded in
terms of the basis function as follows:

X
N X
N
pa = Pm sin(m x), ua = Um cos(m x) (3.22)
m=1 m=1

85
where m = m, N is the number of modes chosen in the Galerkin expansion. The
dynamical evolution equation for Pm and Um is obtained by projecting Eqn. (3.21) on to
the Galerkin basis, after substituting the expansion for pa and ua from Eqn. (3.22). The
application of the Galerkin technique to the present problem is similar to that performed
by Balasubramanian and Sujith (2008b). The final ODEs are of the following form:

X
N


u
us Im,n ds Im,n
d
U m + Pn n = 0 (3.23a)
m=1

Pn n Un + n n Pn = 2q sin (n xf ) (3.23b)

where


sin ((m + n ) xf ) sin ((m n ) xf )
+ , m 6= n
u u m + n m n
Im,n = s

sin (2n xf )
+ xf , m = n
2n


sin ((m + n ) (1 xf )) sin ((m n ) (1 xf ))
+ , m 6= n
d d m + n m n
Im,n = s

sin (2n xf )
1 xf , m = n
2n
 p .
n = C1 (n /1 ) + C2 1 /n (2), n = n, n represents damping in the
acoustic zone due to viscosity and end losses. Here, C1 , C2 are the coefficients that
determine the amount of damping and whose numerical value is given by Matveev and
Culick (2003a). The system of equations (Eqn. 3.23) is solved using the fourth order
Runge-Kutta (RK4) method (Riley et al., 2006). The value of the unsteady heat release
rate q is obtained at each substep of RK4 by solving the equations corresponding to the
hydrodynamic zone (Eqn. 3.24), which is discussed in the following section.

3.4.2 Equations governing the hydrodynamic zone

The governing equations for the hydrodynamic zone are summarised below:

p
+ c (p up ) = 0 (3.24a)
tc

86
     
1 1 1 ua 
p + u p c u p + c pc = c + c (c ) up + s
2
(3.24b)
tc Rec 3 tc xf

1 2
c up = Tp (3.24c)
P ec c
1
p = (3.24d)
Tp
As described in Section 3.2, the heater in its primitive form is a thin wire filament. To
simulate the dynamics of the heater, Selimefendigil et al. (2008) have analysed the un-
steady convective heat transfer from a heated circular cylinder. Following this approach,
the above system of equations (Eqns. 3.24) is solved for the unsteady convective heat
transfer over the circular cylindrical heated wire filament. The flow field over the heater
wire filament is assumed to be two dimensional. The governing equations are solved
by CFD. The details of the geometry of the heat source and the boundary conditions
imposed on the hydrodynamic zone are discussed in Appendix C. The unsteady heat
release rate q is obtained from the temperature field Tp . The obtained q is then used as
the source term for Eqn. (3.23b) at each substep of RK4. Thus the system of equations
in acoustic (Eqn. 3.23) and hydrodynamic zone (Eqn. 3.24) is solved simultaneously.

3.5 Results and discussions

Numerical simulations are performed with the parameters shown in table 4.1. In the
following simulations, the number of the Galerkin modes is chosen as N = 100 (to
capture the acoustic velocity jump across the heat source), such that a further increase
in N leads to less than 5% variation in the evolution of the all the state space variables
(Um , Pm , u, v, p, & T ). Unless otherwise specified, the parameter values in table
4.1 are used for the simulations. The numerical values of the damping coefficients
used for the present simulations are C1 = 0.27 and C2 = 0.03 (Sterling and Zukoski,
1991; Matveev and Culick, 2003a). In the acoustic zone upstream and downstream
of the heater, one dimensional governing equations are used. The hydrodynamic zone
equations in the present case are solved in a two dimensional domain (see Appendix C).
0d ) in the hot side of the acoustic zone is obtained by averaging the
Hence the density (
steady state density s at the far downstream end of the hydrodynamic zone. The value

87
 Acoustic zone Hydrodynamic zone CFD simulations

la = 1 m, Sc = 0.01 m2 Red = 20, Number of grids (r, ) = 100 120

P = 1 bar lc = 2.6 mm Residue for continuity = 106
u0 = 1.18 kg m3 lw = 10 m Residue for momentum = 106
T0u = 295 K Tw = 700 K Residue for energy = 106
d0 = 0.84 kg m3 M = 5 104 ta = 5 104

Table 3.1: Physical parameters in acoustic zone, hydrodynamic zone and numerical pa-
rameters used for CFD simulations

of 0d thus obtained is listed in table 4.1.

3.5.1 Stability regimes

The experimental results of Matveev and Culick (2003b) indicate that the system be-
comes linearly unstable beyond some critical value of the heater power. As the heater
power is increased, the present numerical simulations show two stability regimes: lin-
early stable regime and linearly unstable regime. The non-dimensional heater power
.
(K) is defined as K = (lw lc ) (P ec Sc ) (see Section 3.3.6 for the expression of q). The
average of the acoustic velocity upstream and downstream of the heat source is taken
as the acoustic velocity (uf ) at the heat source. In the subsequent sections, the vari-
ables uf and q are considered as the representative variables to illustrate the dynamical
behaviour of the system in the acoustic and hydrodynamic zones, respectively. The
subsequent sections discuss the results of the numerical simulation.

Linearly stable regime

For low values of the heater power K, the system is stable to small amplitude initial
perturbations. In Fig. 3.2 (a) the initial perturbation in acoustic velocity (uf = ua |xf )
at the heater location is around 5% of the mean flow velocity. The perturbation decays to
zero in the asymptotic time limit. The phase plot between uf and the non-dimensional
fluctuating heat release rate (q) shows the evolution of the system in Fig. 3.2 (b) towards

88
Figure 3.2: Linearly stable system: (a) evolution of acoustic velocity (uf ) at the heater
location (xf ), (b) phase portrait between uf and unsteady heat release rate
q, xf = 0.25, K = 0.10, U1 (t = 0) = 0.05, Um6=1 (t = 0) = 0, Pm (t =
0) = 0.

a stable focus. The arrows indicate the direction of the evolution of the system in the
phase plane.

Linearly unstable regime

In the second stability regime, the system is unstable for small amplitude perturbation,
uf (t = 0) = 0.4, which is 10% of the limit cycle amplitude, see Fig. 3.3 (a). The figure
shows that uf grows exponentially, eventually reaching a limit cycle. The limit cycle is
a closed curve in the phase portrait as shown in Fig. 3.3 (b). The acoustic velocity (ua )
distribution along the duct at various instances over a period, as marked in Fig. 3.3 (c)
is shown in Fig. 3.3 (d). It is observed that the acoustic velocity jump is captured by the
Galerkin technique. The stability regimes of a system with the variation of a parameter
can be represented using a bifurcation diagram.

Bifurcation diagram

Simulations are performed with the heater power as the control parameter and the peak
 
to peak value of acoustic velocity uf |pp at the heater location (xf ) in the asymp-
totic time limit is chosen as the representative variable. The resolution in the control
parameter K for Fig. 3.4 is 2 103 , which is 1.8% of the value of K = 0.11 at the
Hopf point (B). In the bifurcation diagram shown in Fig. 3.4, solid lines indicate sta-

89
Figure 3.3: Linearly unstable system: (a) evolution of acoustic velocity (uf ) at the
heater location (xf ), (b) phase portrait between uf and unsteady heat re-
lease rate q (only limit cycle is shown, transients are not shown for clarity),
(c) evolution of uf during a period of the limit cycle, (d) distribution of
the acoustic velocity ua in the Rijke tube during a period of the limit cycle.
xf = 0.25, K = 0.1785, U1 (t = 0) = 0.5, Um6=1 (t = 0) = 0, Pm (t =
0) = 0.

Figure 3.4: Bifurcation diagram with the non-dimensional heater power K as the con-
trol parameter. The other parameters are, xf = 0.25. The two regimes are
R1 - linearly stable regime and R2 - linearly unstable.

90
Figure 3.5: (a) Comparison of the evolution of acoustic velocity (uf ) at the heater lo-
cation (xf ) with and without the global-acceleration term in Eqn. (3.15),
(b) Evolution of uf for a longer period of time with the global-acceleration
term, zoomed out view of (a). xf = 0.25, K = 0.1785, U1 (t = 0) =
0.5, Um6=1 (t = 0) = 0 and Pm (t = 0) = 0.

ble solutions, while dashed line indicates unstable solutions. For low values of K (say,
K = 0.06), the asymptotic state (uf = 0) is the stable fixed point. This corresponds to
the first stability regime (Section 3.5.1), where the system is linearly stable.

As the heater power is increased, beyond point B (K = 0.11), the system becomes
linearly unstable and reaches a limit cycle. The above transition happens via Hopf bifur-
cation. Also the amplitude of the limit cycle increases as the heater power is increased
further. This corresponds to the second stability regime.

3.5.2 Effect of global-acceleration on the stability of the system

An asymptotic analysis is performed to obtain separate systems of equations governing

the dynamics in the acoustic and hydrodynamic zones. The critical outcome of the
above analysis is the presence of the additional term ua /tc |xf in Eqn. (3.16). The
importance of the global-acceleration term is shown in Fig. 3.5.

In Fig. 3.5 (a), evolution of uf is shown with and without this term for identical
system parameters and initial conditions. The system parameters are chosen, such that

91
Figure 3.6: Bifurcation diagram with non-dimensional heater power K as the control
parameter without the global-acceleration term in Eqn. (3.15). The param-
eters chosen are the same as for the simulation shown in Fig. 3.4.

the simulation performed is in the linearly unstable regime (K = 0.1785, see Fig. 3.4).
With the initial perturbation U1 (t = 0) = 0.5, the simulation with the ua /tc |xf term
indicates that the system reaches a limit cycle eventually. The evolution of the system
to a limit cycle is shown in the inner Fig. 3.5 (b). However, for the same system
parameters and initial conditions, if the above term is dropped from Eqn. (3.16), the
simulation indicates that the system reaches a stable focus in the asymptotic time limit.

To analyse the behaviour of the system in the absence of the global-acceleration

term, the bifurcation diagram is computed without the term ua /tc |xf in Eqn. (3.16),
as shown in Fig. 3.6. A significant difference to be observed between the bifurcation
diagram with (Fig. 3.4) and without (Fig. 3.6) the inclusion of the global-acceleration
term is that the system becomes linearly unstable at K = 0.11 (see Fig. 3.4) for the
simulation performed with the global-acceleration term, whereas the same behaviour
happens at K = 2.89 (see Fig. 3.6) for the simulation performed without the same
term. The numerical simulation without the global-acceleration term predicts linear
instability at a value of K which is one order of magnitude higher than that obtained
from the numerical simulation with the inclusion of the same term. The term ua /tc |xf
acts as a forcing term for Eqn. (3.16) from the acoustic zone. Hence, dropping the
above term breaks the feedback from the acoustic to the hydrodynamic zone apart from

92
the feedback from the free stream boundary condition for Eqn. (3.16). The strength
of the feedback loop between the acoustic and hydrodynamic zones is thus weakened.
Therefore, it is important to perform the asymptotic analysis to obtain the correct system
of equations for the two zones in solving the two length scale problems.

An analysis of thermoacoustic instability using the response function (to capture

the dynamics of the heat source) is termed as a two-part approach (Candel, 2002). In
this approach, the response function of the heat source to velocity fluctuations is first
obtained. The response function thus obtained is used in the acoustic energy equation.
The acoustic equations are then solved with the appropriate boundary conditions and
the stability of the system is determined. When the response function is obtained nu-
merically, for example Preetham and Lieuwen (2008), the global-acceleration term is
not taken into account in the governing equations for the dynamics of the heat source.
If the prediction of thermoacoustic instability is performed based on the response func-
tion of the heat source, the effect of the above term is absent. As a consequence, the
coupling between the acoustic and hydrodynamic zones is weakened, which leads to
over prediction of the stability of the thermoacoustic system.

Furthermore, it is important to analyse the dynamics of the system in the asymptotic

time limit. As fluid convection is the source of energy transfer from the heated wires to
the flow, the flow field during a limit cycle will reveal important insights about them.

3.5.3 Unsteady flow field in the hydrodynamic zone

The flow field in the hydrodynamic zone during a limit cycle is investigated in this
section. Figure 3.7 shows the streamlines of the velocity field up for the flow over the
heater wire at various instants of a limit cycle. Only one half of the flow field is shown in
the figure due to the symmetry condition (see Appendix C). The acoustic velocity ua |xf
at the heater location ranges from -4 to +4 during one period of the limit cycle (Fig.
3.7f ). Hence, during the first half of the limit cycle, the flow is from left to right and
during the next half, it is from right to left. A complete flow reversal in the freestream
happens, as shown in Fig. 3.7 (ae). Labels (a-e) illustrated in Fig. 3.7 (f ) indicate the
 
flow field at various instants of the acoustic velocity ua |xf shown in Fig. 3.7 (ae).

93
Figure 3.7: Streamlines (a-e) in the hydrodynamic zone (corresponding to up ) at various
instants during one period of the limit cycle, (f ) evolution of ua at xf during
the limit cycle: K = 0.1785, xf = 0.25, lw = 10 m.

When ua |xf = 0 (Fig. 3.7a), the flow field resembles the steady flow over the
cylinder. As ua |xf increases and reaches a maximum, the recirculation zone is pushed
further downstream (Fig. 3.7b) and during subsequent time, ua |xf decreases, followed
by flow reversal in the freestream direction (Fig. 3.7e). Thus the fluctuations in the
velocity field (up ) during the limit cycle are comparable to the steady base flow (us ).
Hence nonlinear effects such as steady streaming (Telionis, 1981) will be predominant.
In the present case, since the acoustic velocity (ua ) is responsible for flow oscillation
in the hydrodynamic zone, the steady streaming thus obtained is termed as acoustic
streaming (Andres and Ingard, 1953).

Figure 3.8 (ac) shows the effects of acoustic streaming during the limit cycle. Fig.
3.8 (a) shows the averaged streamlines of up during one period of the limit cycle (Fig.
3.7f ). The streamlines of the steady state flow us are shown in Fig. 3.8 (b). There are
visible differences between Figs. 3.8 (a & b) and hence, apart from the steady base
flow, a non-zero mean flow arises during the limit cycle due to the nonlinearity. The
streamlines obtained from the difference of the above two flow fields (Fig. 3.8a & b)

94
Figure 3.8: Flow streaming in the hydrodynamic zone, (a) streamlines averaged over
one period of the limit cycle, (b) streamlines of the steady base flow, (c)
streamlines of the velocity difference between (a) and (b), (d) evolution
of the non-dimensional unsteady heat release rate (q), showing a signifi-
cant mean shift: K = 0.1785, xf = 0.25, lw = 10 m, U1 (t = 0) =
0.5, Um6=1 (t = 0) = 0 and Pm (t = 0) = 0.

are shown in Fig. 3.8 (c). The streamlines are slightly tilted towards the right due to the
steady base flow from left to right. Also, note that in Fig. 3.8 (c) the streaming velocity
is towards the cylinder along the direction of the propagation of the sound waves (X
direction). In the present case, the Strouhal number (St) and the streaming Reynolds
number (Res ) are calculated as,

St = f lc /
u = /2M = 1/2 =1
(3.25)
Res = (ua |xf )max 2lc / 80

where f = c0 /2la is the fundamental frequency of the natural duct mode. The exper-
iments performed in the above regimes of St and Res (Andres and Ingard, 1953)
indicate that the steady streaming velocity is directed towards the cylinder in the di-
rection of oscillation of the free stream flow field. The same is observed in the present
simulation, as shown in Fig. 3.8 (c). Streaming velocity field obtained from this simula-
tion cannot be directly compared with the experiments with externally excited acoustic
field due to the presence of the global-acceleration term. The same term does not van-
ish even during limit cycles and therefore, only qualitative behaviour of the streamlines

95
corresponding to the streaming velocity field can be compared with the experiments in
acoustic streaming (Andres and Ingard, 1953).

A non-zero averaged mean flow, which appears above the steady base flow due to
acoustic streaming results in non-zero averaged unsteady heat transfer from the heater.
Figure 3.8 (d) shows the evolution of the unsteady, non-dimensional heat transfer rate
(q, see Section 3.3.6) from the heater. The unsteady heat transfer rate q eventually
reaches a limit cycle for linearly unstable systems (Section 3.5.1) where the oscillations
during the limit cycle are about a non-zero mean. Acoustic streaming leads to a shift in
the mean value of q.

3.6 Interim summary

An analysis of thermoacoustic instability in an electrically heated horizontal Rijke tube

is performed. The analysis started with an examination of the conservation equations
for fluid flow. In the limit of zero Mach number of the steady flow and compact size
of the heat source compared to the acoustic length scale, the equations become stiff.
Therefore, solving the governing system of equations by the CFD technique is a dif-
ficult task. Hence, an asymptotic analysis is performed, which gave further physical
insight into the problem. The flow variables are expanded in powers of Mach number.
The equations thus obtained are identified as governing equations for the acoustic and
hydrodynamic zones. An additional non-trivial term that has serious consequences on
the stability of the system appeared in the momentum equation for the hydrodynamic
zone, which cannot be obtained without performing the asymptotic analysis. The ad-
ditional term is the global-acceleration term, which acts as a pressure gradient applied
from the acoustic zone onto the hydrodynamic zone.

Numerical results show two stability regimes. In the first regime, the system is
linearly stable. In the second regime, the system is linearly unstable and the pertur-
bations eventually reach a limit cycle. Bifurcation diagram is then obtained with the
heater power as the control parameter. The effect of global-acceleration term is investi-
gated using bifurcation diagrams. This term acts as one of the coupling terms (the other

96
one is from the boundary condition for the momentum equation in the hydrodynamic
zone) from the acoustic to the hydrodynamic zone. Therefore, the absence of the same
term weakens the coupling between the acoustic and hydrodynamic zones. Without
the global-acceleration term, the transition from linearly stable to unstable behaviour
occurs for a value of the non-dimensional heater power, which is one order of mag-
nitude higher than the value of the heater power corresponding to the simulation with
the above term. Thus the linear stability of the system is predicted incorrectly in the
absence of the term. The same term appears due to the two length scale nature of the
Rijke tube system, which is generic in thermoacoustic systems. The numerical compu-
tation of response functions in the past did not take into account the global-acceleration
term. Hence the evaluation of the stability of thermoacoustic system using the response
function has to be performed carefully in the future.

Finally the flow field during limit cycle oscillation is analysed. The limit cycle
amplitude is observed to be comparable to the steady base flow. Flow reversal occurs
in the hydrodynamic zone during part of a period of the limit cycle oscillation. This
nonlinear behaviour of the unsteady flow is observed as acoustic streaming during the
limit cycle. A mean shift in the unsteady heat release rate from the heater is observed
due to acoustic streaming.

In brief, asymptotic analysis gives the correct system of equations for the dynamics
of the acoustic field and the heater. Also using any response functions for the dynamics
of the heater without rigorous mathematical justifications can lead to incorrect govern-
ing equations, which will lead to erroneous results.

97
 CHAPTER 4

Non-modal stability analysis in a Rijke tube system

4.1 Introduction

In the previous chapter, asymptotic analysis is used to obtain the governing equations
for the dynamics of the acoustic field and heat source. The asymptotic stability of the
system is determined and plotted in a bifurcation diagram (Fig. 3.4). As an extension,
the present chapter deals with the non-modal stability analysis of this Rijke tube system.
This analysis is performed on the obtained governing equations (Eqns. 3.21 and 3.24).

As explained in the introduction chapter, the non-normal nature of the thermoa-

coustic interaction leads to transient growth in the amplitude of the oscillations. The
amount of transient growth depends on the choice of the initial condition. Hence it is
important to find the initial condition (t = 0), which produces the maximum possible
transient growth at a given time (t = Topt ). The conventional technique of obtaining the
maximum transient growth and the corresponding initial condition is by performing sin-
gular value decomposition (SVD) on the linear operator governing the linearised system
(Farrell and Ioannou, 1996a). This technique is used in the analysis of thermoacoustic
instability in SRM. (see Section 2.4 in Chapter 2) and is suitable only for small systems
(degrees of freedom 1000). Hence the non-normal nature of small model problems
(Baggett et al., 1995; Balasubramanian and Sujith, 2008b) can be investigated easily
using the SVD technique.

In the present analysis, thermoacoustic instability in a Rijke tube is investigated by

simultaneously solving the governing equations for the acoustic field and the unsteady
heat transfer from the heat source. In this manner, the dynamics associated with the
heat source is also taken into account. CFD techniques are used to solve the equations
governing the dynamics of the heat source. Hence the number of grid points for numer-
ically solving the governing equations is of the order 104 (see table 4.1), where the use
of SVD to obtain the optimum initial condition for maximum transient growth becomes
computationally costly.

In this thesis, another technique known as the adjoint optimisation technique

(Gunzburger, 2003) is used to obtain the maximum transient growth and the corre-
sponding initial condition for the linearised evolution equations. In the past, adjoint
optimisation technique has been employed successfully in fluid flow problems such
as flow over a swept wing (Guegan et al., 2008), cylinder (Abdessemed et al., 2009),
stenotic flow (Mao et al., 2009) etc. Though adjoint optimisation technique is an opti-
misation algorithm, the adjoint equations obtained in this process yield physical insights
into the problem.

Hill (1995) has explained that the eigenmodes of the adjoint system (adjoint modes)
represent the sensitivity of the flow to perturbation based on the energy (norm, see Sec-
tion 4.3) defined for obtaining the adjoint system. This physical interpretation is used in
flow control. The location of the actuator and sensor are determined from the shape of
the eigenmodes of the adjoint and direct system respectively (Barbagallo et al., 2009).
The adjoint system is also used in the model reduction for control applications (Rowley,
2005; Barbagallo et al., 2009). In some cases, the mechanism of instability in some fluid
flow problems are identified with the adjoint modes along with the eigenmodes of the
direct system (Marquet et al., 2009). Further, nonlinear adjoint optimisation technique
was used to identify the optimal path for the system to undergo subcritical transition to
instability in the context of stability of thermoacoustic interactions (Juniper, 2011b) and
shear flows (Monokrousos et al., 2011). An introduction to the non-modal stability the-
ory applied to the stability of shear flows in fluid mechanics followed by the explanation
of various tools (such as adjoint optimisation) to obtain the optimum initial condition is
given in the review article by Schmid (2007).

4.2 Direct equations

Earlier investigations of the non-normal nature of thermoacoustic systems in Rijke tube

dealt with a simple algebraic model (Balasubramanian and Sujith, 2008b) for the re-

100
 Figure 4.1: Flow domain and boundary conditions in the hydrodynamic zone.

sponse of the heat source to acoustic oscillations. Moreover, the results discussed in
Section 2.7.3 of Chapter 2 have shown that including the dynamics of the heat source
leads to a higher transient growth compared to that using a response function. In the
previous chapter, thermoacoustic instability in the Rijke tube is modelled, taking into
account the dynamics of the heat source. The governing equations for the acoustic and
hydrodynamic zones are rewritten from Chapter 3 (Eqns. 3.21 & 3.24) as follows.

ua 1 pa
s + =0 (4.1a)
ta xa
  Z 2 
1 pa ua lw lc _
+ = (c (Tp Tc ))er d (x xf ) (4.1b)
ta xa P e c Sc 0

p
+ c (p up ) = 0 (4.1c)
tc
    
1 1 1 ua 
p + up c up + c p c = c + c (c ) up + s
2
(4.1d)
tc Rec 3 tc xf

1 2
c up = Tp (4.1e)
P ec c
1
p = (4.1f)
Tp
The governing equations described above are also called as direct equations, as they
temporally advance the system from an initial condition. As discussed in Appendix C,
the electrical heater in its primitive form is modelled as a circular cylinder. The equa-
tions governing the hydrodynamic zone are solved for a geometry of the flow over a

101
heated circular cylinder. Due to low Reynolds number (Red 20) of the flow, sym-
metry boundary condition is applied (Fig. 4.1). The governing equations of the hydro-
dynamic zone in the plane polar co-ordinates, along with the boundary conditions are
shown below.

u u 1 v
Continuity : + + + =0 (4.2a)
t r r r
   
u u v u v 2 p 1 1 rrr 1 r
r Momentum : +u + = + +
t r r r r Re r r r r
uar
+
t
(4.2b)
   
v v v v uv 1 p 1 1 r2 r 1
Momentum : +u + + = + +
t r r r r Re r2 r r
ua
+
t
(4.2c)

u u 1 v 1
Energy : + + = 2 T (4.2d)
r r r Pe
1
State : = (4.2e)
T
with the following boundary condition

u(1, ) = v(1, ) = 0




p(1, )
Cylinder surface r = 1 : =0 0
r




(1, ) = 1/Tw

u(r , )
u(r , ) = 1 + ua |xf cos
= 0




r


v(r , )

v(r , ) = 1 ua |xf sin
= 0

Far field r : p(r , ) , r 0<

2 2
=0
p(r , ) = 0


r





(r , )
(r , ) = 1/Tu = 0
r

102

= 0
u(r, )








v(r, ) = 0
Symmetry = 0, : 1r<
p(r, )
= 0






(r, )
= 0

where rr = 2u/r 2/3 up , = 2 (1/rv/ + u/r) 2/3 up , r =
r (v/r)/r + 1/ru/, up = 1/r (ru) /r + 1/rv/, 2 = 2 /r2 +
/(rr) + 1/(r2 2 ), u, v - velocity components of up in r and directions. The
subscript c is dropped in the derivatives for notational simplicity.

The system of equations in the acoustic zone (Eqns. 4.1 a, b) is shown to be non-
normal due to the inclusion of the unsteady heat release term by Balasubramanian and
Sujith (2008b). Also the Eqns. 4.1 (c-f ) is nothing but the Navier-Stokes equation for
variable density flows, with an additional global acceleration term ua /tc |xf and are
also shown to be non-normal (Waleffe, 1995). Hence the coupled system of equations
(3.21 & 3.24) is non-normal. The amount of transient growth to be obtained in a non-
normal system depends on the initial condition.

One of the important quantity to study non-normal systems is the maximum possible
transient growth at a given time Topt and the corresponding initial condition. Linear op-
timal is obtained in the present investigation; i.e. the optimum initial condition obtained
from the linearised governing equations. Obtaining nonlinear optimal; i.e., the optimal
initial condition obtained from the nonlinear governing equations is beyond the scope of
the present investigations. It is recommended to first determine the linear optimal, study
its evolution and understand the role of transient growth on the asymptotic dynamics of
the system. Obtaining the nonlinear optimal can then be performed as the subsequent
step. Equations governing the acoustic zone (Eqns. 4.1 a, b) are already linear, whereas
that of the hydrodynamic zone (Eqn. 4.2) are nonlinear. The latter is linearised with the
following decomposition, u = u + u0 , v = v + v 0 , p = p + p0 , = + 0 & T = T + T 0 ,
where over bar () indicates the steady state variables and prime (0 ) indicates the fluctu-
ations from the steady state. The decomposition of the variables is applied to Eqns. 4.2

103
to give the following linearised equations (primes 0 are removed for convenience).
 
ua pa
F1 = s + =0 (4.4a)
ta xa

    Z  
pa ua pa T 
k d (x xf ) = 0
_
F2 = + + 
ta xa r (1,,ta )
0

(4.4b)
u + u 1
F3 = + ( u + u) + + (
v + v) = 0 (4.4c)
tc r r r
   
u u u v u v u 2v v u v u v2
F4 = + u +u + + + u +
tc r r r r r r r r
 
p 1 1 rrr 1 r uaf
+ + cos = 0 (4.4d)
r Re r r r r ta
   
v v v v v v v uv + u v
v v v uv
F5 = + u +u + + + + u + +
tc r r r r r r r r
 2

1 p 1 1 r r 1 uaf
+ 2
+ + sin = 0 (4.4e)
r Re r r r ta
 
u u 1 v 1 2 T
F6 = + + + =0 (4.4f)
r r r P e

where, k = 2lw lc /(P ec Sc ) = 2k. The presence of factor 2 is due to the enforcement
T + T = 0) is
of symmetry boundary condition. The linearised equation of state (
used in the elimination of T . The following are the linearised boundary conditions.

