Docslide - Us Combustionj PDF

Combustion Instabilities in Gas
Turbine Engines:
Operational Experience,
Fundamental Mechanisms,
and Modeling
Edited by
Timothy C. Lieuwen
School of Aerospace Engineering
Georgia Institute of Technology
Atlanta, Georgia
Vigor Yang
Department of Mechanical Engineering
The Pennsylvania State University
University Park, Pennsylvania
Volume 210
PROGRESS IN
ASTRONAUTICS AND AERONAUTICS
Frank K. Lu, Editor-in-Chief

University of Texas at Arlington
Arlington, Texas
Published by the
American Institute of Aeronautics and Astronautics, Inc.
1801 Alexander Bell Drive, Reston, VA 20191-4344
Copyright C 2005 by the American Institute of Aeronautics and Astronautics, Inc. Printed
in the United States of America. All rights reserved. Reproduction or translation of any part
of this work beyond that permitted by subsections 107 and 108 of the U.S. Copyright Law
without the permission of the copyright owner is unlawful. The code following this statement
indicates the copyright owner’s consent that copies of articles in this volume may be made
for personal or internal use, on condition that the copier pay the per-copy fee ($2.00) plus the
per-page fee ($0.50) through the Copyright Clearance Center, Inc., 222 Rosewood Drive,
Danvers, Massachusetts 01923. This consent does not extend to other kinds of copying,
for which permission requests should be addressed to the publisher. Users should employ
the following code when reporting copying from this volume to the Copyright Clearance
Center:
1-56347-669-X/05 $2.50 + .50
Data and information appearing in this book are for informational purposes only. AIAA is
not responsible for any injury or damage resulting from use or reliance, nor does AIAA
warrant that use or reliance will be free from privately owned rights.
ISBN 1-56347-669-X
Progress in Astronautics and Aeronautics
Editor-in-Chief
Frank K. Lu
University of Texas at Arlington
Editorial Board
David A. Bearden Richard C. Lind
The Aerospace Corporation University of Florida
John D. Binder Richard M. Lloyd

viaSolutions Raytheon Electronics Company
Steven A. Brandt Ahmed K. Noor

U.S. Air Force Academy NASA Langley Research Center
Fred R. DeJarnette Albert C. Piccirillo

North Carolina State University Institute for Defense Analyses
Philip D. Hattis Ben T. Zinn

Charles Stark Draper Laboratory Georgia Institute of Technology
Abdollah Khodadoust Peter H. Zipfel

The Boeing Company Air Force Research Laboratory
Preface
Gas turbines have made substantial gains in performance since their initial demon-
stration in jet powered aircraft and power turbines. The performance, noise char-
acteristics, and pollutant emissions of gas turbines for propulsive applications
continue to improve. On the ground, contemporary gas turbines produce higher
operating efficiencies and emit fewer pollutants than other major chemical-energy
conversion devices. In addition, the low capital investment, ease of permitting, and
quick installation have made them attractive to investors. As a result, gas turbines
have become a dominant technology for new power generating capacity in the
United States and worldwide.
A variety of factors have contributed to the popularity of gas turbine technology.
Financing considerations are the key high-level driver. Pollutant emissions play
another important role, particularly in motivating the specific technology improve-
ments and innovations over the last decade. For example, in the United States, the
Clean Air Act Amendments of 1990 imposed strict guidelines on the control of
nitrogen oxides, NOx, which, along with SO2 , is a major contributor to acid rain
This book focuses on a particularly serious difficulty in low emissions gas tur-
bines: combustion-driven oscillations. These instabilities routinely constrain the
operating envelope and power output of fielded machines and, in some cases, lead to
serious damage of hot section components. Gas turbine users have found that com-
ponents such as combustor liners, transition pieces, and fuel nozzles need routine
examination for part cracking or excessive wearing because of vibration-induced
fretting. At a minimum, this requires downtime for inspections and part repair,
thereby reducing machine availability. At the worst, a cracked piece may be liber-
ated into the hot gas path, potentially requiring replacement of expensive turbine
components. In addition, users in certain geographic areas have found that engines
must be seasonally retuned to eliminate oscillations due to ambient temperature
changes. The cost for the repair and replacement of hot section components, much
of which is directly attributable to the combustion instability problem, exceeds
$1 billion annually and constitutes up to 70% of the nonfuel costs of F-class gas
turbines. Major power generating companies have suffered losses in the hundreds
of millions of dollars because of lost revenue from forced outages, resulting in a
number of lawsuits.
Although instabilities have not been nearly as severe a problem in nonpremixed
aero engine combustors, they have appeared in a few cases and posed serious
challenges in the development stage. Military engines, however, have experienced
major problems with low-frequency instabilities in augmentors. A large-scale ef-
fort is currently underway at several gas turbine manufacturers in the United States,
in cooperation with the U.S. Air Force, to overcome such difficulties.
Over the last decade, substantial efforts have been expended in the industrial,
government, and academic communities to understand the unique issues associated
with combustion instabilities in low-emissions gas turbines. The objective of this
book is to compile these results into a series of chapters that address various
xiii
xiv PREFACE
facets of the problem. In planning this volume, it was decided to include a few
comprehensive chapters, rather than a large number of more narrowly focused
contributions. As such, it was not possible to solicit articles from every contributor
to the field, although it is certainly our hope that all relevant works are appropriately
represented in the book.
Following the overview, the book is organized into four basic sections: The
Case Studies section compiles chapters from gas turbine manufacturers and users
that detail specific experiences with combustion instabilities in the development
stage and in fielded turbine engines. These chapters describe the basic instability
mitigation approaches that were developed and the tradeoffs encountered between
instabilities and other performance metrics, such as NOx emissions. The Funda-
mental Processes and Mechanisms section addresses the basic phenomenology
of combustion instabilities in premixed and nonpremixed combustors, the mech-
anisms through which unsteady heat release processes may become self-excited,
and measurement techniques for characterizing them. Next, the Modeling and Di-
agnostics section describes analytical and computational approaches to model the
complex acoustic characteristics of combustor geometries and the interactions be-
tween flames and acoustic waves. Finally, the Combustion Instability and Control
section addresses active and passive control of combustion instabilities, including
an industry perspective into approaches for incorporating instability considerations
into the design process.
Publication of this volume was made possible through the substantial contri-
butions of a number of individuals. We would like to first thank the authors for
sharing their time and talent in preparing their manuscripts and carefully revising
them. The invaluable assistance of Rodger Williams, Heather Brennan, and Janice
Saylor of the AIAA in the preparation of the volume for publication is gratefully
acknowledged. Last, but by no means least, we wish to thank Danning You and
Yanxing Wang for providing the technical drawing services.
Timothy C. Lieuwen
Vigor Yang
July 2005
Table of Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
I. Overview
Chapter 1. Combustion Instabilities: Basic Concepts . . . . . . . . . . . . . . . . . 3

Ben T. Zinn and Timothy C. Lieuwen, Georgia Institute of Technology,
Atlanta, Georgia
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Historical Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Causes of Instabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Growth and Saturation of Instabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
II. Case Studies
Chapter 2. Combustion Instabilities in Industrial Gas Turbines:

Solar Turbines’ Experience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Kenneth O. Smith and James Blust, Solar Turbines, Inc.,
San Diego, California
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Lean Premixed Combustion System Configurations and
Operating Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Commercial Introduction at 42 ppmv NOx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Emissions Reduction to 25 ppmv NOx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Combustor Pressure Oscillation Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Centaur CPO Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Mars CPO Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Recent Experience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Conclusion: Needs and Future Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
vii
viii
Chapter 3. Incorporation of Combustion Instability Issues into

Design Process: GE Aeroderivative and Aero Engines Experience . . . 43
H. C. Mongia, T. J. Held, G. C. Hsiao, and R. P. Pandalai, GE Transportation,
Cincinnati, Ohio
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Fundamental Causes of Combustion Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Control Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Examples of Combustion Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Combustion–Acoustic Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Acoustic Modeling Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Chapter 4. Combustion Instability and Its Passive Control:

Rolls-Royce Aeroderivative Engine Experience . . . . . . . . . . . . . . . . . . . . . . 65
Tomas Scarinci, Rolls-Royce Canada, Quebec, Canada
Overview of the Trent 60 Aeroderivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Oscillatory Combustion in the Trent 60 DLE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Combustion System Design Modifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Chapter 5. Thermoacoustic Design Tools and Passive Control:

Siemens Power Generation Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Werner Krebs, Sven Bethke, Joachim Lepers, Patrick Flohr, and Bernd Prade,
Siemens AG, Mülheim, Germany and Cliff Johnson and Stan Sattinger,
Siemens AG, Orlando, Florida
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Siemens Gas-Turbine Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Phenomenological Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Solution Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Chapter 6. Characterization and Control of Aeroengine Combustion

Instability: Pratt & Whitney and NASA Experience . . . . . . . . . . . . . . . . . 113
Jeffrey M. Cohen and William Proscia, Pratt & Whitney, East Hartford,
Connecticut and John DeLaat, NASA Glenn Research Center,
Cleveland, Ohio
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Engine Combustion Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Engine Acoustic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Fuel Injector–Air Swirler Dynamic Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Subscale Combustor Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
ix
Active-Control Demonstration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
Chapter 7. Monitoring of Combustion Instabilities:

Calpine’s Experience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Jesse B. Sewell and Peter A. Sobieski, Calpine Turbine Maintenance Group,
Pasadena, Texas
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Combustion-Dynamics Monitoring System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
General Instability Characteristics and Tuning Considerations . . . . . . . . . . . . . . 151
Detrimental Impacts of Combustion Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
CDM for Combustor Health Monitoring: Case Studies . . . . . . . . . . . . . . . . . . . . 154
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Chapter 8. Monitoring Combustion Instabilities: E.ON

UK’s Experience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Catherine J. Goy, Stuart R. James and Suzanne Rea, E.ON UK, England,
Nottingham, United Kingdom
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Why Monitor Combustion Dynamics? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Description of the On-Line Combustion-Monitoring System . . . . . . . . . . . . . . . 164
Benefits of Combustion-Dynamics Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Case Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
Impact of Ambient Conditions on Dynamic Response . . . . . . . . . . . . . . . . . . . . . 166
Impact of Operating Regime on Dynamic Response . . . . . . . . . . . . . . . . . . . . . . . 167
Combustion Liner Failure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Burner Assembly Failure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
III. Fundamental Processes and Mechanisms
Chapter 9. Combustion Instability Mechanisms in

Premixed Combustors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Sébastien Ducruix, Thierry Schuller, Daniel Durox, and Sébastien Candel,
CNRS and Ecole Centrale Paris, Châtenay-Malabry, France
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Acoustics for Reacting Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Heat Release as a Pressure Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Heat-Release Fluctuations Driven by Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
x
Chapter 10. Flow and Flame Dynamics of Lean Premixed

Swirl Injectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
Ying Huang, Shanwu Wang, and Vigor Yang, Pennsylvania State University,
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
Cold Flow Characteristics of Swirl Injectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Flame Dynamics of Axial-Entry Swirl Injector . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
Chapter 11. Acoustic-Vortex-Flame Interactions in Gas Turbines . . . . 277

Suresh Menon, Georgia Institute of Technology, Atlanta, Georgia
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Length and Time Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
Theoretical Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
Factors Affecting AVF Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
Chapter 12. Physics of Premixed Combustion-Acoustic

Wave Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
Timothy C. Lieuwen, Georgia Institute of Technology, Atlanta, Georgia
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
Heat-Release Response to Flow and Mixture Perturbations . . . . . . . . . . . . . . . . 323
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
IV. Modeling and Diagnostics
Chapter 13. Acoustic Analysis of Gas-Turbine Combustors . . . . . . . . . . . 369

Ann P. Dowling and Simon R. Stow, University of Cambridge, Cambridge,
England, United Kingdom
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Linearized Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
One-Dimensional Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
Modal Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
Application to Gas-Turbine Combustors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
Modal Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
Acoustic Absorbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
xi
Limit-Cycle Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
Appendix: Derivation of Eq. (13.41) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
Chapter 14. Three-Dimensional Linear Stability Analysis of Gas

Turbine Combustion Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Danning You, Vigor Yang, and Xiaofeng Sun, Pennsylvania State University,
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
Theoretical Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
Solution Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
Sample Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
Chapter 15. Implementation of Instability Prediction in Design:

ALSTOM Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
Christian Oliver Paschereit, Hermann-Föttinger-Institute, Berlin
University of Technology, Berlin, Germany and Bruno Schuermans, Valter
Bellucci, and Peter Flohr, ALSTOM Power Ltd, Baden, Switzerland
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
Network Representation of Thermoacoustic Systems . . . . . . . . . . . . . . . . . . . . . . 447
Experimental Determination of Transfer Matrices and Source Terms . . . . . . . . 449
Modeling the Burner Transfer Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454
Reduced-Order Modeling of Complex Thermoacoustic Systems . . . . . . . . . . . . 461
Application to a Gas-Turbine Combustor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
Chapter 16. Experimental Diagnostics of Combustion Instabilities . . . 481

Jong Guen Lee and Domenic A. Santavicca, Pennsylvania State
University, University Park, Pennsylvania
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
Pressure Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Chemiluminescence Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
Infrared-Absorption Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
Laser-Induced Fluorescence Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 506
Laser Mie Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
Phase Doppler Particle Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524
xii
V. Combustion Instability Control
Chapter 17. Passive Control of Combustion Instabilities in

Stationary Gas Turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
Geo A. Richards and Douglas L. Straub, U.S. Department of Energy,
Morgantown, West Virginia and Edward H. Robey, Parsons Project
Services, Morgantown, West Virginia
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
Control-System Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
Methods to Improve Combustion Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550
Acoustic Dampers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 570
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575
Chapter 18. Factors Affecting the Control of Unstable

Combustors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
Jeffrey M. Cohen, Pratt & Whitney, East Hartford, Connecticut and Andrzej
Banaszuk, United Technologies Research Center, East Hartford,
Connecticut
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 582
Description of the Combustor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585
Actuated Fuel Mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 588
Actuation Time Delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594
Fundamental Limitations of Achievable Performance . . . . . . . . . . . . . . . . . . . . . 601
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 606
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 607
Chapter 19. Implementation of Active Control in a Full-Scale

Gas-Turbine Combustor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 611
Jakob Hermann, If TA GmbH, Groebenzell, Germany and Stefan Hoffmann,
Siemens AG, Mülheim, Germany
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 611
Implementation of AIC on Siemens-Type Vx4.3A Land-Based
Gas Turbines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613
Results and Experiences with AIC during Gas-Turbine Operation . . . . . . . . . . . 620
AIC Fault Tolerance and Long-Term Experiences . . . . . . . . . . . . . . . . . . . . . . . . . 631
Advantages of Active Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 632
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 632
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 657
Supporting Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 659
I. Overview
Chapter 1
Combustion Instabilities: Basic Concepts
Ben T. Zinn∗ and Timothy C. Lieuwen†

Georgia Institute of Technology, Atlanta, Georgia
I. Introduction
C OMBUSTION instabilities are characterized by large-amplitude oscillations

of one or more natural acoustic modes of the combustor. Such instabili-
ties have been encountered during the development and operation of propulsion
(e.g., rockets, ramjets, and afterburners), power generation (e.g., land-based gas
turbines), boiler and heating systems, and industrial furnaces. These instabilities
are spontaneously excited by a feedback loop between an oscillatory combustion
process and, in general, one of the natural acoustic modes of the combustor. In
general, the occurrence of instabilities is problematic because they produce large-
amplitude pressure and velocity oscillations that result in thrust oscillations, severe
vibrations that interfere with control-system operation, enhanced heat transfer and
thermal stresses to combustor walls, oscillatory mechanical loads that result in
low- or high-cycle fatigue of system components, and flame blowoff or flashback.
These phenomena may result in premature component wear that could lead to
costly shutdown or catastrophic component and/or mission failure. Consequently,
considerable research and development efforts have been invested during the past
half-century to elucidate the processes responsible for the excitation of these in-
stabilities and the development of approaches for their prevention. The objective
of this chapter is to provide an overview of the causes, characteristics, and control
of these instabilities.
Figure 1.1 summarizes the conditions under which combustion instability spon-
taneously occurs. The cartoon on the top shows an unstable combustor with re-
actants entering on the left and combustion products leaving through the nozzle
on the right. Interaction between one of the combustor’s acoustic modes and heat-
release oscillations transfers or removes energy from the acoustic mode. It can
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ David S. Lewis Jr. Chair and Regents’ Professor, School of Aerospace Engineering.
† Associate Professor, School of Aerospace Engineering.
3
4 B. T. ZINN AND T. C. LIEUWEN
What Causes Combustion Instabilities?

System
Combustion
Reactants Acoustic
Flame products
Oscillations
Feedback
Driving Flame adds energy to acoustic field when

Rayleigh’s Criterion is satisfied: ∫ p′(t)q′(t)dt > 0
t
phase between heat

q'
addition and pressure
p' oscillations θpq< 90°
Damping Oscillations damped by viscosity, heat transfer,

sound radiation...
Condition for Driving of Damping of

Instability : Oscillations
>
Oscillations
Fig. 1.1 Summary of conditions required for a combustion instability to occur.
be shown that the combustion process adds (removes) energy from the acoustic
oscillations locally if the integral in Fig. 1.1, which is often referred to as Rayleigh’s
integral,1 is positive (negative). The sign of this integral depends on the phase dif-
ference between the heat-release and pressure oscillations and is positive (negative)
when this phase difference is smaller (larger) than 90 deg. As shown in Fig. 1.1,
combustion instability spontaneously occurs only if the energy supplied to the
acoustic mode by the combustion process exceeds the energy losses of the mode
caused by, for example, radiation and convection of acoustic energy out of the
combustor (e.g., through the nozzle), viscous dissipation, and heat transfer. Thus,
as long as the magnitude of the driving exceeds the magnitude of the damping
process, the energy of the mode will increase with time. In such a case, the ampli-
tude of oscillations initially increases exponentially with time until it saturates at
some limit-cycle amplitude. When this occurs, the time averages of the driving and
damping processes are equal and no net energy is added to the oscillating mode.
In general, combustion instabilities occur at frequencies associated with natural
acoustic modes of the combustor. These include, for example, bulk (i.e., Helmholtz-
type oscillations), axial, and transverse (i.e., tangential and/or radial) modes (see
Fig. 1.2). On occasion, however, the oscillations are not associated with a purely
acoustic mode and are excited by a coupled convective-acoustic mode similar to
cavity tones,2 which often occur at frequencies lower than those of purely acoustic
modes. Such oscillations occur when an entropy wave (i.e., a hot-gas packet) or
a vortex generated in the flame region is convected toward (and impinges on) the
nozzle, at which it excites an acoustic wave that propagates back to the flame,3,4
exciting another convected wave and thus repeating the process. These types of
modes are often encountered in systems that are operating at conditions close to
flame blowoff.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 5
Transverse Radial Mode
Longitudinal
Transverse Azimuthal Mode

Fig. 1.2 Examples of longitudinal and transverse acoustic modes that are excited in
cylindrical combustors.
Because the initial amplitudes of most instabilities are generally quite small,
their characteristics are described by the linear-wave equation.5 The frequencies
and mode shapes of these oscillations and the conditions under which they sponta-
neously occur are determined by the solutions of these equations. Linear analyses
cannot, however, predict the magnitude of the limit-cycle amplitude attained by the
instability because it is controlled by nonlinear processes. Furthermore, nonlinear
processes may allow a large-amplitude disturbance whose amplitude exceeds a
certain threshold value, A T , to trigger instability in a system that is linearly stable;
that is, one in which low-amplitude oscillations are not spontaneously self-excited.
Consequently, both the characteristics of the limit-cycle oscillations and conditions
under which finite amplitude disturbances trigger instabilities can only be deter-
mined by solving the nonlinear equations that describe the system’s dynamics.6,7
To prevent the onset of detrimental combustion instabilities, the processes re-
sponsible for their driving and damping must be understood. This chapter presents
an overview of the current understanding of these processes with the objective of
providing the reader with the background needed for the more detailed discussions
of these and related subjects in the remaining chapters of this volume. To attain this
goal, the following topics are discussed in this chapter: the history of this problem,
driving and damping of combustion instabilities, common instability mechanisms
in gas turbines, the initial growth and saturation of the oscillations, and some basic
characteristics of limit-cycle oscillations.
II. Historical Overview

This section provides a brief overview of prior experience with combustion insta-
bilities. Although instabilities have been observed in a variety of combustion sys-
tems, related phenomena have also been excited in other systems by heat-transfer
processes.8 For example, glass blowers have reported observations of spontaneous
excitation of acoustic oscillations during the heating of the closed end of a blown-
glass tube. Also, acoustic oscillations can be spontaneously excited inside tubes
with sharp-temperature gradients, as has been observed in cryogenic systems.9
Because in all these cases the acoustic oscillations are excited by thermal sources,
the resulting phenomena are often referred to as thermoacoustic instabilities.
p' – pressure
oscillation
u' – velocity
oscillation
Fuel+Air
Fig. 1.3 An example of a flame-driven instability in a tube open at both ends.
With a focus on combustion-driven oscillations for the remainder of the chapter,

the first observation of flame-driven oscillations (referred to as singing flames)
dates back to 1777.10 This and subsequent studies found that spontaneous acoustic
oscillations of considerable amplitude might be generated when a gas flame is
placed inside a larger-diameter tube, as illustrated in Fig. 1.3. The sensitivity of
flames to music at a musical party was noted by Le Conte11 in 1858 and described
in the following quotation: “Soon after the music commenced, I observed that
the flame exhibited pulsations exactly synchronous with the audible beats. This
phenomenon was very striking to everyone in the room, and especially so when
the strong notes of the violoncello came in. . . . A deaf man might have seen the
harmony. . . .”
Combustion oscillations moved beyond an academic curiosity with the advent of
high-intensity combustion systems. Detrimental, combustion-driven oscillations
have been observed in boilers, blast furnaces, and a variety of other oil, coal,
and gas-fired heating units.12 The occurrence of these instabilities was generally
unexpected as demonstrated, for example, by recent experience with one of the
nation’s largest landfill sites in Los Angeles County (Fig. 1.4). The site has two
45-ft-high, 12-ft-diam flares, each designed to flare excess landfill gas at rates of up
to ∼50 MW. However, at loads greater than about 50% of full capacity, instability
of the quarter-wave mode of the flare, oscillating at about 10 Hz, is excited in the
system.13
Instabilities have also significantly hindered the development of various liquid-
fueled rockets. Notable were the instabilities encountered during the development
of the F-1 engine that powered the Saturn rockets, which were used in the first
manned mission to the moon. The F-1 encountered instabilities with amplitudes up
to 100% of the mean combustor pressure (i.e., more than 2000 psi) with frequen-
cies in the 200- to 500-Hz range. These instabilities caused significant damage to
the combustor, and their elimination required a costly trial and error, development
program that included ∼2000 full-scale tests (of a total of 3200). One of the so-
lutions developed involved welding a system of baffles to the injector face. These
baffles prevented the excitation of the transverse acoustic oscillations that could
be driven by the combustion process near the injector face (Fig. 1.5).
Fig. 1.4 Landfill gas flares that experienced low-frequency, combustion-driven

oscillations.13
Fig. 1.5 Photograph of rocket injector with welded baffle plates that was tested to
prevent transverse mode instabilities during F-1 engine development.26
Combustion instabilities have also been encountered in numerous solid-

propellant rockets, including the Space Shuttle solid-propellant rocket boosters,
the Minuteman intercontinental ballistic missile, and the Mars Pathfinder descent
motor.14 The Minuteman missile provides a good example of the enormous dif-
ficulties that instabilities generated in solid-propellant rocket development and
testing programs. In 1968, the U.S. Air Force experienced five flight failures of
the Minuteman Wing I missiles during routine tests. This rate of failure raised
concerns about the condition of the systems that were already fielded in missile
silos and resulted in costly removal and modifications of many of the systems.
Combustion instabilities in ramjet-powered missiles have also been problem-
atic because they cause thrust oscillations and shock-system oscillations in the
inlet diffuser, which lead to a reduced stability margin of the inlet flow.15 Sim-
ilar problems have also been encountered in afterburners, in which instabilities
of transverse acoustic modes, generally referred to as “screech,” and axial modes
damage flame holders, liner sections, and other engine components.15
III. Causes of Instabilities