Acoustic zone:

Ha1 = pa (0, t) pa0 = 0 (4.5a)

Ha2 = pa (1, t) pa1 = 0 (4.5b)

where, subscript a represents acoustic zone. 0 & 1 indicate the upstream and down-
stream end of the tube.

Hydrodynamic zone:

In the hydrodynamic zone, the following convention for identifying the boundary
conditions is followed. The first subscript h represents hydrodynamic zone. The
numerical subscript indicates the location of the boundary condition in the physical do-
main (Fig. 4.1), whereas the last subscript represents the variable to which the boundary

104
condition is applied. For eg. Hh1u has the numerical subscript 1, which corresponds
to the surface of the cylinder and the last subscript u represents the variable u.

Hh1u = u(1, , t) u0 = 0, 0 (4.6a)

Hh2u = u(, , t)/r ud = 0, 0 /2 (4.6b)

Hh3u = u(, , t) uaf cos = 0, /2 < (4.6c)

Hh4u = u(r, 0, t)/ us1 = 0, 1r< (4.6d)

Hh5u = u(r, , t)/ us2 = 0, 1r< (4.6e)

Hh1v = v(1, , t) v0 = 0, 0 (4.7a)

Hh2v = v(, , t)/r vd = 0, 0 /2 (4.7b)

Hh3v = v(, , t) + uaf sin = 0, /2 < (4.7c)

Hh4v = v(r, 0, t) vs1 = 0, 1r< (4.7d)

Hh5v = v(r, , t) vs2 = 0, 1r< (4.7e)

Hh1p = p(1, , t)/r p0d = 0, 0 (4.8a)

Hh2p = p(, , t) p = 0, 0 /2 (4.8b)

Hh3p = p(, , t)/r pd = 0, /2 < (4.8c)

Hh4p = p(r, 0, t)/ ps1 = 0, 1r< (4.8d)

Hh5p = p(r, , t)/ ps2 = 0, 1r< (4.8e)

105
 Hh1 = (1, , t) 0 = 0, 0 (4.9a)

Hh2 = (, , t)/r d = 0, 0 /2 (4.9b)

Hh3 = (, , t) = 0, /2 < (4.9c)

Hh4 = (r, 0, t)/ s1 = 0, 1r< (4.9d)

Hh5 = (r, , t)/ s2 = 0, 1r< (4.9e)

The corresponding initial conditions are given by

Gua = ua (x, 0) u0a = 0 (4.10a)

Gpa = pa (x, 0) p0a = 0 (4.10b)

Gu = u (r, , 0) u0 = 0 (4.10c)

Gv = v (r, , 0) v 0 = 0 (4.10d)

Gp = p (r, , 0) p0 = 0 (4.10e)

G = (r, , 0) 0 = 0 (4.10f)

where pa0 = pa1 = u0 = ud = us1 = us2 = .... = s1 = s2 = 0.

Having obtained the linearised governing equations for the Rijke tube system, the
next step is to define an energy, which measures the growth or decay of disturbances.
The disturbance energy thus chosen is used to measure the non-normal nature of the
system (see Section 1.5). There has been a significant discussion in the past on choosing
the correct energy in order to describe the non-normal nature of the system (Hanifi et al.,
1996; Sameen and Govindarajan, 2007; Guegan et al., 2008). For incompressible flows
the total kinetic energy of the perturbations over the domain is used as a measure (norm)
in order to quantify the non-normal nature of the fluid flows (Schmid and Henningson,
2001). Also, when the total kinetic energy of the perturbations is used as a norm for
incompressible flows, the contribution to the norm from nonlinear terms vanishes to
zero. This is due to the fact that the nonlinear terms in the Navier-Stokes equation
for incompressible flows are energy conserving (provided the perturbations vanishes
at the boundary, like in plane Poiseuille flow). Since the nonlinear terms are energy

106
conserving, the transition to instability happens via linear mechanism - transient growth
due to the non-normal nature of the system. Also it was shown that non-normal nature of
the governing linearised operator is a necessary condition for the subcritical transition to
instability in shear flows (Henningson et al., 1993; Henningson and Reddy, 1994). The
same conclusion can be extended to any system, where the nonlinear terms conserve
energy (Krechetnikov and Marsden, 2009).

In the analysis of thermoacoustic instability in a model problem, Rijke tube, acous-

tic energy is chosen as the norm to study the non-normal nature of the thermoacoustic
interaction (Balasubramanian and Sujith, 2008b; Nagaraja et al., 2009). In the above
model problem of thermoacoustic interaction in a Rijke tube, an algebraic model (quasi-
steady) for the unsteady heat release rate from the electrical heater due to acoustic ve-
locity is used. Hence the degrees of freedom of the dynamical system (modeling the
Rijke tube) is restricted to the acoustic variables. Instead of an algebraic model for the
response of the heat source, solving the equations which governs the dynamics (both
transient and asymptotic) of heat source gives a more representative solution for the
physical problem. The inclusion of the dynamics of the heat source increases the num-
ber of degrees of the freedom of the thermoacoustic system, which now consists of not
only the acoustic variables but also the variables representing the dynamics of the heat
source. As established in Chapter 2, the degrees of freedom in the unsteady heat release
rate are shown to be important in the context of the non-normal nature of thermoacous-
tic instability. The energy defined not only included acoustic energy, but also the total
contribution from the variables representing the dynamics of the heat source. Although,
this disturbance energy was not obtained from rigorous mathematical arguments, that
chapter explained the importance of the variables representing the dynamics of the heat
source in the context of the non-normal nature of the system. Further in a recent pa-
per, Subramanian and Sujith (2011) also observed similar effects to those explained in
Chapter 2 in the context of ducted premixed flame

As an extension of acoustic energy (which gives the energy in isentropic disturbance

Rienstra and Hirschberg 2008), Myers energy (Myers, 1991) is used as the disturbance
energy in the present thesis. Myers energy gives the energy in an arbitrary disturbance
(not necessarily an isentropic disturbance) in non-reacting flows. An extension of My-

107
ers energy for reacting flows is given by Nicoud and Poinsot (2005); Giauque et al.
(2006). The nonlinear terms in the governing equations (Eqn. 3.2) do not conserve
Myers energy. This is in contrast to the nonlinear terms in the Navier-Stokes equa-
tion governing incompressible flows, which conserves kinetic energy (Henningson and
Reddy, 1994). Although the property of energy conservation by the nonlinear terms is
not present in Myers energy (Myers, 1991), it is the most appropriate form of distur-
bance energy that is available at present to analyse the non-normal nature of Rijke tube
system. Myers energy for isentropic and incompressible flows simplifies to acoustic
and kinetic energy respectively (Chu, 1965; Myers, 1991). It was shown by George and
Sujith (2011a) that Myers is not positive definite and hence fictitious non-normality
or transient growth even in the absence of source terms is possible. This fictitious
growth is due to the presence of terms associated with steady state velocity in Myers
energy. Since, the present investigation is performed in the limit of zero Mach number,
this effect is insignificant. In fact it was shown by George and Sujith (2011b), that My-
ers energy is positive definite when the Mach number of the steady state flow is less

than 1/ , which is satisfied in the present case. Moreover, the results of the present
simulations shown in Section 4.7.4 indicate that the there is negligible change in the op-
timum initial condition obtained using Chus and Myers energy. Hence, Myers energy
is continued to be used as the measure for the amplitude of the disturbance.

In the present problem, there are two length scales involved (see Section 3.2). Hence
the derivation of disturbance energy from Myers energy has to be performed carefully.

4.3 Myers energy

In the present thesis, disturbance energy corresponding to the second order energy
corollary from Myers (1991) is used. This is because as the asymptotic expansion (Eqn.
3.4) has terms up to second order in M . Moreover the leading order (in powers of M )
representation of the disturbance energy corresponding to the exact energy corollary is
precisely same as the disturbance energy (second order disturbance energy) correspond-
ing to the second order energy corollary. Thus it is sufficient to use the second order
disturbance energy expression for the definition of disturbance energy in the present

108
  
context. Disturbance energy per unit volume Ev , as defined for (Eqn. 3.3) is

 2  2 ! !
U 2 P 0 1 T0
Ev = + + s u02 + 2us 0 u0 (4.11)
2 U 2 s 1 Ts

where prime (0 ) represents the fluctuating quantities from the steady state. The fluc-
tuating quantities in the above equation are identified from the ansatz (Eqn. 3.4) as
following:

0 = c + M a , u0 = ua + uc ,
(4.12)
0 2 0
p = M pa + M pc , T = Tc + M Ta

Substituting the above expressions in Eqn. 4.11 gives

 2  2 !
P c + M a 1 Tc + M Ta
U 2
U 2 s
+
1 Ts

Ev = (4.13)
2
2
+ s (ua + uc ) + 2us (c + M a ) (ua + uc )

The total disturbance energy (Ed ) is the integral of Ev over the entire volume. The
volume is divided into three regions. The first one is the acoustic zone upstream of the
heater, second one is the hydrodynamic zone and the third one is again the acoustic zone
downstream of the heater. Disturbance energy in these three zones are then summed up
to get the total disturbance energy.

xZf ZK
ZZ Zla
Ed = Sc Ev d
xa + Ev dVc +Sc xa , 0
Ev d (4.14)
0 Vc xf +

The above integrals are evaluated in the appropriate zones as explained in the following
subsections. Disturbance energy in the acoustic zone is first evaluated.

4.3.1 Acoustic zone

The first and last terms in Eqn. 4.14 represent the energy stored in the acoustic
! field.
xfR R
l a
The expression for acoustic energy is Eac = Sc Ev d
xa + xa , 0.
Ev d
0 xf +

109
Eac is evaluated with the following additional equations.

Ta a
c = uc = 0, 0, = pa equation of state
Ts s
(4.15)
a 1
= pa isentropic relation
s

Using the above equation, acoustic energy (Eac ) is obtained as:

xZf  2  !
pa pa
+ s u2a + 2M s us ua dxa


0
Eac
= U Sc la 2
2
 2  ! , 0
Z1
pa p
+ + s u2a + 2M s us ua
a
dxa

xf +
(4.16)
Now, as M 0, the last term in the integrals drops out. Further, non-dimensionalising

Eac using U 2 Sc la 2 gives the final form of non-dimensional acoustic energy (Eac ).

xZf  2 ! Z1  2 !
Eac pa pa
Eac = 2  = 2
+ s ua dxa + + s u2a dxa , 0
U Sc la 2
0 xf +
(4.17)
It is important to note that the effect of mean flow does not appear in the acoustic energy,
because the flow is at very low Mach number.

4.3.2 Hydrodynamic zone

Evaluation of disturbance energy in the hydrodynamic zone is performed as follows

!
2 2
ZK
ZZ ZZZ 1 ( c + M a ) (T c + M T a )
U 2 K M 2 s
+
( 1)Ts
dVc
Ec = Ev dVc =
2
Vc Vc 2
+ (ua + uc ) + 2us (c + M a ) (ua + uc )
(4.18)
where, Vc is the volume of hydrodynamic zone. In obtaining Ec in the hydrody-
namic zone, the leading order contribution for Ec from all the state space variables
(c , uc , & Tc ) in the hydrodynamic zone are taken into account. It is important to in-
clude the contribution (upto the leading order) from all the state space variables in Ed ,

110
as otherwise the mapping between the direct and adjoint variables (see Eqns. 4.32 &

4.33) will be inconsistent. Non-dimensionalising Ec using U 2 Sc la 2 as before for Eac
leads to the following expression for the non-dimensionalised disturbance energy.

Z Z2   2  
lw l2
Ec = c s uc + ua |xf + 2us c uc + ua |xf rdrd
Sc la
r=1 =0
(4.19)
Z Z2  
lw lc2 s 1
+ Tc2 + 2 rdrd
Sc la ( 1)M 2 M 2 s c
r=1 =0

Since linear equations are considered, the equation of state (Eqn. 4.1f) is also linearised
to obtain Tc = c /2s . Further, according to our convention and linearisation of the
variables as in Section 4.2, Ec takes the following form

Z Z s u2 + v 2 + 2(uu + vv)
2lw lc2  
Ec = 1 1 rdrd (4.20)
Sc la + 2 1 + 2
r=1 =0 M 2 ( 1)

As before, the factor 2 is included in the coefficients of above expression, to en-

force the symmetry boundary condition. In the above equation, the limits of integration
represent the boundary of the hydrodynamic zone in plane polar co-ordinates. The den-
sity fluctuations () in the hydrodynamic zone vanishes as one moves away from the
heater (r ) due to the boundary conditions (Eqns. 4.9b & 4.9c). Thus the integrand
2 M 2 ) (1 + 1/ (( 1)
1/ ( )) 2 (third term) in the integral of (Eqn. 4.20) vanishes
as r . Furthermore, it is also observed from the present numerical simulation that
this integrand decays to zero faster than 1/r2 for large values of r (r 50). Hence
the associated integral in converges.

On the other hand, the free stream of the hydrodynamic zone is perturbed by the
acoustic velocity ua |xf (see boundary conditions 4.6b, 4.6c, 4.7b & 4.7c). Hence the
quantity u2 + v 2 and uu + vv in the first two integrands do not vanish as r ,
2
leaving a non-convergent integral. A convergence factor (r) = e((r1)/Lcu ) is used
to make the integral convergent and Ec takes the form as shown in Eqn. 4.21. In the
expression for (r), Lcu is the cut off radius, which limits the contribution of u2 + v 2
to Ec beyond the cut off radius. It is also verified from the present simulation that

111
optimum initial condition obtained does not vary significantly beyond some value of
Lcu (see Fig. 4.5 & Section 4.7.3). It is to be noted that including a Gaussian weight
function in the cost functional (Eq. 4.24) to make the integral convergent is a standard
technique in the optimisation procedure. In the past, Guegan et al. (2008) have used
the Gaussian weight function to determine the optimum freestream condition to obtain
a maximum spatial growth. The flow is over a swept wing and the domain considered is
unbounded. Similar situation arises in the present case, where the hydrodynamic zone
is considered to be unbounded. Further, the convergence in the shape of the optimum
initial condition in the hydrodynamic zone (Fig. 4.5) with respect to Lcu indicates that
the weight function used is merely for the optimisation procedure and the physics of the
problem remains unchanged. The final expression for Ec is given as follows:

Z Z
2lw lc2


Ec = s u2 + v 2 + 2(
uu + vv) rdrd
Sc la
r=1 =0
   2 (4.21)
Z Z
2lw lc2 1
+ 1+ rdrd
Sc la M 2 ( 1)

r=1 =0

Now all the terms in Eqn. 4.14 are evaluated. The sum of Eqns. (4.17 & 4.21) gives
the total non-dimensionalised disturbance energy (Ed ) and is given by:

xZf  2 ! Z1  2 !
pa pa
Ed = Eac + Ec = + s u2a dxa + + s u2a dxa

0 xf +
| {z }
acoustic zone
Z Z
2lw lc2


+ s u2 + v 2 + 2(
uu + vv) rdrd (4.22)
Sc la
r=1 =0
Z Z    2
2lw lc2 1
+ 1+ rdrd, 0
Sc la M 2 ( 1)

r=1 =0
| {z }
hydrodynamic zone

The above expression for disturbance energy (Eqn. 4.22) is used as the norm to
study the non-normal nature of the present Rijke tube system. This norm is used in the

112
cost functional, which is to be maximised over all possible initial conditions.

4.4 Definition of the cost functional

The amount of transient growth is measured with some scalar measure. In this thesis,
amplification of energy (from t = 0) in an arbitrary perturbation (Myers energy, Sec-
tion 4.3) is used as the measure for quantifying transient growth. The cost functional
(=) is defined as the ratio of Edt at time t = T and t = 0.

T
Ed opt (pa , ua , u, v, p, , T )
== (4.23)
Ed0 (pa , ua , u, v, p, , T )

where,

Z1  2 !

pa 
Edt = + s u2a  dx

0 (x,t)
Z Z



+ 1 (r) u2 + v 2 + 2 (r) u + 3 (r) v (r,,t) rdrd (4.24)
r=1 =0
Z Z

+ (r) 2 (r,,t) rdrd
r=1 =0

where the coefficients, 1 = 2lw lc2

/Sc la , 2 = 4lw lc2
u/Sc la , 3 = 4lw lc2
v /Sc la
and = 2lw lc2 (1 + 1/ (( 1)
)) / (Sc la M 2 2 ) (see Eqn. 4.22). Since pa / occurs
together in both the governing equations (Eqns. 4.4 a, b) and cost functional, the vari-
able combination can be treated as a single variable. The same convention is followed
in the rest of the analysis. Thus = represents the amplification of the disturbance en-
ergy at t = Topt . The above quantity (=) depends on the initial condition of the system.
The aim of the present study is to identify that initial condition, which maximises = at
time t = Topt . This an optimisation problem, with constraints as the governing equa-
tions (Eqn. 4.4), boundary (Eqns. 4.5 4.9) and initial conditions (Eqn. 4.10). This
constrained optimisation problem is converted into an unconstrained problem by intro-
ducing the Lagrange function (). The above technique is explained in Gunzburger
(2003). The expression for is given by the following:

113
= =
|{z} hF1 , ua ia hF2 , pa ia hF3 , ih hF4 , uih hF5 , vih hF6 , pih
| {z }
cost functional governing equations
X n o X n o X n o
am
Ham , H h1n
Hh1n , H h2n
Hh2n , H
a hrs hr1
m=0,1 n=u,v,p, n=u,v,p,
X n o X n o X n o
h3n
Hh3n , H h4n
Hh4n , H h5n
Hh5n , H
hr2 h h
n=u,v,p, n=u,v,p, n=u,v,p,
| {z }
boundary conditions
X h i X h i
n
Gn , G n
Gn , G
a h
n=ua ,pa n=u,v,p,
| {z }
initial conditions

(4.25)

where

ZTopt Z1 ZTopt Z Z
hq1 , q2 ia = q1 q2 dxdt, hq1 , q2 ih = q1 q2 rdrddt,
t=0 x=0 t=0 =0 r=1
ZTopt ZTopt Z
{q1 , q2 }a = q1 q2 dt, {q1 , q2 }hrs = q1 q2 ddt,
t=0

t=0 =0

ZTopt Z/2  ZTopt Z 
 

{q1 , q2 }hr1 = q1 q2 rddt , {q1 , q2 }hr2 = q1 q2 rddt ,
 
t=0 =0 r t=0 =/2 
r
ZTopt Z Z1 Z Z
{q1 , q2 }h = q1 q2 drdt, [q1 , q2 ]a = q1 q2 dx, [q1 , q2 ]h = q1 q2 rdrd
t=0 r=1 x=0 =0 r=1

Tilde ( ) represents the adjoint variables and are in general called Lagrange multipli-
ers. The next section deals with the derivation of adjoint equations and corresponding
adjoint boundary conditions.

4.5 Adjoint system

Maximisation of = requires the first variation of with respect to direct and adjoint
variables to be zero (Gunzburger, 2003). The zeros of the first variations of with re-

114
  . . 
T = 0 , give back
spect to the adjoint variables / ua = pa = .... = G
the original governing equations (Eqn. 4.4) along with the initial (Eqn. 4.10) and
boundary conditions (Eqns. 4.5 4.9). The zeros of the first variations of with re-
spect to the direct variables (/ua = / (pa /) = .... = /GT = 0), give rise
to the adjoint equations with the adjoint initial and boundary conditions. A brief in-
dication of the steps involved in deriving the adjoint equations and the corresponding
boundary conditions are given in Appendix D.

4.5.1 Adjoint equations

In the present thesis, the derivation of the adjoint equations and the corresponding
boundary and initial conditions is similar to that in Corbett and Bottaro (2001) and Gue-
gan et al. (2008). The adjoint equations obtained after performing the steps indicated in
Appendix D are

ua pa
F1 = s +
ta xa

Z Z  
u
v
rdrd (x xf ) = 0
_
cos sin (4.26a)
tc tc
=0 r=1
pa ua
F2 = + pa = 0 (4.26b)
ta xa
 2


1 1 u
v
u
v

F3 = + (ru) + (v ) u + u
tc r r r r r r
 

v v v uv T 2
u + + v p = 0 (4.26c)
r r r P e
u u 1 v ( u v)
F4 = + u + u u + vu) v +
( u
tc r r r r r
 
p 1 u 2 v
+ + + u 2 2
2
=0 (4.26d)
r r Re r r
 
v 2
v u v v v

F5 = + u + (
uv) + + v
tc r r r r r
 
1 p 1 2 u v
+ + + v + 2
2
=0 (4.26e)
r r Re r r2
u u 1 v
F6 = + + =0 (4.26f)
r r r

115
There are two significant differences between the direct (Eqn. 4.4) and adjoint (Eqn.
4.26) equations. The first difference is that the direct energy equation is an algebraic
constraint that determines the hydrodynamic pressure (pc ). On the other hand, one ob-
tains divergence free condition on the adjoint velocity field (Eqn. 4.26f) as the algebraic
P e) 2 p acts as a source in the adjoint continu-
constraint. Furthermore, the term T/ (
ity equation.

The second difference is the following. Laplacian operator 2 is self-adjoint, while

other operators such as gradient changes its sign in their adjoint forms (Appendix
D.2.1). Hence the diffusion terms in the adjoint equations (Eqns. 4.26 d-f, adjoint
equations are multiplied by 1 to make the term /t positive) have opposite sign
in comparison to the that in the direct equations (Eqns. 4.4 d-f ). Also note that the
damping term n n pa in the direct acoustic energy equation (Eqn. 4.4b) changes its sign
in the adjoint acoustic energy equation (Eqn. 4.26b). Thus the diffusion terms, which
dampen the perturbations during the forward evolution of the direct equations (Eqn.
4.4), dampen the perturbations during the backward evolution of the adjoint equations
(Eqn. 4.26). This observation is consistent with the solution procedure for solving the
direct and adjoint equations, where the direct equations are marched forward in time,
while the adjoint equations are marched backward. The above solution procedure for
the direct and adjoint equations is discussed later in Section 4.6.

It is shown already in the previous chapter that the inclusion of global accelera-
tion term alters the asymptotic stability of the system. In the present chapter, it is
observed that this term acts as a coupling term between the two zones in the adjoint
R R
equations. The term v /tc ) cos ( u/tc )) rdrd appearing in the
(sin (
=0 r=1
adjoint acoustic momentum equation (Eqn. 4.26a) acts a source term from the hydro-
dynamic zone to the acoustic zone. The above term can be identified as an adjoint
unsteady heat release rate term q corresponding to the unsteady heat release rate term
R
q = k 0 (T /r)|(1,,ta ) d in the direct acoustic energy equation (Eqn. 4.4b). The
origin of the adjoint heat release term q is associated with the global acceleration term
in the direct equations. During one of the steps in the extremisation of the Lagrangian
(Section 4.23), the global acceleration term in Eqns. 4.4d & 4.4e are integrated by parts
(as briefly explained in Appendix D). The integrals obtained after the process (bound-

116
ary terms also appear, which enter the mapping between the direct and adjoint variables
and will be explained in Section 4.5.3) appear as q. Thus the role of global acceleration
term is identified as the coupling term from the adjoint equations in the hydrodynamic
zone to the adjoint equations in the acoustic zone.

4.5.2 Adjoint boundary conditions

The boundary conditions corresponding to Eqn. (4.26) are the following:

Acoustic zone:
pa (0, t) = pa (1, t) = 0 (4.27)

Hydrodynamic zone:

(1, , t) = (, , t) = 0, 0 (4.28a)

(r, 0, t) = (r, , t) = 0, 1r< (4.28b)

u(1, , t) = 0, 0 (4.29a)

u(, , t) = 0, 0 (4.29b)

u(r, 0, t)/ = u(r, , t)/ = 0, 1r< (4.29c)

v(1, , t) = 0, 0 (4.30a)

v(, , t) = 0, 0 (4.30b)

v(r, 0, t) = v(r, , t) = 0, 1r< (4.30c)

117
 p(1, , t) = kP ePa (xf , t), 0 (4.31a)
1 u(, , t)
p(, , t) = , 0 (4.31b)
Re r 
1
v (r, 0, t)
p(r, 0, t) = u(r, 0)
u(r, 0, t) + , 1r< (4.31c)
Re r
 
1
v (r, , t)
p(r, , t) = u(r, )
u(r, , t) + , 1r< (4.31d)
Re r

As described earlier (Section 4.26) about the difference between the direct and ad-
joint equations, the difference in the direct and adjoint equations are discussed as fol-
lows. One of the coupling from the acoustic zone to the hydrodynamic zone in the direct
equations is through the freestream boundary condition for u & v (Eqns. 4.6c & 4.7c),
which acts as the source for the direct momentum equations (Eqns. 4.4d & 4.4e). In
contrast the coupling from the acoustic zone to the hydrodynamic zone in the adjoint
equations is through the adjoint pressure p at the surface of the heater (Eqn. 4.28a).

4.5.3 Mapping between direct and adjoint variables

The mapping (D A) from direct to adjoint variables at time t = Topt is given by

T Z Z  
2ua opt u Topt
cos vTopt
sin
+ rdrd (x xf )
_
uTa opt =
E0 s
r=1
 =0
T
2 pa opt /
pTa opt =
E0 (4.32)
Topt

= 0 2Topt + 2 uTopt + 3 v Topt
E


uTopt = 2 1 u Topt
+ 2 Topt
E 0


vTopt = 2 1 v Topt
+ 3 Topt
E 0

118
The return mapping (A D) from adjoint to direct variables at time t = 0 is given by

0 2 Z Z  
(E ) 0 u cos v sin
0 0
rdrd (x xf )
_
u0a = ua
2E Topt s
=0 r=1
0 2
p0a (E ) p0a
=
2E Topt
 



=
0 0 0
2 u + 3 v0
(4.33)
21
!
2
1 (E 0 )
0
u = u 2
0 0
21 E Topt
!
2
1 (E 0 )
0
v = v 3
0 0
21 E Topt

 
0 2  
where, = (E ) /E opt / 2(22 +23 )/21 . As mentioned in the last paragraph of Sec-
T

tion 4.5.1, the boundary terms obtained during the integration by parts applied to the
global acceleration term appear inside the integrals in Eqns. 4.32 & 4.33. If one ne-
glects the global-acceleration term, the mapping simplifies to just rescaling the variables
(Corbett and Bottaro, 2001). Another important point to be noted here is that the ho-
mogeneous boundary condition for the adjoint velocity field as r (Eqns. 4.29b
& 4.30b) allows one to expect the integral over the hydrodynamic zone in (Eqns. 4.32
& 4.32) to converge. The convergence of the same has been checked from the present
simulations.

4.6 Solution procedure

The system of equations (Eqns. 4.4 & 4.26) are then solved with the corresponding
boundary conditions (Eqns. 4.5-4.9 & 4.27-4.28) with the above mappings (Eqns. 4.32
& 4.33) between the direct and adjoint variables. In the this thesis, power iteration
procedure (Guegan et al., 2008) is used to solve the direct and adjoint equations. The
algorithm followed is represented schematically in Fig. 4.2. An initial suitable guess
value for the state space variables (u0a , p0a , u0 , v 0 , p0 & T 0 ) is used as the starting value
for the algorithm. Then the direct equations (Eqn. 4.4) is marched forward in time upto
t = Topt . The direct variables at t = Topt is mapped to the adjoint variables at t = Topt

119
 Figure 4.2: Schematic of adjoint looping to determine the optimum initial condition.

using the mapping (Eqn. 4.32). Then the adjoint equations (Eqn. 4.26) is marched
backward in time up to t = 0. Then the adjoint variables at t = 0 is mapped back to
the direct variables at t = 0 using the mapping (Eqn. 4.33), which is again fed as the
initial condition to the direct equations. The process is repeated until convergence in the
initial condition in the direct variables are obtained. This process is referred as adjoint
looping. The final converged initial condition gives the optimum initial condition for
maximum transient growth in the disturbance energy Edt . The power iteration algorithm
fails if the initial guess value for the initial condition is orthogonal (corresponding to
the inner product defined in Section 4.4) to the optimum initial condition, which is a
rare scenario. Both the direct (Eqns. 4.4a & 4.4b) and adjoint (Eqns. 4.26a & 4.26b)
acoustic equations are solved by Galerkin technique, while the direct (Eqns. 4.4c -
4.4f) and adjoint (Eqns. 4.26c - 4.26f) equations in the hydrodynamic zone are solved
by finite difference technique. The above procedure for solving the direct equations is
discussed in detail in Appendix C.

4.7 Results and discussion

Numerical simulations are performed with the parameters shown in Table (4.1). Unless
specified, the parameter values in table (4.1) are used for the simulations.

120
Table 4.1: Physical parameters in acoustic zone, hydrodynamic zone and convergence
parameters in for solving the direct and adjoint equations

Acoustic zone Hydrodynamic zone Convergence parameters

la = 1 m, Sc = 0.01 m2 Red = 20, Hydrodynamic zone grids

P = 1 bar lc = 1 mm Residue Direct Adjoint

u0 = 1.025 kg m3 lw = 16.4 m Continuity 105 102

Tu = 295 K
0 Tw = 1000 K Momentum 105 105

T0d = 500 K M = 5 104 Energy 105 105

Figure 4.3: Evolution of the disturbance energy (Ed ) obtained from nonlinear simula-
tions. a) Globally stable system, C1 = 0.27, C2 = 0.03, xf = 0.25, K =
0.1, 1 = 1.9 103 , 2 = 3 = 3.7 103 , = 5.28 103 , Lcu = 35,
u0a (xa ) = 1.5cos (xa ) , p0a (xa ) = u0 = v 0 = T 0 = 0. b) Linearly un-
stable system, C1 = 0.27, C2 = 0.03, xf = 0.25, K = 0.1785, 1 =
3.3 103 , 2 = 3 = 6.6 103 , = 9.4 104 , Lcu = 35,
u0a (xa ) = 0.5cos (xa ) , p0a (xa ) = u0 = v 0 = T 0 = 0.

121
Figure 4.4: Convergence of the variables during each adjoint looping for a) state space
variables, U i , Pi , u, v & and b) cost functional =. Residue is defined
PNgrid  Nadj   Nadj1   Nadj 
as i=1 i  i  / i  to measure the convergence to-
wards the optimum initial condition. i represents any one of the state space
variables Ui , Pi , u, v & . The summation index i represents the value of
the variables at the ith spatial location. Nadj represents the adjoint loop
number. The threshold of residue is chosen as 0.1% for convergence, which
corresponds to Nadj = 23. The area shaded in grey indicates converged
solution. The parameters are Red = 10, C1 = 0.27, C2 = 0.03, xf =
0.25, K = 0.2856, 1 = 1.06 102 , 2 = 3 = 2.11 102 , =
2.49 103 , Lcu = 35.

4.7.1 Evolution of disturbance energy

One of the important part in the present chapter is the derivation of the energy in the dis-
turbances from Myers energy (Section 4.3). The final form of the disturbance energy
Ed for the Rijke tube system comprises of the disturbance energy in the acoustic zone
(Eac ) and in the hydrodynamic zone (Ec ) (Eqn. 4.22). Further, Ed is used as a measure
to determine the growth or decay of the perturbations. Figure 4.3 shows the evolution of
Ed for globally stable and unstable systems. In the former, disturbance energy asymp-
totically decays (Figs. 4.3 a, b), while it grows and reaches limit cycle oscillations for
the latter (Figs. 4.3 c, d). Hence, using Ed as a scalar to measure transient growth is
justified.

122
Figure 4.5: Streamlines of u & v (top half) and contour of temperature T (bottom
half) corresponding to the optimum initial condition for maximum transient
growth, a) Lcu = 15, b) Lcu = 20, c) Lcu = 25, d) Lcu = 30, e) Lcu = 35,
f ) Lcu = 40. g) Variation of = with Lcu . Other parameters are same as in
Fig. 4.4.

123
4.7.2 Convergence studies

Since adjoint looping is an iterative process, convergence of the solution with the num-
ber of loops is presented in Fig. 4.4. Residue indicates the fractional difference in the
initial condition between successive adjoint loops (Fig. 4.4 a). The values of the residue
for all the state space variables (Ui , Pi , u, v & ) are chosen as 1 103 to ensure con-
vergence. In this simulation, it is observed that a total of 23 adjoint loops are required
to achieve convergence from a random initial condition. Further, the cost functional (=)
reaches a steady value after the same number of loops. Hence, in all the simulations to
follow, a convergence criteria of residue = 1 103 is followed.

A convergence factor (r) is used during the derivation of Ed (Section 4.3). The
convergence on the shape of the optimum initial condition with the cut off radius Lcu is
investigated. Figs. 4.5 (a-f ) show the streamlines corresponding to velocity field u & v
(top half) and contour of temperature distribution T (bottom half) near the cylinder for
various values of Lcu . It can be observed that the shape of the optimum initial condition
does not vary significantly beyond Fig. 4.5 (d) (Lcu = 30). Moreover, this convergence
is also reflected in the cost functional (Fig. 4.5 g, Lcu = 30). Hence for the subsequent
simulations the value of Lcu = 35 is chosen. Further, the patterns in the optimum initial
condition confine to a non-dimensional radius r 5 ensures the convergence of the
optimum initial condition with the domain size (in the present case r = 50).

4.7.3 Optimum initial condition

In this subsection, the optimum initial condition for the maximum transient growth at
t = Topt is analysed. Figure (4.6) shows the distribution of the optimum initial condition
in the state space variables. One can observe in the flow field corresponding to u & v,
distinct patches of vortical structures are present (Fig. 4.6 a). The temperature (T )
contours (Fig. 4.6 b) also shows a similar concentrations of high temperature region
near the heater. The initial acoustic velocity (u0a ) distribution shows distinct peaks along
the length of the duct.

As earlier in Chapter 2, the optimum initial condition is not only distributed among

124
Figure 4.6: Distribution of optimum initial condition in the state space variables, (a)
streamlines and magnitude of velocity corresponding to u & v, (b) temper-
ature field T in the hydrodynamic zone, (c) acoustic field in the acoustic
zone. The parameters are same as that in Fig. 4.4 e.

the acoustic variables, but also among the variables in the hydrodynamic zone. The
distribution of the optimum initial condition in the state space variables shows that for
obtaining the maximum transient growth a non-acoustic disturbance (disturbance in the
hydrodynamic zone) is also required. Hence in the Rijke tube system, it is important
to include the dynamics of the heat source apart from the dynamics of the chamber
acoustic field in order to understand the non-normal nature of the system. The same
idea has been emphasized in the previous chapter on solid rocket motors.

4.7.4 Comparison with Chus energy

As a comparison with obtaining the optimum initial condition using other norms, Chus
energy (Chu, 1965) is used. Chus energy is chosen in particular, as it is positive definite
and George and Sujith (2011a) recommends it for the use in measuring transient growth.
Chus energy is obtained from the expression for Myers energy (Eqn. 4.24) by setting
2 = 3 = 0.

Both the optimum initial conditions obtained from Myers and Chus energy are

125
Figure 4.7: Comparisons of the optimum initial condition, computed using (a) Myers
and (b) Chus energy in the definition of the cost functional =. In the case
of Chus energy, 2 = 3 = 0 in Eqn. 4.24. =max using Myers and
Chus energy are 1843 and 1861 respectively. The legends for the plots are
as follows. Top: streamlines corresponding to u & v & contours of tem-
perature T , bottom: distribution of acoustic velocity and pressure. Other
parameters are same as that in Fig. 4.4 (e).

Table 4.2: Relative comparison of the optimum initial computed using Chus energy
with the results obtained by using Myers energy as the reference.

Percentage difference
in the property Value

=max 0.98
ua 0.85
pa 0.81
u 0.66
v 0.64
T 2.66

126
shown in Fig. 4.7. One could identify that the two optimum initial conditions look
almost identical. Further, the relative difference in the optimum initial condition is tab-
ulated in Table 4.2. It is observed that a relative difference in the optimum initial con-
dition obtained using Myers and Chus energy are less than 1% in all the state space
variables, except T . Hence, one obtains almost the same optimum initial condition from
using the above two energies. This can be attributed to the fact that at low Mach num-
bers (which is the case in the present situation), energy contributions from the velocity
fluctuations are less compared to that from temperature and density fluctuations. Hence
one obtains a quantitatively close match between the two initial conditions.

4.7.5 Evolution of the optimum initial condition

Figure 4.8 (a & b) shows the evolution of the amplification of disturbance energy
(Edt /Ed0 ) for the optimum initial condition from a linear simulation. As expected, Edt
grows initially due to non-normal nature of the system. The parameters are chosen
such that the system is linearly stable and hence Edt decays asymptotically. Another
important observation is that, although the optimisation procedure is performed for
Topt = 0.5, a small amount of energy growth occurs beyond that time. Maximum
transient growth occurs at ta = 0.63, where Edt /Ed0 = 888. While, the transient growth
at Topt = 0.5 is 866. Similar behaviour is observed by Guegan et al. (2008), in their
optimisation problem for the flow over the leading edge of a swept wing.

Figure 4.8 shows the evolution of the optimum initial condition in the acoustic zone.
The initial distribution of the acoustic velocity and pressure are shown in Figs. 4.8 (c &
d). The subsequent evolutions of them are indicated in the Figs. 4.8 (e & f ) respectively.
At t = 0, the amplitude of the acoustic field is small (appears to coincide with the zero
line) and as time progresses, their amplitudes start to grow (compare the scales in Figs.
(c-d & e-f ). This in turn leads to the transient growth in Edt . Furthermore, the jump in
ua across the heat source is also evident.

The observed transient growth is not restricted to the acoustic zone, but also to the
hydrodynamic zone. Figure 4.9 shows the evolution of the flow field in the hydrody-
namic zone. It can be observed that the amplitude of the initial disturbance in the tem-

127
Figure 4.8: Evolution of the optimum initial condition. a) Evolution of the ampli-
fication of disturbance energy (Edt /Ed0 ), b) zoomed in view. The sym-
bol indicate the selected time instants, for which the flow fields are
shown in subsequent subfigures. Optimum initial condition in the acous-
tic zone, c) acoustic velocity & d) acoustic pressure. Evolution of the
acoustic field, e) acoustic velocity & f ) acoustic pressure. The parame-
ters are Red = 35, C1 = 0.31, C2 = 0.1, xf = 0.25, K = 0.06, 1 =
2.12 104 , 2 = 3 = 4.24 104 , = 7.41 102 , Lcu = 35. Other
parameters are same as that in Fig. 4.4 (e).

128
Figure 4.9: Evolution of the optimum initial condition in the hydrodynamic zone, a)
t = 0, b) t = 0.25, c) t = 0.5 & d) t = 1.0. In each subfigure, top half:
velocity field corresponding to u & v, bottom half: temperature field T . The
parameters are are same as that in Fig. 4.8.

perature field is of the order of 103 . As the system evolves, this amplitude increases
to of the order of 101 . Furthermore, there is a slight increase in the amplitude of the
velocity (u & v) perturbations as well. The vortical structures are observed to convect
along with the steady flow. During their convection, they grow in their size and ampli-
tude. This is also reflected in the perturbation temperature field. The size of the patches
of high and low temperature disturbance increases and are convected downstream. To
summarize, a overall increase in the amplitude of the disturbance associated with all the
state variables can be observed from the evolution of optimum initial condition.

129
Figure 4.10: Variation of the optimum initial condition with the time scale ratio . a
& c)  = 0.68, b & d)  = 1.10. =max for the above  are 1843 and
1253 respectively. Top: streamlines corresponding to u & v & contours
of temperature T , bottom: distribution of acoustic velocity and pressure.
Other parameters are same as that in Fig. 4.4 (e).

4.7.6 Parametric study

A parametric study is performed to analyse the non-normal nature of the system with
the variation of the important parameters in the system. In the present case, one such
parameter is the ratio of the time scales ( = tac /tcc , see Section 3.2) associated with
the acoustic (tac ) and hydrodynamic (tcc ) zones.

The optimum initial conditions for  = 0.68 & 1.10 are shown in Fig. 4.10. The
qualitative nature of the optimum initial condition in the hydrodynamic zone does not
vary significantly with  and hence can be thought of as a generic initial condition,
which promotes transient growth in this Rijke tube system. Furthermore, the values of
=max , for both the cases are found to be 1843 & 1253. Maximum amplification are in
the same order, which further enforces the generic nature of these initial conditions.

130
4.8 Interim summary

Non-normal nature of the thermoacoustic interaction in an electrically heated horizontal

Rijke tube is investigated. The dynamics of the heat source is taken into account and the
coupled problem (dynamics of the acoustic field and the heat source) is solved simulta-
neously (two length scale problem). The analysis started from the equations governing
the dynamics of the chamber acoustic field (acoustic zone) and the heat source (hy-
drodynamic zone). In order to study the non-normal nature of the system, a norm is
defined to measure the transient growth observed in the non-normal systems. Since the
dynamics of the heat source is taken into account, the degrees of freedom of the system
not only comprised of the acoustic variables, but also the variables determining the dy-
namics of the heat source. Hence the norm defined has to be a physical energy, which
should take into consideration the energy in the disturbance present both in the acoustic
and hydrodynamic zone. A norm is defined based on the disturbance energy proposed
by Myers, which gives the energy in an arbitrary disturbance with a non-zero steady
base flow and no chemical reaction. Myers energy simplifies to acoustic and kinetic
energy for isentropic and incompressible perturbations respectively. The norm defined
by this manner can, in principle be extended to more complex thermoacoustic problems
and the non-normal nature of the system can be investigated.

In order to study the non-normal nature of the present problem, the optimum ini-
tial condition, which produces the maximum transient growth (measured in the norm
defined in the above paragraph) is obtained. In the present case, adjoint optimisation
technique is applied to obtain the optimum initial condition as the number of degrees
of freedom of the system is large ( 104 ). As the first step in understanding the non-
normal nature of the system, linear optimal (optimum initial condition for maximum
transient growth for the linearised coupled problem) is obtained. In the process of the
application of the adjoint optimisation technique, adjoint equations are obtained. The
coupling between the two length scales (acoustic and hydrodynamic zone) in the adjoint
equations occurs in an elegant way. The adjoint temperature in the hydrodynamic zone
acts as a source for the adjoint velocity in the adjoint momentum equations, which is
in the opposite way as in the direct equations in the hydrodynamic zone. The global
acceleration term in the direct equations in the hydrodynamic zone is responsible for

131
forming the coupling from the adjoint hydrodynamic to the adjoint acoustic zone. The
above term in the adjoint acoustic energy equation is equivalent to the unsteady heat re-
lease rate term in the direct acoustic energy equation. Thus the global acceleration term
present in the direct momentum equations in the hydrodynamic zone produces terms
in the adjoint acoustic energy equation, which can be regarded as the adjoint unsteady
heat release rate term. The global acceleration term also appears in the mapping be-
tween the direct and the adjoint variables. Power iteration algorithm is used to solve
both the direct and the adjoint equations to obtain the optimum initial condition.

The optimum initial condition obtained has components both in the acoustic and the
hydrodynamic zone. In the acoustic zone, the optimum initial condition associated with
the acoustic variables show distinct peaks along the length of the duct. In the hydrody-
namic zone, optimum initial condition projects as distinct patches of high vorticity and
corresponding patches of high temperature regions near the heat source.

132
 CHAPTER 5

Experimental investigations of non-normality

5.1 Introduction

The earlier chapters dealt with the theoretical investigation of non-normality of two
thermoacoustic systems; solid rocket motor and Rijke tube. In order to complete the
analysis and gain further understanding of the phenomenon, we continue to perform
experiments in the Rijke tube system. The present chapter focusses on the experimental
observation of some of the phenomenon that shows the evidence of non-normality and
the associated transient growth.

The concept of transient growth first arose in the stability analysis of shear flows
(Schmid, 2007), where the essential role of short-term energy amplification based on
a linear process has been established by theoretical calculations and numerical simula-
tions (Henningson and Reddy, 1994; Hanifi et al., 1996). Substantially fewer studies
have been experimental. The earliest experimental results have been reported by Mayer
and Reshotko (1997), who observed transient growth of disturbance amplitudes in a
pipe flow experiment. The formation of streaky structures (i.e., streamwise elongated
regions of high and low velocities) in boundary layers, which can be attributed to the
non-normal nature of the linearized system, is another example of experimental evi-
dence for transient growth (Matsubara and Alfredsson, 2001). In the same experimen-
tal configuration, stable laminar streaks have been generated and transient spatial am-
plitude growth has been observed in the streamwise direction (Fransson et al., 2004).
Furthermore, passive control of transition to turbulence in a boundary layer has been
attempted by exploiting the non-normal nature of the system (Fransson et al., 2006).

The experimental confirmation of theoretical and numerical observations gives greater

weight to the role of non-normal effects in wall-bounded shear flows and establishes this
concept as a valid component in transitional shear flows. The same confirmation for
thermoacoustic systems is as yet missing (although experimental observations of Kim
and Hochgreb (2012); Zhao (2012) show the presence of transient growth, the results
were incomprehensive) ; but the present study is intended to fill this gap. In particular,
we will identify the non-normal nature of thermoacoustic interactions in an experimen-
tal Rijke tube and quantify the amount of transient energy growth in acoustic energy it
yields. To this end, the present chapter focusses on two main issues. The first one is
concerned with the experimental extraction of eigenmodes from the Rijke tube which
is accomplished by the Dynamic Mode Decomposition (DMD) introduced in Schmid
(2010). This is followed by confirming the non-orthogonality of the extracted eigen-
modes. The second issue deals with the quantification of transient growth, measured in
the acoustic energy, by generating specific initial conditions with an acoustic driver.

The remaining chapter is structured as follows: the experimental configuration of

the Rijke tube system is described in Section 5.2, after which the employed experi-
mental procedures (Section 5.3) and their theoretical background are introduced. These
tools are then be employed on the Rijke tube experiment and the obtained results are
discussed in light of the two main issues of this inquiry (Section 5.4.4 and Section
5.4.5). This chapter is then concluded with interim summary Section 5.5.

5.2 Experimental configuration of the Rijke tube

In the present case, Rijke tube (made of aluminium sheet with 7 mm thick) is 1 m
long with 9.3 9.3 mm cross section area. A schematic diagram of the experimental
setup is shown in Fig. 5.1. A mean flow is established in the tube, using a blower (1
HP, Continental Airflow Systems, type CLP-2-1-650) operated in suction mode, and
the volume flow rate is measured by a compact-orifice mass-flow meter (Rosemount
3051 SFC) located just upstream of the blower. The measurement range of the flow
meter is 0 5g/s with an uncertainty of 2.1% of the mass flow rate. Two decouplers
(120 45 45cm3 ) one upstream, one downstream of the tube are located so as
to provide acoustic open-open boundary conditions.

Details of the Rijke tube are shown schematically in Fig. 5.2. A mesh-type (size

134
Figure 5.1: Schematic diagram of the Rijke tube experiment. All dimensions are given
in mm.

Figure 5.2: Configuration of the Rijke tube showing the heater, the microphone ports
and the acoustic drivers. The locations of the microphone ports are num-
bered, and their distances from the upstream edge of the tube are listed in
table 5.1. All dimensions are given in mm.

135
30) electrical heater can be located at any desired axial location, and a programmable
DC power supply (TDK-Lambda, GEN8-400, 0 8V, 0 400A) is used to power the
electrical heater. Mesh type heater is chosen in contrast to the wire type used in the
theoretical analysis (see Appendix C). The advantages are as follows. a) Mesh type
heater provide a uniform heating across the cross section of the tube, b) the heater can
deliver large amount of power (800 W ) for a long period of time (6 hours). Thick
copper rods are used to supply electrical power to the mesh. The electrical resistance
of the mesh is of the order of milli Ohms and hence, the DC power supply is required
to supply high currents. The heater is housed in a ceramic stand and the configuration
is shown in Fig. 5.3. The detailed drawing of the individual components are given in
Appendix E.

The horizontal Rijke tube configuration described in this paper is similar to the one
reported in Matveev (2003b) and Song et al. (2006). Two thermocouples have been
installed, one at the inlet of the tube and the other downstream of the heater; these are
used to monitor the steady state temperature in the duct. Four acoustic driver units
(Ahuja AU 60) are mounted on the walls of the tube to impose a prescribed acoustic
initial condition on the system. A total of 15 microphones are placed along the duct
in order to record the evolution of acoustic pressure. Although, both transducers and
microphones are used to measure acoustic pressure, microphone is chosen as the repre-
sentative term that will be used in this thesis. The microphones are distributed such that
the first two eigenmodes of the system can be determined efficiently. The details of the
relative calibration among the microphones are explained in Appendix F. The precise
locations of the heater, the thermocouples, the acoustic drivers and the microphones
are schematically displayed in figure 5.2 and tabulated in table 5.1. Data are acquired
simultaneously using the programmable National Instruments PCI 6221 & 6251 data
acquisition cards at 8kHz.

5.3 Experimental procedures and theoretical background

The non-normal nature of our system results in non-orthogonal eigenmodes which, in

turn, can lead to transient growth of the thermoacoustic oscillations. Transient growth

136
Figure 5.3: Top: electrical heater along with the ceramic holder. All dimensions are in
mm. Bottom: operating at 1 kW

137
Table 5.1: The locations of the microphones, heater, thermocouple and loudspeakers,
measured from the upstream edge of the Rijke tube.

Microphone no. Location (mm) Microphone no. Location (mm)

1 35 10 675
2 70 11 750
3 100 12 800
4 150 13 825
5 225 14 900
6 300 15 965
7 375 Heater 125
8 450 Thermocouple 150
9 525 Acoustic driver unit 600

is a linear phenomenon and any experimental analysis of this phenomenon has to en-
sure operation in the linear regime (a more precise definition will be given in Section
5.3.2). An attempt is then made to determine the non-orthogonality between the first
two eigenmodes of the system and to experimentally observe and quantify transient
growth associated with these two modes. For the extraction of the eigenmodes, the
Dynamic Mode Decomposition (DMD) technique (Schmid, 2010) will be used which
yields eigenmodes of the system as long as the experiment is performed in the linear
regime. In the following subsections, we introduce tools, techniques and experimental
procedures, such as bifurcation analysis (Section 5.3.1), linear regime detection (Sec-
tion 5.3.2), extraction of acoustic dynamic modes (Section 5.3.3) and validation of the
technique (Section 5.3.4 and Section 5.3.5) and experimental exploration of transient
growth potential (Section 5.3.6), before these methods will be applied to measurements
from our experiments (in Section 5.4).

5.3.1 Bifurcation diagram

In a first step, the bifurcation behavior of the full, nonlinear system is obtained which
will guide us in parameter space, establish various stability limits and, in particular,
determine the regions where a (predominantly) linear behavior can be expected or as-
sumed. A bifurcation diagram consists of a relation between a pertinent control pa-

138
rameter and the asymptotic state of the system (properly parameterized by a specified,
representative variable, such as, e.g., the energy or disturbance amplitude). As the
power supplied to the heater (our control parameter) is increased, it has been observed
in earlier investigations (Matveev and Culick, 2003b; Song et al., 2006) that the transi-
tion to instability occurs via a subcritical Hopf bifurcation. The subcritical nature of the
bifurcation allows the co-existence of a stable fixed point and a limit cycle, and this par-
ticular parameter regime is also referred to as the bistable region (Subramanian et al.,
2010). In this region, the system is linearly stable but can nevertheless be forced into a
limit cycle, if the initial condition falls above some threshold amplitude. The minimum
amplitude that ultimately yields a limit-cycle behavior is known as the triggering am-
plitude of the system, and specific experimental details of its determination are given in
Section 5.4.1. The obtained triggering amplitudes provide an estimate of the acoustic
pressure amplitudes at which nonlinear effects play a significant role in the qualitative
behavioral change (limit cycle or fixed point) of the system.

5.3.2 Regime of linear behavior

A linear system can be defined as one in which, when the amplitude of the input is
scaled by a given amount, the amplitude of the output is scaled by the same amount. In
the present case, the input variable is the voltage supplied to the acoustic drivers, while
the output variable is the acoustic pressure measured at a given location in the tube.
For a user-specified heater power, the above input is varied and the output is recorded;
the response curve of the system can thereby be obtained. For our purposes, the limit
of linearity is reached when the deviation of the systems response from the linearized
response function (see Appendix H for details) exceeds 5% of the corresponding trigger-
ing amplitude. The experiments probing the non-normal nature of our thermoacoustic
system are performed with amplitude levels below the limit of linearity.

5.3.3 Dynamic Mode Decomposition

In pursuit of the first goal of the present investigation, the acoustic eigenmodes of the
system have to be extracted. This can be accomplished, for example, by using net-

139
work models for each acoustic component (Bellucci et al., 2004) where each individual
acoustic elements is described by a transfer matrix, which can be determined experi-
mentally (Bellucci et al., 2004) or numerically (Schuermans et al., 2005). The eigen-
frequencies and corresponding eigenmodes of the system can then be calculated from
the assembled network matrix. The transfer matrices associated with the flame elements
are obtained by forcing the flame continuously and measuring its response experimen-
tally. The required experiment has to be performed at low amplitudes of forcing, where
the system behaves linearly, and the accurate identification of the transfer matrices in
this regime may prove difficult (Hosseini and Lawn, 2005). In addition, the acoustic
fields upstream and downstream of the heat source have to be modelled as separate
acoustic elements, making this process rather time consuming. Due to its composite
nature, any error in the calculation of any one of the transfer matrices will propagate
and reflect on the final result. Finally, the transfer matrix technique has been developed
for a frequency-domain analysis; extending it to the time-domain is nontrivial. To cir-
cumvent the above issues, we choose to extract the eigenmodes via a Dynamic Mode
Decomposition (Schmid, 2010) which isolates the modes from a time series of acoustic
pressure measurements simultaneously acquired at various locations on the Rijke tube.

Dynamic Mode Decomposition (DMD) is a data-based technique that extracts dom-

inant dynamic features from a time-resolved sequence of flow field measurements. The
identified dominant flow structures are termed dynamic modes; data from numerical
simulations and experiments can be processed. When the flow process, from which
the snapshots are sampled, is linear, the application of DMD is equivalent to a global
stability analysis (Rowley et al., 2009), and the dynamic modes thus obtained are the
eigenmodes of the system. This technique has been applied to a variety of flow con-
figurations, such as flow in a lid driven cylindrical cavity (Schmid et al., 2009) us-
ing time-resolved particle-image velocimetry (PIV) measurements in a cross-sectional
plane, flow of a helium jet Schmid et al. (2010) using Schlieren images as snapshots, or
numerical simulations of a jet flame and experiments for a water jet (Schmid, 2011). In
all cases, the dominant frequencies and the associated flow structures could be detected.
A brief description of this technique is made in Appendix I.