A. Combustion Process: Acoustic Modes Energy Transfer
This section discusses the mechanisms through which the combustion process
can drive acoustic modes of the system. Lord Rayleigh1 was the first to state
the so-called Rayleigh criterion, which describes the conditions under which a
periodic heat-addition process adds energy to acoustic oscillations. Paraphrased,
the criterion states that a periodic heat-transfer process adds energy to the acoustic
field if the heat is added to or removed from the gas when its pressure is above or
below its mean value. This statement is described mathematically by the integral
in Fig. 1.1. It basically says that the heat-addition process locally adds energy
to the acoustic field when the magnitude of the phase between the pressure and
heat-release oscillations, θpq , is less than 90 deg (i.e., 0 < |θ pq |< 90). Conversely,
when these oscillations are out of phase (i.e., 90 < |θ pq |< 180), the heat-addition
oscillations damp the acoustic field.
The physical reason for this energy exchange follows from determining the
conditions under which the unsteady heat release performs work on the gas. Heat
release at constant pressure results in gas expansion, analogous to blowing up a
balloon. The Rayleigh criterion states that the unsteady heat release performs work
on the gas when this expansion occurs in phase with the pressure. This statement is
analogous to problems in mechanics in which an unsteady force (i.e., the pressure)
must be in phase with the velocity (i.e., the gas-dilatation rate) if net work is to be
performed.
One may also gain physical insight into the Rayleigh criterion by considering
its similarities to the manner in which the combustion-process energy is converted
into work in a Brayton cycle, the thermodynamic cycle used in gas turbines. In
this cycle, thermal energy is added to the working fluid at high pressure after
it has been compressed in the diffuser and compressor. Work is subsequently
extracted from the working fluid at low pressure as it is expanded through the
turbine and nozzle. The similarity between the Brayton cycle and the Rayleigh
criterion becomes apparent if one recalls that according to the Rayleigh criterion
one can drive acoustic pressure oscillations in a system by heating the gas with
an instantaneous “heat-addition pulse” when its pressure is maximum and cooling

it with an instantaneous “cooling pulse” when its pressure is minimum. Such a
process essentially describes the manner in which heat is added and removed in a
Brayton cycle to do work.
The Rayleigh criterion describes the conditions under which unsteady heat re-
lease adds energy to the acoustic field. However, even if energy is transferred from
the combustion process to the acoustic field, this does not necessarily imply that
the combustor is unstable. As described in Fig. 1.1, acoustic oscillations are spon-
taneously excited in a combustor only when the rate of energy supplied by the
periodic combustion process to the acoustic field is larger than the rate at which
acoustic energy is dissipated within the combustor and/or transmitted through its
boundaries. This statement is summarized by the following expression:

p (x, t) q (x, t) dt dV ≥ L i (x, t) dt dV (1.1)
i
VT VT
where p (x, t), q (x, t), V, T , and L i are the combustor pressure oscillations, heat-
addition oscillations, combustor volume, period of the oscillations, and the ith
acoustic energy loss process (e.g., viscous dissipation, radiation of acoustic en-
ergy out of the combustor through its boundaries), respectively. The equal sign in
Eq. (1.1) describes conditions when limit-cycle oscillations are attained and the
time average of the energies added and removed from the oscillations are equal.
The integral on the left side of Eq. (1.1) is referred to as the Rayleigh integral and
is often used in experimental or numerical studies to quantify the energy transfer
from the combustion process to the acoustic field. Note that the inner integral on
the left side of Eq. (1.1) is the integral shown in Fig. 1.1, which describes the
local driving/damping of the acoustic oscillations by the heat-addition process
q (x, t).
B. Instability-Driving Mechanisms
As discussed earlier, combustion instabilities are excited by feedback between
the combustion process and acoustic oscillations that depends on the system char-
acteristics and operating conditions. This section provides a brief overview of
common instability mechanisms and the conditions under which they are self-
exciting.
Figure 1.6 describes the generic feedback loop responsible for combustion in-
stabilities. It consists of the following sequence of events: 1) Fluctuations in the
velocity and/or thermodynamic-state variables induces a fluctuation in the heat-
release rate, 2) the heat-release fluctuation excites acoustic oscillations, and 3) the
acoustic oscillations generate the velocity and thermodynamic-state variable fluc-
tuations that are described in step 1, thus closing the feedback loop. Depending
on the relative magnitudes of the energy added and removed from the acoustic
oscillations, the amplitude of oscillations may decrease, remain constant, or grow
during each cycle of this loop.
Several mechanisms capable of driving combustion instabilities in gas turbines
have been identified, as indicated in Fig. 1.7. Because most of these mechanisms
Heat Release
Oscillations
Flow and
Mixture Acoustic
Perturbations Oscillations
Fig. 1.6 Illustration of the feedback processes responsible for combustion instability.
are discussed in more detail in Chapter 9, the remainder of this section only pro-
vides a brief description of each.
1) Fuel Feed Line–Acoustic Coupling.5,16 Pressure oscillations in the combustor

modulate the pressure drop across unchoked fuel nozzles. The pressure drop, in
turn, modulates the fuel-injection rate into the system, causing an oscillatory heat-
release process that drives the acoustic oscillations.
2) Equivalence-Ratio Oscillations.17 Combustor pressure oscillations propagate
into the premixer section in which they modulate mixing processes and fuel and/or
air supply rates, thus producing a reactive mixture whose equivalence ratio varies
periodically in time. The resulting mixture is convected into the flame in which it
produces heat-release oscillations that drive the instability.
3) Oscillatory Atomization, Vaporization, and Mixing.5,16 Interactions of the
acoustic field with the fuel spray produce periodic variations of the fuel-spray
shape, droplet sizes, evaporation rates, and mixing rates of the fuel vapor with
surrounding gases. These variations, in turn, could result in periodic supply rates
of fuel to the flame and/or periodic variations of the equivalence ratio that produce
heat-release oscillations that drive the acoustic field.
Fuel/air ratio
Fuel flow rate oscillations
oscillations
Flame area and
reaction rate
oscillations
Flow rate
oscillations Combustion
Products
Unsteady mixing,
vaporization,
atomization Vortex/flame
interactions
Fig. 1.7 Flow and flame processes that can cause combustion instabilities in
gas turbines.
Fig. 1.8 Computed image of swirling flame distorted by vortical structures. Courtesy
of Y. Huang and V. Yang.40
4) Oscillatory Flame-Area Variation.18 Interactions of acoustic velocity oscilla-

tions with the flame cause periodic variation of the flame area and, thus, a periodic
heat-addition process that drives the acoustic field.
5) Vortex Shedding.19,20 Large-scale, coherent vortical structures caused by flow
separation from flame holders and rapid expansions, as well as vortex break-
down in swirling flows, are often present in gas-turbine combustors, as shown in
Fig. 1.8. In the initial stage of their formation, these vortices generally consist of
combustible gases. As they form, these vortices entrain hot products and ignite.
This ignition is followed by rapid combustion of the reactants within the vortex and
sudden breakdown of the coherent vortical structure into small-scale turbulence.
Alternatively, the vortical structures may distort the flame and cause its surface
area to oscillate, thus producing an oscillatory heat-release rate process that can
also drive the acoustic field if the heat addition and pressure oscillations are in
phase. These dynamics are also discussed in Chapters 10 and 11 of this volume.
Heat-release oscillations add energy to an acoustic field if the magnitude of

the characteristic timescales of the heat-addition process are of the order of some
integer multiple of the acoustic period. This point is illustrated in Fig. 1.9, which
shows the hypothetical response of 1) the fuel-injection rate, 2) the rate of change
of droplet-size distribution, 3) the rate of heat transfer to the fuel, and 4) the rate
of fuel–air mixing to a pressure pulse with period T . The fluctuations in each of
the rates of the preceding four processes causes a fluctuation in the heat-release
rate after a time delay τ , which depends on the characteristics of the analyzed
process, the combustor-operating conditions, and design (Fig. 1.9). For example,
the magnitude of the time delay τ1 depends on various geometric and operating
parameters of the fuel-delivery system; for example, the fuel-line length and fuel
gas temperature.
T
a. Pressure
T
at flame
τ1
b. Process 1
τ2
Heat Release
Perturbation c. Process 2
due too: τ3
d. Process 3
τ4
e. Process 4
time
Fig. 1.9 Hypothetical responses of various combustor processes to a periodic pressure

disturbance.
According to the Rayleigh criterion, the heat-release oscillations shown in

Fig. 1.9 add energy to the acoustic field if the time average of the product of
the pressure and heat-release oscillations is greater than zero. Figure 1.9 shows
that the heat release attributable to process 1 satisfies this criterion and that the heat
release by processes 2–4 does not. The heat release of process 1 and the pressure
pulses are positively correlated because the characteristic time of the process τ1
is of the order of the period of the acoustic waves, T . Combustion instability will
thus occur if the rate of energy addition to the disturbance by process 1 exceeds
its rate of damping.
Keep in mind that any of a number of the natural acoustic modes of the combus-
tion system can be excited. To illustrate the relationships between different acoustic
modes and various driving mechanisms, assume that the pressure disturbance of
period T in Fig. 1.9 is caused by oscillations of the first longitudinal acoustic
mode of the combustor. If we now repeat the preceding analysis to determine the
driving of the second longitudinal acoustic mode of the combustor with a period of
T /2, then τ1 = 2T and τ2 = T . Consequently, in the example shown in Fig. 1.9,
energy is added to this mode by both processes 1 and 2. This example shows that
an instability could be excited if the characteristic time of the combustion process
equals T , 2T , or any other integer multiple of T .
The preceding examples illustrate several important points. First, they show that
different mechanisms may play different roles in the stability of different modes of
the combustor. For example, the mechanism(s) responsible for exciting a 100-Hz
longitudinal mode of a combustor will significantly differ from those that excite a
5000-Hz transverse-mode instability. Second, these examples show that the role of
various mechanisms may change with operating conditions. For example, suppose
changes in combustor-operating conditions change the timescales of processes 1
and 2. Such a change in timescales could affect the coupling between the pressure
and heat-addition disturbances and, thus, the role of each process in driving various
combustor modes. Finally, the preceding examples demonstrate that the charac-
teristic times associated with processes that are responsible for the excitation of
0.08
p' (arbitrary units)

0.06
0.04 430 Hz
630 Hz
0.02
0
15 20 25 30 35 40
Premixer Velocity (m/s)
Fig. 1.10 Measured25 dependence of the instability amplitude on the premixer

velocity.
combustion instability must have magnitudes that are of the order of some integer
multiple of the acoustic period of the modes.
Note also that combustor pressure oscillations generally vary harmonically with
time and do not exhibit the pulselike behavior used in the examples in Fig. 1.9.
Consequently, the time delays and acoustic periods of the various modes need
not be exactly equal for the process to add energy to the acoustic field; instead,
these characteristic times must satisfy a relationship of the form T − T /4 < nτ <
T + T /4, where n is an integer; that is, n = 1, 2, . . . 17
Some of these points are illustrated by the results in Fig. 1.10,17 which describe
the dependence of the amplitudes of a 430- and 630-Hz combustor mode on the
mean velocity of the reactants in the combustor premixer. The instabilities in this
combustor were driven by a mechanism that depends on the time required to trans-
port the reactants from the fuel-injection point to the combustor,17 a time delay that
is inversely proportional to the mean velocity of the gases in the premixer. Con-
sequently, this time delay decreases as the gas velocity in the premixer increases.
Figure 1.10 shows that, as the velocity in the combustor premixer increased and
the convective time delay decreased, the amplitude of the lower-frequency mode
(i.e., 430 Hz with the longer period T ) decreased and the amplitude of the higher-
frequency mode (i.e., 630 Hz with the shorter period T ) increased. This finding
indicates that when the premixer velocity was low the driving process coupled
with the 430-Hz mode and this coupling “switched” to the 630-Hz mode as the
premixer velocity increased.
C. Damping Processes
As noted earlier, acoustic damping processes play an important role in the
determination of the conditions under which combustion instability occurs. This
section summarizes important damping mechanisms in combustors and the key
parameters that influence their magnitude.
Energy can be dissipated/removed from an unstable mode via the following three
processes: 1) transfer of acoustic energy to vortical or entropy disturbances through
viscous and heat-transfer processes, respectively; 2) convection and/or radiation
of acoustic energy out of the system; and 3) transfer of energy between acoustic
modes. Note that in all these cases, the “dissipation process” refers to the transfer
of acoustic energy out of the combustor, frequency regions, or modes of oscillation
in a manner that reduces the acoustic energy of the unstable mode.For example,
in the second mechanism, the acoustic energy is simply radiated or convected out
of the combustor through its boundaries and represents part of the noise heard
outside the system. The rest of this section describes each of these mechanisms in
further detail.
1. Viscous and Heat-Transfer Damping Mechanisms

This mechanism can be further subdivided into two submechanisms: boundary-
layer losses and flow-separation losses. Boundary-layer losses occur when acoustic
motions are present in the vicinity of surfaces on which viscosity and thermal-
dissipation effects dominate (see Ref. 21). Suppose that an acoustic wave that
perturbs the flow velocity and temperature impinges obliquely on a rigid wall.
Because of the no-slip boundary condition at the wall, the energy in the acoustic
mode is partially transferred into vortical velocity fluctuations. In an analogous
manner, the temperature-boundary condition at the wall (e.g., zero-amplitude tem-
perature oscillations) causes some acoustic energy to be converted into entropy
fluctuations. Consequently, the energy of the acoustic wave reflecting from the
wall is smaller than that of the incident wave, because some of its energy is con-
verted into vorticity and/or entropy fluctuations. These processes are analogous to
those occurring in a steady pipe flow in which heat transfer and viscosity result in
stagnation-pressure losses. The magnitudes of these dissipation mechanisms in-
crease with frequency as ( f τv )1/2 , where τv is the viscous or thermal transport
timescale.
Flow separation at sharp edges or during rapid flow expansions also damp acous-
tic waves by converting acoustic energy into vorticity. This damping mechanism is
analogous to the stagnation-pressure losses in steady, separated flow downstream
of sharp corners or edges, tabulated in many engineering handbooks and fluid me-
chanics texts.22 This dissipation mechanism has nonlinear-amplitude dependence
and is discussed further in Sec. IV.C.
2. Convection and/or Radiation of Acoustic Energy

Acoustic energy inside a duct can leave the system by propagation and/or con-
vection by the mean fluid motion out of the system. For example, the tone heard
from an organ pipe is caused by some of the sound energy in the standing wave
inside the pipe propagating out through its open end. In general, this damping mech-
anism scales with frequency as ( f D/c)2 , where f , D, and c are the frequency, pipe
diameter, and sound speed, respectively. Sound energy is also convected out of the
system by fluid motion with a magnitude that roughly scales with the mean flow
Mach number. Because the magnitude of acoustic damping in a typical system is
quite small, in general, the presence of low-velocity mean flow can have a signif-
icant impact on the system’s damping level. To illustrate this point, consider the
dependence of the magnitude of a reflected acoustic wave incident on the open
end of a pipe upon the flow Mach number in Fig. 1.11. It shows that the reflection
coefficient equals 0.95 in the no-flow case, implying that the wave amplitude is re-
duced by 5% because of acoustic radiation. This reflection coefficient is reduced to
Reflection Coefficient
0.8
0.6
0.4
0.2
0
0 0.1 0.2 0.3 0.4
Mach Number
Fig. 1.11 Dependence of the reflection coefficient from an open-ended pipe on the
flow Mach number. Taken from Ingard and Singhal.41
0.85 at M = 0.05, implying that the presence of a very-low-velocity flow increases

the acoustic damping by a factor of 3.
3. Transfer of Energy Out of Natural Acoustic Frequencies

to Other Frequencies
Oscillations in an unstable combustor, in general, are composed of one or more
nearly pure tones at specific frequencies. This composition occurs because com-
bustors are generally lightly damped acoustic resonators that respond to external
disturbances over very-narrow-frequency ranges. Consequently, mechanisms that
do not directly dissipate acoustic energy but transfer energy from the excited
modes to other modes that oscillate at frequencies that are either not amplified
in the combustor or at which the energy is more readily dissipated, effectively
constitute “dissipation” mechanisms for a given natural acoustic mode. The energy
transfer can be to either narrowband, coherent, fluctuations at other frequencies or
to incoherent, broadband fluctuations. Nonlinear combustor processes are a key
mechanism for enabling the former; that is, they transfer energy from a certain fre-
quency f o , to higher harmonics (2 f o , 3 f o , . . .) or subharmonics ( f o /2, f o /3, . . .).
The energy at these higher frequencies is usually dissipated more rapidly by the pre-
viously discussed viscous and radiation mechanisms whose magnitudes increase
with frequency. Nonlinear combustor processes are discussed further in Sec. IV.
Acoustic energy is transferred from narrowband, coherent oscillations to spec-
trally distributed, incoherent oscillations by random modulation processes. For
example, such spectral broadening occurs during reflection and scattering of a
sound wave from turbulent eddies23 or a randomly flapping flame front.24 The
energy-transfer mechanism in this case can be purely linear and is essentially
caused by a random Doppler shift of the reflected and transmitted waves. For ex-
ample, when a pure tone with a temporal dependence given by sin(ωt) impinges on
a randomly moving flame front or a medium with a random velocity field, reflected
and transmitted waves with a random phase shift are generated. The temporal de-
pendence of the reflected/transmitted waves is given by sin[ωt + φ(t)], where φ(t)
is a random phase shift. This mechanism is discussed further in Chapter 12.
IV. Growth and Saturation of Instabilities

As indicated by Eq. (1.1), the amplitude of the instability grows if the rate
of energy addition to the oscillations exceeds the rate of energy dissipation by
damping processes. As the amplitude of the oscillations increases, the energy
addition and dissipation processes become amplitude dependent and the amplitude
of the oscillations attains its maximum value when the time average of the energy
addition and removal equal one another. The resulting oscillations are referred
to as a limit cycle. The objective of this section is to consider in more detail the
growth and saturation of the instability amplitude.
A. Linear and Nonlinear Stability

As discussed in the introduction, the mechanisms that initiate combustion in-
stabilities are typically grouped into linear and nonlinear categories. A linearly
unstable system is one that is unstable with respect to infinitesimally small dis-
turbances. Thus, because linearly unstable systems can not exist, they are never
observed in nature. An example of a linearly unstable situation is a ball perfectly
balanced at the crest of a hill, where any small disturbance will cause it to roll
away from this unstable equilibrium point.
To further illustrate the dependence of the limit cycle on the amplitude of the
oscillations A consider the hypothetical, amplitude-dependent, driving, H(A), and
damping, D(A) processes, which are described in Fig. 1.12 and the following
expressions:
H(A) = εH A + Hn (A) (1.2)

D(A) = εD A + Dn (A)
where subscript n denotes the nonlinear part of the transfer function.

As shown in Fig. 1.12, the driving and damping curves intersect at the origin,
indicating that a zero-amplitude oscillation is a potential equilibrium point. This
equilibrium point is unstable, however, because any small disturbance that moves
the system away from the origin produces a condition in which H( A) is larger than
εH
Driving/Damping
1 H(A)
D(A)
εD
1
A ALC
Fig. 1.12 Hypothetical dependence of the acoustic driving, H( A), and damping, D( A),
processes on the instability amplitude A.
D(A), resulting in further growth of the disturbance. Because these two curves
diverge near the origin, their difference increases with amplitude, implying that
the amplitude growth rate increases with amplitude. When the amplitudes of the
oscillations are small, the driving and damping processes can be approximated
by linearized expressions; that is, H(A) = εH A and D(A) = εD A. Thus, linear
combustor processes (i.e., processes whose magnitude is directly proportional to
amplitude A), in general, control the balance between driving and damping pro-
cesses when the amplitude of oscillations is small and thus determine the frequency
and growth rate A ∼ eαt of inherent combustor disturbances. It can be shown that
the initial growth rate of the instability α is proportional to the difference between
the driving and damping processes in the linear regime; that is, α ∼ εH − εD .
For this reason, linear combustor stability models (such as those described in
Chap. 13) are routinely used to determine whether a given operating point is stable
or unstable.
Nonlinear combustor processes control the dynamics of the oscillations as the
driving and damping processes become amplitude dependent. Figure 1.12 de-
scribes a situation in which H(A) saturates and D(A) increases linearly with the
amplitude A, thus resulting in an intersection of the two curves at the limit-cycle
amplitude A LC . Note that this limit-cycle amplitude is stable; that is, a perturbation
of the amplitude to the left (right) of this intersection point causes H(A) to become
larger (smaller) than D(A), thus causing the amplitude to increase (decrease) and
return to its limit-cycle value A LC .
Next, consider a situation in which some combustor parameter is systematically
varied in such a way that εh increases while εd remains constant (see Fig. 1.12).
For εh < εd , A = 0 is the stable solution; that is, the system is linearly stable be-
cause all perturbations imposed on the system will decay. However, when εh > εd ,
the solution A = 0 becomes unstable, causing the amplitude of the disturbance to
increase toward a new stable, limit-cycle equilibrium. The εh = εd condition sep-
arates two regions of fundamentally different dynamics and is referred to as a
supercritical bifurcation point. These ideas are illustrated in the bifurcation dia-
gram in Fig. 1.13, which shows the dependence of the amplitude A LC on εh − εd .
Figure 1.13 shows that as εh − εd becomes positive, the system becomes linearly
Stable
Unstable
Amplitude
εh - εd
Fig. 1.13 An example of a supercritical bifurcation.

0.02
Normalized Pressure (p'/p)

0.015
0.01
0.005
0
18 21 24 27 30
Mean Inlet Velocity (m/s)
Fig. 1.14 Measured data describing the occurrence of supercritical bifurcation at

combustion inlet velocity of 23.5 m/s.25
unstable. Although εh − εd describes variation along the x-axis in Fig. 1.13, it

could be replaced in practice by any parameter that affects the system’s stability,
for example, the air velocity or temperature as demonstrated by the measured data
in Fig. 1.14.25 It shows a smooth, monotonic dependence of the amplitude on the
premixer velocity, which is indicative of the presence of the previously discussed
supercritical bifurcation in this combustor.
A nonlinearly unstable system differs from a linearly stable one in that it is sta-
ble with respect to small-amplitude disturbances but is unstable when subjected to
disturbances whose magnitude exceeds a certain threshold value A T . This type of
instability is sometimes referred to as subcritical. A simple example of a nonlin-
early unstable system is shown in Fig. 1.15, which shows a ball in a depression on
the top of a hill. When pushed, this ball returns to its equilibrium point as long as it
is subjected to disturbances with amplitude that does not get it over the sidewalls
of the depression. However, for a sufficiently large disturbance amplitude, the ball
will roll out of the depression and down the hill. As discussed earlier, such behav-
ior is often referred to as “triggering” in the context of combustion instabilities in
rockets. A simple example of a nonlinearly unstable combustion system is one in
which the response of the heat-release process to flow perturbations depends non-
linearly on the amplitude of the disturbance. For example, a small perturbation in
pressure may cause the rate of heat release to fluctuate slightly. However, pressure
perturbations of sufficiently large amplitudes may cause the combustion process
to temporally extinguish. The large-amplitude heat-release oscillations induced by
Fig. 1.15 A simple example of a nonlinearly unstable state.

H(A)
Driving/Damping
1
εD
D(A)
1 εH
A AT ALC
Fig. 1.16 Hypothetical dependence of the acoustic driving, H( A), and damping, D( A),
processes on amplitude A, that produce the triggering of instabilities.
these transient extinguishment and reignition processes may cause instability in

the combustor. Rockets are prone to nonlinear triggering for reasons that are not
fully understood.
Although large-amplitude disturbances are generally required to initiate unsta-
ble oscillations in nonlinearly unstable systems,26 a system may be nonlinearly
unstable at low-amplitude disturbances that are of the order of the background-
noise level. This scenario is somewhat analogous to the hydrodynamic stability
of a laminar Poiseuille flow,27 which is linearly stable but becomes increasingly
susceptible to destabilization by nonlinear mechanisms with increasing Reynolds
numbers. It is important to recognize this point because it indicates that it is difficult
to determine from measured data whether the combustor is linearly or nonlinearly
unstable.
Figure 1.16 provides an example of the amplitude dependences of H( A) and
D(A) that produce the behavior discussed earlier. In this case, the system has three
equilibrium points at which the driving and damping curves intersect. Specifically,
the damping exceeds the driving when A < A T , indicating that A = 0 is a stable
fixed point, because all disturbances in the range 0 < A < A T decay to A = 0.
The next equilibrium amplitude at which the driving and damping curves intersect
is at the triggering amplitude A = A T . This amplitude is an unstable equilibrium
point because any disturbance that shifts the system from this point continues to
increase in time. The third equilibrium point, A = A LC , is a stable limit cycle.
Thus, in such a system all disturbances with amplitudes A < A T return to the
stable solution A = 0 and disturbances with amplitudes A > A T grow until their
amplitude attains the value A = A LC . Consequently, two stable solutions exist
for this operating condition. The one observed at any point in time will depend
on the history of the system. This triggering behavior is probably not limited to
rockets. As discussed in Chapter 12, nonlinearities in the response of premixed
flames to acoustic oscillations result in H( A) curves resembling those shown in
both Fig. 1.12 and Fig. 1.16; thus, both subcritical and supercritical bifurcations
should be expected in gas turbines as well.
A typical bifurcation diagram for this type of system is shown in Fig. 1.17.
It shows that for εh < εd , the A = 0 and A = A LC are stable solutions, as noted
Stable
Unstable
Amplitude
AT
εh-εεd
Fig. 1.17 An example of a subcritical bifurcation.
before. For εh > εd , the A = 0 solution becomes unstable (as indicated by the solid
line becoming dashed), and only a single stable solution is present. In this case, if
a system parameter is monotonically increased to change the sign of εh − εd from
a negative to a positive value, the system’s amplitude will jump discontinuously
from A = 0 to A = A LC at εh − εd = 0. Hysteresis is also present in the system,
because if the system parameter is subsequently decreased, the system’s amplitude
decreases as it follows the stable branch on top, even for a range of εh < εd
values, before it discontinuously “jumps” to the A = 0 solution. Experimental
data exhibiting such behavior are presented in Fig. 1.18.25
B. Other Characteristics of Limit-Cycle Oscillations

The preceding discussion has shown that both linear and nonlinear combustion
and acoustic processes control the dynamics of unstable combustors. In general,
linear processes control the balance between driving and damping processes for
small-amplitude disturbances and, thus, determine the conditions under which
spontaneous self-excited instabilities occur. On the other hand, nonlinear combus-
tor processes control the dynamics of finite amplitude oscillations.
0.015
Normalized Pressure (p'/p)
0.0125
0.01
0.0075
0.005
0.0025
0
13 13.5 14 14.5 15 15.5
Mean Inlet Velocity (m/s)
Fig. 1.18 Experimental data showing evidence of a subcritical bifurcation. Taken

from Lieuwen.25
Nonlinear processes are also responsible for two other phenomena observed in
unstable combustors: generation of harmonics and changes in the mean value of
certain system properties. These phenomena can be understood by considering
the response of a nonlinear system to harmonic forcing. Consider, for example,
the gas-dynamic nonlinearity generated by the isentropic relationship between
pressure and density; that is,
γ
p(t) ρ(t)
= (1.3)
p̄ ρ̄
Assume that the density oscillates harmonically with an amplitude A and angular
frequency ω, that is,
ρ (t)
= A sin ωt (1.4)
ρ̄
Substitute this expression into Eq. (1.3) and expand the resulting expression in
a Taylor series about A = 0 to obtain the following expression for the pressure
oscillations:

p (t) (γ − 1) 2 (γ − 1)(γ − 2) 3 (γ − 1) 2
= A + A+ A sin ωt − A cos 2ωt
γ p̄ 4 8 4
(γ − 1)(γ − 2) 3
− A sin 3ωt + · · · (1.5)
24
Equation (1.5) shows that the expansion for the pressure consists of a linear term
that is proportional to A (the second term on the right), and a collection of nonlinear
terms that are proportional to higher powers of the amplitude A. The latter includes
a time-independent term (first term on the right) that is proportional to A2 and
represents the change in the mean pressure caused by the purely oscillatory density
disturbance. Furthermore, the expansion includes harmonics that are proportional
to A2 , oscillating at twice the disturbance frequency, and terms proportional to
A3 , oscillating with frequencies of 3ω and ω. Such harmonics of the unstable
combustor mode are routinely observed.
Changes in mean pressure of up to several hundred pounds per square inch have
been observed during very-large-amplitude instabilities in solid rockets. On the
other hand, instabilities in gas turbines generally have much smaller magnitudes,
and no reports have been published of significant shifts in mean pressure because
of these instabilities. However, changes in mean flame location and length are
routinely observed.28,29 These changes, in turn, can have an impact on the static
pressure distribution in the combustor because it depends on the distribution of the
combustion-process heat release.
C. Causes of Nonlinearities
Numerous processes that drive and damp oscillations in unstable combustors de-
pend nonlinearly on the instability amplitude. These nonlinearities may be caused
by flow and combustion processes that occur either within the combustor volume
or at its boundaries.
Gas-dynamic nonlinearities within the combustor volume are introduced by
processes described by nonlinear terms in the Navier–Stokes and energy equations,
for example, convective terms such as ρ u · ∇ u in the Navier–Stokes equation
or the previously discussed nonlinear pressure–density relationship. Such terms,
in general, become significant when the amplitudes of the fluctuating pressure,
density, or velocity become on the order of the mean pressure, density, or speed of
sound, respectively. Consequently, these terms are generally not important when
the relative amplitudes of the acoustic disturbances are low (e.g., p / p̄ < 10%).
On the other hand, when very-large-amplitude oscillations are encountered, these
nonlinear processes strongly affect the characteristics of the instabilities in these
systems. Examples of the treatments of these nonlinear processes can be found in,
for example, the works of Zinn and coworkers,30 Culick and coworkers,31,32 and
Yang.33
Combustion-process nonlinearities are introduced by the nonlinear dependence
of the heat-release oscillations on the disturbance amplitude. Additionally, flow
oscillations may control the response of the combustion process, resulting (in
these cases) in nonlinearities that become important when the ratio u /ū ∼ O(1).25
Consequently, in these cases, the relevant velocity scale that determines when
nonlinearities are important is the mean velocity and not the sound speed (as in
the gas-dynamic nonlinearities). In fact, it has been shown that such combustion-
process nonlinearities play a key role in the stability of lean premixed combus-
tion systems.34 Further discussion of these nonlinearities can be found in Chap-
ters 12 and 13.
The nonlinearities in processes that occur at or near the combustor bound-
aries also affect the combustor dynamics as they are introduced into the anal-
ysis of the problem through nonlinear boundary conditions. Such nonlinearities
are caused by, for example, flow separation at sharp edges or rapid expansions,
which, as discussed earlier, cause stagnation-pressure losses and a corresponding
transfer of acoustic energy into vorticity. The resulting nonlinear damping is pro-
portional to ρ| u |2 , resulting in an unsteady damping process that is proportional
u |2 + 2|
to ρ̄(| u | · |u¯ |) + ρ (|u¯ |2 + |
u |2 + 2|
u | · |u¯ |).35 This expression indicates
that the presence of mean flow introduces linear and nonlinear damping terms and
that the damping is proportional to the square of the magnitude of the velocity am-
plitude | u |2 in the absence of mean flow (i.e., |u¯ | = 0). Such nonlinear damping
has been well characterized in experiments (see Ref. 35, for example).
Also, wave reflection and transmission processes through choked and unchoked
nozzles become amplitude dependent at large amplitudes. The effect of these
processes on the instabilities is described by a nonlinear boundary condition.36
V. Conclusion
This chapter concludes with a summary of the state of the art of current model-
ing capabilities of combustion instabilities. From a practical point of view, three
basic instability characteristics must be predicted and understood: 1) frequency of
oscillations, 2) conditions under which the oscillations occur, and 3) their final,
limit-cycle amplitude, which are listed in increasing order of predictive difficulty.
Specifically, the frequency and mode shape are easier to predict because they gen-
erally only require an understanding of the system’s linear dynamics, whereas
determination of the limit-cycle amplitude requires an understanding of nonlinear
system characteristics.
Consider first the prediction of instability frequencies. Although some fun-
damental problems remain, such as analytical descriptions of the combustor’s
acoustic boundary conditions, prediction of instability frequencies is a relatively
mature area. In fact, capabilities for modeling the acoustics of combustors are rea-
sonably well developed, as described in Chapter 13, in this volume and also, for
example, in recent proceedings of the American Society of Mechanical Engineers
Turbo Expo conferences or in Munjal’s book.37 These descriptions indicate that
the frequencies and mode shapes of the excited instabilities can often be predicted
with good accuracy. In general, accurate predictions of these properties simply
require knowledge of the geometric characteristics of the system and average tem-
perature distributions. Although unsteady heat-release effects, which are much
harder to predict, have some impact on instability frequencies, this impact is often
small.
Predicting the conditions under which instabilities occur (the second issue) is
considerably more difficult than predicting acoustic mode shapes and frequencies,
because it requires knowledge and modeling of the interactions of flow and mixture
disturbances with flames and damping processes. In recent years, the focus of
much of the research in this area has been on developing these understanding
and modeling capabilities. Examples of the findings of some of these studies that
investigated, for example, acoustic wave–flame interactions and other mechanisms
that can drive these instabilities are discussed Chapters 9 and 12. Much progress
has been made in these areas to date and many gas-turbine manufacturers have
reported success in predicting instability frequencies and mode shapes, and the
conditions under which they occur, as discussed in the subsequent case-study
chapters in this book. Furthermore, even in cases in which predictive capabilities
do not exist, reasonable understanding of qualitative combustor stability trends
can often be obtained through analysis of combustor data after they have been
obtained; for example, rational correlations of the data can be developed.
Note that direct experimental verification of linear combustor dynamics models
and mechanisms of instability is difficult, because instabilities, in general, are
experimentally studied under limit-cycle conditions (with the notable exception
of Ref. 38). For example, consider the role of vortex shedding, which is thought
to be an important instability mechanism in gas turbines. The mere observation
of vortex shedding at the instability frequency under limit-cycle conditions does
not necessarily imply that it is also responsible for initiating the instability. Vortex
shedding can be forced at the frequency of an external excitation when the forcing
occurs at sufficiently large amplitude, even if this frequency does not coincide with
the natural shedding frequency.39 Consequently, flow oscillations may be excited
by some other instability mechanism, which subsequently forces the shedding of
vortices at the instability frequency.
In contrast to the prediction of the instability characteristics and the condi-
tions under which it spontaneously occurs, which can be determined with linear
combustor models, the prediction of the limit-cycle amplitude of the instability
and the conditions under which large-amplitude disturbances destabilize a linearly
stable system (triggering) requires solution of nonlinear models of the combustor

processes. Amplitude-prediction capabilities are critical because it would tell en-
gineers whether an instability can be tolerated in the system under consideration
if its amplitude is not too large, or whether they need to take steps to decrease its
magnitude to prevent damage to the combustor. Such capabilities require under-
standing of the flame and gas-dynamic response to large-amplitude disturbances
and the ability to model these nonlinear phenomena. Progress in this area has
largely been limited to simple, laminar flames and little is known, even in a qual-
itative sense, about the key parameters controlling nonlinear flame dynamics. As
such, instability-amplitude prediction capabilities remain a key challenge for future
work.
References
1
Rayleigh, J. S. W., The Theory of Sound, Vol. 2, Dover, New York, 1945.
2
Crighton, D., “Airframe Noise,” Aeroacoustics of Flight Vehicles, edited by H. H.
Hubbard, Acoustical Society of America, New York, 1995.
3
Yu, K., Trouve, A., and Daily, J., “Low-Frequency Pressure Oscillations in a Model
Ramjet Combustor,” Journal of Fluid Mechanics, Vol. 232, 1991, pp. 47–72.
4
Marble, F., and Candel, S., “Acoustic Disturbance from Gas Non-uniformity Convected
Through a Nozzle,” Journal of Sound Vibrations, Vol. 55, 1977, pp. 225–243.
5
Crocco, L., and Cheng, S., Theory of Combustion Instability in Liquid Propellant Rocket
Motors, Butterworths Scientific Publications, London, 1956.
6
Yang, V., Kim, S. I., and Culick, F. E. C., “Triggering of Longitudinal Pressure Os-
cillations in Combustion Chambers: I: Nonlinear Gasdynamics,” Combustion Science and
Technology, Vol. 72, 1990, pp. 183–214.
7
Wicker, J. M., Greene, W. D., Kim, S. I., and Yang, V., “Triggering of Longitudinal
Combustion Instabilities in Rocket Motors: Nonlinear Combustion Response,” Journal of
Propulsion and Power, Vol. 12, 1996, pp. 1148–1158.
8
Swift, G., Thermoacoustics, Acoustical Society of America, New York, 2002.
9
Rott, N., “Damped and Thermally Driven Acoustic Oscillations,” Zeitschrift fuer Ange-
waxdte Mathematik und Physik, Vol. 20, 1969, p. 230.
10
Jones, A. T., “Singing Flames,” Journal of the Acoustical Society of America, Vol. 16,
No. 4, 1945, pp. 254–266.
11
Le Conte, J., Philosophical Magazine HP, 235.
12
Putnam, A., Combustion Driven Oscillations in Industry, American Elsevier Publishers,
New York, 1971.
13
Pun, W. Ph.D. Thesis, California Inst. of Technology, Pasadena, CA, 1991.
14
Blomshield, F. S., “Historical Perspective of Combustion Instability in Motors: Case
Studies,” AIAA Paper 2001-3875, 2001.
15
Culick, F., “Combustion Instabilities in Liquid-Fueled Propulsion Systems-An
Overview,” AGARD, 1977.
16
Kendrick, D. W., Anderson, T. J., and Sowa, W. A., “Acoustic Sensitivities of Lean-
Premixed Fuel Injectors in a Single Nozzle Rig,” American Society of Mechanical Engi-
neers, Paper 98-GT-382, 1998.
17
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism for Combustion
Instabilities in Premixed Gas Turbine Combustors,” Journal of Engineering for Gas Turbines
and Power, Vol. 123, No. 1, 2001, pp. 182–190.
18
Candel, S., “Combustion Dynamics and Control: Progress and Challenges,” Proceed-
ings of the Combustion Institute, Pittsburgh, PA, Vol. 29, 2002.
19
Hegde, U. G., Reuter, D., Daniel, B. R., and Zinn, B. T., “Flame Driving of Longitudinal
Instabilities in Dump Type Ramjet Combustors,” Combustion Science and Technology,
Vol. 55, 1987, pp. 125–138.
20
Schadow, K., and Gutmark, E., “Combustion Instability Related to Vortex Shedding
in Dump Combustors and Their Passive Control,” Progress in Energy and Combustion
Science, Vol. 18, pp. 117–132, 1992.
21
Temkin, S., Elements of Acoustics, Wiley, New York, 1981.
22
Roberson, J., and Crowe, C., Engineering Fluid Mechanics, Houghton Mifflin, New
York, 1993.
23
Kim, J. S., “Effects of Turbulence on Linear Acoustic Instability: Spatial Inhomogene-
ity,” Liquid Rocket Engine Combustion Instability, edited by V. Yang and W. Anderson,
AIAA, Washington, DC, 1994, Chap. 16.
24
Lieuwen, T., Neumeier, Y., and Rajaram, R., “Measurements of Incoherent Acoustic
Wave Scattering from Turbulent Premixed Flames,” Proceedings of the Combustion Insti-
tute, Pittsburgh, PA, Vol. 29, 2002.
25
Lieuwen, T., “Experimental Investigation of Limit Cycle Oscillations in an Unsta-
ble Gas Turbine Combustor,” Journal of Propulsion and Power, Vol. 18, No. 1, 2002,
pp. 61–67.
26
Oefelein, J. C., and Yang, V., “Comprehensive Review of Liquid-Propellant Combustion
Instabilities in F-1Engines,” Journal of Propulsion and Power, Vol. 9, 1993, pp. 657–
677.
27
Drazin, P. G., and Reid, W. H., Hydrodynamic Stability, Cambridge Univ. Press,
Cambridge, England, U.K., 1981.
28
Polifke, W., Fischer, A., and Sattelmayer, T., “Instability of a Premix Burner with Non-
Monotonic Pressure Drop Characteristics,” Journal of Engineering for Gas Turbines and
Power, Vol. 125, No. 1, 2003, pp. 20–27.
29
Broda, J. C., Seo, S., Santoro, R. J., Shirhattikar, G., and Yang, V., “An Experimental
Investigation of Combustion Dynamics of a Lean, Premixed Swirl Injector,” Proceedings
of the Combustion Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1849–1856.
30
Zinn, B. T., and Powell, E. A., “Nonlinear Combustion Instability in Liquid- Propellant
Rocket Engines,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 13, 1970.
31
Culick, F. E. C., Burnley, V., and Swenson, G., “Pulsed Instabilities in Solid-Propellant
Rockets,” Journal of Propulsion and Power, Vol. 11, No. 4, 1995, pp. 657–665.
32
Culick, F. E. C., “Nonlinear Growth and Limiting Amplitude of Acoustic Oscillations
in Combustion Chambers,” Combustion Science and Technology, Vol. 3, No. 1, 1971.
33
Wicker, J. M., Greene, W. D., Kim, S.-I., and Yang, V., “Triggering of Longitudinal
Combustion Instabilities in Rocket Motors: Nonlinear Combustion Response,” Journal of
Propulsion and Power, Vol. 12, No. 6, 1996.
34
Peracchio, A. A., and Proscia, W. M., “Nonlinear Heat Release/Acoustic Model for
Thermo-Acoustic Instability in Lean Premixed Combustors,” Journal of Engineering for
Gas Turbines and Power, Vol. 121, 1999.
35
Zinn, B. T., “A Theoretical Study of Nonlinear Damping by Helmholtz Resonators,”
Journal of Sound Vibrations, Vol. 13, No. 3, pp. 347–356.
36
Zinn, B. T., and Crocco, L, “Periodic Finite Amplitude Oscillations in Slowly Con-
verging Nozzles,” Astronautica Acta, Vol. 13, Nos. 5 and 6, Aug. 1968, pp. 481–488.
37
Munjal, M., Acoustics of Ducts and Mufflers, Wiley, New York, 1987.
38
Poinsot, T., Veynante, D., Bourienne, F., Candel, S., Esposito, E., and Surget, J., “Ini-
tiation and Suppression of Combustion Instabilities by Active Control,” Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 22, pp. 1363–1370, 1988.
39
Blevins, “The Effect of Sound on Vortex Shedding from Cylinders,” Journal of Fluid
Mechanics, Vol. 161, 1985, pp. 217–237.
40
Huang, Y., and Yang, V., “Effect of Swirl on Combustion Dynamics in a Lean-Premixed
Swirl-Stabilized Combustor,” Proceedings of the Combustion Institute, Pittsburgh, PA,
Vol. 30, 2004, pp. 1771–1778.
41
Ingard and Singhal, “Effect of Flow on the Acoustic Resonances of an Open Ended
Duct,” Journal of the Acoustical Society of America, Vol. 58, No. 4, 1975, pp. 788–793.
II. Case Studies
Chapter 2
Combustion Instabilities in Industrial Gas Turbines:

Solar Turbines’ Experience
Kenneth O. Smith∗ and James Blust†

Solar Turbines, Inc., San Diego, California
I. Introduction
G AS TURBINE manufacturers have developed and continue to improve lean

premixed (LP) combustion systems to meet emissions regulations for NOx,
carbon monoxide, and unburned hydrocarbons. With LP combustion, high levels
of combustion air are introduced into the gas turbine combustor primary zone (the
flame zone) to produce a leaner fuel–air mixture than is typical of “conventional”
diffusion flame combustors. The high airflow reduces the LP combustor flame
temperature and, in turn, the NOx formation rate, which is an exponential func-
tion of temperature. A second characteristic of LP combustion is the mixing (or
“premixing”) of the fuel and air upstream of the primary zone. Premixing per-
mits combustion of a uniform fuel–air mixture, thus preventing the locally high
temperatures that can occur within a diffusion flame combustor.
The development of LP combustion-based products at Solar Turbines Incor-
porated (Solar) started in the mid-1980s.1,2 The two-tiered development goal for
natural gas combustors was to meet emissions regulations of 42 ppmv NOx (at
15% O2 , dry) first, with a subsequent reduction to 25 ppmv. Maximum allow-
able CO and hydrocarbon emissions were set at 50 ppmv. The development effort
eventually expanded to meet the need for reduced emissions with no. 2 diesel fuel.
By the early 1990s, work had progressed to rig testing of full-scale LP com-
bustors and in-house engine tests of prototype systems. In 1992 Solar placed the
first two engines with LP combustors at customer sites for field trials. Centaur
Type H and Mars turbines were installed for these initial field evaluations.3 Since
that time, more than 1100 Solar gas turbines with LP combustors (trademarked as
SoLoNOx) have been commissioned.
The development of LP combustion has led to the commercialization of a new
generation of low-emissions gas turbines, but LP combustion has also brought
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
∗ Manager, Advanced Combustion.
† Senior Principal Combustion Engineer.
29
30 K. O. SMITH AND J. BLUST
with it new technological challenges. Perhaps most difficult is the avoidance of

high-amplitude, combustor pressure oscillations (CPOs) that can severely damage
turbine components.
High CPOs occur in LP gas turbines because combustor flame temperatures
have been reduced to reduce NOx emissions. At lower flame temperatures, the
combustor design point is closer to the lean blowout condition, where flame stabil-
ity is reduced and small temporal variations in the fuel–air ratio can significantly
affect the heat release rate. This increases the likelihood that the combustion pro-
cess will drive CPOs at a resonance frequency of the combustor liner or the fuel
system.
The CPO phenomenon was initially observed at Solar in early rig and engine
tests of full-scale Mars and Centaur combustors. Since then, the SoLoNOx product
line has grown to include Taurus and Titan turbines. During the development of
each of these products, CPOs have been observed and corrective actions have been
taken to ensure that customers’ durability expectations are met.
With the early SoLoNOx development work as a foundation, effective methods
have evolved to prevent the occurrence of damaging CPOs. Although these meth-
ods have been effective, in general, they are not fool-proof nor are they necessarily
a cost-effective means of avoiding CPOs. CPO reduction is currently a retrofit
or redesign activity, and none of the methods used has proved to be a “universal
fix.” Even ten years after the commercialization of LP combustion systems, the
gas turbine industry still lacks a robust methodology for reducing CPOs. More
significantly, no a priori design methodology has evolved to prevent high CPOs.
The severity of the CPO problem for a specific combustor design is still not known
until testing has started. As a result, it is not uncommon for CPOs, once reduced,
to reemerge as a problem if combustion system changes are made as part of engine
performance uprates or to introduce greater fuel flexibility.
Gas turbine manufacturers continue to work to reduce NOx emissions in re-
sponse to tighter air quality regulations. The most common approach is through
further reductions in primary zone temperature. Because primary zone tempera-
tures are designed to lie nearer the lean blowout condition, it is likely that CPOs
will become a more critical issue. The gas turbine industry will need a better un-
derstanding of the phenomenon and cost-effective methods (passive or active) to
keep CPOs manageable. Ideally, industry would like to develop design tools that
reduce the probability of CPOs early in the product development cycle, before
hardware is fabricated. However, despite the lack of such design tools, Solar’s
natural gas–fired engines are meeting NOx regulations of 25 ppmv and lower on
a production basis, with only infrequent issues associated with CPOs.
Discussed next are the actions taken and methodologies used by Solar as the
SoLoNOx product line was developed and large-amplitude CPOs were dealt with
for the first time.
II. Lean Premixed Combustion System Configurations

and Operating Conditions
Solar’s LP combustion systems share several common features. Annular com-
bustors (Fig. 2.1) are standard. Multiple fuel injectors, from 8 to 14, are inserted
into the combustors through the liner combustor dome (Fig. 2.2). Combustor liners
COMBUSTION INSTABILITIES IN INDUSTRIAL GAS TURBINES 31
Fig. 2.1 Typical annular combustors (view of upsteam ends).
range in size (outer diameter) from approximately 30 in. (Centaur) to 38 in. (Titan).
Table 2.1 presents a summary of Solar’s turbine product line.
At the time of LP combustor development, the Centaur engine full-load com-
bustor inlet conditions were 100 psia and 600◦ F (690 kPa and 590 K). The corre-
sponding Mars conditions were 220 psia and 800◦ F (1520 kPa and 700 K). The
Centaur combustor has outer and inner diameters of 30 and 24 in. (76 and 61 cm),
respectively; 12 fuel injectors are used. The outer and inner diameters of the Mars
liner are 36 and 28 in. (91 and 71 cm), respectively; the Mars uses 14 fuel injectors.
Fuel Injector Combustor Housing
Combustor Liner
Turbine Nozzle
Fig. 2.2 Annular combustor liner/fuel injector configuration.