In the present investigation, experiments are performed in the linear regime to ex-

140
tract the first two eigenmodes of the system. The Rijke tube system is initially perturbed
using acoustic driver units in a manner that excites the dominant eigenmodes of the
system. The temporal evolution of the acoustic pressure is recorded by microphones,
after which the DMD technique is applied to the acquired data. The first two acoustic
pressure eigenmodes of the system are then isolated from the DMD spectrum. After
obtaining the eigenmodes of the system, their mutual non-orthogonality is quantified
RL
by computing the inner product hpi , pj i = pi pj dx, where pi represents the ith eigen-
0
mode of the system, properly normalized according to hpi , pi i = 1. The length of the
tube is denoted by L, and the superscript stands for the complex conjugate. The above
definition of the inner product is identical to the one used by Culick (2006) and Nicoud
et al. (2007). The inner product is calculated for various values of heater power levels;
the results are presented and discussed in Section 5.4.4.

5.3.4 Theoretical investigation of the eigenmodes

To the best of our knowledge, the DMD technique has so far not been applied to acous-
tically active or thermoacoustic systems. It thus seems prudent to ensure and validate its
applicability to this type of flow and data. To this end, we generate synthetic data from
a theoretical model, extract the eigenvalues and eigenmodes using DMD and compare
them to the theoretical results. Thermoacoustic interactions in the Rijke tube system can
be represented by the following linearized acoustic momentum and energy equations,
given in non-dimensionalised form (Balasubramanian and Sujith, 2008b),

u p
+ = 0, (5.1a)
t x

p u
+ = Q (x xf ) , (5.1b)
t x
where, u and p are the non-dimensionalised acoustic velocity and pressure, respectively,
Q is the non-dimensionalised unsteady heat release rate and xf is the location of the heat
source in the tube. The effect of steady-state flow and temperature gradients in the duct
are neglected. The heat source is assumed to be compact compared to the length of
the tube and is thus represented by a Dirac-delta function ((x)). The jump conditions

141
across the heater are as follows (Poinsot and Veynante, 2005):

p(x+
f , t) = p(xf , t) (5.2a)

f , t) u(xf , t) = Q(t)
u(x+ (5.2b)

where the superscript indicates the value of the variables upstream or downstream of
the heat source. The unsteady heat release rate from the heat source is modelled using
the common n model (Crocco, 1956) and is of the form

Q = nu(x
f , t ) (5.3)

with n as the interaction index, which indicates the power supplied to the heater, and
as the time lag associated with the acoustic velocity fluctuations upstream of the heat
source. The Rijke tube is assumed open at both ends. The eigenmodes of the system are
p) corresponding
then obtained from the above equations (5.1-5.3). The eigenmodes (
to the acoustic pressure of the system, defined in the form p(x, t) = p(x)et , are given
by

sinh (x) , 0 x xf ,
p(x) = sinh (xf ) (5.4)

sinh ((x 1)) , xf < x 1,
sinh ( (xf 1))
and the corresponding eigenvalues satisfy the transcendental dispersion relation

n
sinh () e sinh ( (2xf 1)) = 0. (5.5)
2

The theoretical pressure eigenmodes of the above system are shown in the left col-
umn of Fig. 5.4. The heater is located at x = 0.125, where the first two eigenmodes
of the system becomes less stable (Matveev, 2003b). The eigenmodes are numbered
as p1 , p2 , . . . , in ascending order of the imaginary part (frequency) of the associated
eigenvalues 1 , 2 , . . . . Each plot contains the real and imaginary part of the first two
eigenmodes. When the heater is switched off (n = 0), the eigenmodes coincide with
the natural duct modes. As the value of n is increased, the eigenmodes progressively
deviate from the natural duct modes. A discontinuity in the slope of the eigenmodes
is clearly visible at the location of the heat source, a consequence of the acoustically

142
compact nature of the heat source. This change in slope becomes more pronounced as
the value of n increases, with a more marked discontinuity in the imaginary part than
in the real part. For each subfigure, the value of the non-dimensionalised heater power
n as well as the degree of non-orthogonality given by the inner product |hp1 , p2 i| (as
described in Section 5.3.3) are displayed in the title. In cold flow (n = 0), the eigen-
modes of the system (shown in figure 5.4a) are a simple sine function for the real part
and zero for the imaginary part. The inner product hp1 , p2 i is very small, indicating that
the eigenmodes are orthogonal in this case. As the value of n is increased, the imag-
inary part of the respective eigenmodes becomes non-zero; the eigenmodes become
increasingly non-orthogonal (see Figs. 5.4a, c, e, g). This non-orthogonality between
the eigenmodes is caused by the discontinuity in the slope at the location of the heater;
the value of the discontinuity, and thus the degree of non-normality, is proportional to
n. The above theoretical findings motivate the placement of three microphones on the
upstream side of the heater and the remaining twelve microphones downstream of the
heater (see table 5.1), which should facilitate the detection of the modal shapes from
the data acquired from the microphone array.

5.3.5 Extraction of dynamic modes

Data produced by the theoretical model (Eqn. 5.1) will now be used as input for DMD
analysis, and we extract the acoustic pressure data at locations identical to the micro-
phone positions in the experiment (see table 5.1). A total of 700 snapshots, equispaced
in time with a non-dimensionalised time-interval of 0.025, have been used.

As only acoustic pressure is measured along the length of the duct and no direct
measurements related to the heat source dynamics is taken into account, any dynamic
information about the heat source is lost. It is advisable then to apply DMD separately
to data from the upstream and the downstream side of the heater. The resulting dynamic
modes are suitably scaled so as to match the acoustic pressure on either side of the heater
(see Eqn. 5.2a).

For the extraction of the first eigenmode of the system, the model equations (5.1)
evolved forward in time, starting at t = 0 with the analytical shape of the first eigen-

143
Figure 5.4: Comparison of the first two eigenmodes obtained analytically (see Eqn. 5.4)
from the n model (left column) with the eigenmodes obtained by per-
forming DMD on the data obtained from the same model (right column).
The value of the non-dimensionalised heater power n and the absolute value
of the inner product hp1 , p2 i are indicated in the title of each subfigures.
The eigenmodes are displayed following the scheme: real part of
first eigenmode, real part of second eigenmode, imagi-
nary part of first eigenmode, imaginary of second eigenmode. The
governing parameters are xf = 0.125 and = 0.2 (prescribed by Lighthill
1954).

144
Figure 5.5: Eigenvalues obtained by applying DMD to data produced from (Eqn. 5.1).
The decomposition is applied separately to data from the upstream and
downstream side of the heater; the corresponding eigenvalues are displayed
by symbols, and , respectively. Using Eqn. (5.4), the initial condi-
tion consists of a first theoretical eigenmode in (a) and a second theoretical
modes in (b). A magnified view of the dominant eigenvalues is given in
the right column. Finally, a comparison of the eigenvalues from theoretical
investigations in -symbols and from the DMD in -symbols is presented
in (e). The governing parameters have been chosen as n = 0.5, = 0.2 and
xf = 0.125.

145
mode. The resulting data set, again separated into upstream and downstream of the
heater location, are then processed by the DMD algorithm. The eigenvalues obtained
from data of the upstream (symbol ) and downstream (symbol ) side of the heater
are shown in Fig. 5.5a, for a parameter setting of n = 0.5, xf = 0.125 and = 0.2. Two
dominant, complex conjugate eigenvalues, close to the imaginary axis, are identified to-
gether with spurious eigenvalues displaying high frequencies and large decay rates; Fig.
5.5b shows an enlarged view of the relevant part of the complex plane. A very close
match between the dominant eigenvalues from the upstream and downstream data set
is observed. After the upstream and downstream modes have been frequency-matched,
continuity of the acoustic pressure across the heater is enforced and a composite eigen-
mode of the full system results. The second eigenmode of the system is calculated in
an analogous manner from a data sequence, where the system (5.1) is initialized by the
theoretical second eigenmode. Results of this calculation are displayed in Fig. 5.5 (c,
d).

Figure 5.5 (e) shows the eigenvalues of the first two modes, with the -symbol
representing the analytical eigenvalues and the -symbol illustrating the eigenvalues
obtained from a data sequence via DMD. Good agreement between the two results is
observed. The first two eigenmodes obtained by DMD are depicted in the right column
of Fig. 5.4, where black dots are included to indicate the locations of the microphones.
A discontinuity in the slope of the eigenmodes at the heater location is clearly visible,
and an excellent match in the shape of the eigenmodes with the theoretical ones (see
left column of figure 5.4) is observed. Furthermore, the amplitude of the inner product
|hp1 , p2 i| is in very good agreement between the extracted and theoretical eigenmodes.

Additional validation tests have been performed by extracting the dynamic modes,
computing their rms-amplitude in the original data sequence, reconstructing the data
sequence by a superposition of the most dominant dynamic mode and calculating the
relative error between the original and reconstructed data sequence. The results are sum-
marized in Fig. 5.6 for the first (left column) and second (right column) eigenmodes.
The raw signal (Fig. 5.6a,b) is processed by the DMD which identifies two strong com-
ponents for both the data set upstream (subfigure c,d) and downstream (subfigure e,f )
of the heater. Higher dynamic modes are not represented in the original data sequence.

146
Figure 5.6: Validation of the extraction of eigenmodes via the Dynamic Mode Decom-
position. The subfigures on the left (right) hand side are associated with
initial conditions along the first (second) mode. The initial condition is (a)
p(x, 0) = 0.6p1 (x) 0.6p1 (x), and (b) p(x, 0) = 0.3p2 (x) 0.3p2 (x).
The relative contribution (root-mean-square value) of the dynamic modes
in the data sequence from the (c,d) upstream and (e,f ) downstream side
of the heater. The first two dynamic modes in each subfigure corre-
spond to the actual eigenmodes. (g,h) The percentage difference (xj ) =
(||p(xj , t) pr (xj , t)||) /||p(xj , t)|| 100 between the input data and the
reconstructed data sequence using the eigenmodes of the system at all mi-
crophone locations. The parameters are the same as in Fig. 5.5.

147
Figure 5.7: Comparison of the non-orthogonality measure |hp1 , p2 i| obtained from the-
oretical calculations (solid line) and from processing data by the DMD tech-
nique (symbols).

As indicated in the second paragraph of this subsection, each eigenmode is extracted

separately. The first two dominant dynamic modes represent a single eigenmode, along
with its complex conjugate. Further, the initial conditions are chosen to have equal con-
tributions in the eigenmode and its complex conjugate (see the caption of figure 5.6).
Hence the amplitudes ||Aj || of the first two dynamic modes are same in figures 5.6(c-f ).
The temporal data sequence reconstructed with only the most dominant dynamic modes
(and their frequencies) is then compared to the original data sequence. In particular, the
normalized difference between the original (p(xj , t)) and reconstructed (pr (xj , t)) sig-
nal is calculated according to (xj ) = ((kp (xj , t) pr (xj , t)k)/kp (xj , t)k) 100.
The results are plotted in Fig. 5.6(g, h). Less than a 0.4%-error is observed at all xj -
locations of the microphones for the first eigenmode; a maximal percentage error of
0.15% is observed for the second eigenmode.

A final test concerns the dependence of the non-orthogonality on the non-dimensionalised

heater power n. The comparison between the theoretical results (solid line in Fig. 5.4)
and the results obtained by DMD (symbols in Fig. 5.4) shows very good agreement
over the entire range of heater powers, with a maximal deviation of 3.7% in the value
of (|hp1 , p2 i|) for n = 0.55.

Before applying DMD to the experimental data, it is important to assess the robust-
ness of the above techniques with respect to noise in the measurements. In particular,

148
the uncertainty in the final results as noise is added to the processed measurement data
should be determined. Details of this procedure are given in Appendix J.

5.3.6 Transient growth experiment

The amount of transient growth depends on the exact shape of the initial condition.
To observe transient growth, the initial condition given to the linear system must have
non-zero projections onto at least two non-orthogonal eigenmodes (Schmid, 2007). In
our case, this is achieved by exciting the system initially for a few cycles by a signal
composed of two eigenfrequencies, obtained from the measured data by the DMD tech-
nique. The relative amplitude of the two frequencies can then be varied to construct
various initial conditions; their transient amplification in time can then be monitored.

5.4 Results and discussions

The techniques, tools and procedures introduced in the previous section will now be
applied to our experimental setup of the Rijke tube. The experiments are performed at
an ambient temperature of 296 1K, with a relative humidity of 73 1% such that the
acoustic damping remains within required limits (Kinsler et al., 2000). The exponential
decay rate of the system in cold flow at 156 Hz, which corresponds to the eigenfre-
quency of the first natural mode has been determined as -11.5 /s (see Appendix G).
This decay rate is measured at the beginning, and the experiments are performed only
if the decay rate is within 5% of the previously stated value. This will ensure that the
acoustic damping does not change appreciably from one experiment to another. The
heater is located at xf = 0.125m for a Rijke tube of length 1m. Following the series of
steps outlined above, experiments are performed to initially obtain the bifurcation dia-
gram, followed by the determination of the triggering amplitudes, after which the linear
regime for the system is identified, the two most dominant eigenmodes are extracted,
their non-orthogonality is established and the transient growth potential contained in
these two modes is quantified.

149
Figure 5.8: Bifurcation diagrams displaying the rms values of the acoustic pressure at
x = 0.15 versus the power supplied to the heater. The mass flow rate is (a)
m = 2.34 g/s, (b) m
= 2.19 g/s, (c) m
= 2.03 g/s.

5.4.1 Bifurcation diagram

For the bifurcation diagram, the electrical power K supplied to the heater is chosen
as the control parameter and the root-mean-square (rms) value of the acoustic pressure
fluctuations (Prms |x=0.525 ) at x = 0.525 is chosen as the variable that measures the state
of the system. Figure 5.8 present the bifurcation diagrams for various mass flow rates
m.
Initially, the power supplied to the heater is low (point A in figure 5.8a) and the
system is linearly and nonlinearly (globally) stable: the system reaches a steady state
which is confirmed by monitoring the temperature at the inlet and further downstream
(x = 0.15) of the heater. After about 30 minutes of preheating, the values of K and
Prms |x=0.525 are noted. Next, the voltage supplied to the heater is increased by 0.05V,
which amounts to an increase in power of less than 20W. This level of heating is main-
tained for 4 minutes to eliminate transients and allow the system to settle into a steady
state (Matveev, 2003b). In this manner, the power supplied to the heater is increased in
a quasi-steady manner; raising the heater power more rapidly leads to nonlinear trigger-
ing of instabilities (Matveev, 2003b).

150
 The time-asymptotic state of the system is recorded for increasing values of K and
is termed the forward path. Analogously, the term return path is associated with de-
creasing values of K. The behavior of the system, as K is varied through a forward and
backward path, is shown in Fig. 5.8 (a). For low values of K (line AB), the system is
globally stable. As the value of K is increased beyond point C (854 W ), the system
becomes linearly unstable and enters into a limit cycle (point D). Further increasing K
yields progressively larger amplitudes of the limit cycle (line DE). On the return path,
as the value of K is reduced from point E, the bifurcation curve retraces the forward
path until point D after which it remains in the limit-cycle state up to point F (571
W ). Beyond point F, the system falls back to the steady state. In summary, the return
path (ECFBA) and forward path (ABCDE) are distinct, establishing the existence of
common hysteresis behavior. Similar results have been reported by Matveev (2003b).

From dynamical-systems point of view, the above transition scenario between glob-
ally stable and globally unstable behavior is referred to as subcritical Hopf bifurca-
tion (Strogatz, 2000). Within the hysteresis zone (BCDF), the system is linearly sta-
ble. However, an initial threshold amplitude exists (termed triggering amplitude) above
which the system becomes unstable and approaches limit-cycle behavior. This regime
is also termed as bistable region, as two stable solutions i.e., steady state and limit cycle
are possible.

Experiments at different mass flow rates m

reveal that the hysteresis zone becomes
smaller (Fig. 5.8 b,c) as the flow rate decreases (see also Matveev (2003b)). For our
study of the non-normal nature of the system, we will concentrate on the case with the
largest hysteresis zone and choose m
= 2.34 g/s. The steady state flow velocity (uu )
upstream of the heater and the Reynolds number (Red ) based on the diameter of the
wire in the mesh are 0.24 m/s and 4.2 respectively.

5.4.2 Triggering amplitudes

The triggering amplitudes have been obtained by first allowing the system to attain a
steady state within the hysteresis region after which an initial acoustic pulse train is fed
to the system using the acoustic driver units. The pulse train consists of 14 cycles of

151
Figure 5.9: Determination of triggering and the corresponding limit cycle amplitudes.
(a & b) Evolution of the system that is just triggered and (c) just decayed.
K = 782W , m = 2.34 g/s.

a sinusoidal wave at a frequency equal to the first eigenfrequency of the system. The
amplitude of the input wave is increased until the system changes from a linearly stable
state to a nonlinearly unstable state. For excitations with the frequency of the second
eigenmode, higher triggering amplitude is required to render the system nonlinearly
unstable; hence, the triggering amplitudes have been estimated by exciting only the
first eigenmode. Figure 5.9 (a) presents the evolution of the system for a case where
the initial excitation is sufficient to trigger the system into a nonlinearly unstable state.
The inset confirms that the excited system ultimately reaches a limit cycle. The initial
Prms |x=0.525 (after the acoustic drivers are switched off) is noted as the triggering am-
plitude. The associated resulting limit cycle is shown in Fig. 5.9 (b) and the value of
Prms |x=0.525 is recorded as well. On the other hand, figure 5.9(c) represents the evolu-
tion of the system that just fail to trigger and eventually decay to a stable fixed point.

The above procedure for finding the triggering and limit cycle amplitudes has been
repeated for various values of K, and the obtained results are shown in Fig. 5.10. The
bifurcation diagram is labeled, as before, by the characteristic points ABCDEF. In ad-
dition, the triggering and corresponding limit cycle amplitudes are designated by GH

152
Figure 5.10: Bifurcation diagram showing hysteresis behavior along with the triggering
(GH) and limit cycle (IJ) amplitudes. The experiments have been repeated
four times to ensure repeatability of the results. A mass flow rate of m
=
2.34 g/s has been chosen.

and IJ, respectively. For each value of K the experiment has been repeated four times
to control and quantify measurement inaccuracies. Consequently, the limit cycle ampli-
tudes in the bifurcation diagram have been found to vary by less than 0.5%. Results
near the Hopf (C) and fold (F) points of the diagram could be determined within a band
of 40 W and 26 W , which amounts to less than 4.5% of the power levels at the respec-
tive points. Furthermore, the spread in the triggering amplitudes and corresponding
limit cycle amplitudes is less than 6% of the respective mean values.

Triggering amplitudes beyond H have not been attempted since the spread pre-
scribed above could not be maintained in this parameter regime due to an increasing
sensitivity near the Hopf point (Strogatz, 2000). Also, the acoustic drivers failed to
trigger the system into a nonlinearly unstable state below point G; thus no results are
reported below point G. The amplitude of the limit cycle (IJ) obtained by triggering the
system is smaller than that obtained from the bifurcation diagram (FD). This discrep-
ancy can be attributed to the difference in the mean temperature in the duct during the

153
Figure 5.11: Steady state temperature recorded downstream of the heater during experi-
mental run 1. The steady-state temperature during increasing and decreas-
ing values of K are shown. Vertical dashed lines indicate the power level
associated with the region IJ shown in Fig. 5.10.

phase of increasing and decreasing (path ABCDE and EDFBA in Fig. 5.8) heater power
as shown in Fig. 5.11. The higher mean temperature is due to non-zero mean heat trans-
fer effects occurring during intense thermoacoustic oscillations; this phenomenon has
also been observed by Matveev (2003b).

5.4.3 Regime of linearity

The bifurcation diagram is further augmented by determining the parameter regime

where the system dynamics can be well approximated by a linear process. This is
accomplished by measuring the response of the system to continuous excitation by the
acoustic driver units and comparing it to expected results based on theoretical scalings.
The details of the procedure are presented in Appendix H. As a result, we obtain critical
pressure amplitudes below which the system performs nearly linearly and above which
nonlinear effects become no longer negligible. These threshold amplitude levels (as
the power supplied to the heater is varied) define the limit of linearity. It is shown in
Fig. 5.12 along with the triggering amplitude and labeled by LN. Curves GH and IJ

154
Figure 5.12: Bifurcation diagram showing averaged triggering amplitudes (GH) and av-
eraged limit cycle amplitudes (IJ). The limit of the linearity (LN) defines
the linear regime (in gray); all experiments investigating the non-normal
nature of the thermoacoustic system are performed in this regime.

represent the mean value of the triggering and the corresponding limit cycle amplitudes
averaged over different experiments (shown previously in figure 5.10). It is evident
from Fig. 5.12 that the linear regime, indicated in gray, covers a rather large area in
the bifurcation diagram measurements can thus be obtained with high signal to noise
ratio.

5.4.4 Non-orthogonality of the eigenmodes

Acoustic pressure data recorded by the microphones are used for the DMD analysis
where a total of 742 snapshots (with 0.125 ms as the time interval between them)
have been processed. The amplitudes of the recorded acoustic pressure is below the
linearity limit defined in Appendix H. As described in Section 5.3.3, a measure of non-
orthogonality of the eigenmodes is given by the expression |hp1 , p2 i|. The effect of
measurement noise on the value of |hp1 , p2 i| is discussed in Appendix J.

155
Figure 5.13: (a-c) First two dominant eigenmodes obtained by applying DMD to the
experimental acoustic pressure data. The respective values of the heater
power K and the absolute value of the inner product hp1 , p2 i are indicated
in the title of each subfigure. The use is symbols is identical to the one in
Fig. 5.4. (d) Dependence of |hp1 , p2 i| on the power supplied to the heater
K, together with the associated error bars.

The eigenmodes and corresponding values of |hp1 , p2 i| obtained from experimental

data are displayed in Fig. 5.13 (a-c). The first and second eigenmodes are obtained
through individual self-excited experiments. The reason is as follows. As indicated in
the second paragraph of 5.3.5, DMD technique is applied separately in the sections
upstream and downstream of the heater. In the upstream side, there are only three mi-
crophone ports (table 5.1) and hence only three dynamic modes are obtained. Among
this, the two dominant ones represent a single eigenmode, along with its complex con-
jugate. This eigenmode is chosen by the frequency of the initial excitation. Hence two
separate experiments are conducted to determine the first two eigenmodes.

As the power supplied to the heater is increased, the discontinuity in the slope of
the eigenmodes at the heater location becomes more noticeable. In addition, the non-
orthogonality measure |hp1 , p2 i| increases steadily with increasing values of K. The
uncertainty in the value of |hp1 , p2 i| is at most 2.5% for the noise levels encountered
in our experiments (see Appendix J for details). The variation of |hp1 , p2 i| with heater

156
power K is shown in Fig. 5.13d along with the associated error bars and confirms a
significant rise in non-orthogonality as the power supplied to the heater is increased.

As in Section 5.3.5, we verify that the extracted dynamic modes indeed represent
the eigenmodes of the system. Figures 5.14 (a, b) show typical acoustic pressure sig-
nals measured at the upstream and downstream side of the heater which are used for
the extraction of the second eigenmode of the system. The system is initially forced
in a manner that excites (nearly exclusively) the second eigenmode of the system. The
contribution of the most relevant dynamic modes to the data sequence is shown in Fig.
5.14 (c, d); only the first two dynamic modes figure significantly in each data sequence.
The spatial structure (with real and imaginary components) is depicted in Fig. 5.14 (e).
Lastly, acoustic pressure data are reconstructed based on only the second eigenmode
and are compared to the original data sequence. The relative error between the two time
series, evaluated at each microphone location, is displayed in Fig. 5.14 (e), where an
error of less than 7% is observed at all but one microphone locations. The 9th micro-
phone is located close to the acoustic pressure node of the second eigenmode and thus
experiences a substantially lower signal-to-noise ratio and a relative error of 19%. A de-
viation of less than 7% suggests that the dynamics is well represented by the extracted
second eigenmode of the system (Podvin et al., 2006).

5.4.5 Evidence of transient growth

After the non-orthogonality of the two dominant eigenmodes has been established, the
amount of transient amplification of acoustic energy will be determined. The total
RL

acoustic energy (dimensional) is defined as E(t) = S2c ( u2 + p2 / ( p)) dx (Rien-
0
stra and Hirschberg, 2008) and has previously been used in theoretical studies on the
same Rijke tube system (Balasubramanian and Sujith, 2008b). In the above expression,
Sc denotes the cross-sectional area of the tube and the overbar represents steady-state
quantities. The acoustic pressure p is measured at the 15 locations (see Fig. 5.2 and
table 5.1), and the acoustic velocity u is obtained by a two-microphone technique (Bel-
lows, 2006). The largest spacing between any two microphones in the present experi-
ment is 75 mm and the cross-sectional dimension is 92 mm, which allows us to mea-

157
Figure 5.14: Experimental acoustic pressure data at (a) location x = 0.1 and (b) loca-
tion x = 0.8 which is used in the DMD-analysis. The initial condition
is closely aligned in the direction of the second eigenmode of the system.
The relative contributions (root-mean-square value) of the dynamic modes
to the full data sequence for the (c) upstream and (d) downstream side of
the heater. The first two dynamic modes, along with their complex conju-
gates, represent the second eigenmode of the system. (e) The spatial struc-
ture of the first dynamic mode with signifying the real part and
the imaginary part. (f ) The relative error (in percent) at all loca-
tions of the microphones, (xj ) = (||p(xj , t) pr (xj , t)||) /||p(xj , t)||
100, between the input data p and a reconstructed data sequence pr using
the first two dynamic modes.

158
Figure 5.15: Three kinds of initial perturbation imposed on the system through the
acoustic driver. Sinusoidal input of the form (a) 0.54sin(2f1 t), (b)
0.54sin(2f2 t), and (c) 0.3sin(2f1 t) + 0.3sin(2f2 t), where f1 and f2
are the frequencies of the first and second eigenmodes of the system. The
vertical black line indicates the instant in time when the acoustic driver is
switched off.

sure acoustic velocity fluctuations in the frequency range 100 Hz to 1.76 kHz (Abom
and Boden, 1988). For our interest in the first ( 175Hz) and second ( 350Hz)
eigenmodes of the system, the inter-microphone spacing is suitable for our experimen-
tal investigation. A detailed description and application of the technique to the present
investigation is given in Appendix K.

Three kinds of initial perturbations are considered in our analysis, as shown in Fig.
5.15. The first two initial conditions correspond to a harmonic excitation with a pure
frequency of the first two respective eigenmodes, while the waveform of the third kind
of initial perturbation consists of a linear combination of the first two eigenfrequencies
of the system. The vertical black line indicates the instant in time, when the acoustic
driver is switched off.