Table 2.1 Characteristics of Solar products
Turbine model Nominal hp Airflow, lb/s Pressure ratio
Centaur 40 4,700 41.3 10.3

Centaur 50 6,130 40.6 10.3
Taurus 60 7,700 47.0 12.2
Taurus 70 10,300 57.9 16.0
Mars 90 13,200 84.0 16.0
Mars 100 15,000 92.0 17.4
Titan 130 19,800 105 16.0
The basic SoLoNOx fuel injector configuration includes an axial combustion

air swirler and a series of radial fuel injection spokes downstream of the swirler
(Fig. 2.3). The spokes are located within a premixing duct where the natural gas
and swirling combustion air mix to a near-homogeneous state before entering the
combustor. The swirl intensity of the axial swirlers varies across the product line.
For example, the Centaur has a swirl blade angle of approximately 48 deg whereas
the Mars is closer to 60 deg.
The injectors include a pilot fuel injector that is integrated into the injector cen-
terbody. The pilot injects a small amount of fuel and air (only partially premixed)
AXIAL GAS GAS INJECTION

SWALOR MANIFOLD SPOKES
AIR
INLET
MARS
INJECTOR
PREMIX
DUCT
Fig. 2.3 Cross section of typical lean premixed gaseous fuel injector.
into the combustor. The pilot flame, therefore, has stability characteristics that
are more like those of a diffusion flame. The pilot injector was originally used to
provide a stable flame for engine startup, low-load operation, and off-load tran-
sients. The pilot has since proved invaluable for overcoming CPOs. In fact, Solar’s
experience indicates that the most productive means of achieving reductions in
CPOs is through optimization of the pilot specifically and the fuel injector more
generally.
Solar’s initial LP combustor liner designs used traditional film cooling to keep
liner temperatures low. Since then, liner cooling has evolved to either effusion
cooling or convective backside cooling, in which no cooling air is injected into
the combustion zone. During this period of development, the number and size of
holes in the combustor liners have varied widely. To date, however, no signifi-
cant relationship has been identified between the method of liner cooling and the
occurrence of high CPOs.
III. Commercial Introduction at 42 ppmv NOx

The initial release of the SoLoNOx technology occurred with the Centaur Type
H and Mars engines in the early 1990s. These combustion systems were initially
guaranteed to meet NOx emissions of 42 ppmv (at 15% O2 ) on natural gas and
then 25 ppmv NOx as the products matured.
Preliminary development testing of both the Mars and Centaur combustion
systems brought focus on the CPO challenge. Although there was no indication
of troublesome CPOs in single-injector rig tests (can combustor) and a sequence
of three-injector-sector tests, high CPOs were observed in the initial rig testing of
the full annular combustors.
Rig tests at both atmospheric and high pressures, as well as in-house engine
tests, demonstrated the CPO phenomenon. Both the Mars and Centaur combustion
systems exhibited operating regions in which oscillations occurred and were
unacceptably high. The CPO magnitudes were typically of the order of 1–3 psi
(rms). In one extreme case, magnitudes as high as 6 psi (rms) were observed.
The oscillations resulted in cracking of the dome’s internal splash plates from
high-cycle fatigue (Fig. 2.4) and fretting of the fuel injector’s outer barrel where
the barrel contacts the combustor dome. This damage occurred within a period of
approximately 30 min. The CPO frequency spectra were characterized by distinct
high-amplitude, narrow-band spikes. The largest spike in the Centaur frequency
spectrum was at approximately 440 Hz (with harmonics) (Fig. 2.5), whereas the
larger Mars combustion system oscillated at approximately 350 Hz (again with
harmonics). The Centaur oscillation tendency was greatest at full-load operation.
The Mars was quiet at and near full load but tended to oscillate during part-load
operation.
Simple one-dimensional frequency calculations using estimated average com-
bustor gas temperatures showed good agreement between the calculated and ob-
served CPO frequencies for both engines. This agreement, however, was not suf-
ficient to characterize the nature of the instability. Additional calculations showed
that each of these liners could experience circumferential instabilities at nearly
identical frequencies as in the axial direction. In actuality, the situation was found
Fig. 2.4 Liner internal splash plate damaged by high-amplitude combustor pressure
oscillations.
5
T6102 S/N 001 BLD 3 3/24/92
4021 KW 100 PCT LOAD
Pressure Oscillation (arbitrary units)
LOG
5E-4
0 1000 2000
Fig. 2.5 Typical experimental pressure frequency spectrum showing amplitude peak
at 440 Hz.
4 50
DYNAMIC PRESSURE psi (ms)

Dynamic Pressure
40
NOx (ppm,O 15% O2)

3 NOx
30
NOx Target
2
20
1
Dynamic Pressure 10
New Design
0 0
0 2 4 6 8 10 12 14 16
PILOT FUEL, %
Fig. 2.6 Pilot fuel effect on combustor pressure oscillations and NOx emissions.
to be more complex than either of these scenarios. Subsequent detailed pressure-

wave measurements conducted in a Centaur engine at Solar indicated that the
instability was of a mixed mode, with both axial and circumferential components.
Early in the annular combustor testing of the Mars and Centaur LP systems,
it was established that the operation of the pilot injector, even with just a small
percentage of the total engine fuel flow, was extremely effective at reducing CPOs.
By injecting approximately 10% pilot fuel, CPOs could be reduced to well below
1 psi (rms) while maintaining NOx emissions below 42 ppmv. Consequently, this
became the interim CPO control strategy for the first SoLoNOx production units.
The availability of the pilot for CPO control was to a degree coincidental. The
pilot was originally designed to enhance engine light-off, part-load operation, and
off-load response. Since the first recognition of the pilot as a means of reducing
CPOs, the pilot has remained the primary means of controlling CPOs at Solar;
however, pilot fueling for CPO control increased NOx emissions (Fig. 2.6). To
achieve lower NOx levels, combustion systems that were inherently less prone to
CPOs (and therefore less reliant on pilot flames) would be needed.
IV. Emissions Reduction to 25 ppmv NOx

As development work continued to reduce NOx emissions to the longer-term
goal of 25 ppmv, the limitations of using high pilot injector fueling for CPO control
were seen. The level of pilot fueling needed to control CPOs with the then-current
injector designs resulted in NOx levels that were too high to consistently meet a
25-ppmv guarantee. Combustion system design modifications were clearly needed
to achieve lower emissions.
The approach taken at Solar to reduce CPOs for 25-ppmv NOx products was
to focus on the impact of injector design and operation on CPOs. Combustor liner
modifications were viewed as an alternative approach because of the time and cost
associated with liner redesigns.
With little in the way of analytical tools for guidance, CPO reduction work
primarily involved testing modified injectors. The injector design elements that
could most easily be varied were rig tested to quantify their effect on CPOs. These
elements included injector premixing barrel inner diameter and the distance from
the fuel-injection spokes to the injector exit plane. These parameters affect the
injector exit velocity, the overall fuel–air mixedness, and the fuel transport time
from the spokes to the flame.
A reduction in exit velocity was expected to have a stabilizing effect as long
as flashback into the premixing channel was avoided. Similarly, moving the fuel
spokes downstream would reduce premixedness, which would enhance flame sta-
bility but at the cost of increased NOx emissions. Also, changes in the transport
time might reduce any fuel system–liner coupling that was driving CPOs, as dis-
cussed subsequently.
Testing of modified injectors for both the Mars and Centaur engines demon-
strated that instabilities strongly depended on these design features. Small shifts
in either the fuel-spoke location or exit velocity were found to impact CPOs
significantly.
Injector screening for CPO characteristics was conducted in two ways. For rela-
tively minor changes in injector configuration, full sets of injectors were modified
and then tested in engines or in high-pressure rigs that attempted to simulate the
acoustics of the engine environment. In some cases, injector modifications were
substantial. In those cases, a single injector was fabricated for testing in a single-
injector test rig that contained a can combustor. The can combustors had the same
lengths as their annular counterparts and had diameters equal to the annulus height
of their corresponding annular liner. Testing duplicated the engine-operating con-
ditions, but flow rates were scaled for single-injector operation.
Special provision was made in an attempt to duplicate the instability frequency
of the full annular combustor in the single-injector rig. Following the approach of
Richards et al.,4 the combustor was fitted with a refractory plug similar to the one
shown in Fig. 2.7. The combustion zone and the plug form a Helmholtz resonator.
The plug was sized to match the resonance frequency experienced in the engine
environment.
Comb
Zone
Fig. 2.7 Can combustor with refractory plug to simulate a Helmholtz resonator.
Clearly the single-injector rig was unable to duplicate the acoustic characteristics
of the annular combustor, in which circumferential and mixed-mode oscillations
could appear. The rig proved valuable as a screening tool, however. Injectors that
tended to drive oscillations in the rig environment almost always showed similar
tendencies in an engine. Injectors that operated without high CPOs in the single-
injector rig were frequently stable in the engine, although this was not universally
the case. Thus, the single-injector rig helped identify potentially better injector
configurations without fabricating complete sets of injectors and conducting costly
engine tests.
V. Combustor Pressure Oscillation Model

5
Putnam describes high-amplitude CPOs in the context of a sinusoidal pressure
wave and a fluctuating heat release (caused by pressure-driven fuel flow variations).
High amplitude CPOs occur when the pressure and heat release are phased to
peak simultaneously. Simplistically, in the SoLoNOx injector, the heat-release
fluctuation frequency can be tied to the time τ required for the fuel to travel
from the fuel spoke to the injector exit (or more accurately to the flame front).
Thus, according to this model, CPOs (of frequency f C ) are less likely to occur
if the fuel transit time does not align with a resonant acoustic frequency of the
combustor (τ = 1/[2 f C ]). This model, though overly simple, has proved valuable
in understanding the physical mechanisms that can contribute to CPOs. In addition,
the model has been used with some success in determining fuel-spoke locations
within LP injectors that are less likely to cause high-amplitude CPOs.
In practice, the heat release and pressure oscillations need not be exactly in
phase for high-amplitude CPOs to occur. In theory, oscillations can occur when
heat release rates lead or lag the pressure fluctuations by as much as one-fourth of
the acoustic cycle.6 This, at least mechanistically, can be used to explain the results
of the Mars injector optimization (discussed subsequently) where ranges of stable
and unstable operation were mapped out as a function of fuel spoke location. In
this simplistic model, high oscillations are expected to occur at values of (τ f C )
from 0.25 to 0.75. Not addressed in a simple one-dimensional model such as this
are the effects of velocity and fuel–air profiles at the exit of the premixer, the
evolution of these profiles once the flow exits the injector, the possible interaction
of multiple injectors, and the three-dimensional and unsteady nature of the flame
front downstream of the injector.
VI. Centaur CPO Reduction

The Centaur fuel injector was the focus of Solar’s first efforts to reduce CPOs.
With little analytical guidance, a decision was made to reduce the average full-load
exit velocity of the fuel injector to match that of the Mars injector (which at that
time had not exhibited CPOs of any significance at full load). Thus, the cross-
sectional area of the injector premixing channel was increased just downstream of
the fuel injection plane. Although a reduction in velocity was the goal, the change
also had an impact on τ , the injector’s radial velocity profile, St/Uax (the ratio
of turbulent flame speed to injector axial exit velocity), and, quite possibly, the
average fuel/air mixedness and the exit plane fuel–air profile.
Subsequent testing of the modified injector indicated that the reduction in Uax
was an extremely robust solution to CPOs in the Centaur combustor. No significant
CPOs occurred at any point within the engine’s operating envelope. NOx levels
below 25 ppmv were readily achieved without high levels of pilot fueling. Because
of the minimal effort required to reduce CPOs, no further effort was made to
assess whether any of the other injector design parameters were actually of greater
importance than Uax in terms of CPO levels.
VII. Mars CPO Reduction

The ease of resolving the Centaur CPO issue was not duplicated on the Mars
engine. Much more extensive work was necessary to reduce the high amplitude
CPOs that occurred in the Mars engine at part-load conditions. This may have
been caused partly by the more geometrically complex configuration of the Mars
injector, whose design incorporated features for varying the combustor airflow
during operation (air bleed) (Fig. 2.8).
The modified Centaur injector and the Mars injector had the same Uax and
were stable during full-load operation, but the latter engine exhibited high CPOs
at part-load conditions. Therefore, following the success of the Centaur work, a
reduction in Mars Uax was the first modification assessed.
Decreasing Uax in the Mars injector did reduce CPOs but an increase in CO
emissions occurred simultaneously. The magnitude of the CO increase was de-
pendent on the axial location at which the injector flow area was increased. The
further downstream from the swirler the increase occurred, the smaller the effect
on CO emissions. Consequently, an area expansion was incorporated very near the
exit plane of the injector.
These initial Mars results reinforced the fact that the impact on CPOs of an area
expansion within the injector was complex and not solely caused by a change in
Uax. For example, the area expansion located at the exit plane of the Mars injector
almost certainly has an impact on the flowfield near the injector outer barrel more
than it reduces velocities near the injector centerbody.
Fig. 2.8 Cross section of Mars injector, showing provision for air bleed.
Table 2.2 Comparison of CPO amplitudes for different spoke locations
Axial length, Axial length,

cm in. τ, s Frequency, Hz Noise τ × frequency
3.5 1.38 0.76 × 10−3 390 No 0.30

4.1 1.63 0.90 × 10−3 370 Yes 0.33
4.5 1.75 0.97 × 10−3 360 Yes 0.35
5.7 2.25 1.25 × 10−3 333 Yes 0.42
5.8 7.30 1.28 × 10−3 313 Yes 0.40
6.9 2.70 1.50 × 10−3 315 Yes 0.47
7.6 3.00 1.67 × 10−3 290 No 0.48
Although the modification to the Mars injector did reduce CPOs during part-
load operation, further reductions were sought. Work was continued to define the
role of fuel spoke location on CPOs.7 Development injectors were fabricated with
different axial distances from the fuel spokes to the exit plane of the injector.
Spokes were located at seven different distances from the exit, ranging from 1.38
inches (3.85 cm) to 3.0 inches (7.6 cm). The first observation was that small axial
changes in spoke position had a significant effect on the amplitude of any CPOs.
Certain spoke locations did not lead to coupling of the flame and the combustion
system. According to the simplistic model described, changing the axial location
of the spoke altered the relationship between τ and f C such that coupling was
avoided. Data from these Mars injector tests are presented in Table 2.2.
The data indicate that discrete regions of instability exist and that these regions
can be defined by the magnitude of the nondimensional number τ f C . As seen in
Fig. 2.9, for these Mars injectors, a region of instability was bounded by values
of τ f C of 0.30 and 0.48. These values suggest that significant coupling occurred
for approximately one-third of the duration of one pressure cycle. Whether this
finding is of general validity or specific to the Mars combustion system remains
to be determined.
On the basis of the preceding tests, a modified Mars injector was developed
that included both a change in flow cross-sectional area and a new spoke location.
Subsequent tests on production Mars engines confirmed that these changes, in
L = 1.38 in = 3.0
stable unstable stable
0.30 0.48
τ fc
Fig. 2.9 Mars CPO amplitudes as a function of fuel-spoke location and τ fC .
conjunction with the pilot injector, were effective in reducing the severity of CPOs
without compromising the NOx emissions goal of 25 ppmv.
VIII. Recent Experience

In the 12 years since the introduction of Solar’s LP combustors, the perva-
siveness of CPOs has become apparent. As product improvements are made on
mature turbine products, “quiet” combustors can unexpectedly display unaccept-
able CPOs. In general, any change that affects the combustor airflow, the airflow
distribution, or the stoichiometry of the combustor primary zone has the potential
to trigger CPOs. Improvements in this category include the following: 1) increased
compressor flow, 2) increased primary zone airflow for NOx reduction, 3) liner
cooling or dilution flow modification that alters the combustor flow split, 4) in-
creased primary zone flow (even at constant flame temperature) to increase turbine
inlet temperature, 5) injector modifications to incorporate a liquid backup fuel to
natural gas and 6) use of a nonstandard fuel, even without physical changes to the
injector.
Paralleling the CPO challenges associated with product improvements are those
arising through new product development. The design of a new combustion system
requires that the CPO issue be addressed anew. As a consolation, this provides the
chance to apply lessons learned in the past and to continue expanding the tools
available for CPO reduction.
As an example, during the development of the new Titan gas turbine, studies on
the impact of fuel spoke design, swirl vane configuration, and spoke location were
conducted. This work has demonstrated that details of the spoke design that affect
the injector fuel–air profile can have a significant effect on CPO amplitudes. For
the geometries studied, spoke configuration had a greater impact than the other
parameters. Although these results are probably engine dependent and thus cannot
be generalized, this work has helped further the development of a methodology
for overcoming high CPOs.
IX. Conclusion: Needs and Future Challenges

If Solar’s experience with CPOs is common, CPOs will continue to be a chal-
lenge for the gas turbine industry as new and uprated engines are developed to
provide higher efficiency, higher output, lower emissions, and greater fuel flexi-
bility. Of particular importance is the continuing need for lower NOx emissions
because this need may require combustor design changes in mature products. The
drive for lower NOx emissions will result in combustors that are increasingly fuel-
lean and are therefore more likely to exhibit high amplitude CPOs. Combustors
with lower emissions will not be able to rely on NOx-producing pilot injectors for
stability. Low NOx pilots and more optimized (for quiet operation) fuel injector
designs will be required.
An improved understanding of the mechanisms and design features that drive
CPOs would streamline gas turbine combustor development in several areas. Im-
proved analytical tools to support the combustor design process are certainly re-
quired. The ultimate need is for design tools that effectively prevent (or at least
minimize) CPOs at the design stage, before hardware has been fabricated. Meeting
this need is probably an unrealistic hope in the near term, given the complexity
of CPOs. More realistic might be the development of models that can predict the
onset of CPOs (rather than limit cycle behavior). Similarly, a more rigorous means
of extrapolating CPO data from test rigs to the engine environment would help
reduce the time and costs associated with CPO mitigation.
References
1
Smith, K. O., Angello, L. C., and Kurzynske, F. R., “Design and Testing of an Ultra-
Low NOx Gas Turbine Combustor,” American Society of Mechanical Engineers, New York,
Paper 86-GT-263, 1986.
2
Etheridge, C. J., “Mars SoLoNOx: Lean Premix Combustion Technology in Production,”
American Society of Mechanical Engineers, New York, Paper 94-GT-255, 1994.
3
Rawlins, D. C., “Dry Low Emissions: Improvements to the SoLoNOx Combustion
System,” 11th Symposium on Industrial Applications of Gas Turbines, Canadian Gas As-
sociation, Banff, Alberta, Canada, Oct., 1995.
4
Richards, G. A., Gemmen, R. S., and Yip, M. J., “A Test Device for Premixed Gas
Turbine Combustion Oscillations,” Journal of Engineering for Gas Turbines and Power,
Vol. 119, 1997, pp. 776–782.
5
Putnam, A., Combustion Driven Oscillations in Industry, Elsevier, New York, 1971.
6
Richards, G. A., and Janus, M. C., “Characterization of Oscillations During Premix
Gas Turbine Combustion,” American Society of Mechanical Engineers, New York, Paper
97-GT-244, 1997.
7
Steele, R. C., Cowell, L. H., Cannon, S. M., and Smith, C. E., “Passive Control of
Combustion Instability in Lean Premixed Combustors,” Journal of Engineering for Gas
Turbines and Power, Vol. 122, 2000, pp. 412–419.
Chapter 3
Incorporation of Combustion Instability Issues

into Design Process: GE Aeroderivative and Aero
Engines Experience
H. C. Mongia,∗ T. J. Held,† G. C. Hsiao,‡ and R. P. Pandalai§

GE Transportation, Cincinnati, Ohio
I. Introduction
T HE occurrence of combustion dynamics in aircraft propulsion and aeroderiva-

tive industrial engine combustors presents a great challenge for combustor
designers. Not only are combustion dynamics detrimental to the operation of the
engine and combustor, but the difficulty in predicting and remedying dynamics
problems can lead to significant costs and delays in engine development.
Combustion dynamics are objectionable for at least two primary reasons. Under
some circumstances, the activity within the combustor generates an externally
audible tone at intolerable levels. More frequently, the pressure oscillation can
also drive resonant vibrations in mechanical components, resulting in significant
hardware damage.
A fundamental issue in designing gas-turbine combustors is the late stage of the
development process at which combustion-dynamics phenomena become appar-
ent. Many of the critical performance parameters of a combustor can be determined
analytically or through a combination of analysis and component testing. Although
some indication of the susceptibility of a combustion system to high levels of dy-
namics can sometimes be inferred from component tests, the behavior of the full
system cannot be predicted on the basis of either analysis or component testing.
The severity and character of the problem is not fully determined until an engine
test is conducted, at which point significant changes to component design are very
expensive and are likely to have major impacts on schedule and development cost.
∗ Section Manager, Advanced Combustion Engineering.
† Subsection Manager, Advanced Industrial Aeroderivative Combustor Design.
‡ Lead Engineer.
§ Staff Engineer.
43
44 H. C. MONGIA ET AL.
Fig. 3.1 Framework for combustion dynamic modeling.
Various acoustic-control strategies, both passive and active, are then applied to
deal with the unacceptable levels of dynamics. However, the application of these
strategies is a largely empirical process with little assurance of success.
The conventional approach to this problem is mostly based on empirical cor-
relations and design experience. Attempts are made to predict frequencies of
combustion–acoustic waves, such that none of the subsystems (fuel nozzle, heat
shield, and combustor liners, etc.) of the total combustion system have natural fre-
quencies that can couple with the combustion–acoustic frequency. Although good
success has been obtained in predicted acoustic mode shapes and frequencies, the
amplitude of the oscillation(s) is not easy to predict.
To reduce the risk of uncontrollable combustion-acoustic behavior, detection
and abatement of combustion–acoustic susceptibility is required in the early
stages of a design. A comprehensive strategy to predict, avoid, and/or improve
the combustion-acoustic performance of a combustion system includes both ana-
lytical and experimental determination of system and component properties and
interactions. The basic framework of such a strategy is shown in Fig. 3.1. In brief,
a semianalytical model is used to link the acoustic characteristics of the sub-
component parts of the combustion system and boundaries. The characteristics of
these subcomponents can be derived either from analytical models or from well-
characterized empirical testing. A more detailed description will be given in a later
section.
This chapter is intended to provide an overview of the combustion-dynamics
problems observed in aeroderivative industrial gas-turbine engines and flight-
propulsion engines. The fundamental issues driving combustion dynamics in prac-
tical gas-turbine combustors are reviewed as a means to interpret the observed
combustion behavior. Several methods used in the laboratory and in production
engines for controlling combustion dynamics are described, along with two exam-
ples of combustion-dynamics control in production gas-turbine engines. Finally,
a framework for analysis and design of gas-turbine combustors to mitigate the
occurrence and impact of combustion dynamics is presented.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 45
II. Fundamental Causes of Combustion Dynamics

In general, the occurrence of combustion dynamics is understood as depending
on a coupling between pressure oscillations and energy-release rate, often referred
to as the Rayleigh criterion. A pressure disturbance affects the local instantaneous
heat-release rate, which in turn creates a pressure disturbance with some time
(or phase) lag to the initial disturbance. The pressure disturbance is reflected at
the boundaries of the combustion chamber, thus potentially closing the feedback
loop that causes excessive (and destructive) pressure oscillations. Two parameters
critical in determining the overall feedback-loop stability are the relative phase
lag between pressure and heat-release oscillations and the amount of damping
present. These parameters are functions of the combustion-chamber geometry and
boundary conditions and of the nature of the coupling mechanism itself.
The first category of coupling mechanisms is the pressure-disturbance inter-
action with the instantaneous flame position and shape. As the flame surface re-
sponds to the pressure disturbance, its response can generate an acoustic wave of its
own. These coupling mechanisms should be relatively insensitive to details in the
fueling system and are often identified with high-frequency (>1 kHz) acoustic ac-
tivity. Lieuwen1 (see also Chap. 12) gives an excellent review on the modeling pro-
cesses of premixed combustion–acoustic wave interaction. Based on analytical and
experimental observations, it appears that the combustion process is increasingly
sensitive to perturbation in the equivalence ratio φ under lean operating conditions.
The second category of coupling mechanisms is often termed fuel–air wave
coupling. Within this category are several submechanisms, which share a com-
mon interaction between a pressure oscillation and the local fuel–air ratio of the
combustor. The physical interaction can take place through effects on local airflow
rate, fuel flow rate, or fuel-spray characteristics. This type of interaction is often
identified with midrange frequencies (100–1000 Hz).
The third category is incipient blowout coupling. This mechanism is unusual
in that it is an engine-system-level coupling mechanism. An example is when a
segment of a combustor locally reaches a low enough fuel–air ratio to extinguish.
The energy reaching the turbine immediately reduces, causing the rotor speed to
decrease. The airflow through the engine is reduced, causing the fuel–air ratio to
increase and the blownout segment to reignite. These oscillations are often de-
tected as a very-low-frequency mode (<30 Hz) and is not typically classified as a
combustion–acoustic phenomenon.
III. Control Strategies

Methods for controlling combustion dynamics in practical gas-turbine com-
bustors fall into three basic categories. Fundamental design changes to the fuel–
air mixing device have been shown to have significant impact on combustion-
dynamics behavior.2,3 These changes can influence either or both the fuel and
airflow paths4 and are generally derived empirically though component and en-
gine tests.
Control of combustion dynamics can often be achieved through manipulation
of the operating characteristics of the combustor. This manipulation can take the
form of adjustments in radial staging of fuel flow (and thus flame temperature)
distribution,5 in axial fuel staging adjustments,6 or in asymmetric fuel distribution.7

Such fuel-distribution strategies generally deteriorate the nitrous oxide (NOx)
emissions performance of the combustor, because they shift the flame temperatures
away from the ideal uniform distribution.
Finally, combustion dynamics can be affected by the use of both passive- and
active-control devices. The passive controls are typically Helmholtz resonators8 or
quarter-wave tubes,5 and serve as damping devices in the oscillating system. These
devices have demonstrated successful suppression of acoustics in gas-turbine com-
bustors. A disadvantage of these devices is that they only operate over a limited
frequency range, thus requiring empirical selection of the number and configura-
tion of the devices.
Active controls also take multiple forms, the most common being modulation
of the fuel flow with a frequency and phase relationship designed to destructively
interact with the combustion-dynamic oscillation. This modulation can take the
form of the main fuel flow,9,10 or a smaller secondary fuel-injection site,11 possibly
intended to act more directly on regions in which the Rayleigh index is largest.12
Although significant suppression of combustion dynamics has been demonstrated
in several subscale laboratory combustors, several barriers to implementation in
full-scale gas-turbine combustors still exist. First, the geometry of an annular com-
bustor is far more complex than that of a simple “can”-type laboratory combustor.
The control algorithms for controlling the complex acoustic mode structures in an
annular combustor still need to be developed and proven. Also, the reliability and
durability of the sensors, actuators, and control systems need to be comparable to
those of the gas turbine itself to avoid causing unscheduled maintenance events.
IV. Examples of Combustion Dynamics

A. Conventional Combustor
Combustion instabilities encountered in conventional, rich-dome combustors
are generally observed at frequencies between 150 and 700 Hz. Typical ampli-
tudes of combustion dynamics range from 1 to 3 psi peak-to-peak during transition
from start to idle and/or during transition from idle to a higher-power setting. De-
sign guidelines have been developed for designing new diffusion-flame combustor
concepts to eliminate potential instability problems before they appear during the
development phase. These guidelines are not universally successful in eliminating
the problem. Once encountered, however, instability problems, in general, can be
mitigated by relatively simple modifications as summarized in Fig. 3.2. The basic
concept in these fixes is to change the interaction between the fuel-injection system
and the combustor to eliminate coupling between the two systems.
Two recent examples of postcertification combustion instability problems were
encountered on double annular combustors (DACs). In one case, the instability
problem consisted of a low-frequency “growl” near ground-idle condition and
higher-frequency instability at approximately 600 to 630 Hz at higher operating
conditions. Subsequent analysis of test data from the engines showed two distinct
frequencies associated with each of these instabilities. The first growl instability
appeared at 200 Hz followed by a second tone at 300 Hz near ground idle. Only the
200-Hz mode was audible outside the engine. A detailed evaluation of the growl
instability and the potential root causes of the problem identified combustor–fuel
Fig. 3.2 Control of combustion dynamics in conventional gas-turbine combustors
system coupling as the most probable source. Altering the fueling strategy during
the engine start transient eliminated the objectionable tone.
Two distinct frequency peaks (signals) were also observed from pressure trans-
ducers mounted on the combustor near the flame zone for the higher-frequency
dynamics. The two primary frequencies observed were instability at 580 Hz and
a second tone at approximately 700 Hz. These acoustic modes were collectively
called organ tone because of their proximity in frequency to the quarter-wave axial
mode. This instability occurred in an operating mode in which only the outer (pi-
lot) dome of the DAC combustor was fired near its maximum fuel flow. Because
reliable analytical design tools were not available, an iterative empirical approach
was used to fix the instability problem. Several candidate approaches were identi-
fied and screened for ease of implementation, including alternate fueling modes,
Helmholtz resonators, fuel-nozzle cavity modifications, and spray-angle changes.
Variation in the fueling mode was selected as the prime approach. This approach
involved controlling certain main burner fuel nozzles in various circumferential
patterns. The strategy was to reduce the local pilot stoichiometry and to introduce
circumferential nonuniformity in the main stage to reduce dynamics.
Two separate engine tests were conducted to document the engines combustor-
dynamics characteristics. The first ground engine test was used to establish the
organ-noise threshold limits as a function of combustor-operating conditions and
for a range of core engine speeds. Analysis clearly showed the influence of pilot
fuel–air ratio on organ tone amplitude. Data from the ground engine tests repli-
cated earlier engine results, where two pure tones, one tone at 600 Hz and a second
tone at 680 Hz, were again observed when operating in the pilot-only fueling
mode at greater than 2000 rpm engine fan speed. A second engine test on a flying
test bed was subsequently conducted to get additional data from an altitude en-
gine environment. Data from the flying test bed engine were also used to map
the organ-noise envelope in the pilot-only fueling-mode. Extensive testing of the
fueling-mode changes identified a new fueling mode as the optimal configuration
for minimizing the instability. This fueling mode incorporates two main-stage fuel
nozzles operating adjacent to each other at two circumferential locations in the
operating range where the instability was previously observed. Subsequent engine
tests validated the fueling-mode change as the most practical and effective way to
eliminate the dynamics problem in DAC combustors.
B. Dry Low-Emissions Combustor

Increasingly stringent emissions regulations, along with the disadvantages of
using water or steam injection for NOx control for land-based industrial gas
turbines drove the development of the dry low-emissions (DLE) lean premixed
combustion system for the LM6000.5,13,14 The combustion technology was
systematically developed using empirical and analytical design processes and ex-
tensive component testing, including single- cup, two-cup, and full-scale annular
combustors. The rig tests were followed by final combustion-system refinements on
an engine test. The LM2500 and the LM1600 combustors were developed in quick
succession to the LM6000 utilizing the technology developed in the process.15
A cross section of the LM6000 combustor is shown in Fig. 3.3. The combustor
has three domes arranged radially to permit parallel staging of the three domes. The
middle and the outer domes each consist of 30 premixers, whereas the inner dome
has 15 premixers. This arrangement permits the use of standard premixer sizes in
the three domes. The inner dome is at about half of the radius of the outer dome,
and so circumferential spacing can accommodate only 15 standard premixers. To
Fig. 3.3 LM6000 DLE combustor.

Fig. 3.4 Schematic illustration of the fuel-delivery system of the LM2500/6000 DLE
combustion system.
increase the air available for combustion, the combustor liners are convectively
cooled by using turbine-cooling air.
The 75 premixers are arranged on 15 two-cup and 15 three-cup assemblies. The
two-cup assemblies do not have the innermost premixer. Each “cup” consists of
a double annular counterrotating swirler (DACRS) premixer, in which two axial
counterrotating coaxial swirlers are mounted with a hub separating them, followed
by a mixing duct.5 The middle and inner dome premixers are identical whereas
the outer premixers are somewhat larger, such that the dome reference velocities
are all similar.
The fuel-delivery system consists of individual fuel controls for each dome of
the combustion system as described by Joshi et al.5 (Fig. 3.4). Independent control
of the fuel flow to each of the three domes is used to operate within the combined
constraints of emissions, combustion acoustics, and demanded total fuel flow rate
(shaft power). In addition, compressor air is bled from the engine at part-power
operation to extend the range over which the flame temperatures can be maintained.
As power is further reduced, radial and circumferential staging modes are also used
to bring flame temperatures within their operating limits (Fig. 3.5.)
Combustion dynamics can create substantial limitation to the operation of a lean
premixed combustor. Three design features of the system are used to expand the
range over which dynamics are within acceptable limits. First, a small fraction of
the fuel is injected into the combustor from holes in the walls of the mixing duct
(Fig. 3.6). This method of fuel injection has been shown in several instances to
provide improved performance of combustion dynamics in premixed combustion
devices.
Second, a set of quarter-wave tubes is provided on the premixers outside the en-
gine to absorb combustion-generated noise. These damper tubes, of three different
Fig. 3.5 Schematic illustration of the fuel-staging strategy for the LM2500/6000 DLE
combustion system.
lengths, are installed on the fuel nozzles on the engine as shown in Fig. 3.3. Each
damper tube length is designed for critical operating band dynamic frequencies.
The damper tubes open into the diffuser cavity and communicate with the com-
bustor through the premixers.
Finally, the combustor-dome flame temperatures are empirically scheduled
to achieve low emissions and stable operation of the gas-turbine engine over its
entire operating range. The stable values of dome flame temperatures are affected
by variation in fuel property, ambient conditions, and load changes; thus, unstable
operation of the combustor can develop from changes in these parameters.
Fig. 3.6 The LM6000 DLE premixer showing ELBO fuel injection for controlling
dynamics.
Combustion-dynamic pressures are measured continuously and monitored by the

acoustics and blowout avoidance logic (ABAL) within the control system. If the
monitored dynamic pressures exceed factory-set limits for more than a set period,
the control system takes action to alter flame temperatures based on algorithms
developed in factory testing of the engines to reduce dynamic pressures to
acceptable levels. If the control system is unable to effect a reduction in dynamic
pressures, then it commands the gas turbine to step to idle as a precautionary
measure. The ABAL logic within the control system can also detect incipient
lean blowouts by comparing measured and calculated fuel flows for the operating
conditions based on a cycle model calibrated for each engine. The ABAL control
system increases the flame temperature in the appropriate dome when an incipient
lean blowout is detected.
In summary, premixed low-emissions combustors have been designed and op-
erated in the field for nearly a decade with combustion-dynamics as a controlled
but persistent issue. Further development of DLE combustion systems is limited
by the risk of introducing new combustion-dynamic behavior as a result of design
changes. As emissions regulations become more stringent, the evolution of flight-
engine combustors will likely follow the path set by the DLE combustor toward
increasing levels of premixing and lower flame temperatures.16 Thus, a strong need
exists to drive the design process toward a more predictive capability of identifi-
cation and mitigation of combustion-dynamics phenomena in such combustors.
V. Combustion–Acoustic Modeling
A framework for eventually developing this predictive capability is shown in
Fig. 3.1. The basic philosophy behind this approach is the combination of analytical
modeling with experimentally determined boundary and submodel behavior. The
fundamentals of acoustic-wave propagation are well known; thus, determination of
available acoustic modes is straightforward once a temperature field is established
through the use of either empirical means or by computational fluid dynamics. The
remainder of the modeling problem consists of determining boundary conditions
and defining the submodel behavior. In particular, the accurate modeling of the
interaction of the combustion process with pressure waves is critical to the success
of the overall approach.
A. Time-Lag Combustion–Acoustic Model

The modeling of combustion acoustics in practical combustion devices is quite
complex and, in general, simplifying approximations are necessary to yield com-
putationally tractable solutions. One commonly employed approximation is the
use of a single convective delay time to characterize the time of flight of individual
fuel parcels from the point of injection to the flame front. The primary underly-
ing assumption here is that the turbulent flame brush can be modeled as a steady,
contiguous thin reaction zone anchored at fixed locations within the combustion
chamber. In reality, the location of the reaction sheet varies in an unsteady manner,
in part, because of instabilities in fuel and airflow rates.
Even when such fluctuations are ignored, the single convective delay-time ap-
proximation assumes that the flame sheet is positioned in such a manner that the
time of flight for each fuel parcel is equivalent. The local flow velocity and turbulent
a) b) c)
Fig. 3.7 Contour plots: a) axial velocity, b) temperature, and c) reaction progress
variables from a steady CFD simulation of the LM6000 DLE combustor. The views
shown are on the plane that intersects the three-cup premixer.
flame speed determine the location of the flame sheet. Given the complexities of
the flowfield aft of the swirl-stabilized premixers, there can be little expectation
that convective times of individual fuel parcels originating within the same pre-
mixer will be confined to a narrow range. In addition, variation between premixers,
whether by design or chance, can result in further disparity in convective times.
A steady-state computational fluid dynamics (CFD) simulation17 of a 24-deg
sector of the LM6000 DLE combustor provides an illustrative example of the
variation in convective timescales. Plots of axial-flow velocity, temperature, and
reaction progress variables are shown in Fig. 3.7 for a typical high-power operating
point. In each annulus, corner recirculation zones anchor the flame front. The
extent of vortex breakdown and subsequent recirculation zone aft of the premixer
strongly depends on the premixer swirl number and expansion ratio. The lower
swirl number in the outer premixers results in a flame surface that is elongated
relative to those found in the middle and inner premixers. The swirl numbers of the
latter two premixers are equivalent; however, the inner premixer expansion ratio
is much larger because of increased premixer spacing.
To computationally determine the location of the flame front, a threshold value
of unity is chosen for the reaction progress variable G(x, t). This choice defines the
flame surface as the locus of points where the local mixture is in the burned state
50% of the time. The resulting distribution of calculated convective timescales
is provided in Fig. 3.8. Fuel–air mixing is not modeled; instead, experimentally
derived profiles of velocity and mixture fraction are used as inlet boundary condi-
tions. As such, the convective delay times represent only the time of flight from the
premixer exit planes to the flame sheet. The time of flight from the injection points
to the premixer exits can be determined by using through-the-vane CFD simula-
tions but, for the purposes of this example, convective times within the combustor
are sufficient to illustrate the complications of choosing a single timescale.
As seen in Fig. 3.8, convective times between the premixer exit and the flame
front range from 0.1 to more than 0.7 ms. The median, the mass-weighted mean, and
Fig. 3.8 Distribution of convective times from the premixer exit plane to the flame
front. Also shown is a log normal distribution.
the standard deviation are 0.40, 0.46, and 0.18 ms, respectively. For comparison
proposes, the average residence time of the fuel–air mixture in the premixer is
approximately 0.45 ms. Statistical differences between individual premixers are
significant because median times for individual premixers range from 0.25 to
0.47 ms. The overall distribution is not well fit by using either normal or log normal
distributions (for the log normal distribution, see Fig. 3.14). These time lags can
also be expressed as phase lags in the frequency domain. At a frequency of 500 Hz,
the convective time-delay range of 0.1–0.7 ms is equivalent to a 18- to 126-deg
phase lag, which covers both positive and negative Rayleigh index values. To
account for the multiple convective timescales associated with the mixer, the single-
time-lag-modeling approach is oversimplistic and may need to be expanded to
reproduce observed behavior. An approach for deriving complex premixer transfer
functions from steady-state CFD simulations has recently been proposed.18
B. Semiempirical Combustion–Acoustic Model

Rather than utilize a fully analytical description of the combustion–acoustic
interaction process, judicious use of subcomponent data may provide a means
for overcoming the modeling issues. By themselves, component tests (for in-
stance, single-cup tests of an individual fuel injector–swirler combination) cannot
reproduce full-engine behavior for at least three reasons. First, a full-scale annular
combustor has several different natural acoustic mode shapes (and thus resonant
frequencies) that can have significant radial and circumferential components. A
single-cup test can only support a first- or second-order mode, providing only one
or two resonant frequencies with which the fuel nozzle–swirler combination can
interact. Second, the fuel system can potentially also play a role in affecting the
Fig. 3.9 Tunable combustion acoustics (TCA) rig conceptual view.
phase relationship between a pressure disturbance and the resulting equivalence-

ratio fluctuation that reinforces the pressure wave. Third, in general, the acoustic
inflow and outflow boundary conditions are not well represented by a single-cup
test. The choked outflow boundary of a turbine nozzle can be simulated with a re-
stricted test combustor exit, but the inflow boundary of the final compressor stages
is more difficult to simulate.
Many of these objections can be overcome by application of a computational
model that encompasses the entire combustion system, including appropriate
modeling of inflow and outflow boundaries, fuel-system dynamics, aerodynamic
pressure–equivalence ratio interactions, flame dynamics, and acoustic wave prop-
agation. Unfortunately, not only is a full model of the dynamic system beyond
current computational capabilities, many of the physical phenomena are insuffi-
ciently understood to permit first-principles modeling. In particular, the interaction
of acoustic waves with the flame front, fuel flow, and airflow rate is complex and
not well understood. Thus, the combustion–acoustic phenomena are represented
by a transfer function, whose parameters are unknown but are expected to be a
function of the design details of the fuel and air injection and mixing devices. To
circumvent the modeling difficulty, an experimental means of determining these
interactions has been developed in a relatively small-scale component test. The
data obtained from this test can then be used to determine the input parameters
for the full-combustor simulation. The test facility is called the tunable combustor
acoustic (TCA) test rig (Fig. 3.9), and has been applied to several GE industrial
aeroderivative combustor components. The philosophy behind the experimental
design is to provide a single-fuel nozzle–swirler assembly with an acoustic cham-
ber that has well-characterized boundary conditions and a continuous range of
natural frequencies that covers the resonant modes typically found in full-annular
gas-turbine combustors.
The TCA consists of an upstream chamber, a single-cup fuel nozzle–swirler
assembly, and a downstream combustion chamber. Both chambers are continu-
ously tunable by use of a perforated piston and actuators. Air is preheated to
typical compressor-discharge temperatures, passes through the upstream piston,
the swirler, and the downstream piston. Cooling water is introduced only near
the backpressure valve, beyond which it should have an effect on the combustion
acoustics process. The acoustic response of the system is characterized by sev-
eral dynamic pressure transducers located down the length of the chambers on
both sides of the pistons and within the fuel-system feed tubes. Dynamic pressure
data are collected while the combustor length is varied continuously to give a first
axial acoustic mode that covers a natural frequency range consistent with engine
experience. The piston is moved in both directions, thus giving an indication of
hysteresis of the system response. For relatively small acoustic levels (<190 dBa),
no hysteresis is detectable. At higher levels of response, some amount of
“mode-locking” is evident. However, these levels are beyond the levels sustainable
in an engine without damage and are thus of little practical interest.
Although simple in concept, several issues make the interpretation of test results
less simple than one might wish. Because of overall pressure-drop limitations and
mechanical durability issues, the flow through the pistons is not choked. How-
ever, by measuring dynamic pressures on both sides of the pistons, the complex
impedance of the pistons can be determined analytically. The geometry of the
combustion chamber is quite simple, only supporting axial acoustic modes of
well-defined frequencies. Although the axial mode shape of the TCA is desirable
from an analytical perspective, the measured and calculated mode shapes in an an-
nular combustor often contain a significant circumferential component. Provided
that the primary acoustic coupling mechanism is through pressure-wave-induced
fluctuations in equivalence ratio, lack of circumferential modes developed in the
test rig will be unimportant. The typical length scale of a pressure wave is on the
order of 1 m, whereas the length scale of the swirler is typically an order of mag-
nitude less. From the perspective of the pressure fluctuation, the mode shape will
not significantly alter the response of the fuel nozzle–swirler. However, the flame
dynamic response will also depend on the velocity fluctuation, which is sensitive
to orientation and thus mode shape. This limitation cannot be easily overcome
through this type of test and would require a very different approach.
A typical result from the TCA test rig is shown in Fig. 3.10. The vertical axis is
clock time, during which the downstream piston is moved from its maximum to
Fig. 3.10 Typical results from a tunable combustion acoustics rig.

minimum length position, and then back to its original position. The piston velocity
is constant, and thus the axis can also be interpreted as effective combustor length.
The horizontal axis is acoustic frequency, and the color scale indicates amplitude
at the corresponding frequency. An envelope of the maximum response is shown
at the bottom of the figure. As is evident, the system response strongly depends
on the combined natural resonant frequency of the combustion chamber and the
response of the fuel nozzle–swirler. Because of the relative simplicity of the phys-
ical experiment, computational modeling of the test vehicle can be accomplished
with confidence. The combustor–acoustic interaction is still treated as a transfer
function; in this case, the experimental data are used to determine the parame-
ters of the transfer function. Under the assumption of the mode shape–combustor
response independence described earlier, this transfer function is expected to be
applicable to the full annular model of the combustor as well.
A significant advantage of this approach is that it permits evaluation of the impact
of swirler and fuel nozzle design variables on combustion acoustics, without the
expense and delay associated with construction of a full-engine set of hardware and
test. It also avoids the circumstance under which a design change only moves the
maximum response frequency of the fuel nozzle to a nonresonant mode of a fixed-
geometry single-cup test rig. As stated earlier, a large-scale annular combustor
has numerous mode shapes covering a wide range of frequencies available to it,
whereas a fixed-geometry single-cup rig generally has only one or two natural
acoustic frequencies. The design strategy described is still being validated and has
not yet been demonstrated to completion. Given the risk to combustor and engine
development schedules represented by combustion acoustics, the value of such a
strategy is clear.
C. CFD Modeling of Combustion Dynamics

Detailed CFD modeling of the combustion process also holds promise in the
determination of combustion–pressure wave interactions. Significant progress has
been made in time-dependent modeling of the combustion process through large-
eddy simulation (LES) for modeling combustion-generated flow instability.19–21
A detailed discussion of CFD tools and their relationship to combustion acoustics
is included in Chapters 10 and 11. The current discussion addresses the challenges
in the use of these calculations as design tools.
The computational demands of a LES simulation are considerable, in general,
limiting the computational domain to a single fuel nozzle or premixer and simpli-
fied combustor geometry. The compromises made in this approach are similar to
the compromises made in the previously discussed TCA test device. Only axial
acoustic modes can be simulated because the computational expense of a full an-
nular calculation. The inflow and outflow boundary conditions are approximations
to those in a gas turbine. Fuel-system interactions and feedback are either ignored
or reduced to simplified models that require their own assumptions of physical
coupling mechanisms.
Despite these difficulties, early results with these models are encouraging, and
they will likely be applied as submodels within a system-level acoustic model as
depicted in Fig. 3.1. Advanced CFD can certainly be used as an analytical tool to
obtain new insights into the physical interactions that drive combustion acoustics
and to identify and evaluate potential design avenues to improve combustion-

system performance.
VI. Acoustic Modeling Results

Preliminary results of combustion acoustic modeling of the LM6000 DLE com-
bustor provide some level of confidence in the strategy described in the preceding
section. In high-power operation, two discrete frequencies at 450 Hz and 600–
650 Hz are typically observed. Figure 3.11 shows the comparison between mea-
sured and predicted dynamic response by using the single-time-lag approach. The
predicted response is obtained by taking the maximum values of the individual re-
sponse from each dome and normalizing with a reference value (peak value) from
the measured spectrum plot. Characteristic convective times associated with each
combustor dome are estimated based on the flame location provided by the CFD
analysis. Within the frequency range of interest, the model predicts two dominant
frequencies, one at 500 Hz and the other at 600 Hz. These frequencies are reason-
ably close to those observed from engine tests at the specified operating condition.
These figures also show that the 500-Hz mode is more active than the 600-Hz
dynamics. The predicted resonant frequency of the 500-Hz mode is approximately
50 Hz higher than the frequency observed in the engine test. This deviation may
be attributed to two possible sources of error, which are inherently present in the
computation. First, the accuracy of the predicted flame temperature is a critical
factor because of its influence on the speed of sound; therefore, it contributes to
errors in frequency prediction. Second, the uncertainty in estimating the location
Fig. 3.11 Comparison of measured and predicted combustion acoustic response from
LM6000 DLE.
a)
b)
c)
d)
Fig. 3.12 Effect of radial temperature nonuniformity on combustion dynamics.