For each setting of the heater (on and off), the evolution of the system excited
by the three initial conditions is analyzed. When the heater is switched off, the non-
[Sc u u2 /2] with subscript
dimensional acoustic energy (E = E/ u representing the
u

steady-state quantities on the upstream side of the heater) decays continuously in time

159
Figure 5.16: Evolution of non-dimensional acoustic energy, which is defined as E(t) =
RL
0
(u2 + p2 /(pu )) dx/ (u u2u ) . For (a), the heater is switched off; for
(c), the heater is switched on (K = 747 W ). Evolution of the normalized
amplification of acoustic energy E(t)/E(0) with heater switched off (b)
and switched on (d). The colors indicate the three kinds of initial condition
shown in Fig. 5.15; the time interval during which the acoustic driver is
active is indicated in gray.

after the excitation from the acoustic driver ceases (see Fig. 5.16a). The grey area indi-
cates the region, where the forcing from the acoustic driver is present. The normalized
amplification of acoustic energy E(t)/E(0) (where the origin of the time-axis is shifted
to the instant when the excitation from the acoustic speaker stops) is plotted in Fig. 5.16
(b). No significant transient growth can be observed in cold flow for any of the initial
conditions, indicating that the dynamics of the pure acoustic field is normal.

For the case with a heater power of 747 W, no significant transient growth can be
detected when the system is excited by the first or second kind of initial condition (as
shown in Fig. 5.16c) since almost a pure eigenmode is excited. On the other hand,
when the initial condition contains non-zero components in the two eigenmodes (as
shown in Fig. 5.15c), transient growth (indicated by the solid curve in figure 5.16c)

160
Figure 5.17: Evolution of the acoustic pressure at x = 0.3 in the linear regime. The
two horizontal dash-dotted lines in all subfigures represent the ampli-
tudes delimiting the linear regime. The amplitude of the initial pertur-
bation is scaled and the subsequent evolution of the system is recorded for
the following: (a) 0.1 sin(2f1 t) + 0.1 sin(2f2 t), (b) 0.075 sin(2f1 t) +
0.075 sin(2f2 t), (c) 0.05 sin(2f1 t) + 0.05 sin(2f2 t). In (d) the acous-
tic pressure signals from subfigures (a)-(c) are rescaled and superimposed.
Circle and cross symbols are associated with data from subfigures (b) and
(c), respectively. The duration of the systems evolution is taken as 0.2s,
which corresponds to a non-dimensional time span t/(L/a) of 69. The
supplied heater power is K = 747W.

is clearly visible. The relative energy amplification can be seen in Fig. 5.16d; for
the specific initial condition, it amounts to 1.8 times the initial energy. This growth
in energy is comparable to numerical studies by Balasubramanian and Sujith (2008b),
Nagaraja et al. (2009) and Juniper (2011b).

To rule out nonlinear effects and be able to attribute the measured energy amplifi-
cation to the non-normal nature of the thermoacoustic system, we have to ensure that
the measured pressure amplitudes remain within the linear regime for all times. Figure
5.17 (a) displays the acoustic pressure signal which corresponds to the transient growth
in figure 5.16c for the composite initial condition (third kind). The two horizontal dash-
dotted lines indicate the limit of linearity. To further confirm the linear nature of the
system output, a simple scaling experiment has been conducted: the initial perturbation

161
amplitude is scaled by a factor 0.75 and 0.5 and the associated evolution of the pressure
is recorded (see Fig. 5.17b and c, respectively). These two pressure measurements are
then rescaled and compared to the original pressure measurements. The close match (of
less than 3%), displayed in Fig. 5.17d, confirms that the transient effects shown above
are void of nonlinear contributions and can thus be ascribed to the non-normal character
of our system.

5.4.6 Optimal initial condition

Motivated by the results of the previous section, it remains to determine the maximum
transient growth potential contained in the two dominant eigenmodes by varying their
relative amplitude and phase angle in the initial condition. To this end, a more general
input perturbations is forced upon the system; we take it of the form A1 sin(2f1 t) +
A2 sin(2f2 t + ). As before, f1 and f2 stand for the frequencies of the first and second
eigenmodes of the system; the amplitudes of the respective eigenmodes are denoted
by A1 and A2 . The phase angle between the two sine waves is given by the variable
. The maximum of the energy amplification E(t)/E(0) over time is represented by
Emax = maxt E(t)/E(0).

Figure 5.18a displays the variation of Emax with the amplitude ratio A2 /A1 for a
fixed phase angle of = 0o . The symbols () represents the experimental data; the
continuous line shows an Gaussian fit of the data. Maximum amplification occurs at
A2 /A1 = 0.85, i.e., at an amplitude ratio where the initial condition has comparable
contributions from either eigenmode. The value of Emax decreases markedly away
from the optimal value. This observation also reinforces the fact that the amount of
transient growth has to diminish as the initial conditions progressively contains only
contributions from one of the two eigenmodes. To complement, the variation of Emax
with phase angle for a fixed amplitude ratio of A2 /A1 = 1 is shown in Fig. 5.18 (b) in
form of a polar plot. For this parameter setting, the value of Emax is nearly independent
of . The complete parameter dependence (simultaneously varying A2 /A1 and ) is
depicted as a contour map in Fig. 5.18 (c). Two local maxima have been found; the
global maximum of transient growth appears for an initial condition with an amplitude

162
Figure 5.18: Variation of the maximum energy amplification Emax with (a) amplitude
ratio A2 /A1 for fixed phase angle = 0o and (b) with phase angle for
fixed amplitude ratio A2 /A1 = 1. The symbols () indicate the experi-
mental data point; the continuous line represents a Gaussian fit. (c) Maxi-
mum transient growth over initial conditions of the form A1 sin(2f1 t) +
A2 sin(2f2 t + ). The optimal initial condition for maximum transient
growth is identified for the parameter combination A2 /A1 = 0.93 and
= 90o . The corresponding maximum transient growth is 2.3. The re-
maining parameters are f1 = 173Hz, f2 = 346Hz, m = 2.34g/s and
K = 747W.

163
ratio of A2 /A1 = 0.93 and a phase angle of = 90o . The associated maximum transient
growth is Emax = 2.3.

It should be emphasized that the maximum transient growth observed in this exper-
iment is three orders of lower than that obtained from the numerical simulations in the
previous chapter (Section 4.7.3). The reason is that in the experiment, only the initial
conditions associated with the acoustic variables are combined to obtain the optimum
initial condition. While, in the numerical simulations, additional degrees of freedom
associated with the dynamics of the heat source (hydrodynamic zone), apart from those
associated with the acoustic variables are present. Hydrodynamic systems have large
transient growth. For example, wall bounded shear flows have transient growth in the
perturbations of the order of 105 (Trefethen et al., 1993). Hence, the numerical sim-
ulations show higher transient growth than those observed in experiments. Recently,
Foures et al. (2012) have developed an optimisation procedure, where the cost func-
tional can be a semi-norm and constraints are imparted on the other degrees of freedom.
This procedure can be adapted to our system discussed in the previous chapter, where
the acoustic energy (comprising of ua & pa ) is the semi-norm and other state space vari-
ables (u, v & ) can be constrained. This can be taken as a possible extension of this
thesis.

It should be further noted that, earlier theoretical investigation in Rijke tube by Na-
garaja et al. (2009), with the degrees of freedom restricted only to acoustic variables
indicates a maximum transient growth of 4 6 (Fig. 7 in the same paper). However, a
maximum transient growth of 2.3 (Fig. 5.18c) is observed in the present experimental
investigation. The difference may be attributed to the following. The optimum initial
condition in the experiments is restricted to linear combinations of the first two eigen-
modes. However, the theoretical analysis by Juniper (2011b) indicates a significant
contributions from the third and fourth eigenmodes. Hence restricting the optimum ini-
tial condition to be obtained from the first two eigenmodes may lead to a lower value of
transient growth.

164
Figure 5.19: Evolution of (a) acoustic pressure P (x = 0.525) and (b) acoustic energy
E(t) for the triggering of an instability. The inset (c) shows the initial evo-
lution of E(t). Zones A, B, C and D represent the time intervals where the
system is excited by the acoustic driver (A), near the triggering amplitude
(B), in the nonlinearly unstable state (C) and in the final limit cycle (D).
The gray area represents the region when the acoustic driver is switched
on. The governing parameters are A2 /A1 = = 0o , K = 782W and
m = 2.34g/s.

5.4.7 Triggering energy

In a final step, we explore the triggering energies for initial conditions composed of the
two eigenmodes with varying amplitude ratio A2 /A1 and phase angle by increasing
the amplitude of the initial amplitude until a limit cycle is reached. Figure 5.19 (a)
(see also Fig. 5.9a) shows a typical time history of acoustic pressure for such an ex-
periment; in the companion Fig. 5.19 (b) the acoustic energy E(t) is displayed. The
various phases toward a final limit cycle are also indicated qualitatively in both subfig-
ures. For slightly lower amplitude, but otherwise constant parameters, the oscillations
decay asymptotically to zero. The experiment can thus be assumed near the triggering
threshold. As before, the grey area indicates the time span during which the acoustic
driver is active.

The value of E(t) increases steadily in region A as the system is excited during
this period. In zone B, the energy E(t) remains nearly constant. In the subsequent

165
zone C, the energy increases further as the trajectory is repelled from the unstable limit
cycle (Strogatz, 2000). Eventually, the system reaches a stable limit cycle, marked as
zone D, where E(t) remains almost constant at a higher value compared to that of the
other zones. In the above case, the initial condition has been selected with A2 /A1 = 0
and = 0o . The inset in Fig. 5.19b shows more details of the evolution of E(t) in zone
A. Since E(t) is strongly fluctuating, the triggering energy Etrig is determined by time-
averaging E(t) over a period 1/f1 immediately before the instant when the acoustic
driver is switched off.

The variation of Etrig with A2 /A1 for fixed = 0o is shown in Fig. 5.20 (a).
Since the determination of Etrig is rather sensitive to the ambient noise in the system,
experiments are repeated three times (Section 5.4.1) and the mean value, along with
the associated deviation, is reported. For low values of A2 /A1 , i.e., initial conditions
with predominant contributions from the first eigenmode, less energy is required to
trigger the system into a limit cycle. The value of Etrig increases rapidly beyond an
amplitude ratio of A2 /A1 = 3, and the system requires higher triggering energies as the
contribution from the second eigenmode becomes more prevalent. A close-up view of
the region before the sharp rise in Etrig is presented separately in Fig. 5.20 (b). It can
be established that Etrig is nearly constant until approximately A2 /A1 = 1.6; the lowest
triggering energy Etrig is recorded for A2 /A1 = 0.4 with a spread of 6% around its
mean value.

From Fig. 5.18 (a), the optimal initial condition (with = 0o ) for maximum tran-
sient growth occurs for A2 /A1 = 1.6 and while the triggering energy Etrig in Fig. 5.20
(a) shows no particular preference for this value of the amplitude ratio A2 /A1 . It appears
that the role of transient growth around the stable fixed point, undoubtedly present in the
thermoacoustic system, plays a minor or insignificant role in the subcritical transition
regime. The following may be possible reasons for this observation.

First, the true difference in value of Etrig as a function of A2 /A1 (which is respon-
sible for the selection of a preferred amplitude ratio) might be noticeably smaller than
the error bar obtained in the present investigation. Secondly, and more importantly,
our optimal initial condition has been restricted to linear combinations of the first two
eigenmodes of the system. For the particular case examined by Juniper (2011b), the op-

166
Figure 5.20: (a) Acoustic energy Etrig required for triggering an instability as a func-
tion of amplitude ratio A2 /A1 . (b) Magnified version of (a) for a value of
A2 /A1 between 0.1 and 3.0. This parameter interval is shown in gray in
(a). The remaining parameters are = 0o , K = 747W and m = 2.34g/s.

timal initial condition had distinct contributions from the third and fourth eigenmodes,
but little from the second. It is possible, however, that different parameters (such as
heater position) could change this. This has yet to be tested. Theoretical investiga-
tions by Juniper (2011b) however suggest that the optimal initial condition has distinct
contributions from the third and higher eigenmodes of the system. For our experimen-
tal setting, the third eigenmode is highly stable compared to the first two eigenmodes
(see also Matveev 2003b) and extracting this mode, simultaneously with the first two
eigenmodes, in the linear regime will encounter very low signal-to-noise ratios and thus
prohibit any definite conclusions regarding its role in subcritical transition. In view
of these difficulties, the role of transient growth in the subcritical transition regime re-
mains open from an experimental point of view. Nonetheless, the non-normality of
our thermoacoustic system, the notable increase in the non-orthogonality of the eigen-
modes with heater power and the evidence of associated transient growth have been
clearly identified in our experiments.

167
5.5 Interim summary

Experiments have been performed to identify and probe the non-normal nature of ther-
moacoustic interactions in an electrically heated, horizontal Rijke tube. As the power
supplied to the heater is increased, the transition from linearly stable to unstable behav-
ior occurs via a subcritical Hopf bifurcation. This transition is aided by transient growth
of acoustic energy in the tube, a phenomenon that has been established and investigated
theoretically. Since transient growth is a linear phenomenon, experiments have to be
performed in a parameter regime where the system behaves nearly linearly. In addition,
only the linearly stable regime can be explored to ensure linearity for all times.

Measurements of the acoustic pressure have been processed via the Dynamic Mode
Decomposition (DMD), a data-based technique that extracts dominant eigenmodes from
measurement sequences of a linear process. It has been demonstrated that the non-
orthogonality of the obtained eigenmodes increases as more power is supplied to the
system by the heater. This behavior confirms the non-normal nature of thermoacous-
tic interactions in the Rijke tube. The amount of resulting transient growth has been
measured in terms of the acoustic energy composed of the measured acoustic pressure
and the acoustic velocity, calculated via a two-microphone technique. With the heater
switched on, no significant transient growth is observed when the system is excited
along one of the two dominant eigendirections; transient growth, causing a two-fold
increase in acoustic energy, has been observed, however, when the excitation contains
contributions from both eigenmodes. For cold flow (with the heater switched off), no
significant transient growth has been found for any initial perturbations, indicating that
the classical linear acoustic system is normal.

A parameter study, varying amplitude ratio and phase angle of the two initial eigendi-
rections, revealed a maximum transient amplification of acoustic energy of 2.3 which
has been obtained for an amplitude ratio A2 /A1 = 0.93 and a phase angle of 90o . An
analogous parameter study of the threshold acoustic energy Etrig for triggering limit-
cycle behavior (for = 0o ) showed a low and constant value for 0.1 < A2 /A1 < 1.6,
after which it rises steadily. A link between parameter combination, for which max-
imum transient growth can be observed, and parameter combinations, for which the

168
triggering energy is particularly low, could not be established. The restriction to only
two dominant eigendirections (which could be extracted from measurements with suffi-
cient confidence in their accuracy) is expected as the reason for this discordance and is
corroborated by previous theoretical studies. Nonetheless, the non-normal nature of the
thermoacoustic system and the associated transient growth in acoustic energy has been
clearly identified by experimental means.

169
 CHAPTER 6

Comparison of experimental and theoretical results

In this chapter, the bifurcation diagram from experiments is compared with the numeri-
cal simulations described in Chapter 3. In particular, simulations are performed in order
to compare with the bifurcation diagram shown in Fig. 5.8 (a). In order to ensure that
the simulation parameters are same as that encountered in the experiments, the follow-
ing procedure is used.

The geometry of the heater in the simulation is a rack of wire filaments, whereas a
mesh type geometry is used in the experiments (Fig. 5.3). In order to obtain a corre-
spondence between the two geometries, an effective Reynolds number (Reef f ) is intro-
duced based on Laws and Livesey (1978); Das and Chhabra (1989). Reef f is defined as
(
u/) 2lc /, where represents the fraction of the free cross section area available for
the flow. For the mesh used in the experiments, = 0.46, lc = 0.14 mm and therefore,
Reef f = 9.13.

The second one is to determine the damping coefficient to be used in the simulation.
Exponential decay rates during cold flow conditions are determined as illustrated in
Appendix G. The decay rates for the first and second modes are 11.5 /s (at 156 Hz)
and 12.7 /s (at 312 Hz). The corresponding values of 1 & 2 (see Eqn. 3.23) are
0.023 and 0.013. For higher modes, the expression from Matveev and Culick (2003b)
is used, with C1 = 0.27 & C2 = 0.03 as that used in Chapter 3.

6.1 Comparison of theoretical and experimental bifur-

cation diagram

Figure 6.1 compares the bifurcation diagram from numerical simulations and exper-
iments. The control parameter is the non-dimensionalised K(Tw T0u )/T0u and the
Figure 6.1: Bifurcation diagram with the non-dimensional heater power K(Tw
T0u )/T0u as the control parameter, a) with out corrections and b) with cor-
rections for the heat transfer losses from Matveev (2003a). Solid lines and
triangles indicate the results from numerical simulations and experiments
respectively. The parameters for the experiments are xf = 0.125, Reef f =
9.13, N ud = 1.84 (1.83 from Collis and Williams 1959), kth = 2.4 102
w/m K (Kreith, 2000), T0u = 295 K, M = 5 104 , lc = 1.4 104 m,
Sc = 0.01 m2 .

representative variable is the rms value of the non-dimensional acoustic pressure, de-
fined as Prms |x=0.525 /M P , as t . The non-dimensionalised acoustic pressure is
same as pa defined in Chapter 3.

6.1.1 Nature of the bifurcation

Figure 6.1 (a) represents the comparison without any corrections for the heat losses in
the heater. There are two significant differences in the comparison. First, in the nature
of Hopf bifurcation. Theoretical predictions indicate a supercritical Hopf bifurcation,
while the experiments indicate a subcritical Hopf bifurcation. Furthermore, theoretical
Hopf point (Hth = 0.01) occur at lower values of heater power than the experimental
Hopf point (Hth = 0.09). This may be due to the heat loss by conduction and radiation
through the walls of the heater, subsequently to the tube walls. A correlation given by
from the heater to
Matveev (2003a), which relates the steady state heat transfer rate Q

171
the surrounding fluid as a functions of the supplied electrical power (Pelec ) and mass
flow rate m
is employed.


= h1 P h2 1 eh3 m
Q (6.1)
elec

where, Pelec is in Watts, m

is in g/s, empirical constants h1 = 2.07, h2 = 0.805 &
h3 = 1.01.

Bifurcation diagram, along with the modification for the heat loss is shown in Fig.
is
6.1(b). As expected, Hexp is shifted to the right as the effective heat transfer rate Q
used for the calculation of K. Further, the subcritical transition region shrinks with the
inclusion for the corrections for heat losses. This is due to the nonlinearity in the heat
loss (Eqn. 6.1).

6.1.2 Amplitude levels

Although the nature of bifurcation is not captured by the simulation, the amplitude
levels during instability are captured reasonably well Fig. 6.1(a). Further, with the
inclusion for the corrections for heat loss, the comparison deteriorates. The amplitude
levels are under predicted.

A possible reason for the discrepancy may be the following. Both the nature of
bifurcation and the amplitude levels of the oscillations are determined by the nonlin-
earities present in the system. In this case, nonlinearity arises from the unsteady heat
release rate term (Q, see Eqns. 3.21 & 3.24). A mesh type heater geometry is used in the
experiments (Fig. 5.3), while the simulations are performed for the unsteady heat trans-
fer over a single cylinder (Fig. C.1). Both the flow (Kang, 2003; Nicolle and Eames,
2011) and heat transfer (Stanescu et al., 1996; Mandhani et al., 2002) characteristics
are different for a single and rack of cylinders. A computation by considering a more
appropriate geometry might bring the match together.

172
 CHAPTER 7

Conclusions and outlook

The present thesis is focussed on the development of a theoretical framework to inves-

tigate the non-normal nature of thermoacoustic interaction and provide an experimental
evidence for the non-orthogonality of eigenmodes, along with the existence of transient
growth. Theoretical framework to analyse thermoacoustic systems are reformulated
to adapt the techniques of non-modal stability analysis developed already in hydrody-
namic stability theory. Analysis are performed in two thermoacoustic systems: 1. solid
rocket motor (SRM) and 2. Rijke tube. The following are the salient conclusions of the
doctoral investigations.

7.1 Conclusion

7.1.1 Non-modal stability analysis in a solid rocket motor

To begin with, the case of a SRM is particularly chosen, as orthogonality of the eigen-
modes (normal system) was assumed explicitly in the earlier investigations and some
of the crucial results regarding the stability were on the assumption of normal modes.
Along the axis, the steady state flow velocity in the combustion chamber increases due
to addition of mass along the length of the propellant. This gradient in the velocity ren-
ders the system non-normal. In addition to the acoustic equations, a governing equation
describing the dynamics of the burn rate further increases the non-normal nature of the
system. The optimum initial condition for maximum transient growth has significant
contributions from the degrees of freedom associated with the dynamics of the burn
rate. The amount of transient growth significantly reduces, when the dynamics of the
burn rate are modeled using a response function.

Two scenarios for pulsed instability are identified. In the first scenario, a large am-
plitude initial condition triggers the system to instability, which is the classical route. In
the second scenario, a small (relative to the earlier one) but finite amplitude initial condi-
tion along or close to the direction of the optimal initial condition, causes large transient
growth. The transient growth thus obtained, along with the nonlinearities present in the
system, leads to instability. This prediction may be supported by the experiments in
SRM performed by Blomshield et al. (1997a). The results indicate the occurrence of
pulsed instability by an initial perturbation of pressure amplitude, which is only 4% of
the steady state pressure in the system.

Although there are some indirect experimental observations from Blomshield et al.
(1997a), which could be related to the non-normal nature of thermoacoustic system,
there is no firm evidence. Measurements in SRM, are difficult to perform in consid-
eration of the harsh environment encountered. In most cases, only a few variables of
the system can be measured, which is not sufficient to characterize transient growth.
Hence, a simpler thermoacoustic system; electrically heated horizontal Rijke tube is
chosen for further analysis. In the past, apart from the non-normal nature, the coupling
used between the acoustic field and the heat source was ad hoc. It is necessary to estab-
lish the above coupling with mathematical rigor, as it forms the basis for the non-modal
stability analysis.

7.1.2 Asymptotic formulation of thermoacoustic interaction

In the effort towards theoretical analysis, the governing equations for the fluid flow be-
come stiff, as the Mach number ( 103 ) of the steady state flow and the thickness of
the heat source (compared to the acoustic wavelength) are small. As a consequence,
solving the system by computational fluid dynamics (CFD) technique is difficult. Alter-
natively, asymptotic analysis is performed in the limit of small Mach number and com-
pact heat source to eliminate the above stiffness problem. After applying this technique,
two systems of governing equations are obtained: one for the acoustic field and other
for the unsteady flow field near the heater, which is termed as the hydrodynamic zone.
By this way, the coupling between the acoustic field and the unsteady heat release rate
is established with mathematical rigor. Further, a non-trivial additional term, referred to
as the global-acceleration occurs in momentum equation of the hydrodynamic zone,

175
which has serious consequences for the stability of the system. It is observed that exclu-
sion of this term over predicts the stability of the system, as the coupling between the
acoustic field and the heat source is weakened. The global acceleration term appears
due to the two length scale nature of the Rijke tube system, which is generic in thermoa-
coustic systems. Moreover, numerical computations of response functions in the past
do not consider global-acceleration, which in turn leads to erroneous predictions of the
stability.

7.1.3 Transient growth and nonlinear instabilities

In non-modal stability theory, the amount of transient growth is measured using a scalar
(norm), which in general is related to the energy of the perturbations. There has been a
significant debate on the proper choice of perturbation energy. The current framework
based on asymptotic analysis allows one to obtain a norm systematically from the def-
inition of disturbance energy. In the present case, the disturbance energy defined by
Myers (1991) is used. It is shown further that the results vary by a fractional amount if
the energy defined by Chu (1965) is used as the disturbance energy. As the dynamics
of heat source is also included in the analysis, the number of degrees of freedom of the
system is large ( 104 ). In such cases, conventional method of obtaining the optimum
initial condition by singular value decomposition (SVD) is computationally expensive.
Alternatively, the technique of adjoint optimisation is used to determine the optimum
initial condition. As before in SRM, significant contributions (vortical structures) from
the degrees of freedom associated with the dynamics of the heat source is observed.

7.1.4 Experimental investigation of non-normality

Till now, the concept of non-normality and the associated transient growth from a theo-
retical point of view. In order to gain a more complete understanding, it is important to
perform experimental investigations. New experimental procedures have been devised,
which do not assume the conventional view of a normal system. For example, since
transient growth is a linear process, the threshold amplitude level below which the sys-
tem behaves almost linearly is identified. All experiments are performed such that the

176
amplitude levels are in the linear regime.

Acoustic pressure is measured along the length of the duct. The first two eigen-
modes, associated with the acoustic pressure are extracted by a recent data processing
technique, dynamic mode decomposition (DMD, Schmid 2010). The inner product
between the two eigenmodes, which measures the non-orthogonality is observed to
increase with increase in the power supplied to the heater, thus confirming the non-
normal nature of the thermoacoustic interaction. Furthermore, a two fold amplification
of acoustic energy is observed when the system is excited with an initial condition that
has contributions from the two eigenmodes. The results form the direct evidence for the
presence of transient growth. However, no such significant transient growth is observed
during the excitations with a single eigenmode, which is in line with the non-modal
stability theory. The present study represents the first experimental confirmation of
non-normality in thermoacoustic systems.

7.2 Scope for future work

As the present thesis is focussed both on theoretical and experimental investigations,

recommendations for the future work are made on the same classification. A schematic
representation of the current investigation and future outlook are shown in Fig. 7.1.

7.2.1 Theoretical investigations

In this thesis, it is shown in the context of Rijke tube that an efficient coupling between
fluid mechanics and acoustics arises when the associated length scales are disparate.
This disparity translates mathematically to the stiffness problem. Moreover, inclusion
of combustion process introduces a third length scale, which further increases the stiff-
ness. An extension will be to continue the asymptotic formulation to include flame -
acoustic interaction. A series of papers by Wu (Wu et al., 2003; Wu and Law, 2009; Wu
and Moin, 2010) describe the response of a premixed flame to inlet acoustic, vorticity
and enthalpy fluctuations using asymptotic analysis. In particular, one could identify
the generation of acoustic waves due to the fluctuations in the heat release rate from

177
Figure 7.1: An outlook and possible future extension of the present thesis. The plot
is similar to the one shown in Fig. 1.8 described in the introduction chap-
ter. The axis start from the current investigations and recommendations for
future work are arranged according to the expected difficulty level.

premixed flame, thereby closing the feed back loop.

Non-modal stability theory can then be applied and determination of various crucial
information such as, distribution of the optimum initial condition along various length
scales. These information will be of valuable input regarding the sensitive areas in the
flow field, which can be manipulated to prevent the occurrence of instabilities.

In experiments with dump combustors, the length of the flame is comparable to

the length of the duct during the normal operation. While during the onset of in-
stability, flame shrinks in length and becomes compact compared to the duct length
(Shreenivasan, 2008). Hence modelling multiple length and time scale problems in the
investigation of combustion instabilities is a crucial step. Such a generalized formu-
lation is currently under development and is applied to dump combustor (Balaji and
Chakravarthy, 2010). Application of perturbation methods to these problems will allow
to identify the mechanisms of instability in various configurations.

Another tool to determine the nonlinear characteristics of thermoacoustic interac-

tions is numerical continuation (Allgower and Georg, 2003). This method is superior
as it determines the complete bifurcation diagram at once, rather than from a num-
ber of simulations performed in the traditional time marching techniques. Numerical

178
continuation was applied to investigate thermoacoustic systems (Anathakrishnan et al.,
2005; Subramanian et al., 2010) and were largely restricted to low dimensional mod-
els. Recently, with introduction of matrix free methods, it became possible to apply
continuation to large degrees of freedom (Georg, 2010). Therefore, the accelerated
continuation methods can be adapted to investigate thermoacoustic systems. Using the
combination of system identification techniques and continuation, bifurcation diagrams
can be determined in a cost effective way for practical systems.

7.2.2 Experimental investigations

The non-orthogonality of the eigenmodes and the associated transient growth are shown
experimentally in the present investigation. As a next step, one could perceive to iden-
tify similar evidences in the context of flame - acoustic interaction. In the case of
the present Rijke tube system, no measurements associated with the dynamics of the
electrical heat source is performed. On the other hand, the sequence of image from a
premixed flame during oscillations, along with the measurement of acoustic variables
perhaps throw more insight into the non-normality of the system.

Another aspect that was given less importance during earlier investigation is the role
of initial conditions associated with the hydrodynamic zone. In the present experiment,
acoustic drivers are used to supply a precise initial condition and no provision is made
to provide an initial condition for the flow near the heater. To begin the investigation,
an initial condition in the form of vortex can be generated (Allen and Auvity, 2002) and
its role in the onset of instability can be probed.

Recently, it was shown that the trajectories of thermoacoustic oscillations are not
always limit cycles. Chaotic and quasi-periodic oscillations were observed during insta-
bility in a ducted premixed flame acoustic interactions (Kabiraj et al., 2011). Nonlinear
time series analysis (Abarbanel et al., 1993) can be used to investigate these oscillations
and thereby develop lower order models that display the above characteristics.

179
 APPENDIX A

Coupling terms in Eqns. (2.9) and (2.10)

The linear and nonlinear coupling terms used in Eqns. (2.9) and (2.10) are given below:
Z 
1
dU
1
In,m =
M U m cos(m x) + M sin(m x) km sin(m x) sin(n x)dx
0 dx
Z 1 2 
M U dU
2
In,m = cos(m x) m sin(m x) sin(n x)dx
0 dx
Z 1
In,m =
3
km U cos(m x) sin(n x)dx
0
Z 1
In,m =
4
km U sin(m x) sin(n x)dx
Z 10
5
In,m = M m cos(m x) sin(n x)dx
Z 1
0

2 dU
6
In,m = (M ) U cos(m x) M U m sin(m x) cos(n x)dx
2
0 dx
Z 1
In,m =
7
ke cos(m x) cos(n x)dx
Z 1 "X #
0
N XN
Nn1 = (Rm c
cos(m x) + Rms
sin(m x)) Uk sin(k x) sin(n x)dx
0 m=1 k=1
Z " #
1 X
N X
N
Nn2 = Um sin(m x) Uk k cos(k x) sin(n x)dx
0 m=1 k=1
Z " #
1 X
N X
N
Nn3 = Pm cos(m x) Pk k sin(k x) sin(n x)dx
0 m=1 k=1
Z " #
1 X
N XN
Nn4 = Um sin(m x) Pk k sin(k x) cos(n x)dx
0 m=1 k=1
Z " #
1 X
N XN
Nn5 = Pm cos(m x) Uk k cos(k x) sin(n x)dx
0 m=1 k=1

(A.1)
 APPENDIX B

Linearised equations governing thermoacoustic

interactions in SRM

The linearised evolution equation, corresponding to Eqns. (2.25, 2.26, 2.29 & 2.30) for
the coupled acoustic burn rate equation is

d
= L
dt

A2N 2N B2N (2N (Mg 1))
L=
C(2N (Mg 1))2N D2N (Mg 1)2N (Mg 1)
(2N Mg )(2N Mg )

1 2 1 2
I1,1 I1,1 I1,2 I1,2 . .

5
I1,1 /M I1,1
6
/M (N O + 1 ) /2 I1,2
5
/M 6
I1,2 /M . .


I2,1 1 2
I2,1 1
I2,1 2
I2,2 . .

5
I2,1 /M 6
I2,1 /M 5
I2,1 6
/M I2,2 /M (N O + 2 ) /2 . .

A = 2

. . .


. . .


.

. . . . . .
2N 2N

Ec c
E1,2 c
. . E1,N s
E1,1 s
E1,2 s
. . E1,N
1,1
c
E2,1 E2,N
c
. . c
E2,N s
E2,1 s
E2,1 . . s
E2,1

2m

B= . . .
1

.

. . . . . . . . . .
2N (2N (Mg 1))

c 3 c 7 s 4
Ej,n (1, 1) = Ij,n , Ej,n (2, 1) = Ij,n , Ej,n (1, 1) = Ij,n .
 c
All other entries are zero. Note that Ej,n s
and Ej,n are 2 (Mg 1) matrices.

S1


S2


S3


.


.  

2Bk
C= SN Sq (2, 2) = M F 1 B(1 k) +


0


0


0


.

.
(2N (Mg 1)2N )

Sq is (Mg -1)2N matrix. All other entries are zero.

G 0 0 . .
1

0 G2 0 . .
Dc
0
D= , Ds = Dc = .
Ds
0
. .

0 0 . . GN
(Mg 1)N (Mg 1)N

B1c
0 0 B2c 0 0 . .

1
H2 + H24 H22 H23 0 0 0 . .


H34 H31 H32 H32 0 0 . .


.
GN = F



. . .


. . . .


. . . 0

4 1 2 3
HM g 1
. . 0 0 HM g 1
HM g 1
HM g 1 (Mg 1)(Mg 1)

184
  2    2
k 2A 1 k
B1c = A(k 1) 1+ mB2 =
c
k 3
  !  2
2
i kq k 2 q kq kq
1
Hq = + Hq = 2
2
i1 2
 2 !
i kq kq k 2 q i 1/
Hq3 = Hq4 = mp q k
i+1 2 1

where q is the co-ordinate at qth discretised point. Note that is the same between
all successive grid points.

185
 APPENDIX C

Governing equations and boundary conditions involved

in the hydrodynamic zone:

An unsteady heat transfer problem is solved with the system of equations (Eqn. 3.24).
The heater in its primitive form is a thin wire filament wound around the heater frame.
The heater wire filament is arranged in the form of a rack (a schematic diagram of the
same is shown in Fig. C.1a). The typical spacing between the racks of the wire filament
is 50 times larger than the wire radius ( mm). Hence, a two-dimensional flow over
a single cylinder is considered, as shown in Fig. (C.1 (b). The heat transfer from the
single cylinder is multiplied by the effective length of the wire filament lw to obtain the
total heat transfer from the hydrodynamic zone.

Typical Reynolds number Red (= 2Rec ) of the base flow, based on the diameter
of the wire filament is around 20. Also, it is observed from Fig. 3.3 (a) that the max-
imum non-dimensional acoustic velocity at the heater location uf is approximately
four. The fluctuating flow over the heater wire experiences a freestream flow with a
maximum flow velocity of five times the base flow during one cycle. Hence, the maxi-
mum Reynolds number (Rem ) to be obtained is 100 during one cycle. Sarpkaya (1986)
experimentally obtained the condition for the above oscillatory flow to become unstable
and shed vortices based on the non-dimensional number; Keulegan-Carpenter number
Kc (Kc = Um T /D) for a given viscous scale parameter ( = Rem /Kc ), where Um is
the maximum velocity of the free stream encountered during a cycle and T is the time
period of oscillation. For the present problem (parameter values from table 4.1), the
values of the above non-dimensional numbers are Kc = 1.25 and = 80. For = 80,
the critical value of Kc above which the oscillatory flow over the cylinder becomes un-
stable to shed vortices is 2.11 (Sarpkaya, 1986), which is larger than that investigated
in the present case. Hence vortex shedding does not occur. Owing to this, only one half
of the flow domain is considered for the present simulation and the symmetry boundary
condition is enforced.
Figure C.1: Schematic representation of the heat source, (a) rack of the heater wire
filament (b) boundary conditions for the two dimensional flow over a heated
circular cylinder.

Figure C.2: Grid generated (121 101) for flow in the hydrodynamic zone (a) Physical
domain with grid clustering near the cylinder surface with k = . Flow
domain is shown only up to r = 25 so that the presence of the cylinder can
easily be visible in the figure. Numerical simulations are performed for the
domain size r = 50. Convergence tests are performed and it is found that
there is less than 5% change in the results with the domain size for r = 50,
(b) Computation domain with uniform grids.

188
Figure C.3: Staggered grid arrangement, indicating density, pressure, temperature and
velocity nodes. (a) Physical domain, (b) Computational domain.

Figure C.4: (a-b) r direction velocity component discretisation stencil. (c-d) di-
rection velocity component discretisation stencil. (a & c) Physical domain,
(b & d) Computation domain.

189
 A plane polar coordinate system is used to implement the no-slip boundary condi-
tion on the surface of the wire filament (circular cylinder). Moreover, fluid viscosity and
thermal conductivity are assumed to be independent of temperature. The flow domain
and the boundary conditions are shown in Fig. (C.1). The switch from Dirichlet to Neu-
mann boundary condition at = /2 (from upstream to downstream in Fig. C.1) for
up , Tp is to have the numerical solvability of the hydrodynamic equations. The above
switch is performed in Abu-Hijleh (2003) for a similar problem. The Dirichlet boundary
condition (specified by the upstream acoustic zone) is applied in the farfield upstream
boundary and the Neumann boundary condition is applied in the farfield downstream
boundary for p , up and Tp . The average of the above flow variables in the downstream
boundary of the hydrodynamic zone is used in the one dimensional acoustic zone down-
stream of the heat source.

Near the cylinder surface, the gradient is large and it is important to cluster more
number of grids near the cylinder surface. Therefore, grid clustering is incorporated
by the following transformation r = ek , where k is the grid clustering parameter,
which determines the rate at which the grid grows as one moves away from the cylinder
surface. A uniform grid in , plane (computational plane) will give a stretched grid
in r, plane (physical plane) with more grids clustered around the cylinder surface.
Fig. C.2 (a) shows the generated grid for the present problem, where the grid size near
the cylinder surface is very small compared to the far field and the grid size increases
exponentially as r increases. The corresponding grid in the computational domain is
uniform and is shown in Fig. C.2 (b). The system of equations (3.24) with the above
transformation is solved by using semi-implicit method for pressure linked equation
(SIMPLE) algorithm (Patankar, 1980) with fast Poisson solver (Press et al., 2007) for
solving the pressure correction equation in the SIMPLE algorithm.

Staggered grid arrangement is implemented in the present computation as recom-

mended by Patankar (1980) and is shown in Fig. C.3. The location indexed as (rd , d )
represents a grid point. The grid point is shifted by half grid length to obtain the stag-
gered grid point, which is indexed as (rds , ds ). The values of the variables associated
with density, pressure and temperature are stored in the node, which is in the middle
of the primitive cell (ABCD) as shown in Fig. C.3. On the other hand, the velocity

190
 Property Red = 20 Red = 30

Recirculation zone length (in terms of 2lc ) 1.24 (1.15) 1.54 (1.54)
Separation angle from trailing edge (in radians) 0.79 (0.78) 0.87 (0.87)
Nusselt number 2.50 (2.40) 2.98 (2.83)

Table C.1: Comparison of various steady state flow properties between the present sim-
ulation and experiments. Numerical values in parentheses indicate val-
ues obtained from experiments. Experimental results for the recircula-
tion zone length and the separation angle are obtained from Coutanceau
and Bouard (1977), whereas
 2 the Nusselt number, defined as N ud =
u R  
2T0
2(T Tu )
T
r
p
d is obtained from Collis and Williams
w 0 r=1
0
(1959).

Figure C.5: Response of the unsteady heat transfer from the heated cylinder for the
forcing of the freestream velocity, up (r , /2 < < ) = 1 +
0.1 sin(ta /2.43). The parameters are, Red = 10, Tu = 295 K, Tw =
700 K.

191
components are stored in the edges of the cell. The continuity (3.24a) and energy (4.1e)
equations are discretised with its centre as the grid point (rd , d ), whereas, the momen-
tum equation (3.24b) in the r and direction are discretised around (rds , d ) and
(rd , ds ) nodes. This is shown in Fig. C.4 for both physical and computational domain.
First order up winding scheme is used for the continuity equation, while second order
central difference scheme is used for the rest.

The results in the present numerical simulation (without the global-acceleration

term) of the hydrodynamic zone are compared with the existing results for validation.
The steady state properties are compared in table (C.1) with experimental results from
Coutanceau and Bouard (1977) and Collis and Williams (1959). A good agreement
(with less than a 10% difference) is observed for the steady state properties. The re-
sponse of the unsteady heat transfer from the heated cylinder to sinusoidal forcing is
compared with the numerical simulation performed by Apelt and Ledwich (1979) and
is shown in Fig. C.5. A reasonable agreement in the response of the system is ob-
served. The constant density formulation used by Apelt and Ledwich (1979) might be
the reason for the difference in the above two responses of the unsteady heat release
rate.

192
 APPENDIX D

Steps involved in obtaining adjoint equations:

A brief explanation of the steps involved in deriving the adjoint equations (Eqn. 4.26),
the corresponding boundary conditions (Eqns. 4.27-4.28) and the mapping (4.32 &
4.33) between the direct and adjoint equations are given as follows. The adjoint equa-
tions and the corresponding boundary conditions are obtained when the first variation of
the Lagrangian () with respect to the direct variables are set to zero (see Section 4.23).
There are various terms in and the structure of these terms are fewer in number. One
representative member is chosen for each structure and the first variation of the same
member with the direct variables is performed. The first variation thus obtained for the
above member is set to zero (which gives the adjoint system) and the contributions of
the same member to the adjoint equations and the boundary conditions are analysed.
Let represent any one of the direct variables (ua , pa , u, v....) and represents the
adjoint variable corresponding to the direct variable .

D.1 Acoustic zone

D.1.1 Algebraic term

The following generic algebraic term is considered from the expression for Lagrangian
(Eqn. 4.25) in the acoustic zone.

D E ZTopt Z1
, =
dxdt (D.1)
a
t=0 x=0

Taking first variation of the above expression with respect to gives the following

  ZTopt Z1

, =
dxdt (D.2)
a
t=0 x=0
The used in the above expression indicates the variation of the variable . Since in
the right hand side of the above expression is arbitrary, the first variation of the above
expression vanishes when the integrand itself vanishes. Similar terms from which
have the same form as that of the above expression are gathered and the final integrand
obtained is equated to zero. Thus the right hand side of the above expression is one of
the term, which contributes to the adjoint equation (Eqn. 4.26). Other terms in , which
has gradient and laplacian forms are discussed below. The first variation of these terms
generate terms, which contribute to the boundary conditions (Eqns. 4.27-4.28) for the
adjoint equation and the mapping between the direct and adjoint variables (Eqns. 4.32
& 4.33).

Gradient term

A generic gradient term has the following form in .

  ZTopt Z1

, = dxdt (D.3)
t a t
t=0 x=0

Taking first variation with respect to as before in Section D.1.1, one obtains the fol-
lowing:
    ZTopt Z1

, = dxdt (D.4)
t a t
t=0 x=0

Now integration by parts is performed for the variable /t so as to move the operator
The expression thus obtained is as follows.
/t from to .

    Z1 Topt ZTopt Z1

, = 
dx dxdt (D.5)
t a t=0 t
x=0 t=0 x=0

The first integral in the right hand side of the above expression contributes to the map-
ping between the direct and adjoint variables (Eqns. 4.32 & 4.33) and the second inte-
gral contributes to the adjoint equations (Eqn. 4.26). In the above case, gradient in t is
considered. Similarly the first variation of the terms with gradient in x can be performed
with one difference. The difference is that the surface terms obtained after performing

194
integration by parts (like the first integral in Eqn. D.5) contribute to the boundary condi-
tions (Eqns. 4.27-4.28) for the adjoint equations. It is important to note that the second
integral in (D.5) has the same form as the integral in (Eqn. D.4) with a sign difference.
This implies the fact that the gradient term is non-self adjoint. The same conclusion
about the non-self adjoint nature of the gradient operator is applicable to the gradient
terms in the governing equations (Eqn. 4.2) of the hydrodynamic zone. In the present
paper, the gradient terms in the equations (Eqn. 4.2) governing the hydrodynamic zone
contribute to the non-normal nature of the system. The higher order derivatives in space
(e.g. Laplacian) generate more surface terms, which again contribute to the boundary
conditions.

D.2 Hydrodynamic zone

D.2.1 Laplacian term

In this section, the first variation of a term in involving Laplacian operator in the
hydrodynamic zone is performed and is given by the following.

  ZTopt Z Z  2 

1 1 2

, =
2
+ + 2 rdrddt (D.6)
h r2 r r r 2
t=0 =0 r=1

As before in Section D.1.1, integration by parts is performed on the above expression

leaving the expression in the following form:

  Z Topt Z !


r 
2 , =
r
+  ddt
h t=0 r r 
=0 r=1

Zopt Z
T !
1 

+  rdrdt (D.7)
r r 
t=0 r=1 =0

ZTopt Z Z  
2
+ rdrddt
t=0 =0 r=1

The first two integrals in the right hand side of the above expression contributes to

195
the adjoint boundary conditions and the last integral contributes to the adjoint equation.
Note that the last integral in (Eqn. D.7) has the same form as the integral in (Eqn. D.6).
This is due to the fact that Laplacian 2 is a self-adjoint operator. In this way Eqns.
(D.1.1, D.1.1 & D.2.1) the first variation of all the terms in the expression for the first
variation of with respect to the state space variables of the system are evaluated and
the adjoint system is obtained.

196
 APPENDIX E

Detailed drawing of the Rijke tube experimental setup

A detailed drawing for the configuration of the Rijke tube setup shown in Fig. 5.1 is
presented in this section. The material used for making the major elements in the setup
are tabulated in Table E.1.

Description Material

Rijke tube Aluminium

Decoupler Mild steel
Heater stand Cynthanium
Heater mesh Stainless steel
Electrical leads Copper
Insulation Glass fibre

Table E.1: Materials used to make the major components of the Rijke tube setup.
Figure E.1: Detailed drawing for the Rijke tube along with provisions for instrumenta-
tions; a) isometric, b) Orthographic view. Microphones and thermocouples
are mounted in the small holes, while acoustic drivers are mounted in the
large holes. All dimensions are in mm.

198
Figure E.2: Detailed drawing for the decoupler (orthographic projections). All dimen-
sions are in mm.

199
Figure E.3: Detailed drawing for the heater stand; a) isometric, b) Orthographic view.
All dimensions are in mm.

200
Figure E.4: Detailed drawing for the heater mesh brazed with the copper rod. All di-
mensions are in mm.

201
 APPENDIX F

Calibration of microphones

A total of 15 microphones are used to track the evolution of acoustic pressure along
the length of the duct. It is important to calibrate them for their relative sensitivity and
phase in the frequency of interest. Towards this purpose, one microphone, model - PCB
103B02 SN 5236 is chosen as the reference, which has a sensitivity of 223.65 mV /kP a.
A schematic of the calibration setup is shown in Fig. F.3 (a). A calibration tube of length
0.9 m and internal diameter (dcal ) of 0.11 m is used as an acoustic resonator. The above
tube is used to reduce the effect of ambient noise in the measurement. In the present
case, the cutoff frequency (fcut ) for the propagation of a non-planar mode (radial mode)
is 1.8 kHz (fcut = 1.84a/(2dcal /2), referred from Kinsler et al. 2000). The frequency
of interest in the present investigation is in the range 100500 Hz, which is much below
fcut and hence only longitudinal modes propagate in the duct. Four microphones are
placed on the end plate of the calibration tube and the arrangement is shown in Fig.
F.3 (b). A reference microphone is placed at one of the ports and the rest of the three
ports are used to calibrate other microphones. An 18 extended low frequency Ahuja
acoustic driver (L18-SW650) is used for exciting the system.

The calibration is performed as follows. At a given frequency, the system is excited

continuously and the acoustic pressure is recorded by the microphones. The ampli-
tude of pressure oscillations from the microphones are obtained and compared with the
reference. Further, the phase difference between a microphone measurement and the
reference is also noted. The experiment is repeated for various frequencies in the range
100 - 500 Hz, which is of interest in the present investigation (see Section 5.4.5). A
typical calibration data is shown in Fig. F.3 for the microphone 103B02 SN 5235. The
amplitude ratio is defined as Acalib = Am /Aref , where Am & Aref represent the ampli-
tudes measure from the above microphone and the reference. The mean value is used
to obtain the relative sensitivity of the microphone. It is observed from Fig. F.3 (a),
that the variation of Acalib = Am /Aref in the frequency of interest is less than 4% of the
Figure F.1: (a) Schematic representation of the microphone calibration setup. (b) Mi-
crophone arrangements for calibration. All dimensions shown are in mm.

204
Figure F.2: Image showing two models of microphones, 103B02, 377B10 used for the
measurement of acoustic pressure.

Figure F.3: Calibration data for a microphone, 103B02 SN 5235 in the frequency range
50 500 Hz. Calibration is performed twice to ensure repeatability and are
shown as cross and hollow circular symbols. (a) Amplitude ratio Acalib =
Am /Aref , Am & Aref represent the amplitudes measure from the above
microphone and the reference. (b) Phase difference between the microphone
and the reference.

205
 No. location (mm) Model number Serial number (SN) Acalib = Am /Aref

1 0.035 377A50 41064 4.79

2 0.070 377A50 41061 4.02
3 0.010 377B10 111212 3.65
4 0.150 103B02 5246 1.03
5 0.225 103B02 5234 1.00
6 0.300 103B02 4753 0.91
7 0.375 103B02 5245 0.94
8 0.450 377B10 111206 3.72
9 0.525 377B10 111213 3.56
10 0.675 103B02 5238 0.99
11 0.750 103B02 4752 0.95
12 0.800 103B02 5236 (reference) 1.00
13 0.825 377B10 111207 3.62
14 0.900 103B02 5235 1.02
15 0.965 377B11 426E01 23.81

Table F.1: Calibration data, Acalib = Am /Aref for all the microphones used in the
present investigation. All the microphones used belongs to PCB SN se-
ries. The model number along with the location in the Rijke tube are also
shown. The sensitivity of the reference microphone, 103B02 SN 5236 is
223.65 mV /kP a with 1% uncertainty (Piezotronics, 2008).

mean value. Moreover, the phase difference between the above microphone and the ref-
erence is also less than 4 degrees. The above exercise is repeated for other microphones
and the obtained values of Acalib = Am /Aref are tabulated in table F.1. Since the phase
difference is small, the same is neglected. The calibration data of the microphones are
used further in the analysis to obtain the acoustic pressure values.

206
 APPENDIX G

Measurement of acoustic damping

In order to ensure repeatability of the experimental results, the acoustic damping in the
system should be estimated and monitored in the subsequent realisations. The experi-
ments are performed at a prescribed ambient condition (see Section 5.4). The exponen-
tial decay rate of the system is determined for a given frequency as follows.

In the cold flow, the system is excited for a brief period of time at the frequency
of the first eigenmode, 156 Hz. After the acoustic driver is switched off, the system
decays according to the decay rate corresponding to the first eigenfrequency. A typical
acoustic pressure data at x = 0.525 is shown in figure G.1 (a). Data used for the
determination of acoustic damping is shown in light grey region. The instantaneous
amplitude of the signal is obtained by the absolute value of the Hilbert transform of
the signal (King, 2009). The evolution of the logarithmic decay of the instantaneous
amplitude, defined as log (|H(P (t))|/|H(P (tst ))|) is plotted in figure G.1 (b). tst is the
time at the start of the light grey region. The straight line part of the curve indicates the
exponential decay of the instantaneous amplitude, whose slope gives the decay rate ()
of the system. In this case, the value of is 11.5 /s. This indicates that the signal
drops to 1/e of its original value after approximately 13.5 cycles. Data points marked in
figure G.1 (a) are separated by 13.5 cycles and their ratio is 0.34, which is close to 1/e.

Decay rates are measured at the start of the experiments. The experiments are further
continued only when the decay rates are within 5% of the value specified earlier. The
uncertainty of the microphones in the measurement of acoustic pressure in the frequency
of interest, 50-500 Hz is 1% (Piezotronics, 2008). This translates to an uncertainty
of 2% in the determination of damping rate. The above procedure is to ensure that
acoustic damping does not vary significantly during the course of the experiments.
Figure G.1: Estimation of acoustic damping in the system. (a) Acoustic pressure mea-
sured at x = 0.525, which is used for damping measurement. Dark grey
region indicates the period during which the acoustic driver is switched
on at a frequency of 156 Hz. Data used for the determination of acoustic
damping is shown in light grey region. (b) Determination of the decay rate
of the system by plotting the evolution of the logarithm of the instantaneous
amplitude ratio (log (|H(P (t))|/|H(P (tst ))|)). The exponential decay rate
() is obtained by the slope of the above curve.

208
 APPENDIX H

Experimental determination of the linear regime

In order to identify the linear regime for our experimental setup, it is important to first
characterize the acoustic drivers regarding their linearity. The equations governing the
classical acoustic field in a duct are linear (Rienstra and Hirschberg, 2008). The Rijke
tube system is continuously excited at a given frequency using acoustic drivers, with
the heater and blower switched off, and the response of the system is recorded by the
microphones. The voltage supplied to the acoustic driver is also noted and varied. This
way, a response curve for the acoustic driver unit is obtained; it is shown, for three
forcing frequencies, in Fig. H.1. On the horizontal axis, the rms-value of the voltage
supplied to the acoustic driver units is shown; on the vertical axis, the rms-value of
the acoustic pressure fluctuations (Prms |x=0.525 ) is displayed. The symbols indicate
the experimental data, the lines present a linear fit. From the response curve, one can
conclude that the linear fit represents the experimental data with less than a 0.25%
spread. Hence, for the voltage range covered in Fig. H.1, the relation between the
voltage supplied to acoustic driver and the generated acoustic pressure can be assumed
linear.

The heater and blower are now turned on. Due to reasons explained in 5.4.1, the
linear regime is identified only within the zone GH. Fig. H.2 presents the response of
the system to continuous excitation with the acoustic driver, where the output is given
by the rms-value of acoustic pressure fluctuations. As before, the symbols indicate the
experimental data. It can be noted that the symbols do not form a straight line; rather,
a sublinear relation for larger amplitudes of the voltage can be detected. The dashed
line represents a power-law fit to the experimental data (with a spread of 0.15%). A
linear fit, shown as a continuous line, is performed on part of the experimental data for
low amplitudes of VAcoustic driver (in our case less than 3 V ). Additionally, the triggering
amplitude is marked as a dash-dotted horizontal line. The limit of linearity is then
determined as the point when the deviation of the experimental data from the linear fit
Figure H.1: Characteristics of the acoustic driver unit, obtained for three different fre-
quencies. The rms-value of the voltage supplied to the acoustic driver unit
is indicated on horizontal axis. The symbols represent the experimental
data, while the continuous line presents a linear fit of the data.

Figure H.2: Identification of the linear regime. The acoustic forcing is conducted at
300 Hz with K = 764 W and m = 2.34 g/s. A power law is used to fit the
response of the system. The resulting equation for the fit reads 28.36Vp0.67
14.24. The shaded area indicates the linear regime.

210
exceeds 5% of the corresponding triggering amplitude. The amplitude, at which the
nonlinear nature of the system comes into play, is given by the triggering amplitude;
hence, it is taken as the reference quantity. The linear regime defined above is indicated
in grey in Fig. H.2.

211
 APPENDIX I

Dynamic mode decomposition

Dynamic Mode Decomposition (DMD) is a data-based technique that extracts dominant

dynamic features from a time-resolved sequence of flow field measurements (Schmid,
2010). The identified dominant flow structures are termed dynamic modes; data from
numerical simulations and experiments can be processed. This process is equivalent to
performing a global stability analysis. For a linear system, the dynamic modes represent
the global modes (eigenmodes). In case of a nonlinear flow, the results represent a linear
tangent approximation of the underlying nonlinear flow.