of the flame front (which is an input critical to estimating the characteristic convec-
tion time distribution) by examining the steady-state heat-release contours could
be a factor contributing to the discrepancy in frequency calculation. The uncer-
tainty in calculating transport-time distribution associated with the unsteady heat
release will induce a different phase lag into the model and thus predicts a dif-
ferent resonant frequency. Based on the current analysis, the dynamic pressures
response in the vicinity of 500 Hz is mainly contributed by the plane waves from
the middle and inner domes. On the other hand, the source of the 600-Hz acoustic
mode appears to be the first-order circumferential mode arising in the outer ring.
The acoustic analyses of the baseline and uniform-temperature-distribution
case were conducted to investigate the effect of radial temperature nonunifor-
mity on combustion instability. The measured and predicted responses are shown
in Figs. 3.12a and 3.12b, respectively, for baseline operation, and in Figs. 3.12c and
3.12d, respectively, for an operating condition with uniform temperature in all three
domes. The measured spectra shown in these figures were based on dynamic pres-
sure measurements from pressure sensors installed just downstream of the flame
front in the DLE combustor. The predicted response is obtained by taking the max-
imum values of the individual responses from each of the three rings. Two distinct
frequency peaks near 400 and 600 Hz were found from the analysis, similar to those
observed from engine tests. These figures also show that the 400-Hz mode is more
active than the 600-Hz dynamics for both operating conditions. In addition, the
measured amplitudes of the two acoustic modes at the baseline condition showed
relatively low levels. The predicted stability indices for these two peaks were less
than 1 and are both categorized as acoustically inactive. In contrast, an acousti-
cally active mode near 400 Hz with a stability index of 3 is found for the operating
condition with uniform flame temperatures in all three rings. This predicted trend
agrees well with test data and further demonstrates the capability of the current
model to distinguish between acoustically active and inactive regimes of operation.
The methodology required to incorporate passive damping devices was a prime
requirement in this modeling activity.22,23 As mentioned in Sec. III, a set of
22 damper tubes is provided upstream of the premixers, outside the engine to
absorb combustion-generated noise. Because the mean flow Mach number in the
air column of these devices is very small (typically less than 0.05) and the diameter
of these quarter-wave tubes is much smaller than the wavelength of the resonant
frequency of oscillation, the dynamic pressure and mass flow within the damper
tubes can be represented by a one-dimensional analysis.
These devices are quarter-wave resonators installed in the cold section of the
combustor just upstream of the premixers. The installation schematic of the damper
tubes used in the formulation of this analysis is shown in Fig. 3.13. These devices
act to detune the combustor by providing a finite number of discontinuities at the
locations where they are installed. The incident and reflected acoustic waves in the
diffuser cavity are significantly altered to the extent that certain discrete oscillations
are attenuated and therefore become less destructive to the combustor. The acoustic
damping devices in the current analysis are modeled by treating each quarter-wave
tube as a monopole acoustic source characterized by its acoustic impedance. The
effect of damper tubes (tuned for 510 Hz) on combustion dynamics in the pilot
(middle) ring is shown qualitatively in Fig. 3.14. A 7-fold decrease in stability
index at the tuning frequency of the damper tubes is predicted based on this analysis.
Fig. 3.13 Schematic illustration of acoustic damper tube in combustor.
Fig. 3.14 Predicted impact of damper tubes on combustor acoustic response.

Although analytical techniques have improved markedly during the past few
years, the difficulty in representing the combustion–pressure wave interaction will
likely remain a barrier to full predictive capability. Thus, it is likely that a combined
approach, including both experimental and high-fidelity unsteady CFD modeling,
will be required to develop models of the full annular combustor response that will
have utility in defining design direction.
VII. Conclusion
Combustion instability in conventional diffusion-flame gas-turbine combustors,
if encountered, can be eliminated, in general, with simple modifications to the
design or operating parameters. The design guidelines based on empirical know-
how are generally good enough to avoid the unforeseen occurrence of dynamics
during the engine-certification phase. If encountered, simple analysis coupled with
empirical guidelines and a systematic testing and development process are used to
find engineering solutions without adversely affecting the key combustion-system
design requirements.
On the other hand, controlling combustion dynamics in industrial DLE combus-
tion systems remains a substantial challenge for designers. The combustion process
is pushed close to the limits of lean flame stability and/or heat-release rates that
lead to strongly coupled nonlinear interaction between the flame exothermicity,
acoustic behavior of the system, and components. The approach for controlling
dynamics in the LM engines (bleed air, fuel staging, dome flame temperatures
differences, ELBO, acoustic damper tubes, and ABAL control logic) has worked
very well. However, to further improve on the design of lean premixed combustors,
we need to make significant advances in this area.
More recent directions in propulsion engine technology are driving designs that
have more in common with the lean premixed DLE combustors than with the
rich-dome combustors of the past. As these designs progress toward maturity, it
is likely that combustion dynamics will become a limiting factor in the attainable
performance of these combustors. Thus, improved predictive capability, funda-
mental understanding, and control technologies for combustion dynamics need to
be developed. The current state of the art is represented by a combined-systems
approach in which empirically derived submodels are linked with a physics-based
system model. In the future, we anticipate an increased role in developing LES
capabilities as a tool to develop these submodels, rather than relying on component
test or engine data. In the longer term, more direct linkage of the physical coupling
mechanisms between the components of the combustion system (fuel system, fuel
injector, combustor cavity, and combustion–pressure wave interactions) will make
it possible to accurately describe, predict, and control the physical phenomena
associated with combustion dynamics.
References
1
Lieuwen, T. “Analytical Modeling of Combustion-Acoustic Wave Interactions: A Re-
view,” Journal of Propulsion and Power, Vol. 11, No. 22, 2003, pp. 222–444.
2
Combustion Instability in Lean Premixed Combustors,” American Society of Mechanical
Engineers, Paper 99-GT-52, 1999.
3
Straub, D. L., and Richards, G. A., “Effect of Axial Swirl Vane Location on Combustion
Dynamics,” American Society of Mechanical Engineers, Paper 99-GT-109, 1999.
4
Paschereit, C. O., and Gutmark, E., “Passive Combustion Control for Enhanced Stability
and Reduced Emissions in a Swirl-Stabilized Burner,” AIAA Paper 2003-1011, 2003.
5
Joshi, N., Epstein, M., Durlak, S., Marakovits, S., and Sabla, P., “Development of a Fuel
Air Premixer for Aero-Derivative Dry Low Emissions Combustors,” American Society of
Mechanical Engineers, Paper 94-GT-253, 1994.
6
Scarinci, T., and Halpin, J. L., “Industrial Trent Combustor – Combustion Noise Char-
acteristics,” American Society of Mechanical Engineers, Paper 99-GT-9, 1999.
7
James, D., “A Solution for Noise Associated with a Series Staged DLE Combustion
System,” Proceedings of the 4th International Pipeline Conference, 29 Sept. to 3 Oct. 2002.
8
Schlein, B. C., Anderson, D. A, Beukenberg, M., Mohr, K. D., Leiner, H. L., and
Träptau, W., “Development History and Field Experiences of the First F T8 Gas Turbine
with Dry Low NOX Combustion System,” American Society of Mechanical Engineers,
Paper 99-GT-241, 1999.
9
Johnson, C. E., Neumeier, Y., Lubarsky, E., Lee, J. Y., Neumaier, M., and Zinn, B.
T., “Suppression of Combustion Instabilities Using a Fast Adaptive Control Algorithm,”
AIAA Paper 2000-16365, 2000.
10
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Structure and Control of
Thermoacoustic Instabilities in a Gas-Turbine Combustor,” Combustion Science and
11
Magill, J., Bachmann, M., and McManus, K., “Combustion Dynamics and Control in
Liquid-Fueled Direct Injection Systems,” AIAA Paper 2000-1022, 2000.
12
Jones, C. M., Lee, J. G., and Santavicca, D. A., “Closed-Loop Active Control of
Combustion Instabilities Using Subharmonic Secondary Fuel Injection,” Journal of
Propulsion and Power, Vol. 15, No. 2, 1999, pp. 1–7.
13
Leonard, G., and Stegmaier, J., “Development of An Aero-Derivative Gas Turbine Dry
Low Emissions Combustion System,” American Society of Mechanical Engineers, Paper
93-GT-288, 1993.
14
Joshi, N. D., Mongia, H. C., Leonard, G., Stegmaier, J. W., and Vickers, E. C., “Dry
Low Emissions Combustor Development,” American Society of Mechanical Engineers,
Paper 98-GT-310, 1998.
15
Patt, R., “Development and Operating Experience of DLE Combustion Systems,” 12th
Symposium on Industrial Applications of Gas Turbines, 15–17 Oct. 1997.
16
Mongia, H. C., “TAPS: A 4th Generation Low Emissions Propulsion Engine Com-
bustion System,” The AIAA/ICAS International Air and Space Symposium, 2003, AIAA,
Reston, VA, 2003.
17
Held, T. J., Mueller, M. A., Li, S.-C., and Mongia, H. C., “A Data-Driven Model for
NOX , CO and UHC Emissions for a Dry Low Emissions Gas Turbine Combustor,” AIAA
Paper 2001-3425, 2001.
18
Flohr, P., Paschereit, C. O., and Belluci, V., “Steady CFD Analysis for Gas Turbine
Burner Transfer Functions,” AIAA Paper 2003-1346, 2003.
19
Huang, Y., Sung, H. G., Hsieh, S. Y., and Yang, V., “An LES Study of Combustion
Instabilities in Lean-Premixed Gas Turbine Combustors,” Journal of Propulsion and
Power, Vol. 19, 2003, pp. 782–794.
20
Stone, C., and Menon, S., “Open-Loop Control of Combustion Instabilities in a Model
Combustor,” Journal of Turbulence, Vol. 4, 2003, p. 20.
21
Schlüter , J. U., “Static Control of Combustion Oscillations by Coaxial Flows: An
LES Investigation,” Journal of Propulsion and Power, Vol. 20, No. 3, 2004, pp. 460–467.
22
Hsiao, G. C., Pandalai, R. P., Hura, H. S., and Mongia, H. C., “Investigation of Combus-
tion Dynamics in Dry Low Emission Gas Turbine Engines,” AIAA Paper 98-3381, 1998.
23
Pandalai, R. P., and Mongia, H. C., “Combustion Instability Characteristics of
Industrial Engine Dry Low Emission Combustion System,” AIAA Paper 98-3379, 1998.
Chapter 4
Combustion Instability and Its Passive Control:

Rolls-Royce Aeroderivative Engine Experience
Tomas Scarinci∗
Rolls-Royce Canada, Quebec, Canada
Nomenclature
c = speed of sound
d = location of a source of fuel injection
dk = location of the kth fuel-injection source
p = limit-cycle pressure oscillation amplitude
f = frequency
H ( f ) = transfer function
K = regression constant
ki = regression constant
L c = length of combustion chamber
L m = length of fuel–air-mixing region
M = Mach number
p = instantaneous static pressure
P = average static pressure
p = pressure fluctuation
P3 = compressor discharge pressure
Re = real part of a complex number
s = spacing between two points of fuel injection
sk = strength of the kth fuel-injection source
T 3 = compressor discharge temperature
Ti = flame temperature of the ith combustion zone
u = instantaneous velocity
U = average velocity
u = velocity fluctuation
Copyright c 2005 by Rolls Royce Plc. Published by the American Institute of Aeronautics and
Astronautics, Inc., with permission.
∗ Director of Gas Turbine Engineering, Rolls-Royce Energy Business.
65
66 T. SCARINCI
α = FAR-damping coefficient, as defined by Eq. (4.5)

αi = regression constants [Eq. (4.2)]
γ = ratio of heat capacities
λ = sound wavelength
ω = angular frequency
I. Overview of the Trent 60 Aeroderivative
T HE objective of this chapter is to describe the experience of Rolls-Royce with

combustion instabilities, or more accurately, oscillatory combustion, in the
Trent 60 Aeroderivative gas turbine by using dry low-emissions (DLE) combustion
technology. The Trent 60 is a 35 : 1-pressure-ratio, three-shaft engine that produces
nominally 52 MW of power on an International Organization for Standardization
(ISO) standard day (i.e., 15C at sea level) with a simple-cycle efficiency of 43.1%.
Cross sections of the engine with a detail of the DLE combustor (as originally
configured in 1997) are shown in Fig. 4.1. There are eight can combustors per
engine. For size-calibration purposes, each of these “combustor cans” is roughly
0.2 m in diameter and 0.7 m in length.
The combustion chamber consists of three premixing channels, which are re-
ferred to, respectively, as the primary, secondary, and tertiary premixers. The pri-
mary premix system is the only premix system that is self-stabilized. That is, the
primary system can be operated alone, whereas the secondary and tertiary systems
cannot. Furthermore, the primary system is the only system that uses swirling
flow as a means to discharge into the combustion chamber. The secondary (and
tertiary) premixed streams are discharged into the combustion chamber by means
of discrete jets (similar to traditional dilution jets of conventional gas-turbine com-
bustors), which are ignited by the upstream stages, as the jets are mixed inside the
combustor. If the primary stage flames out, the whole combustor flames out. The
practical implication of the design is that the secondary and tertiary stages can
be operated at much lower temperatures than what is normally required for flame
stabilization. The result is a large turndown ratio in the achievable fuel–air ratios
of the secondary and tertiary premixers.
Part of the secondary fuel–air mixture is entrained into the primary zone to
mix with the primary premix stream. Once the flame temperature of the secondary
stream reaches a certain level, this entrainment results in a significant improvement
in the weak-extinction limit of the primary system. This improvement occurs
because the average temperature resulting from the mixing of the two streams
governs the weak-extinction of the primary zone.
At a given power level, the total amount of fuel inside the combustor is pre-
scribed. The combustion engineer is left with the choice of the allocation of the
total fuel between the three premix stages. The possible fuel splits are limited by a
series of constraints: the primary weak-extinction temperature, the maximum tem-
perature for any of the three stages because of nitrous oxide (NOx) reasons, and a
minimum temperature of the last stage (tertiary) because of carbon monoxide (CO)
requirements. The secondary stage does not really have a minimum temperature,
provided the tertiary stage is hot enough to provide CO burnout.
At a given power level, the ensemble of possible ways to allocate the fuel inside
the combustor defines an operating envelope, the axes of which are best defined in
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 67
Fig. 4.1 Cross-section of Trent 60 DLE aeroderivative with details of the original
three-stage dry, low-NOx combustion system.
terms of premixer (or zone) temperatures. For instance, once a primary temperature
is chosen (say, 1750 K), a secondary temperature (say, from 1200 K to 2000 K) can
be chosen. The amount of fuel to be allocated to the tertiary stage falls out from
the total amount of fuel required by the engine. Note that allocating no fuel at all
to the tertiary stage is also (sometimes) an option. Fig. 4.2 shows typical possible
operating envelopes for the Trent combustor at different power levels. The temper-
atures of the y axis and x axis are the primary- and secondary-flame temperatures,
respectively, from which a reference temperature was subtracted. Any point inside
68 T. SCARINCI
250
primary premixer temperature, K
200
80% power
150
100% power
100
50
0
-50
-100
-150
-200
-300 -200 -100 0 100 200 300 400
secondary premixer temperature, K
Fig. 4.2 Typical operating envelopes for the Trent 60 combustor (ISO Day).
the operating envelope defines a unique fuel split between primary, secondary,
and tertiary stages. One can easily imagine that the NOx level at the combustor
exit will depend on the chosen fuel split. By appropriately staging the fuel, NOx
emissions in the 10- to 20-ppm range can be achieved from 50–100% power.
II. Oscillatory Combustion in the Trent 60 DLE

Recently, combustion instability has emerged as a central technical problem in
the design of DLE combustors. Despite the variety of technical approaches used
in the design of DLE combustors (annular vs. can combustors, parallel vs. series
staging, various flame stabilization strategies, etc.) the problem has been experi-
enced by almost all gas-turbine manufacturers. Premixing of fuel and air (before
to combustion) requires the presence of a mixing duct, which itself implies the
existence of a time delay between fuel injection and heat release in the combustion
chamber. It is the presence of this time delay (necessary for mixing fuel and air)
that lies behind most combustion-instability problems.
In general, when a given engine-operating parameter crosses a particular thresh-
old, a spontaneous oscillation of the combustion process manifests itself and
rapidly stabilizes into a limit-cycle oscillation of a given frequency and ampli-
tude. These limit cycles actually tend to be very stable; it takes a significant effort
to break away from the limit cycle, that is, there can be significant hysteresis
moving in and out of these regions of combustion oscillation. Furthermore, the
amplitude and frequency of the limit cycle tend to be a consistent function of the
engine-operating parameters (power level, ambient temperature, fuel splits, etc.).
Over the years, we have come to adopt the term “spontaneous oscillatory combus-
tion” as being more descriptive of the phenomenon, rather than the usual term of
“combustion instability.” The fundamental issue is to determine the conditions un-
der which a limit cycle appears, and if so, to determine its frequency and amplitude
under these conditions.
The amplitude of these limit cycles is typically on the order of 1–2% of the
static pressure inside the combustion chamber, and typical frequencies seem to
range from approximately 100 Hz to roughly 1–2 kHz, although these values
depend on the engine geometry and size. These amplitudes and frequencies can be
highly problematic for the mechanical components of most engines because they
are a source of high-alternating mechanical stresses. In the absence of control or
avoidance procedures (or designs), mechanical components can abruptly fail by
high-cycle fatigue under these conditions.
A. Basic Characteristics of Oscillatory Combustion

To assess the limit-cycle characteristics of the combustion system of the original
Trent, a detailed mapping exercise was undertaken at a variety of power levels and
ambient conditions and on different engines.1 Depending on the ambient condi-
tions, the relationship between combustor inlet and outlet temperatures can vary
quite significantly. Thus, at different engine-operating conditions, the allocations
of fuel among primary, secondary, and tertiary stages that gives the best combina-
tion between NOx emissions and pressure-oscillation amplitude will vary.
Figure 4.3 illustrates the typical results of a measurement of the observed
pressure-oscillation amplitude at 100% power. Again, at a given power level, the
primary and secondary temperatures can be controlled independently, whereas the
tertiary stage takes the balance of the total fuel. The amplitudes in the pressure-
oscillation map are the rms value of the signal from piezoelectric transducers
between 10 and 2000 Hz. The frequencies corresponding to the regions of large-
pressure-oscillation amplitude (or regions of oscillatory combustion) are discussed
subsequently.
Fig. 4.3 Pressure-oscillation amplitude contour map near 100% power.

70 T. SCARINCI
1.4
1.2
1.0
P/ Pref (%)
0.8
0.6
0.4
0.2
0.0
0.00 0.25 0.50 0.75 1.00
L/λ
Fig. 4.4 Measured combustor pressure-oscillation amplitudes and frequencies for
the longitudinal acoustic modes of the combustion chamber.
This contour map, and others at 50% and 80% load,1 shows that it is always
possible to find regions of high- and low-pressure-oscillation amplitude at any
power condition. In other words, the rms level of pressure fluctuations is clearly
affected by the distribution of heat release inside the combustor. In general, the
pressure-oscillation amplitude scales linearly with the combustor pressure. Thus,
the potential for structural damage is much higher at high engine pressure ratios.
The potential for structural damage is a difficult challenge for the Trent, which
operates at pressures up to 40 MPa.
Depending on the power level and the ambient conditions, the regions of large-
pressure-oscillation amplitude will be located in a different region of the operating
envelope. In Fig. 4.3, there are two regions of large-pressure-oscillation ampli-
tude; both are located in the region of low-primary temperatures. However, as the
secondary temperature is varied, it becomes possible to find an optimum condition
that will minimize the pressure-oscillation amplitude. The two distinct regions of
large-pressure-oscillation amplitude in Fig. 4.3 correspond to two different acous-
tic modes of the combustor. In all, a total of three different longitudinal acoustic
modes were seen under different operating conditions. These three modes are il-
lustrated in Fig. 4.4, which condenses the results from more than 300 tests by
plotting the pressure amplitude against a nondimensional frequency. Note that
all results are clearly grouped into three well-defined acoustic modes. At a given
condition, there may no large-amplitude oscillations, that is, there is no clearly ob-
servable limit-cycle oscillation. Nonetheless, one can always identify a dominant
frequency in the pressure spectrum, and this dominant frequency always seems to
correspond to one of the normal acoustic modes of the combustion chamber. In
this case, that is, when there is no clear limit-cycle oscillation, the rms level of the
pressure fluctuations is on the order of 0.1–0.2% of the reference pressure. When
oscillatory combustion sets in, it always appears at the frequency of one of the
100%
primary bias = 0 K
primary bias = +75 K

80%
3rd m ode
(% of maximum)
noise amplitude
60%
40%
2nd m ode
20%
0%
-50 0 50 100 150 200 250
secondary tem perature bias, K
Fig. 4.5 Change in acoustic mode and pressure oscillation amplitude as the
combustor fuel split is varied at a steady-engine-operating condition near base load
(100% power).
first three natural acoustic modes. Depending on the operating conditions, the fuel
split at which the limit cycle appears and its frequency and amplitude will vary.
We discuss this relationship in more detail next.
Figure 4.5 is effectively a slice through the pressure-oscillation amplitude map
of Fig. 4.3. As the secondary bias is increased, the pressure-oscillation amplitude
decreases until it reaches a minimum. At the amplitude minimum, however, a
mode-switching phenomenon occurs and the frequency of the limit-cycle oscilla-
tion switches from the second to the third acoustic mode. It can also be seen that the
overall pressure oscillation levels can be reduced if the primary bias is increased.
This figure provides a good illustration of the flexibility offered by the three-stage
design. At a fixed-engine-running condition, a small change in the primary and/or
secondary temperature can reduce the combustion pressure-oscillation amplitude
by 50% and/or select the frequency of the limit cycle.
Figure 4.6 plots the dependence of the amplitude of the second longitudinal
mode (i.e., L c /λ = 0.5) on combustor-fueling conditions and engine power. The
amplitude of this mode always decreases as the primary bias is increased. What
appears to be large scatter at a given primary bias is actually the effect of the
secondary bias being changed. Even though the pressure-oscillation amplitude is
nondimensionalized by the combustor reference pressure, all the curves do not
collapse into one. Thus, the second-mode amplitude depends to a certain degree
on the total energy being released inside the combustor.
The third unstable mode (L c /λ = 0.75) has amplitude characteristics that are
completely different from those observed with the second mode. As seen from
Fig. 4.7, it is possible to collapse all the data onto a single curve. For the third
mode to appear, the secondary-zone temperature must exceed the primary-zone
temperature.
72 T. SCARINCI
140%
120% 50% power

80% power
100% 100% power
(% of maximum)
noise amplitude
80%
60%
40%
20%
0%
-100 -50 0 50 100 150 200 250
primary temperature bias, K
Fig. 4.6 Effect of primary-zone temperature on the pressure-oscillation amplitude

of the second unstable longitudinal mode (Lc /λ = 0.5).
120%
100%
80%
(% of maximum)
noise amplitude
60%
40%
20%
0%
-100 -80 -60 -40 -20 0 20 40 60 80 100
Secondary minus Primary Zone Temperature (K)
Fig. 4.7 Amplitude characteristics of the third unstable mode (Lc /λ = 0.75). The
data in this figure are obtained at engine power levels ranging from 30% to 100%
power.
Under most combustor-operating conditions, the primary-zone temperature is

slightly above the secondary-zone temperature, because the primary zone needs to
keep a small margin above weak extinction, whereas the secondary does not need
to keep this margin. However, at high power, the primary-zone does have a large
margin above weak extinction, so that it becomes possible to have a condition
in which the secondary-zone temperature exceeds the primary-zone temperature.
Although this is a necessary condition for the existence of the third mode, it
is not a sufficient one. Depending on the power level and ambient temperature,
an unstable mode may or may not manifest itself. It is possible (depending on
combustor-operating conditions) to have situations in which neither the second
nor the third mode (nor the first mode) is present.
These observations on oscillatory combustion on a Trent 60 DLE engine date
back to 1997 and before, and they were obtained with a combustor design like
the one represented in Fig. 4.1. The next sections describe some of the control
approaches that were developed to either eliminate the presence of a limit cycle or
to reduce its amplitude or control (choosing) its frequency.
B. Instability Control Approaches

The simplest control approach to oscillatory combustion does not require any
hardware changes; software changes are made according to how the fuel is sched-
uled between the three stages as a function of various engine-operating conditions.
Changing the fuel splits of the combustor results in varying the heat-release dis-
tribution in the combustor and, as discussed earlier, this variation in itself is a
powerful means of controlling the limit-cycle amplitudes. This approach was in
fact used for many years, before we began to consider more fundamental changes
resulting in the potential elimination of oscillatory combustion altogether. Sec-
tion II.B.1 describes work to control oscillatory combustion amplitudes based on
fuel-staging approaches.
Development of other instability-mitigation approaches is aided substantially by
a deeper understanding of the underlying processes responsible for the resonance.
Richards and Janus2 and Lieuwen and Zinn3,4 suggested that the link between
lean premixing and combustion instabilities was possibly the introduction of a
time delay into the system. This mechanism is also discussed in Chapter 9 This
time delay originates from the need to achieve thorough fuel and air mixing before
combustion, that is, a certain “residence time” is required in the premixing duct to
achieve the required mixing quality. Pressure waves from the combustion zone lead
to airflow modulations in the premixer that, in turn, lead to a fuel–air ratio (FAR)
modulation. The resulting FAR disturbance is convected through the premixer into
the combustion zone, leading to an oscillation in heat release. If the fuel supply rate
is fixed, the relationship between flow-velocity oscillations and FAR oscillations
is given approximately by
F A R u
=− (4.1)
F AR ū
Lieuwen and Zinn3 further noted that, because of the low Mach number of the
flow in the passage (typically, M ∼ 0.05), small pressure fluctuations can result
74 T. SCARINCI
in significant fuel–air ratio oscillations. This can be seen by noting that
u 1 p
= (4.2)
U γM P
Thus, a 1% fluctuation in static pressure inside the combustor can typically result
in a 15% FAR fluctuation from the premixers. In other words, a pressure oscillation
inside the combustor would normally lead to a “FAR-wave” emanating from the
premixer exit and this can in turn modulate the heat release inside the combustion
zone. If the heat release associated with the FAR oscillations is in phase with the
original pressure waves, the oscillating heat release will augment the original pres-
sure waves in the combustion chamber. We will refer to this particular feedback
mechanism of combustion instability in DLE systems as the “Richards–Lieuwen
mechanism.” This expression seems to be useful shorthand and it recognizes the
empirical explanation first proposed by Richards in 1997 and the theoretical foun-
dations subsequently proposed by Lieuwen in 1998.
Although, as indicated in Chapter 9, many other potential mechanisms can lead
to combustion instability in a premixed system, even in the absence of fuel–air ratio
oscillations,5,6 none seem as important, or at least as powerful, as the Richards–
Lieuwen mechanism. At least, the Richards–Lieuwen mechanism seems to be the
most powerful in DLE combustion systems as they apply to most industrial gas
turbines.
In Fig. 4.8, a simplified account (compared with the more fundamental discus-
sion found in Chapters 9 to 12 (i.e., Sec. III of the book) is presented of the key
instability mechanisms believed to be present in most practical lean premixed sys-
tems. In simple terms, large-amplitude pressure waves can be generated either from
heat-release oscillations or the interaction of density oscillations with the pressure
gradient associated with the choked turbine (entropy noise). Note, in Fig. 4.8, that
the presence of fuel–air ratio oscillations emanating from the premixers can lead
to the generation of pressure waves in several ways. First, the oscillating FAR is a
direct source of heat-release oscillations (the Richards–Lieuwen mechanism). The
time-varying FAR from the premixers will also lead to a time-varying density field
inside the combustor, a time-varying flame-surface area, and a time-varying flame-
burning speed. All of these can lead to the generation of pressure waves either by
way of entropy noise or heat-release oscillations. A significant effort has been made
during the past few years, in particular, using active-control methodologies, to find
ways of eliminating FAR oscillations in DLE systems. The approach taken at Rolls-
Royce has been, somewhat analogous to acoustic-damping devices, to damp the
presence of FAR oscillations from the premixers. This is referred to as “FAR-Wave
damping” in Fig. 4.8, where it is quite clear that this sort of damping has the po-
tential to weaken numerous feedback loops leading to oscillatory combustion. The
efforts to develop FAR-wave damping technologies are described in Sec. II.B.2.
Note, in Fig. 4.8, that most instability mechanisms involve a coupling by means
of pressure waves. The aim of any passive-design solution is to reduce the strength
of any of the feedback loops. Irrespective of the mechanism, acoustic-attenuation
devices, such as resonators, always achieve this goal because the strength of pres-
sure waves is reduced. A brief description of our efforts to implement such acoustic
resonators is described in Sec. II.B.3.
Pressure Waves
in Premixer Aerodynamic
Aerodynamic
Damping
Premixer Velocity Fuel Flow

Oscillations Oscillations Supply Plenum
u (t) W f (t) fluctuations
(non-acoustic)
FAR-Wave
Damping
Damping
FAR Oscillations
FAR (t)
Flame Surface Burning Speed Fluid Density

Oscillations Oscillations Oscillations
Af ) Su (t) ρ (t)
Heat Release Interaction with

Oscillations Choked Nozzle
Q (t) (Entropy Noise)
Combustor Velocity
Oscillations Generation of
u (t) Pressure Waves
P (t)
Propagation of Pressure
Pressure Waves Wave
in Combustor Damping
Damping
Fig. 4.8 Driving mechanisms for combustion instability and means for suppression
of instabilities. The thicker line refers to mixing-induced oscillation (R-L mechanism).
Also noted in Fig. 4.8 are aerodynamic fluctuations (i.e., broadband turbulent
fluctuations) which disturb the flame and cause a general broadband level of in-
creased combustor pressure fluctuations. Efforts to minimize these aerodynamic
sources are described briefly in Sec. II.B.4.
1. Fuel Staging
The variable distribution of heat release along the length of the combustor
provides an effective mechanism to control the amplitude of longitudinal resonance
modes of the combustor. Having obtained some indications of the limit-cycle
behavior, an attempt was made to develop empirical correlations that could be
used by a controller to predict the amplitude of each of the unstable modes under
a particular condition.
In developing this correlation, it was assumed that the nondimensional amplitude
of the pressure oscillation is predominantly a function of the temperature ratio
76 T. SCARINCI
across the region of heat release. In a three-stage (axially staged) combustor there
are three regions of heat release, and hence there are at least three important
temperature ratios that will affect pressure-oscillation amplitude. It might be argued
that the overall heat release in the combustor (and the corresponding temperature
ratio from combustor inlet to combustor exit) might also be a controlling parameter.
For these reasons, the functional form of the empirical correlations that were used
for each of the unstable modes was:
N
p (Ti − T3 ) αi
= K 1 + K 2 P3A
(4.3)
P i=1
ki
where K 1 , K 2 , A, ki , and αi are all constants to be fitted to the data. The summation
index i, which runs from 1 to N , refers to each of the combustion zones of the
combustor, including the overall heat release inside the combustor. The arbitrary
constants were obtained from linear regression of engine data for each of the
unstable modes.
Three independent empirical correlations were obtained for each of the longi-
tudinal resonant modes. The assumption that was then made was that the mode
observed would be the one having the highest amplitude of the three, given the
combustor-operating conditions. Note that this approach also permits prediction of
not only the amplitude, but also the frequency of the combustion resonance. These
correlations are able to capture relatively well the effects of engine cycle, ambient
conditions, and changes in fuel schedule (or equivalently, variations in combustor
fuel splits). The correlations were able to reproduce the pressure-oscillation ampli-
tude mapping results within approximately 10% accuracy. These developed corre-
lations were used to design fuel schedules to minimize high-combustion pressure-
oscillation amplitude across the operating range of the engine. Of course, having a
three-stage premix combustor was almost a prerequisite for the implementation of
this type of approach. A three-stage premix combustor offered sufficient fuel-split
options to control pressure oscillations whilst meeting emissions requirements.
2. Fuel–Air Ratio Wave Damping

As previously suggested, the Richards–Lieuwen mechanism is believed to be
the dominant source of instability in most modern DLE combustors. The existence
of a finite time delay between the generation of a pressure wave and the associated
FAR fluctuation is at the heart of this mechanism. Once a perturbation in airflow
(or fuel flow) occurs at the point of fuel injection, a finite time (of the order of
milliseconds) is needed for this disturbance to be convected to the zone of heat
release. This travel time is also the time during which the mixing of the fuel and
air takes place. The time allocated to fuel–air mixing cannot be arbitrarily short,
otherwise the low-emissions objectives will not be met (because the quality of the
mixing will not be adequate). Similarly, the residence time in the premixer cannot
be made arbitrarily long, otherwise autoignition of the fuel–air mixture will occur
within the premixer.
The introduction of time delays in even the simplest of unstable physical systems
can result in the appearance of unstable modes.7 In fact, the number of unstable
modes in the system may often be linearly proportional to the time delay introduced
fuel fuel-1 fuel-2
Air Air
A B
s
x=-d x=0 x=- d x=0
fuel-1 fuel-N
fuel inlet
air jets
Air
C D
Lm Lm
x=-d x=0 x=-d x=0
Fig. 4.9 Schematic illustrations of evaluated premixer concepts.
in the system. Longer time delays provide more multiples of each of the fundamen-
tal unstable modes of the system. However, and in a DLE combustor, if there was a
sufficiently long mixing time (ignoring autoignition concerns for a moment) then
no FAR fluctuations would be coming out of the exit of the premixer, because they
would be eliminated by the mixing process. Hence, there are competing issues: as
the residence time in the premixer increases, there is a large number of possible
unstable modes, but there is no longer any fuel–air ratio (temporal) fluctuations
coming out of the premixer to drive the instability.
Although we refer the reader to our original publications for the details, con-
siderable effort was spent to develop and experimentally calibrate models that
captured the effects of convection and turbulent dispersion upon FAR perturba-
tions. A variety of candidate-injector configurations were examined experimen-
tally, analytically, and computationally. These configurations are shown in Fig. 4.9.
The laboratory tests made use of simple small-scale atmospheric rigs. Hot wires
and pressure transducers were used to monitor unsteady-flow conditions (u and
p ) and a proprietary fast flame-ionization detector (FID)8 was used to measure
instantaneous changes in fuel concentration (FAR ). Bottled ethylene was used
as fuel. The air supply was fan driven, and a downstream variable speed rotary
valve produced premixer airflow oscillations with peak amplitudes (u /U ) of 30%.
The mean airflow rate and the total fuel flow rate were held constant for all test
configurations.
Some of the key considerations associated with the effect of premixer design
variables on combustor stability characteristics can be understood by applying the
characteristic time analysis, as described in Chapter 17. As shown, instabilities
from the Richards–Lieuwen mechanism are excited when the time for the mixture
to travel from the injection point(s) to the flame lies within some range of integer
multiple of the acoustic period. Consider N sources of fuel-injection locations,
78 T. SCARINCI
such as shown in configuration C in Fig. 4.9. Each location has a strength sk and is
located at a position dk . Neglecting turbulent-dispersion effects for the time being,
it can be shown9 that the Lieuwen–Richards mechanism causes amplification of
acoustic waves when the following inequality is satisfied:

N
−i (ωdk ( U1 + 1c )− π2 )
Re sk e ≥0 (4.4)
k=1
This equation assumes for simplicity that the reactive mixture is immediately
consumed at x = 0. Referring back to Eq. (4.1), we also define a parameter α as
the proportional change of FAR oscillations divided by the proportional change in
velocity fluctuation, that is:

F A Rr ms u r ms
α= (4.5)
F AR ū
Parameter α is a useful figure of merit for evaluating different mixer designs. A

good premixer is then one in which α approaches zero, that is, velocity fluctuations
in the premixer do not result in any FAR oscillations at the exit. It is expected that
0 < α < 1, in general, and that α will be frequency dependent.
Note that the convective wavelength of these FAR oscillations is given by λc =
U/ f . When α ∼ 1 it means that all the airflow oscillation in the premixer is
translated into a corresponding FAR oscillation [see Eq. (4.1)]. This occurs at
low frequencies when the convective wavelength is long relative to the mixing
distance, that is, when f d/U 1, in which case the premixer behaves in a quasi-
steady manner. On the other hand, α progressively approaches zero as f d/U
increases. The progression toward zero is because the convective wavelength of
the FAR oscillations becomes very small as frequency increases (λc ∼ U/ f → 0)
and turbulent diffusion in the mixing duct will smooth out any FAR oscillation
before it reaches the premixer exit. Thus, at these high frequencies, the Richards–
Lieuwen mechanism is unlikely to be important.
Consider first the case of one injector, that is, only one point of fuel injection
(N = 1 fuel-injection points), and neglect turbulent-dispersion effects for the time
being. The key premixer “design parameters” in this case are U and d. These
parameters are key because the ratio of the two d/U is in fact the residence time in
the premixer passage—an important design parameter when it comes to ensuring
a good level of mixing and avoiding the problem of autoignition. Figure 4.10
shows the regions of pressure-wave amplification in the range of 0–1000 Hz that
can occur as the residence time in the premixer is varied. For very small residence
times (d/U < 1 ms) there is only one mode that can be excited by this mechanism.
As d/U is increased (d/U > 4 ms) there are up to 5 unstable frequencies in the
range of 0–1000 Hz.
Figure 4.10 thus shows the problem associated with a single-injector location;
it is very difficult to design a static injector position that results in stable operation
across a number of combustor modes, in particular, as the velocity U is varied.
Fig. 4.10 Regions of positive amplification of pressure waves as the parameter d/U
(premixer residence time) is varied.
Potentially in a combustor with only a single unstable mode could such a location be
identified (e.g., see Chap. 2); however, it is much more difficult in the Trent 60 DLE
in which one is dealing with three distinct modes (and frequencies). The only way
that such a configuration could be made sufficiently robust then, would be to posi-
tion the fuel-injection location far enough upstream that the FAR perturbation was
washed out by turbulence. To determine the level of turbulent dispersion of a FAR
disturbance with a delay-time representative of that of the Trent 60 DLE, measure-
ments were made in a representative configuration (configuration A in Fig. 4.9).
Typical raw data obtained for u (t) and F A R (t) are shown in Fig. 4.11. The dashed
horizontal lines in Fig. 4.11 correspond to the peak-to-peak amplitude of u which
would give α = 1.0, that is, the level of FAR fluctuation expected to occur from
the airflow fluctuation u , according to Eq. (4.1). In this case, α was approximately
0.95. As such, very little attenuation of FAR fluctuations occus in the premixer.
Configuration B can be seen as a logical attempt to deal with both problems
described here: time delay and turbulent dispersion. In principle, this approach
could eliminate a FAR wave of a given frequency by using a “wave-cancellation”
approach. The concept consists of adding a second fuel-injection point and dividing
the fuel flow equally between the two points. Such a system can be designed with
the aid of Eq. (4.4) to determine an appropriate d1 and d2 with a wider stable range.
A typical result that includes turbulent-dispersion effects, reproduced from Ref. 9
is shown in Fig. 4.12. The injector spacing is given by s = d2 − d1 . Although this
solution is more robust than the single-injector configuration, it still has problems.
First, it is still difficult to determine optimum injector location that stabilizes
all relevant combustor modes over all operating conditions. Second, the effect
of turbulent dispersion, which does not substantially alter the result shown in
Fig. 4.10 for N = 1, actually broadens amplification regions. Thus, the criterion
summarized in Eq. (4.4) is overly optimistic.
If such a configuration cannot be used for a suitably robust determination of
convective time delays to ensure stable combustion, what impact would it have on
80 T. SCARINCI
12
10
u (m/s), FAR (%)
6 hot wire signal (m/s)
alpha = 1
4
2
FID signal (% fuel)
0
0 10 20 30 40 50
time (ms)
Fig. 4.11 Results obtained for configuration A showing time dependence of air-
velocity oscillations in premixer and FAR oscillation at premixer exit.
the FAR-damping effectiveness parameter, α? As shown in Fig. 4.13 (reproduced

from Ref. 10), it rarely achieves less than α = 0.5 and only so at a few dis-
crete frequencies. Although α would go to zero at certain discrete frequencies in
the absence of mixing, turbulent-dispersion effects cause unexpected problems.
These problems occur because the fuel from the downstream fuel source does not
have time to mix as effectively as the fuel from the upstream source.
The logical way to widen the frequency range of configuration B is to split the
fuel between more injection locations. This solution is shown schematically as
configuration C in Fig. 4.9. In this configuration N locations of fuel injection are
Fig. 4.12 The regions of positive-pressure-wave amplification with two fuel sources
in the premixer in the presence of turbulent mixing (d/U = 4 ms).
1
Configuration A
0.8
0.6
alpha
0.4
Configuration B
Configuration D
0.2
0
0 1 2 3 4 5
fL/U
Fig. 4.13 Analytical comparison of frequency responses for configurations A, B, and

D (configuration B result obtained using d/U = 4 ms and s/d = 0.25).
spread equally over a distance L m . Clearly, FAR oscillations could be damped

over a much wider range of frequencies by using this approach. A fundamental
engineering problem with configuration C is the large number of fuel-injection
holes that will be required to achieve effective FAR-wave damping. One cannot
arbitrarily increase the number of fuel holes because, for the same overall flow
rate, the holes would have to diminish in size. There is a physical dimension below
which the diameter of a hole can no longer be reliably achieved in manufacture.
Also, holes below a certain critical size are vulnerable to blockage from particulate
matter in the fuel. Another practical limitation of configuration C is the previously
mentioned problem of reduced mixing time for the downstream injection points;
the closer the fueling point is to the exit of the mixer, the less effective is the mixing
before combustion. Configuration C, although theoretically a step forward, was
thus found to be unsatisfactory and was abandoned.
Configuration D is a substantially different design than configurations A–C. Here
a large number of air jets are equally spaced on the external walls of the premixer
and distributed over the same distance L m as in configuration C. All the fuel is
now added at the head of the mixer and air is progressively introduced to dilute
the fuel to finally achieve the same FAR as in the previous case. In a volumetric
sense, more air than fuel is flowing through the premixer, and so it makes sense
to put the air through a multitude of small holes and to supply the fuel from a
single source. Unlike configuration C, there is little practical (or manufacturing)
limit to the number of air injection holes with this concept. The concept essentially
injects air into fuel as opposed to the conventional approach, which is to inject fuel
into air.
The superiority of this approach over configurations A and B is illustrated in
Fig. 4.13. The effectiveness of this device was further confirmed from data obtained
82 T. SCARINCI
12
10
u (m/s), FAR (%)
6
hot wire signal (m/s)
alpha = 1
4
2
FID signal (% fuel)
0
0 10 20 30 40 50
time (ms)
Fig. 4.14 Results obtained for configuration D showing time dependence of air-
velocity oscillations in premixer and FAR oscillation at premixer exit.
confirmed in the lab, see Fig. 4.14. These data show that configuration D appeared
nearly free of any FAR oscillations at the exit of the premixer, that is, α < 0.1.
This sort of behavior, in which airflow oscillations become decoupled from the
FAR oscillations emanating from the mixing duct, is exactly what was desired
and this appeared (from Fig. 4.13) to have been achieved over a wide frequency
range.
The process by which the FAR and airflow oscillations are decoupled is not quite
as simple as what was portrayed in the case of progressively adding more points of
fuel injection. As will be seen, the fuel–air mixing process and its dynamics were
fundamentally changed with the new design.
First, consider the case in which the airflow in the premixer is oscillating, for
example, in response to a pressure oscillation in the combustion chamber. The
mixing arrangement may be thought of as a “phased array of air jets,” that is, there
are N rows of air-injection jets, not necessarily equally spaced, over a distance L m .
The upstream end of the mixing duct contains only pure fuel. At the location of the
first row of air jets, because the air jets are modulated, a local FAR oscillation is
generated. This FAR oscillation is convected to the next row of air jets. Depending
on the frequency of the oscillation, the air from the second row of air jets will
either weaken or sustain the FAR oscillation received from the first row of air
jets. This process is repeated again for each of the remaining N − 2 rows of air
jets. With numerous rows, there are numerous convective delays between any
two given rows, and hence a multiplicity of phasing relationships is generated as
the fuel makes its way toward the exit of the mixing duct. Thus, a progressive
smoothing of FAR oscillations occurs along the length of the premixer. As an
order-of-magnitude estimate one would expect that the FAR oscillations will be
successfully “integrated out” if a sufficient number of cycles are experienced in
the mixing process; that is, if the wavelength of the velocity fluctuation is smaller
than the distance L m over which the air jets are distributed:
f Lm
≥1 (4.6)
U
A more detailed analysis was made10 in which configuration D was treated

as a well-stirred mixer, that is, one resulting in an exponential distribution of
time delays. By taking the Fourier transform of the residence time distribution,
the transfer function magnitude H ( f ) (note the frequency dependence) can be
obtained. The transfer-function magnitude H ( f ) shows that an FAR perturbation
is damped by the amount
1
|H ( f )| =
2 (4.7)
1+ 4π 2 f Lm
U
Substituting values into this equation shows that when f L m /D = 1, the FAR
oscillations are 85% attenuated. When f L m /D = 2 the FAR oscillations are 92%
attenuated. These levels of FAR attenuation are close to an order of magnitude
more than anything that could be achieved by using passive turbulent diffusion
alone.
This mixing-duct technology was implemented and tested on the Trent 60 DLE
aeroderivative. For the new mixing ducts to work effectively, it is necessary to
ensure that the FAR-damping process becomes effective at frequencies below the
frequencies of the lowest acoustic mode of the combustor. For effective damping,
the FAR-wave cutoff frequency of the premixer must be lower than the natural
frequency of the lowest acoustic mode of the combustor, that is,
f cutoff (FAR wave) < f (acoustic mode) (4.8)
If the can combustor is (pessimistically) treated as a semiopen tube, in which

the lowest-standing wave mode is a quarter-wave, this implies that
U c
< (4.9)
Lm 4L c