A schematic illustration of the application of DMD technique is shown in Fig. I.1.

A set of temporal snapshots, v1 , v2 , ....vN are arranged as column vectors to construct
a matrix V1N . In turn each vi contains the entries corresponding to the measurements
performed at various spatial locations at the ith time instant. For example, in the present
case, each vi contains 15 entries from the measurement of acoustic pressure. The tem-
poral snapshots are equally spaced, with a time interval t.

From the input data, DMD technique extracts the eigenvalues and eigenvectors
(eigenmodes) of a linear mapping A, defined as follows.

vi+1 = Avi (I.1)

Figure I.1: Schematic illustration of the application of DMD technique.

Matrix A is obtained such that the error in fitting the above equation for all the snapshots
is minimum in a least square sense. In dynamical systems, A is known as the propagator,
which takes the system from ith to i + 1th state (Strogatz, 2000). For a linear system,
the expression for A is the following.

A = eLt (I.2)

where L represents the same linear operator defined in the Eqn. (1.1, Chapter 1).

I.1 Determination of the eigenmodes

As the experiments are performed in the linear regime, we will continue with the va-
lidity of Eqn. (I.3). As mentioned earlier, DMD technique extracts the eigenvalues
and eigenvectors of A. Let them be indexed as m and m respectively. The number
of eigenvectors is equal to the number of spatial measurement locations. Further, if
represent the matrix of the m arranged as the column vector and represent the
eigenvalue matrix, then the following eigenmode decomposition holds.

A = 1 (I.3)

Repeating the eigenmode decomposition for the matrix L leads to the following
equation.
L = 1 (I.4)

where, and represent the matrix containing the eigenvalues and eigenvectors of L
respectively.

Multiplying Eqn. (I.4) by t and exponentiating leaves to

1 t
eLt = e (I.5)

Further, expanding the right hand side of this equation in power series, followed by

214
some matrix manipulations leads to the following equation.

A = eLt = 1 et (I.6)

Comparing Eqns. (I.6) and (I.3), we obtain the following relations.

log ()
= (I.7a)
t

= (I.7b)

Using these, the eigenvalues and eigenmodes of the system are determined.

215
 APPENDIX J

Application of DMD in the presence of noise

The application of the DMD technique to determine the eigenmodes of the system from
synthetic data by solving Eqn. (5.1) is illustrated in Section 5.3.3. Good agreement
between the eigenmodes extracted by the DMD technique and the true eigenmodes is
observed. Applying the DMD technique to experimental data warrants an estimation of
robustness of the DMD algorithm in the presence of noise and of the uncertainty in the
relevant output quantities.

As a first step, the noise in the system has to be characterized. Fig. J.1 presents an
estimate of the noise level in the system. The noise levels measured at locations x = 0.1
and x = 0.8 are shown in Figs. J.1 (a, c). The input waveforms at the same locations,
used to perform DMD, are shown in Figs. J.1 (b, d). These waveforms are mostly
aligned with the second eigenmode of the system. The horizontal line indicates the
rms-level of the corresponding signal which is used to measure the signal strength. The
noise level is then estimated as the ratio of the rms-level of the noise to the rms-level
of the signal at the locations of the microphones. The resulting noise-level data, plotted
along the duct, is shown in Fig. J.1 (e). The maximum noise level is determined as 4.2%
and occurs at x = 0.525. Since the second eigenmode of the system is excited, the above
location is close to the acoustic pressure node and hence has the highest relative noise
level. Similarly, the relative noise level is determined for computing the first eigenmode
(see Fig. J.1f ).

The power supplied to the heater in our case is K = 747W , which is in the subcrit-
ical transition zone (see Fig. 5.12). The rms-amplitude of the signals in Fig. J.1 (b, d)
are smaller than the rms-amplitude (87 P a) of the linearity limit. Since the experiment
is performed for lower values of K, acoustic pressure data with a larger signal-to-noise
ratio can be used as the amplitude of the linearity limit is higher. In the present experi-
ments, the maximum value of K used is 747 W (Fig. 5.13). Hence, the maximum noise
level encountered in the determination of eigenmodes of the system is 4.2%.
Figure J.1: Estimation of noise levels encountered in the experiments. (a) and (c) noise
levels measured at locations x = 0.10 and x = 0.80. (b) and (d) the in-
put waveform measured at the same locations, used to perform DMD. The
horizontal line indicates the rms-level of the corresponding signal. The rel-
ative noise level is estimated as the ratio of the rms-level of the noise to
the rms-level of the signal at the microphone locations. Results are shown
for determining the (e) second and (f ) first eigenmode of the system. The
remaining parameters are m = 2.34g/s and K = 747W.

218
Figure J.2: Illustration of the robustness of the DMD algorithm in the presence of noise.
(a) Histogram plot based on 20 bins for the value of the inner product | <
pi , pj > | from 200 realizations with random noise added to the experimental
data. The added noise level is equivalent to the one shown in Fig. J.1 (e,f ).
(b) Probability density function (PDF) for the above histogram. The cross
symbols indicate the histogram data, and the continuous line indicates a
Gaussian distribution fit. The equation of the fit is also indicated. The mean
value xm of | < pi , pj > | is 0.127, and the standard deviation is 3.124
103 . The percentage error associated with | < pi , pj > | is evaluated
as /xm 100 = 2.46. This value is used as the amplitude of the error
bar (Holman, 1994) indicated in Fig. 5.13d. The governing parameters are
the same as in Fig. J.1.

The uncertainty in the inner product | < pi , pj > |, which determines the non-
orthogonality of the eigenmodes, is determined as follows. For a conservative estimate,
noise at levels shown in Figs. J.1 (e, f ) is added numerically to the data set obtained
from the experiments before the DMD is performed. The data provided to the DMD
are first projected onto the proper orthogonal decomposition (POD) modes to reduce
the effect of noise (Negrete et al., 2008) in the determination of the eigenmodes. A
lower cut-off is defined on the number of POD modes (based on their energy content)
to be used for the projection (Schmid, 2010). In our case, the cut-off is set as 10 for
the application of DMD on the downstream side of the heater, while no such cut-off is
set for the upstream side. Several numerical realizations (in the present case, 200) are
performed and the values of | < pi , pj > | are recorded.

The obtained values of | < pi , pj > | are distributed over 20 bins of size 2 103 for

219
the data range 0.114 0.134. Fig. J.2a shows the histogram for the above realizations.
The histogram is then normalized by the total number of realizations to obtain the prob-
ability density function (PDF) for | < pi , pj > |, see Fig. J.2b. A normal distribution
3 2
is fitted to the PDF and, for our example, is of the form 127.7e((xp 0.127)/3.12410 ) /2
where xp represents the stochastic variable | < pi , pj > |. From the above distribu-
tion, the mean xm and standard deviation are found to be 0.127 and 3.124 103 ,
respectively. The relative normalized standard deviation (in percent) is determined as
/xm 100 = 2.5 and is used as the percentage amplitude of the error bar (Holman,
1994) indicated in Fig. 5.13 (d). The above spread in the value of | < pi , pj > | is
indicated as shaded region in Fig. J.2 (b).

220
 APPENDIX K

Determination of acoustic velocity using two microphone

technique

Two microphone technique uses acoustic momentum equation to determine acoustic

velocity from the spatial measurement of acoustic pressure (Bellows, 2006). One di-
mensional acoustic momentum equation is shown below.

u p
s = (K.1)
t x

The value of the spatial gradient (p/x) is determined using central difference scheme
from the microphone measurements. Further, the largest spacing between any two mi-
crophones in the present experiment is 75 mm and the cross-sectional dimension is 92
mm, which allows us to measure acoustic velocity fluctuations in the frequency range
100 Hz to 1.76 kHz (Abom and Boden, 1988). For our interest in the first ( 175Hz)
and second ( 350Hz) eigenmodes of the system, the inter-microphone spacing is suit-
able for our experimental investigation. Integrating Eqn. K.1, one obtains an expression
for the acoustic velocity (u(x, t)).
Z t
1 p
u(x, t) = dt + u0 (x) (K.2)
s 0 x

where, s represents the steady state flow density. In the present investigation, acoustic
pressure measurements are performed from the steady state. Hence it is possible to
assume u0 (x) = 0. Time integration is performed numerically by Simpsons 1/3rd rule.
The acoustic velocity field thus obtained is used to calculate the acoustic energy in the
system.

Fig. K.1 shows the evolution of the calculated acoustic velocity. Figure. K.1 (a) is
same as Fig. 5.19 (a), which is redrawn to emphasize the evolution of u(x, t) at various
zones. Acoustic pressure and velocity are non-dimensionalised by M Pu and uu re-
spectively. uu and M represent the steady state velocity and Mach number upstream of
Figure K.1: Evolution of (a) acoustic pressure P (x = 0.525) for the case, when the sys-
tem is triggered by external forcing. Evolution of (b & d) non-dimensional
acoustic pressure, P (x = 0.525)/(M Pu ) and (d & e) acoustic velocity
during triggering (initial phase of zone B) and limit cycle (zone D) respec-
tively. The governing parameters are A2 /A1 = = 0o , K = 782 W ,
m = 2.34 g/s, uu = 0.24 m/s, Tu = 298 K, M = 6.9 104 and
Pu = 1.01235 105 P a.

222
the heat source respectively. Pu represent the steady state pressure. Figures. K.1 (b &
c) represent the evolution of the non-dimensional acoustic pressure and velocity during
triggering i.e., in the phase of zone B. It is observed that the amplitude of the acoustic
velocity required for triggering is around 30% of the steady state value. Further during
limit cycle, the amplitude of the acoustic velocity is 1.25 times the steady state value,
indicating a complete flow reversal occurring at part of a cycle (Fig. K.1e).

223
 REFERENCES
1. Abarbanel, H. D. I., R. Brown, J. J. Sidorowich, and L. S. Tsimring (1993). The
analysis of observed chaotic data in physical systems. Review of Modern Physics, 65,
13311392.

2. Abdessemed, N., A. S. Sharma, S. J. Sherwin, and V. Theofilis (2009). Transient

growth analysis of the flow past a circular cylinder. Physics of Fluids, 21(4), 044103.

3. Abom, M. and H. Boden (1988). Error analysis of two-microphone measurements in

ducts with flow. The Journal of the Acoustical Society of America, 83(6), 24292438.

4. Abu-Hijleh, B. A. (2003). Numerical simulation of forced convection heat transfer

from a cylinder with high conductivity radial fins in cross-flow. International Journal
of Thermal Sciences, 42(8), 741 748. ISSN 1290-0729.

5. Agharkar, P., P. Subramanian, N. S. Kaisare, and R. I. Sujith, Thermoacoustic in-

stabilities in a ducted premixed flame: Reduced order models and control. In European
Combustion Meeting. London, 2011a.

6. Agharkar, P., P. Subramanian, N. S. Kaisare, and R. I. Sujith, Control of thermoa-

coustic instabilties in a ducted premixed flame. In 18th International Congress on Sound
and Vibration. Rio de Janeiro, Brazil, 2011b.

7. Allen, J. J. and B. Auvity (2002). Interaction of a vortex ring with a piston vortex.
Journal of Fluid Mechanics, 465, 353378.

8. Allgower, E. L. and K. Georg, Introduction to Numerical Continuation Methods.

SIAM, Philadelphia, 2003.

9. Anathakrishnan, N., S. Deo, and F. E. C. Culick (2005). Reduced-order modeling and

dynamics of nonlinear acoustic waves in a combustion chamber. Combustion Science
and Technology, 177, 221247.

10. Anderson, J. D., Computational fluid dynamics: the basics with applications.
McGraw-Hill, New York, 2001, 8th edition.

11. Andres, J. M. and U. Ingard (1953). Acoustic streaming at low Reynolds numbers.
Journal of Acoustical Society of America, 25, 932938.

12. Annaswamy, A. M., M. Fleifil, J. P. Hathout, and A. F. Ghoneim (1997). Impact

of linear coupling on the design of active controllers for the thermoacoustic instability.
Combustion Science and Technology, 128, 131180.

13. Anthoine, J., J. M. Buchlin, and A. Hirschberg (2002). Effect of nozzle cavity on
resonance in large SRM: Theoretical modelling. Journal of Propulsion and Power, 18,
304311.

225
14. Apelt, C. J. and M. A. Ledwich (1979). Heat transfer in transient and unsteady flows
past a heated circular cylinder in the range 1 Re 40. Journal of Fluid Mechanics,
95(4), 761777.
15. Baggett, J. S., T. A. Driscoll, and L. N. Trefethen (1995). A mostly linear model of
transition to turbulence. Phys. Fluids, 7(4), 833838.
16. Balaji, C. and S. R. Chakravarthy (2010). A simultaneous multiple space-/time-scale
formulation of fluid flow problems with application to combustion thermoacoustics. n3l
- Intl Summer School and Workshop on Non-Normal and Nonlinear Effects in Aero-
and Thermoacoustics, May 17-21, 2010, Munich.
17. Balasubramanian, K. and R. I. Sujith (2008a). Non-normality and nonlinearity in
combustion acoustic interaction in diffusion flames. Journal of Fluid Mechanics, 594,
2957.
18. Balasubramanian, K. and R. I. Sujith (2008b). Thermoacoustic instability in a Rijke
tube: Non-normality and nonlinearity. Phys. Fluids, 20, 044103.
19. Barbagallo, A., D. Sipp, and P. J. Schmid (2009). Closed-loop control of an open
cavity flow using reduced-order models. Journal of Fluid Mechanics, 641, 150.
20. Barkley, D. and L. S. Tuckerman (1999). Stability analysis of perturbed plane Couette
flow. Phys. Fluids, 11, 11871195.
21. Baum, J. D. and J. N. Levine (1986). Modelling of nonlinear longitudinal instability
in solid rocket motors. AIAA. Journal, 13, 339348.
22. Bellows, B. D. (2006). Characterization of Nonlinear Heat Release-Acoustic Inter-
actions In Gas Turbine Combustors. Ph.D. thesis, Georgia Institute of Technology,
Atlanda.
23. Bellucci, V., B. Schuermans, D. Nowak, P. Flohr, and C. O. Paschereit (2004). Ther-
moacoustic modeling of a gas turbine combustor equipped with acoustic dampers. Pro-
ceedings of ASME Turbo Expo, Power for Land, Sea, and Air, GT200453977.
24. Bittanti, S., A. Marco, G. Poncia, and W. Prandoni (2002). Identification of a model
for thermoacoustic instabilities in a Rijke tube. IEEE Transactions on control systems
technology, 10, 490502.
25. Blomshield, F. S., H. B. Mathes, J. E. Crump, C. A. Beiter, and M. W. Beckstead
(1997a). Nonlinear stability testing of full-scale tactical motors. Journal of Propulsion
and Power, 13, 356366.
26. Blomshield, F. S., H. B. Mathes, J. E. Crump, C. A. Beiter, and M. W. Beckstead
(1997b). Stability testing of full-scale tactical motors. Journal of Propulsion and Power,
13, 349355.
27. Brewster, Q. and S. F. Son (1995). Quasi-steady combustion modeling of homogeneous
solid propellants. Combustion and Flame, 103(1-2), 11 26.
28. Bush, W. B. and F. E. Fendell (1972). Asymptotic analysis of turbulent channel and
boundary-layer flow. Journal of Fluid Mechanics, 56(04), 657681.

226
29. Butler, K. M. and B. F. Farrell (1992). Three-dimensional optimal perturbations in
viscous shear flow. Phys. Fluids A, 4(8), 16371650.

30. Candel, S. (2002). Combustion dynamics and control: Progress and challenges. Pro-
ceedings of the Combustion Institute, 29(1), 1 28. ISSN 1540-7489.

31. Carvalho, J. A., M. A. Ferreira, C. Bressan, and L. G. Ferreira (1989). Definition

of heater location to drive maximum amplitude acoustic oscillations in a Rijke tube.
Comb. Flame, 76, 1727.

32. Chagelishvili, G. D., A. G. Tevzadze, G. Bodo, and S. S. Moiseev (1997). Linear

mechanism of wave emergence from vortices in smooth shear flows. Physical Review
Letters, 79, 31783181.

33. Chong, L. T. W., S. Bomberg, A. Ulhaq, T. Komarek, and W. Polifke (2011). Com-
parative validation study on identification of premixed flame transfer function. Pro-
ceedings of ASME Turbo Expo, GT201146342.

34. Chu, B. T. (1965). On the energy transfer to small disturbances in fluid flow (part I).
Acta Mechanica, 3, 215234.

35. Cohen, N. S. and L. D. Strand (1985). Combustion response to compositional fluctu-

ations. AIAA. Journal, 23, 760767.

36. Collis, D. C. and M. J. Williams (1959). Two-dimensional convection from heated

wires at low Reynolds numbers. Journal of Fluid Mechanics, 6(03), 357384.

37. Corbett, P. and A. Bottaro (2001). Optimal linear growth in swept boundary layers.
Journal of Fluid Mechanics, 435, 123.

38. Coutanceau, M. and R. Bouard (1977). Experimental determination of the main fea-
tures of the viscous flow in the wake of a circular cylinder in uniform translation. Part
I. Steady flow. Journal of Fluid Mechanics, 79(02), 231256.

39. Cozzi, F., L. T. Deluca, and B. V. Novozhilov (1999). Linear stability and pressure-
driven response function of solid propellants with phase transitions. Journal of Propul-
sion and Power, 15, 806815.

40. Criminale, W. O. and P. G. Drazin (1999). The initial-value problem for a modeled
boundary layer. Phys. Fluids, 12, 366374.

41. Crocco, L. (1956). Theory of combustion instability in liquid propellant rocket motors.
Technical Report Rep 0429886, RTO AGARDorgraph.

42. Culick, F. E. C. (1963). Stability of high-frequency pressure oscillations in rocket

combustion chambers. AIAA. Journal, 1, 10971104.

43. Culick, F. E. C. (1968). A review of calculations for unsteady burning of a solid

propellant. AIAA. Journal, 6, 22412255.

44. Culick, F. E. C. (1976a). Nonlinear behavior of acoustic waves in combustion cham-

bers. Part I. Acta Astronautica, 3(9-10), 715 734.

227
45. Culick, F. E. C. (1976b). Nonlinear behavior of acoustic waves in combustion cham-
bers. Part II. Acta Astronautica, 3(9-10), 735 757.

46. Culick, F. E. C. (1994). Some recent results for nonlinear acoustics in combustion
chambers. AIAA. Journal, 32, 146169.

47. Culick, F. E. C. (1997). A note on ordering perturbations and the insignificance of

linear coupling in combustion instabilities. Combustion Science and Technology, 126,
359379.

48. Culick, F. E. C. (2006). Unsteady motions in combustion chambers for propulsion

systems. Technical Report AG-AVT-039, RTO AGARDorgraph.

49. Culick, F. E. C., V. Burnley, and G. Swenson (1995). Pulsed instabilities in solid-
propellant rockets. Journal of Propulsion and Power, 11, 657665.

50. Culick, F. E. C. and G. Isella, Modelling the combustion response function with surface
and gas phase dynamics. In Thirty-Eighth Aerospace Sciences Meeting and Exhibit.
AIAA, Reno, NV, 2000.

51. Das, S. and R. Chhabra (1989). A note on very low reynolds number fluid flow through
screens. Chemical Engineering and Processing: Process Intensification, 25(3), 159
161.

52. Deluca, L., R. Disilvestro, and F. Cozzi (1995). Intrinsic combustion instability of
solid energetic materials. Journal of Propulsion and Power, 11, 804815.

53. Dyke, M. V., Perturbation methods in fluid mechanics. The Parabolic press, Stanford,
California, 1975, 2nd edition.

54. Entezam, B., W. K. V. Moorhem, and J. Majdalani, Modeling of a Rijke-tube pulse

combustor using computational fluid dynamics. In Thirty-third AIAA/ASME/SAE/ASEE
Joint Propulsion Conference and Exhibit. 97-2718, Seattle, Washington, 1997.

55. Farrell, B. F. and P. J. Ioannou (1996a). Generalized stability theory. Part I. Au-
tonomous operators. Journal of the Atmospheric Sciences, 53(14), 20252040.

56. Farrell, B. F. and P. J. Ioannou (1996b). Generalized stability theory. Part II. Non-
autonomous operators. Journal of the Atmospheric Sciences, 53(14), 20412053.

57. Flandro, G. A. (1995a). Effects of vorticity on rocket combustion stability. Journal of

Propulsion and Power, 11, 607625.

58. Flandro, G. A., On flow turning. In Thirty-First ASME, SAE, and ASEE, Joint Propul-
sion Conference and Exhibit. 95-2730, San Diego, CA, 1995b.

59. Flandro, G. A., Nonlinear combustion instability data reduction. In Thirty-Second

AIAA/ASME/SAE/ASEE Joint Propulsion Conference and Exhibit. 96-3251, Lake
Buena Vista, Florida, 1996.

60. Flandro, G. A., S. R. Fischbach, and J. Majdalani (2007). Nonlinear rocket motor
stability prediction: Limit amplitude, triggering and mean pressure shift. Physics of
Fluids, 19, 094101.

228
61. Flandro, G. A. and J. Majdalani (1996). Aeroacoustic instability in rockets. AIAA
Journal, 41, 485497.

62. Foures, D., C. Caulfield, and P. Schmid (2012). A variational framework for flow
optimization using semi-norm constraints. arXiv:1202.0650v1[physics.flu-dyn].

63. Fransson, J. H. M., L. Brandt, A. Talamelli, and C. Cossu (2004). Experimental

and theoretical investigation of the nonmodal growth of steady streaks in a flat plate
boundary layer. Physics of Fluids, 16, 36273638.

64. Fransson, J. H. M., A. Talamelli, L. Brandt, and C. Cossu (2006). Delaying transition
to turbulence by a passive mechanism. Physical Review Letters, 96(6), 064501.

65. Friedly, J. C. and E. E. Petersen (1966). Influence of combustion parameters on in-

stability in solid propellant motors. Part I. Development of model and linear analysis.
AIAA. Journal, 4, 16041610.

66. Fu, W. S. and B. H. Tong (2002). Numerical investigation of heat transfer from a heated
oscillating cylinder in a cross flow. International Journal of Heat and Mass Transfer,
45(14), 3033 3043.

67. Gebhardt, T. and S. Grossmann (1994). Chaos transition despite linear stability. Phys-
ical Review E, 50(5), 37053711.

68. Georg, K. (2010). Matrix-free numerical continuation and bifurcation,

http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.34.9184.

69. George, J. and R. I. Sujith (2009). Emergence of acoustic waves from vorticity fluc-
tuations: Impact of non-normality. Physical Review E, 80, 046321.

70. George, J. K. and R. I. Sujith (2011a). On Chus disturbance energy. Journal of Sound
and Vibration, 330(22), 52805291.

71. George, J. K. and R. I. Sujith (2011b). On disturbance energy norms. Journal of

Sound and Vibration, 331(7), 15521566.

72. Giauque, A., T. Poinsot, M. J. Brear, and F. Nicoud (2006). Budget of disturbance en-
ergy in gaseous reacting flows. In Proc. Summer Program 2006, Center for Turbulence
Research, Stanford University.

73. Golub, G. H. and C. E. VanLoan, Matrix Computations. The Johns Hopkins Univer-
sity Press, UK, 1989, 2nd edition.

74. Guegan, A., P. J. Schmid, and P. Huerre (2008). Spatial optimal disturbances in swept
attachment-line boundary layers. Journal of Fluid Mechanics, 603, 179188.

75. Gunzburger, M. D., Perspectives in flow control and optimization. SIAM, 2003.

76. Gusachenko, L. and V. Zarko (2008). Analysis of unsteady solid-propellant com-

bustion models (review). Combustion Explosion and Shock Waves, 44, 3142. ISSN
0010-5082.

229
77. Gustavsson, L. H. (1991). Energy growth of three-dimensional disturbances in plane
Poiseuille flow. Journal of Fluid Mechanics, 224, 241260.

78. Hanifi, A., P. J. Schmid, and D. S. Henningson (1996). Transient growth in compress-
ible boundary layer flow. Physics of Fluids, 8(3), 826837.

79. Hantschk, C. C. and D. Vortmeyer (1999). Numerical simulation of self-excited ther-

moacoustic instabilities in a Rijke tube. Journal of Sound and Vibration, 227, 511522.

80. Harris, P. G. and A. D. Champlain (1998). Experimental database describing pulse-

triggered nonlinear instability in solid rocket motors. Journal of Propulsion and Power,
14, 429439.

81. Heckl, M. (1985). Heat sources in acoustic resonators. Ph.D. thesis, University of
Cambridge, Cambridge.

82. Heckl, M. A. (1990). Nonlinear acoustic effects in the Rijke tube. Acustica, 72, 6371.

83. Heckl, M. A. and M. S. Howe (2007). Stability analysis of the Rijke tube with a Greens
function approach. Journal of Sound and Vibration, 305, 672688.

84. Henningson, D. S., A. Lundbladh, and A. V. Johansson (1993). A mechanism for

bypass transition from localized disturbances in wall-bounded shear flows. Journal of
Fluid Mechanics, 250, 169207.

85. Henningson, D. S. and S. Reddy (1994). On the role of linear mechanisms in transition
to turbulence. Physics of Fluids, 6, 13961398.

86. Hill, D. C. (1995). Adjoint systems and their role in the receptivity problem for bound-
ary layers. Journal of Fluid Mechanics, 292, 183204.

87. Holman, J. P., Experimental methods for Engineers. McGraw-Hill, Inc., USA, 1994,
6th edition.

88. Hoppensteadt, F. (1975). Analysis of some problems having matched asymptotic ex-
pansion solutions. SIAM Review, 17(1), 123135.

89. Hosseini, S. M. R. and C. J. Lawn (2005). Non-linearities in the thermo-acoustic

response of a premixed swirl burner. Twelfth International Congress on Sound and
Vibration, 1114 July, Lisbon.

90. Juniper, M. (2011a). Transient growth and triggering in the horizontal Rijke tube.
International Journal of Spray and Combustion Dynamics, 3, 209224.

91. Juniper, M. P. (2011b). Triggering in the horizontal Rijke tube: Non-normality, tran-
sient growth and bypass transition. Journal of Fluid Mechanics, 667, 272308.

92. Junye, W. (2000). Non-linear analysis of solid propellant burning rate behavior. Inter-
national Journal of Numerical Methods in Fluids, 33(5), 627640.

93. Kabiraj, L., R. I. Sujith, and P. Wahi, Experimental studies of bifurcations leading
to chaos in a laboratory scale thermoacoustic system. In ASME Turbo Expo. GT2011-
46149, Vancouver, Canada, 2011.

230
 94. Kang, S. (2003). Characteristics of flow over two circular cylinders in a side-by-side
arrangement at low Reynolds numbers. Physics of Fluids, 15, 24862498.

95. Kaufmann, A., F. Nicoud, and T. Poinsot (2002). Flow forcing techniques for nu-
merical simulation of combustion instabilities. Combustion and Flame, 131(4), 371
385.

96. Kedia, K. S., S. B. Nagaraja, and R. I. Sujith (2008). Impact of linear coupling on
thermoacoustic instabilities. Combustion Science and Technology, 180, 1588 1612.

97. Kevorkian, L. and J. D. Cole, Perturbation methods in applied mathematics. Springer-

Verlag, New York, 1981, Applied mathematical sciences 34 edition.

98. Kevorkian, L. and J. D. Cole, Multiple scale and singular perturbation methods.
Springer-Verlag, New York, 1996, Applied mathematical sciences 114 edition.

99. Kim, K. T. and S. Hochgreb (2012). Measurements of triggering and transient growth
in a model lean-premixed gas turbine combustor. Combustion and Flame, 159(3), 1215
1227. ISSN 0010-2180.

100. King, F. W., Hilbert Transforms. Cambridge University Press, 2009.

101. Kinsler, L. E., A. R. Frey, A. B. Coppens, and J. V. Sanders, Fundamentals of Acous-

tics. John Wiley and Sons, Inc., USA, 2000, 4th edition.

102. Klebanoff, P. S., K. D. Tidstrom, and L. M. Sargent (1962). The three-dimensional

nature of boundary-layer instability. Journal of Fluid Mechanics, 12(01), 134.

103. Klein, R. (1995). Semi-implicit extension of a Godunov-type scheme based on low

mach number asymptotics: I One-dimensional flow. Journal of Computational Physics,
121, 213237.

104. Klein, R., N. Botta, T. Schneider, C. D. Munz, S. Roller, A. Meister, L. Hoffmann,

and T. Sonar (2001). Asymptotic adaptive methods for multi-scale problems in fluid
mechanics. Journal of Engineering Mathematics, 39, 261343.

105. Kopitz, J. and W. Polifke (2008). CFD based application of the Nyquist criterion to
thermoacoustic instabilities. Journal of Computational Physics, 227, 67546778.

106. Kourta, A. (1997). Shear layer instability and acoustic interaction in solid propellant
rocket motors. International Journal of Numerical Methods in Fluids, 25(8), 973981.

107. Krasnov, D. S., E. Zienicke, O. Zikanov, T. Boeck, and A. Thess (2004). Numerical
study of the instability of the hartmann layer. Journal of Fluid Mechanics, 504, 183
211.

108. Krechetnikov, R. and J. E. Marsden (2009). On the origin and nature of finite-
amplitude instabilities in physical systems. Journal of Physics A: Mathematical and
Theoretical, 42(41), 412004 (18pp).

109. Kreith, F., The CRC handbook of thermal engineering. CRC Press, New York, 2000,
1st edition.

231
110. Krier, H., S. J. Tien, W. A. Sirignana, and M. Summerfield (1968). Nonsteady
burning phenomena of solid propellants: Theory and experiments. AIAA. Journal, 6,
278285.
111. Kulkarni, R., K. Balasubramanian, and R. I. Sujith (2011). Non-normality and its
consequences in active control of thermoacoustic instabilities. Journal of Fluid Me-
chanics, 670, 130149.
112. Kumar, K. R. and K. N. Lakshmisha (2000). Nonlinear intrinsic instability of solid
propellant combustion including gas-phase thermal inertia. Combust. Sci. Tech, 158,
135166.
113. Kuo, K. K. and M. Summerfield (1984). Nonsteady burning phenomena of solid
propoellants: Theory and experiments. Progress in Astronautics and Aeronautics, 90,
760767.
114. Kwon, Y. P. and B. H. Lee (1985). Stability of the Rijke thermoacoustic oscillation. J.
Acoust. Soc. Am., 78, 14141420.
115. Lagree, P. Y. (2011). Multiscale hydrodynamic phenomena: Matched asymptotic ex-
pansions. Lecture notes, http://www.lmm.jussieu.fr/ lagree/COURS/M2MHP.
116. Laws, E. M. and J. L. Livesey (1978). Flow through screens. Annual Review of Fluid
Mechanics, 10(1), 247266.
117. Lee, J. G. and D. A. Santavicca (2005). Experimental diagnostics of combustion in-
stabilities. Progress in Astronautics and Aeronautics, 210, 481529.
118. Levine, J. N. and J. D. Baum (1983). A numerical study of nonlinear instability phe-
nomena in solid rocket motors. AIAA. Journal, 21, 557564.
119. Lighthill, M. J. (1954). The response of laminar skin friction and heat transfer to
fluctuations in the stream velocity. Proceedings of the Royal Society of London. Series
A, Mathematical and Physical Sciences, 224(1156), 123.
120. Lin, A. C. and S. Y. Wang, Investigation of aluminized solid propellant combustion
instability by means of a T-burner. In Thirty-Third Aerospace sciences meeting and
exhibit. 95-0606, Reno, US, 1995.
121. Majdalani, J., B. Entezam, and W. K. V. Moorhem, The Rijke tube revisited via
laboratory and numerical experiments. In Thirty-fifth AIAA Thermophysics Conference.
2001-2961, Anaheim, California, 2001.
122. Mandhani, V., R. Chhabra, and V. Eswaran (2002). Forced convection heat transfer
in tube banks in cross flow. Chemical Engineering Science, 57(3), 379 391.
123. Mangesius, H. and W. Polifke (2011). A discrete-time, state-space approach for the
investigation of non-normal effects in thermoacoustic systems. International Journal of
Spray and Combustion Dynamics, 3, 331350.
124. Mao, X., S. J. Sherwin, and H. M. Blackburn (2009). Transient growth and bypass
transition in stenotic flow with physiological waveform. Theoretical and Computational
Fluid Dynamics.

232
125. Margolis, S. B. and R. C. Armstrong (1986). Two asymptotic models for solid pro-
pellant combustion. Combust. Sci. Tech, 47, 138.

126. Margolis, S. B. and R. C. Armstrong (1988). Diffusional/thermal coupling and intrin-

sic instability of solid propellant combustion. Combust. Sci. Tech, 59, 2784.

127. Marquet, O., M. Lombardi, J. M. Chomaz, D. Sipp, and L. Jacquin (2009). Di-
rect and adjoint global modes of a recirculation bubble: lift-up and convective non-
normalities. Journal of Fluid Mechanics, 622, 121.

128. Matsubara, M. and P. H. Alfredsson (2001). Disturbance growth in boundary layers

subjected to free-stream turbulence. Journal of Fluid Mechanics, 430, 149168.

129. Matveev, K. (2003a). Energy consideration of the nonlinear effects in a Rijke tube.
Journal of Fluids and Structures, 18(6), 783 794.

130. Matveev, K. I. (2003b). Thermo-acoustic instabilities in the Rijke tube: Experiments

and modeling. Ph.D. thesis, California Institute of Technology, Pasadena.

131. Matveev, K. I. and F. E. C. Culick (2003a). A model for combustion instability in-
volving vortex shedding. Combustion Science and Technology, 175(6), 1059 1083.

132. Matveev, K. I. and F. E. C. Culick (2003b). A study of the transition to instability in

a Rijke tube with axial temperature gradient. Journal of Sound and Vibration, 264(3),
689 706.

133. Mayer, E. W. and E. Reshotko (1997). Evidence for transient disturbance growth in a
1961 pipe flow experiment. Physics of Fluids, 9, 242244.

134. McIntosh, A. C. and S. Rylands (1996). A model of heat transfer in Rijke tube burners.
Combustion Science and Technology, 113(1), 273289.

135. Mclntosh, A. C. (1987). Combustion-acoustic interaction of a flat flame burner system

enclosed within an open tube. Combustion Science and Technology, 54(1-6), 217236.

136. Mcmanus, K., T. Poinsot, and S. Candel (1993). A review of active control of com-
bustion instabilities. Progress in Energy and Combustion Science, 19, 1 29.

137. Mellor, G. L. (1972). The large Reynolds number, asymptotic theory of turbulent
boundary layers. International Journal of Engineering Science, 10(10), 851 873.

138. Moeck, J., M. Oevermann, R. Klein, C. Paschereit, and H. Schmidt (2009). A

two-way coupling for modeling thermoacoustic instabilities in a flat flame Rijke tube.
Proceedings of the Combustion Institute, 32(1), 1199 1207. ISSN 1540-7489.

139. Moeck, J. P., H. Schmidt, M. Oevermann, C. O. Paschereit, and R. Klein, An

asymptotically motivated hydrodynamic-acoustic two-way coupling for modeling ther-
moacoustic instabilities in a Rijke tube. In ICSV14. Cairns, Australia, 2007.

140. Monokrousos, A., A. Bottaro, L. Brandt, A. Di Vita, and D. S. Henningson (2011).

Nonequilibrium thermodynamics and the optimal path to turbulence in shear flows.
Phys. Rev. Lett., 106, 134502.

233
141. Mukhopadhyay, B., N. Afshordi, and R. Narayan (2006). Growth of hydrodynamic
perturbations in accretion disks: Possible route to non-magnetic turbulence. Adv. Space
Res., 38(12), 28772879.

142. Myers, M. K. (1991). Transport of energy by disturbances in arbitrary steady flows.

Journal of Fluid Mechanics, 226, 383400.

143. Nagaraja, S., K. Kedia, and R. Sujith (2009). Characterizing energy growth during
combustion instabilities: Singularvalues or eigenvalues? In Proceedings of the Comb.
Institute, 32(2), 2933 2940. ISSN 1540-7489.

144. Nayfeh, A. H., Introduction to perturbation techniques. John Wiley and Sons, Inc.,
New York, 1998, 1st edition.

145. Negrete, D. C., D. A. Spong, and S. P. Hirshman (2008). Proper orthogonal de-
composition methods for noise reduction in particle-based transport calculations. Phy.
Plasmas, 15, 092308.

146. Neuringer, J. L. and G. E. Hudson (1952). An investigation of sound vibrations in a

tube containing a heat source. Journal of Acoustical Society of America, 24, 667674.

147. Nicoli, C. and P. Pelce (1989). One-dimensional model for the Rijke tube. Journal of
Fluid Mechanics, 202, 8396.

148. Nicolle, A. and I. Eames (2011). Numerical study of flow through and around a circular
array of cylinders. Journal of Fluid Mechanics, 679, 131.

149. Nicoud, F., L. Benoit, C. Sensiau, and T. Poinsot (2007). Acoustic modes in combus-
tors with complex impedances and multidimensional active flames. AIAA. Journal, 45,
426441.

150. Nicoud, F. and T. Poinsot (2005). Thermoacoustic instabilities: Should the Rayleigh
criterion be extended to include entropy changes? Combustion and Flame, 142, 153
159.

151. Orr, W. M. (1907). The stability or instability of the steady motions of a perfect liquid
and of a viscous liquid. Part I: A perfect liquid. Proceedings of the Royal Irish Academy.
Section A: Mathematical and Physical Sciences, 27, pp. 968.

152. Padmanabhan, M. S. (1975). The effect of nozzle nonlinearities on the nonlinear

stability of liquid rocket motors. Ph.D. thesis, Georgia Institute of Technology, Atlanta,
USA.

153. Patankar, S. V., Numerical heat transfer and fluid flow. Hemisphere Publishing Cor-
poration, Washington, 1980, 1st edition.

154. Pier, B. (2002). On the frequency selection of finite-amplitude vortex shedding in the
cylinder wake. Journal of Fluid Mechanics, 458, 407417.

155. Piezotronics, P. (2008). Calibration certificate for pressure sensor, model no. 103b02,
serial no. 5236. National Institute of Standards and Technology, NISTCA322.

234
156. Podvin, B., Y. Fraigneau, F. Lusseyran, and P. Gougat (2006). A reconstruction
method for the flow past an open cavity. Journal of Fluids Engineering, 128, 531540.

157. Poinsot, T. and D. Veynante, Theoretical and numerical combustion. R.T. Edwards,
Inc., Philadelphia, 2005, 2nd edition.

158. Prandtl, L. (1928). ber Flssigkeitsbewegung bei sehr kleiner Reibund, Translated
into English. NACA Technical memorandum, 452.

159. Preetham, S. H. and T. Lieuwen (2008). Dynamics of laminar premixed flames forced
by harmonic velocity disturbances. Journal of Propulsion and Power, 24(6), 1390
1402.

160. Press, W. H., S. A. Teukolsky, and B. P. Flannery, Numerical recipes: the art of
scientific computing. Cambridge University Press, UK, 2007, 3rd edition.

161. Price, E. W. (1984). Fundamentals of solid-propellant combustion. Progress in Astro-

nautics and Aeronautics, 90, 479513.

162. Putnam, A. A. and W. R. Dennis (1954). Burner oscillations of the gauze-tone type.
Journal of Acoustical Society of America, 26, 716725.

163. Raun, R. and M. Beckstead (1993). A numerical model for temperature gradient and
particle effects on Rijke burner oscillations. Combustion and Flame, 94(1-2), 1 24.

164. Raun, R., M. Beckstead, J. Finlinson, and K. Brooks (1993). A review of Rijke tubes,
Rijke burners and related devices. Progress in Energy and Combustion Science, 19(4),
313 364.

165. Rayleigh, L. (1878). The explanation of certain acoustical phenomenon. Nature, 18,
319321.

166. Reddy, S. C. and L. N. Trefethen (1994). Pseudospectra of the convection-diffusion

operator. SIAM Journal on Applied Mathematics, 54(6), pp. 16341649.

167. Rienstra, S. W. and A. Hirschberg, An introduction to acoustics. Technische Univer-

siteit Eindhoven, Netherlands, 2004, IWDE 92-06 edition.

168. Rienstra, S. W. and A. Hirschberg (2008). An introduction to acoustics. International

Journal of Spray and Combustion Dynamics, IWDE 9206.

169. Rijke, P. L. (1859). The vibration of the air in a tube open at both ends. Philosophical
Magazine, 17, 419422.

170. Riley, K. F., M. P. Hobson, and S. J. Bence, Mathematical Methods for Physics and
Engineering. Cambridge University Press, UK, 2006, 3rd edition.

171. Romanov, O. Y. (1999). Unsteady burning of solid propellants. Journal of Propulsion

and Power, 15, 823836.

172. Rowley, C. W. (2005). Model reduction for fluids, using balanced proper orthogonal
decomposition. International Journal of Bifurcation and Chaos, 15, 9971013.

235
173. Rowley, C. W., I. Mezic, S. Bagheri, P. Schlatter, and D. S. Henningson (2009).
Spectral analysis of nonlinear flows. Journal of Fluid Mechanics, 641, 115127.

174. Sameen, A. and R. Govindarajan (2007). The effect of wall heating on instability of
channel flow. Journal of Fluid Mechanics, 577, 417442.

175. Sarpkaya, T. (1986). Force on a circular cylinder in viscous oscillatory flow at low
keulegan-carpenter numbers. Journal of Fluid Mechanics, 165, 6171.

176. Schmid, P. J. (2007). Nonmodal stability theory. Annual Review of Fluid Mechanics,
39(1), 129162.

177. Schmid, P. J. (2010). Dynamic mode decomposition of numerical and experimental

data. Journal of Fluid Mechanics, 656, 528.

178. Schmid, P. J. (2011). Application of the dynamic mode decomposition to experimental

data. Experiments in Fluids, 50, 11231130.

179. Schmid, P. J. and D. S. Henningson (1992). A new mechanism for rapid transition
involving a pair of oblique waves. Physics of Fluids, 4, 19861989.

180. Schmid, P. J. and D. S. Henningson, Stability and Transition in Shear Flows. Springer-
Verlag, New York, 2001, 1st edition.

181. Schmid, P. J. and D. S. Henningson (2002). On the stability of a falling liquid curtain.
Journal of Fluid Mechanics, 463, 163171.

182. Schmid, P. J., L. Li, M. P. Juniper, and O. Pust (2010). Applications of the dynamic
mode decomposition. Theoretical and Computational Fluid Dynamics, 641.

183. Schmid, P. J., K. E. Meyer, and O. Pust, Dynamic mode decomposition and proper
orthogonal decomposition of flow in a lid-driven cylindrical cavity. In 8th International
Symposium on Particle Image Velocimetry. Melbourne, Australia, 2009.

184. Schuermans, B., H. Luebcke, D. Bajusz, and P. Flohr (2005). Thermoacoustic anal-
ysis of gas turbine combustion systems using unsteady CFD. Proceedings of ASME
Turbo Expo, Power for Land, Sea, and Air, GT200568393.

185. Schuller, T., D. Durox, and S. Candel (2003). Self-induced combustion oscillations of
laminar premixed flames stabilized on annular burners. Combustion and Flame, 135(4),
525 537.

186. Selimefendigil, F., S. Fller, and W. Polifke, Nonlinear identification of the unsteady
heat transfer of a cylinder in pulsating crossflow. In International Conference on Jets,
Wakes and Separated Flows. Technical University Berlin, Berlin, Germany, 2008.

187. Selimefendigil, F., R. Sujith, and W. Polifke (2011). Identification of heat transfer
dynamics for non-modal analysis of thermoacoustic stability. Applied Mathematics and
Computation, 217(11), 5134 5150. Special issue on Fluid Flow and Heat Transfer.

188. Shimada, T., M. Hanzawa, T. Kata, T. Yoshikawa, and Y. Wada, Stability analy-
sis of solid rocket motor combustion by computational fluid dynamics. In Thirteenth
AIAA/CEAS Aeroacoustics Conference. 2007-3427, 2007.

236
189. Shreenivasan, O. J. (2008). Onset of acoustic instability by vortex shedding in a non-
premixed combustor. Ph.D. thesis, Indian Institute of Technology Madras, Chennai.

190. Song, W. S., S. Lee, D. S. Shin, and Y. Na (2006). Thermo-acoustic instability in the
horizontal Rijke tube. Journal of Mechanical Science and Technology, 20, 905913.

191. Stanescu, G., A. Fowler, and A. Bejan (1996). The optimal spacing of cylinders
in free-stream cross-flow forced convection. International Journal of Heat and Mass
Transfer, 39(2), 311 317.

192. Stephane, L. D. and D. Fabre (2007). Large-Reynolds-number asymptotic analysis of

viscous centre modes in vortices. Journal of Fluid Mechanics, 585, 153180.

193. Sterling, J. D. and E. E. Zukoski (1991). Nonlinear dynamics of laboratory combustor

pressure oscillations. Combustion Science and Technology, 77(4-6), 225238.

194. Strogatz, S. H., Nonlinear Dynamics and Chaos: with applications to Physics, Biology,
Chemistry, and Engineering. Westview Press, Colorado, 2000, 1st edition.

195. Subramanian, P., S. Mariappan, P. Wahi, and R. I. Sujith (2010). Bifurcation anal-
ysis of thermoacoustic instability in a Rijke tube. International Journal of Spray and
Combustion Dynamics, 2, 325356.

196. Subramanian, P., A. Saha, R. I. Sujith, and P. Wahi, Analytical estimates for sta-
bility boundaries and nature of the bifurcation in a simple thermoacoustic system. In
Bifurcations and instabilities in fluid dynamics. Barcelona, Spain, 2011.

197. Subramanian, P. and R. I. Sujith (2011). Non-normality and internal flame dynamics
in premixed flameacoustic interaction. Journal of Fluid Mechanics, 679, 315342.

198. Sutton, G. P. and O. Biblarz, Rocket Propulsion Elements. Wiley-IEEE, New York,
2001, 7th edition.

199. Telionis, D. P., Unsteady viscous flows. Springer-Verlag, New York, 1981, 1st edition.

200. Ting, L., R. Klein, and O. M. Knio, Vortex Dominated Flows: Analysis and Compu-
tation for Multiple Scales. Springer Verlag, New York, 2007, Series in Applied Mathe-
matical Sciences edition.

201. Trefethen, L. N. and M. Embree, Spectra and Pseudospectra, The Behaviour of Non-
normal Matrices and Operators. Princeton University Press, New Jersey, 2005, 1st
edition.

202. Trefethen, L. N., A. E. Trefethen, S. C. Reddy, and T. A. Driscoll (1993). Hydrody-

namic stability without eigenvalues. Science, 261(5121), 578584.

203. Tulsyan, B., K. Balasubramanian, and R. I. Sujith (2009). Revisiting a model for
combustion instability involving vortex shedding. Combustion Science and Technology,
181, 457 482.

204. Vuillot, J. (1995). Vortex shedding phenomena in solid rocket motors. Journal of
Propulsion and Power, 11, 626639.

237
205. Waleffe, F. (1995). Transition in shear flows: non-linear normality versus non-normal
linearity. Physics of Fluids, 7, 30603066.

206. Ward, M. J., S. F. Son, and M. Q. Brewster (1998). Steady deflagration of hmx with
simple kinetics: a gas phase chain reaction model. Combustion and Flame, 114(3-4),
556 568.

207. Waugh, I. C., M. Geu, and M. Juniper (2011). Triggering, bypass transition and the
effect of noise on a linearly stable thermoacoustic system. Proceedings of the Combus-
tion Institute, 33(2), 2945 2952.

208. Waugh, I. C. and M. Juniper (2011). Triggering in a thermoacoustic system with

stochastic noise. International Journal of Spray and Combustion Dynamics, 3, 224
242.

209. Wicker, J. M., W. D. Greene, S.-I. Kim, and V. Yang (1996). Triggering of longitudi-
nal combustion instabilities in rocket motors: nonlinear combustion response. Journal
of Propulsion and Power, 12, 11481158.

210. Wieczorek, K., C. Sensiau, W. Polifke, and F. Nicoud (2010). Assessing non-normal
effects in thermoacoustic systems with non zero baseline flow. n3l - Intl Summer School
and Workshop on Non-Normal and Nonlinear Effects in Aero- and Thermoacoustics,
May 17-21, 2010, Munich.

211. Williams, F. A. (1962). Response of a burning solid to small-amplitude pressure oscil-

lations. Journal of Applied Physics, 33, 31533166.

212. Williams, F. A., Combustion Theory. Addison-Wesley, New York, 1985, 2nd edition.

213. Wu, X. and C. K. Law (2009). Flame-acoustic resonance initiated by vortical distur-
bances. Journal of Fluid Mechanics, 634, 321357.

214. Wu, X. and P. Moin (2010). Large-activation-energy theory for premixed combustion
under the influence of enthalpy fluctuations. Journal of Fluid Mechanics, 655, 337.

215. Wu, X., M. Wang, P. Moin, and N. Peters (2003). Combustion instability due to
the nonlinear interaction between sound and flame. Journal of Fluid Mechanics, 497,
2353.

216. Yang, V., S. I. Kim, and F. E. C. Culick (1990). Triggering of longitudinal pressure os-
cillations in combustion chambers. Part I. Nonlinear gasdynamics. Combustion Science
and Technology, 72, 183214.

217. Yoon, H. G., I. Peddieson, and K. R. Purdy (2001). Non-linear response of a gener-
alized Rijke tube. International Journal of Engineering Sciences, 39, 17071723.

218. Zeytounian, K. H., Asymptotic modelling of fluid flow phenomena. Kluwer academic
publisher, Netherlands, 2002, 1st edition.

219. Zhao, D. (2012). Transient growth of flow disturbances in triggering a rijke tube com-
bustion instability. Combustion and Flame, (0), .

238
220. Zinn, B. T. (1972). Longitudinal mode acoustic losses in short nozzles. Technical
Report AD0744623, Research Report Naval Weapons Center China Lake.

221. Zinn, B. T. and T. C. Lieuwen, Combustion instabilities: Basic concepts - Combustion

Instabilities in Gas Turbine Engines: Operational Experience, Fundamental Mecha-
nisms, and Modeling. AIAA, USA, 2006.

222. Zinn, B. T. and M. E. Lores (1971). Application of the Galerkin method in the solu-
tion of non-linear axial combustion instability problems in liquid rockets. Combustion
Science and Technology, 4, 269278.

239
 LIST OF PAPERS BASED ON THESIS

Referred Journals:
1. Mariappan, S. and Sujith, R. I. Modeling nonlinear thermoacoustic instability
in an electrically heated Rijke tube. Journal of Fluid Mechanics, 680:511533,
2011.
2. Mariappan, S. and Sujith, R. I. Thermoacoustic instability in a solid rocket motor:
non-normality and nonlinear instabilities. Journal of Fluid Mechanics, 653:133,
2010.
3. Mariappan, S., Sujith, R. I. and Schmid, P. J. Experimental investigation of non-
normality of thermoacoustic interaction in an electrically heated Rijke tube, under
revision with Journal of Fluid Mechanics, 2012.

International Conferences:
1. Mariappan, S., Sujith, R. I. and Schmid, P. J. Non-normality of thermoacoustic
interactions: an experimental investigation. 47th AIAA/ASME/SAE/ASEE Joint
Propulsion Conference, San Diego, California, July 31 August 3, 2011.
2. Mariappan, S., Schmid, P. J. and Sujith, R. I. Subcritical transition to thermoa-
coustic instability in a horizontal Rijke tube: non-normal and nonlinear effects.
13th Asian Congress of Fluid Mechanics, Dhaka, Bangladesh, December 17 21,
2010.
3. Mariappan, S., Schmid, P. J. and Sujith, R. I. Role of transient growth in sub-
critical transition to thermoacoustic instability in a horizontal Rijke tube. 16th
AIAA/CEAS Aeroacoustics Conference, AIAA 2010 3857, Stockholm, June 7
9, 2010.
4. Mariappan, S., Sujith, R. I. and Schmid, P. J. Non-normal and nonlinear dynam-
ics of thermoacoustic instability in a horizontal Rijke tube. n3 l - Intl Summer
School and Workshop on Non-Normal and Nonlinear Effects in Aero- and Ther-
moacoustics, Munich, May 17 21, 2010.
5. Mariappan, S. and Sujith, R. I. Modeling nonlinear thermoacoustic instability
in an electrically heated Rijke tube. 48th AIAA Aerospace Sciences Meeting in-
cluding the New Horizons Forum and Aerospace Exposition, AIAA 2010 25,
January 4 7, Orlando, Florida, 2010.
6. Mariappan, S. and Sujith, R. I. Thermoacoustic instability in a solid rocket motor:
non-normality and nonlinear instabilities. 48th AIAA Aero-space Sciences Meet-
ing including the New Horizons Forum and Aerospace Exposition, AIAA 2010
1517, Orlando, Florida, January 4 7, 2010.

241
 CURRICULUM VITAE

NAME : Sathesh Mariappan

DATE OF BIRTH : 30th November 1985

EDUCATIONAL QUALIFICATIONS

2007 Bachelor of Engineering (B.E.)

Institution : Madras Institute of Technology, Chennai

Specialisation : Aerospace Engineering

2012 Doctor of Philosophy (Ph. D.)

Institution : Indian Institute of Technology Madras, Chennai

Registration Date : 20th July 2007

HONOURS & AWARDS

1. Alexander von Humboldt fellowship for Postdoctoral Researchers, 2012

2. Session co-chair on Combustion Instabilities, 47th AIAA/ASME/SAE/ASEE Joint

Propulsion Conference & Exhibit, San Diego, 2011

3. Session co-chair on Aero- and Thermo-Acoustical Coupling in Energy Applica-

tions II, 16th AIAA/CEAS Aeroacoustics Conference, Stockholm, 2010

4. All India first rank out of 1500 (99.9 percentile) in GATE (entrance exam for
Master level programme in India), XE stream, 2006

5. Direct admission to Ph. D., from Bachelor of Engineering (B. E.), 2007

6. University first rank in Bachelor of Engineering, 2007

243
 DOCTORAL COMMITTEE

CHAIR PERSON: Dr. K. Bhaskar

Professor, Department of Aerospace Engineering
Indian Institute of Technology Madras

GUIDE: Dr. R. I. Sujith

Professor, Department of Aerospace Engineering
Indian Institute of Technology Madras

MEMBERS: Dr. S. Narayanan

Professor, Department of Mechanical Engineering
Indian Institute of Technology Madras

Dr. K. Balasubramaniam
Professor, Department of Mechanical Engineering
Indian Institute of Technology Madras

Dr. N. K. Sinha
Associate Professor, Department of Aerospace Engineering
Indian Institute of Technology Madras

Dr. S. Sarkar
Assistant Professor, Department of Aerospace Engineering
Indian Institute of Technology Madras

245