Lm
c
>4 (4.10)
Lc U
Equation (4.10) relates the design parameters of the new mixing ducts to the overall
geometry of the combustion system.
Figure 4.15 shows a comparison between the pressure-oscillation levels with
the original Trent premixer (i.e., a combustor cross section as in Fig. 4.1 and
using a fuel–air concept analogous to configuration A of Fig. 4.9) and a mod-
ified premixer (i.e., one with the design implementation of configuration D of
Fig. 4.9). The pressure-oscillation measurements from the original Trent premixer
were obtained during stable-engine operation, that is, the fuel split between the
84 T. SCARINCI
100%
Noise Amplitude (arb. units) .
80%
Original Trent Premixers
60%
40%
Modified Trent Premixers
20%
0%
0% 20% 40% 60% 80% 100%
Power (% of full load)
Fig. 4.15 Pressure-fluctuation levels as a function of engine power comparing the

original1 and modified Trent premixers.10
primary, secondary, and tertiary premixers was chosen to avoid any serious in-
stability in the combustor. This line thus represents the “background” operating
pressure fluctuations of the original design, in the absence of any observable limit-
cycle combustion oscillation. The pressure-fluctuation levels from the new Trent
60 DLE, which implements the style of the premixer as in configuration D of
Fig. 4.9, are considerably lower even though the fuel split was not purposefully
chosen. If the acoustic properties of the two combustors are relatively similar, then
the gap between the two lines can be seen as a measure of the stability offered by
the new premixers. The new premixers thus offer lower-pressure-fluctuation levels
and improved stability margin.
A more important characteristic of the pressure-fluctuation levels is their ro-
bustness against fuel-split variations. Even though one could design a nominal or
optimized fuel schedule, the engine in the field is exposed to many sources of un-
certainty (e.g., sensor- and fuel-valve accuracy, gas-composition variations, engine
deterioration over time, manufacturing tolerances, etc.) that cause the nominal (or
desired) fuel schedule to never be exactly achieved. Thus, there will be natural
fuel-split (or flame-temperature) variations on the engine. To explore the margins
against combustion instability, a large variety of fuel splits at various engine-power
levels were explored. In the industrial Trent, being a three-stage, premixed com-
bustor, it is possible to vary the flame temperature (i.e., the FAR) of two of the
three premixers independently while keeping the engine at a fixed-power condition.
In Fig. 4.16, the results of base-load, pressure-fluctuation amplitude-mapping ex-
periments with the new premixers are compared with the behavior previously
observed with the original premixer. In both cases the engine is operating at
steady-state base load, and the fuel splits to the primary and secondary pre-
mixers are explored systematically. As before, fuel splits are expressed in terms
of a flame-temperature bias from the nominal (or reference) flame temperature.
10.00
RMS Noise Amplitude (% of Pref) .
Original Trent premixers
1.00
- 50K
REF
+ 60K
+ 75K- REF 75K
0.10
Modified Trent premixers
0.01
-150 -100 -50 0 50 100 150
secondary temperature bias, K
Fig. 4.16 Comparison of the original1 and modified Trent premixers,10 the showing
influence of fuel splits on pressure-fluctuation amplitude, as obtained from a mapping
procedure at engine base load.
Again, Fig. 4.16 clearly shows that the overall pressure-fluctuation levels from
the new premixers (that have FAR-wave damping) is roughly one order of magni-
tude less than those from the original combustor. Also note that the dependence of
pressure-fluctuation amplitude on fuel splits is significantly reduced. Thus, it seems
reasonable to assume that the presence of FAR-wave damping has, in the Trent
combustor, fundamentally addressed what might have been the dominant mecha-
nism of the instability, that is, the Richards–Lieuwen mechanism.
Having established a satisfactory configuration that was validated on develop-
ment engines, the new mixing-duct concept was tested on a real power-generating
site. Over a field trial of 3500 h, the engine showed satisfactory compliance at all
times against the 25-vppm legislation requirement for the site, while keeping the
same low-pressure-fluctuation levels as seen on development engines. Details of
the field-trial results are in Ref. 10.
3. Acoustic Resonators
Because the topic of resonators is covered in detail in several other chapters (e.g.,
see Chaps. 13 or 17), our discussion of their implementation in the Trent engine will
be brief. A suitably designed and well-located pressure-wave attenuator clearly can
significantly reduce the strength of feedback coupling for most if not all feedback
mechanisms that can lead to combustion instabilities. Despite the great care taken
to remove FAR oscillations from the premixers, we nonetheless included several
pressure-wave attenuators as part of the combustor structure. We included these
attenuators as a potential protection against instability mechanisms other than the
Richards–Lieuwen mechanism (see Fig. 4.8).
86 T. SCARINCI
Pressure Wave
Damping
FAR-Wave Damping
Aerodynamic
Damping
Fig. 4.17 Cross section of the modified two-stage Trent 60 DLE combustion system.
The development of a pressure-wave attenuator requires some knowledge about

the acoustic mode being excited. The frequency of the mode affects the sizing of
the attenuator (larger dimensions give lower frequencies), whereas the mode shape
determines the optimal location of the attenuator. The mode shapes and frequen-
cies of the Trent 60 DLE combustor are fairly easy to predict because the geometry
can be reasonably approximated as that of a cylindrical cavity, whose diameter is
roughly 20% of its length. The resulting donut-shaped resonators were developed
and optimized by using a simple acoustic-speaker rig before being validated on
combustion rigs and finally on development engines. Care was taken to always im-
merse resonators within the combustor case plenum. This position ensures that the
air temperature within the resonators is unaffected by ambient conditions around
the engine. Furthermore, all resonators are directly connected to the flame tube,
that is, they are located as close as possible to the combustion zone itself. A signif-
icant number of frequencies are covered by means of several resonating cavities,
all naturally integrated within the combustor architecture. These resonators can be
seen in Fig. 4.17, which will be further discussed subsequently.
4. Aerodynamic Damping Devices
Referring once more to Fig. 4.8, note that one of the mechanisms that can lead to
combustor pressure oscillations is an upstream-velocity fluctuation convected from
the plenum, through the premixer, and then the combustion zone. Potential sources
of plenum-velocity fluctuations are the turbulent-flow structure surrounding the
combustor. Large-scale eddies convected to the premixers can cause significant
velocity fluctuations, which disturb the combustion process.
Although it is difficult to assess the importance of this instability mechanism rel-
ative to other mechanisms, a preventive approach was taken in which the objective
was to damp out these pressure fluctuations as much as possible within the combus-
tion system. Several airflow-damping devices were tested and their configuration
was optimized. These devices are located upstream of the premixers (see Fig. 4.17).
Some of the configurations tested included a number of reticulated materials,
such as honeycomb and metal foams, all of which were tested for various thick-
ness, porosity, and pressure-loss characteristics. The device reduces aerodynamic
plenum fluctuations by more than 20 dB over a wide-frequency range.11 Thus, the
premixers and, hence, the flame experience a reduced level of velocity fluctuation,
resulting in a quieter combustion process. Note that the pressure loss introduced
by the damping device is typically less than 0.2%.
III. Combustion System Design Modifications

The understanding gained from the control approaches discussed in Secs. II.B.2,
II.B.3, and II.B.4 resulted in design rules that could be implemented for passive
control of combustion oscillations of the original Trent 60 DLE combustor shown
in Fig. 4.1. One of the most important consequences of being able to design a
system with inherent damping of combustion oscillations over a wide-frequency
range is that it is no longer necessary to control pressure oscillations via a fuel-
staging approach. This system has two important benefits. First, the fuel-split
flexibility offered by having a third stage is no longer required. In other words, the
control approach explained in Sec. II.B.1 is not deemed necessary, which means
that the third stage could be eliminated, resulting in a much simpler, cheaper
system. Second, fuel splits can be optimized for the sole purpose of achieving low
emissions without instability regions as a constraint on the choice of the optimum
fuel splits.
A two-stage Trent 60 DLE combustor incorporating the control approaches
of FAR-wave damping, pressure-wave damping, and aerodynamic damping, as
discussed earlier, is shown in Fig. 4.17. This demonstration might give the reader
an appreciation of how the basic principles described in preceding sections were
incorporated into a practical design.
Acknowledgments
The investigative work on combustion instability, or combustion oscillations,
summarized here, and the physical and practical understanding that came with
it as a result, was carried over a number of years and by quite a large number
of people. Nonetheless, a few individuals must be singled out. Chris Freeman,
now retired from Rolls-Royce (Derby, United Kingdom) and Ivor Day, from the
Whittle Laboratory of Cambridge University, have been long-standing soul mates
on our quest to find passive design solutions to combustion instability in DLE
systems. Chris was key in formulating the right penetrating questions at times
when only confusion seemed to be the correlating parameter. Ivor, through his
heart-of-the-matter $1 experiments, has a unique ability to redefine what common
sense actually means. I have learned much from these two close friends and this
chapter is really dedicated to the fun we had throughout this project.
My colleagues, past and present, mainly from Montreal, but also Derby, Ansty,
and Indianapolis have often made it possible to translate research ideas into real
88 T. SCARINCI
engine hardware and results and have been a constant source of support. It’s been
a privilege working with them.
Some key senior members of Rolls-Royce Engineering, Chris Barkey, Vic
Szewczyk, Mike Howse, and Phil Ruffles (now retired) have been crucial sup-
porters of the ideas put forward in this chapter. They all offered and created an
appropriate environment for technical innovation to happen.
References
1
Scarinci, T., and Halpin, J. H., “Industrial Trent Combustor–Combustion Noise Char-
acteristics,” Journal of Engineering for Gas Turbines and Power, Vol. 122, No. 2, 2000,
pp. 280–286.
2
Richards, G., and Janus, M. C., “Characterization of Oscillations During Premix Gas
Turbine Combustion,” Journal of Engineering for Gas Turbines and Power, Vol. 120, No. 2,
1998, pp. 294–302.
3
Lieuwen, T., and Zinn, B. T., “A Mechanism for Combustion Instabilities in Premixed
Gas Turbine Engines,” Journal of Engineering for Gas Turbines and Power, Vol. 242, No. 5,
2001, pp. 893–905.
4
Lieuwen, T., and Zinn, B. T., “The Role of Equivalence Ratio Oscillations in Driv-
ing Combustion Instabilities in Low NOx Gas Turbines,” Proceedings of the Combustion
Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1809–1816.
5
Poinsot, T., Trouvé, A., Veynante, D., Candel, S., and Esposito, E., “Vortex Driven
Acoustically Coupled Combustion Instabilities,” Journal of Fluid Mechanics, Vol. 177,
1987, pp. 265–292.
6
McManus, K. R., Poinsot, T., and Candel, S. M., “A Review of Active Control of
Combustion Instabilities,” Progress in Energy and Combustion Science, Vol. 19, No. 1,
1993, pp. 1–30.
7
Manneville, P., “Dissipative Structures and Weak Turbulence,” Perspective in Physics,
edited by H. Araki, A. Libchaber, and G. Parisi, Academic Press, San Diego, 1990, Chap. 1.
8
Cheng, W. K., Summers, T., and Collings, N., “The Fast-Response Flame Ionization
Detector,” Progress in Energy Combustion Science, Vol. 24, 1998, pp. 89–124.
9
Scarinci, T., and Freeman, C., “The Propagation of a Fuel-Air Ratio Disturbance in a
Simple Premixer and its Influence on Pressure Wave Amplification,” American Society of
Mechanical Engineers, Paper 2000-GT-0106, May 2000.
10
Scarinci, T., Freeman, C., and Day, I., “Passive Control of Combustion Instability in a
Low Emissions Aeroderivative Gas Turbine,” American Society of Mechanical Engineers,
Paper 2004-53767, June 2004.
11
Scarinci, T., and Barkey, C., “Dry Low Emissions Technology for the Trent 50 Gas
Turbine,” Proceedings of PowerGen Europe, Pennwell, U.K. 2004.
Chapter 5
Thermoacoustic Design Tools and Passive Control:

Siemens Power Generation Approaches
Werner Krebs,∗ Sven Bethke,∗ Joachim Lepers,∗ Patrick Flohr,∗

and Bernd Prade∗
and
Cliff Johnson∗ and Stan Sattinger∗
Siemens AG, Orlando, Florida
I. Introduction
T HIS chapter provides an overview of the design challenges and methods

to mitigate combustion instabilities in industrial gas-turbine engines. The
application of several design tools for improving combustion stability is shown in
examples.
II. Siemens Gas-Turbine Products
Siemens Powergeneration offers a complete product line of gas turbines ranging
from 4 MW to 278 MW. Details of the gas-turbine products can be found at
www.powergeneration.siemens.com. Table 5.1 contains the performance data of
gas turbines mentioned in this chapter.
The gross power output and the gross efficiency at International Organization
for Standardization (ISO) conditions are given for a single-cycle operation. These
values are taken at the generator terminals. For combined-cycle plants the net
power output and the net efficiency are listed for a single-shaft arrangement. In a
single-shaft, combined-cycle power plant one gas turbine, one generator, and one
steam turbine are arranged along one shaft. The net values are taken at the terminals
of the combined-cycle power plant. The SGT5-4000F is shown in Fig. 5.1. It is
operated in the 50-Hz range. The engine is fired by 24 hybrid burners mounted on
an annular combustion chamber achieving a gross power output of 278 MW. The
SGT-1000F is a scaled version of the SGT5-4000F rotating at 5400 rpm. It can be
operated in the 50- and 60-Hz range by application of a gear box. The SGT6-6000G
∗ Power Generation.
89
90 W. KREBS ET AL.
Table 5.1 Performance data for selected Siemens gas turbines for
large-scale applications
Gas-turbine frame
Values SGT-1000F SGT5-4000F SGT6-5000F SGT6-6000G
General/boundary conditions
Grid frequency 50/60 50 60 60
Rotor speed, rpm 5400 3000 3600 3600
Performance data, single cycle
Gross power output, MW 68 278 198 266
Gross efficiency, % 35.1 39.0 38.0 39.3
Performance data, combined cycle-single shaft
Net power output, MW 101 407 293 391
Net efficiency, % 52.6 57.7 57.0 58.4
with a gross power output of 266 MW is the largest engine in the 60-Hz range. The
cross section of the engine is shown in Fig. 5.2. It is fired by 16 Can-type combus-
tion systems with steam-cooled transitions. The SGT6-5000F is also fired by 16
can-type combustors that are air cooled with a gross power output of 198 MW. All
combustors operate in premix mode at base load to provide low NOx emissions.
III. Phenomenological Description
Combustion-driven oscillations or thermoacoustically induced oscillations (also
called combustion dynamics) are characterized by a feedback cycle that converts
chemical energy to acoustic energy at a rate of about 10−4 . Typical for thermo-
acoustic oscillations is the observation of pronounced peaks at the resonance fre-
quencies of the combustion system.
Fig. 5.1 Cross section of the SGT5-4000F.

THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 91
Fig. 5.2 Cross section of the SGT6-6000G.
Three frequency ranges can be defined to describe combustion instabilities:

1) Low-frequency dynamics (LFD) occur at frequencies <50 Hz. These dynamics
are frequently referred to as “breathing” modes, “bulk” modes, or “Helmholtz”
modes. 2) Intermediate-frequency dynamics (IFD) occur at frequencies between
50 Hz and 1000 Hz. In general, the first natural mode in an industrial gas-turbine
combustor is between 50 and 300 Hz, depending on the geometry and firing tem-
perature. Research efforts in the area of thermoacoustics address the feedback
cycles of this kind of instability. Descriptions of the feedback cycles are given in
Chapters 1 and 14. High-frequency dynamics (HFD) occur at frequencies greater
than 1000 Hz. In general, HFD refers to three-dimensional acoustic modes. To
date, research efforts have not revealed the feedback cycles exciting this type of
combustion instability. Note that the three frequency ranges defined here are used
for convenience in describing various dynamics that may be observed; they should
not be interpreted as rigid boundaries of physical behavior.
Figure 5.3 shows a dynamic-pressure spectrum measured in a single-burner
high-pressure test rig driven to an “unstable” operating condition. Note that, al-
though it is not technically correct, the term unstable is frequently used in the field
to describe dynamic-pressure oscillations that reach their limit cycle or exceed al-
lowable dynamic limits. Figure 5.3 illustrates the three frequency ranges described
earlier. In this case, dynamics are excited in the LFD range of 0–50 Hz, the IFD
range of 50–1000 Hz, and the HFD range of >1000 Hz. The mechanisms that
drive LFD, IFD, and HFD are different, although all involve coupling between the
heat release of the combustion process and the acoustic-pressure field.
Note that different frequencies respond to changes in operating conditions dif-
ferently. Whereas one IFD mode may be damped by an increase in the fuel–gas
fraction in one of the fuel stages, another mode’s amplitude may increase because
92 W. KREBS ET AL.
1.0
HFD
LFD
IFD
0.9
0.8
0.7
Amplitude
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0 500 1000 1500 2000 2500 3000 3500
Frequency (Hz)
Fig. 5.3 Dynamic-pressure spectrum measured in a high-pressure single-burner rig

driven to unstable conditions.
of the same action. Changes in combustor design also have similar tradeoffs for
optimizing the dynamic response of the combustion system.
Thermoacoustically induced pressure pulsations cause liner vibrations that can
only be tolerated up to a certain level. Maximum permissible pressures will be
unique to each design and typically will be a function of frequency. Excessive heat
transfer to surfaces can produce softening or weakening and, when combined with
excessive pressure oscillations, can yield disastrous consequences.
The combustion-driven oscillation feedback cycle, in general, describes all phe-
nomena related to thermoacoustic-stability analysis. The feedback cycle is compli-
cated by the interaction of the influence parameters shown in Fig. 5.4. The acoustic
waves under concern have long wavelengths compared with the dimensions of the
combustion system and they expand over several gas-turbine components, includ-
ing the compressor, burner plenum, and turbine. Therefore, all these components
play a pivotal role in combustion-dynamics analysis. The dimensions and acoustic
Air
Combustor
Plenum
Fig. 5.4 Thermoacoustically relevant influence parameters of the feedback cycle.

properties of the burner plenum have an impact on the impedance of the burner
exit, thus affecting the magnitude of the pressure-induced fluctuations in volume
flow. The acoustic properties of the fuel lines determine the magnitude of fuel flow
fluctuations at the fuel nozzles. The interaction between the fuel flow and airflow
results in equivalence-ratio oscillations, which in turn produce oscillations of the
heat release. The acoustic geometry of the combustor determines the shape of the
acoustic modes, which are also influenced by the acoustic-boundary conditions
at the combustor exit. The source of the instability is the perturbed flame. In that
context it is important to understand and quantify the heat-release fluctuations
induced by dynamic-pressure perturbations.
The technical objectives of thermoacoustic design are 1) Determine the nature
of the thermoacoustic-feedback cycle and investigate the interaction of different
components. 2) Optimize the thermoacoustic-design process: Identify thermoa-
coustically relevant design parameters and evaluate the thermoacoustic impact of
design changes. 3) Optimize the prediction capability of test rigs and develop com-
putational models to predict engine performance based on rig results. For acoustic
energy balance, note that the acoustic energy generated by the flame is mainly lost
at the inlets and outlets of the combustion system under consideration. Because
the test rigs differ especially at their inlets and outlets from the engine design,
the knowledge of these losses is of crucial importance for successful gas-turbine
combustion design. 4) Develop active and passive means for the suppression of
thermoacoustic oscillations.
IV. Solution Methods

During the past decade, numerous analytical and numerical solution methods
have been developed at Siemens to investigate the thermoacoustic properties of
gas-turbine combustion systems. The methods have been applied to achieve the
technical objectives 1–4 that were listed in the preceding section. The methods are
listed in Table 5.2. Methods have been developed for analyzing individual compo-
nents (e.g., fuel lines) and for analyzing the interaction of all relevant components
in a full-stability analysis method (Fig. 5.4).
A. One-Dimensional Acoustic Analysis

In most components and especially for single-burner test rigs, a one-dimensional
pressure field can be assumed because the frequency range under consideration is
well below the cutoff frequency of multidimensional modes. For special purposes
such as the design of resonators, transmission line models1 have been developed,
in which the acoustic properties of serially connected elements are considered. For
systems in which branching of acoustic passages has to be considered, a transfer-
matrix network can be applied (Fig. 5.5). The method has been described in several
textbooks (e.g., Ref. 2).
The sketch of a hybrid burner used in the SGT-1000F is shown in Fig. 5.5. It
consists of two concentrically arranged air passages.3 Through the central axial
swirler passage about 10% of the total airflow is discharged. Ninety percent of
the air flows through the diagonal swirler passage. A branching element is used
to represent this air split. In both passages elements are included to treat the fuel
injection. In simple models, the fuel orifices are represented by the fuel-nozzle
impedance; in more elaborate models, whole fuel lines are explicitly included.
94 W. KREBS ET AL.
Table 5.2 Solution methods applied at Siemens AG for the design of

gas-turbine combustion systems
Type Method Function
Prediction methods
Acoustic properties Transmission line method Resonator design
of combustor parts
One-dimensional Transfer matrix network Evaluate acoustic properties
acoustic analysis like impedances for
gas-turbine components
Three-dimensional Finite element methods, Evaluate acoustic-pressure
acoustic analysis Sysnoise distributions for
a) monitoring
b) developing resonators etc.
Flame response Time-lag models Evaluate impact of design
changes on flame response
Flame response Unsteady computational Evaluate impact of design
fluid dynamics changes on Flame Response
Full-stability analysis Transfer-matrix approach Evaluate impact of design
Galerkin method changes on stability
Interaction of components
Component Testing
Tunable rig with Evaluate impact of design
variable-exhaust- changes on stability
passage impedance Interaction of components
The main advantage of the method is that the acoustic properties of each acousti-
cally relevant component (e.g., a duct) are represented by a separate transfer matrix.
The acoustic properties of the complete system are obtained by connecting these
transfer matrices in a transfer-matrix network. Applications of the transfer-matrix
approach to gas-turbine combustion systems have been described by Krüger et al.4
The transfer-matrix method is successfully applied to identify the properties of cer-
tain components of the gas-turbine combustion system, like fuel lines or exhaust
passages of test rigs, for which the one-dimensional sound propagation is valid.
B. Three-Dimensional Acoustic Analysis

In a combustor and combustor plenum, in general, a three-dimensional acous-
tic field is encountered. This makes the development and application of three-
dimensional codes necessary to investigate the acoustic-pressure distribution. At
Siemens AG three-dimensional finite element codes are applied that solve the three-
dimensional acoustic equations in the presence of nonviscous flow. The governing
equations are described by Bethke et al.5
Figure 5.6 gives an overview on the main tasks of the three-dimensional finite
element acoustic analysis. The acoustic mode shapes resulting from the finite
element analysis are the source for three-dimensional thermoacoustic-stability
analyses. In Fig. 6 the mode shapes are visualized by the distribution of the modulus
of the dynamic pressure.
Fig. 5.5 Representation of a gas-turbine burner by a network of acoustic-transfer

elements representing burner inlet, axial and diagonal burner passages, fuel-injection
locations, and the flame.
Determine Eigenfrequencies and related mode- Estimate Eigenfrequencies and

shapes of acoustic pressure and velocity at low related mode-shapes of acoustic
and intermediate frequencies. pressure and velocity at high
frequencies.
• Input for stability analysis
• Optimized positions of HFD-
• Determination of impedances (e.g. at burner exit) resonators
• Interaction of combustion chamber and plenum • Evaluation of HFD-resonators
• Excitations for combustor mechanical design
• Evaluation of IFD-resonators
Fig. 5.6 Finite element acoustic models for gas-turbine combustor designs.
96 W. KREBS ET AL.
Antinodes of dynamic pressure are located in the medium-grey regions, whereas

nodes of dynamic pressure are represented in the dark-grey regions. The left-hand
side of Fig. 5.6 shows an azimuthal mode shape inside an annular combustor of first
order with two pressure antinodes and two pressure nodes. Such a mode shape be-
longs to frequencies in the range of 100 Hz. The thermoacoustic instability induced
inside the combustor also generates dynamic-pressure oscillations in the plenum,
upstream of the burners. The azimuthal component also dominates the pressure
wave in the plenum as shown in the center. Hence, direct relations exist between
the pressure oscillations inside the combustor and the plenum that are given by the
corresponding transfer matrix of the burner. For continuous online monitoring, this
transfer relation has been utilized by installing monitoring devices in the plenum
instead of using direct measurements of the dynamic pressure inside the combus-
tor. The rightmost figure shows a high-frequency dynamic mode in a can-annular
combustor featuring a combined axial and azimuthal mode, corresponding to a
frequency greater than 2000 Hz.
The distribution of the acoustic pressure on the surface of the combustion cham-
ber is important information for the optimum arrangement of resonators on the
combustor shell. Furthermore, mode shapes of acoustic pressure are needed to
analyze structural vibrations and to determine the life of the combustor shell. As
pointed out before, the acoustic analysis of the full combustion system requires a
large computational domain starting from the compressor outlet and ending at the
turbine inlet. To manage the computational effort, a flexible approach has been de-
veloped that is outlined here. Three different types of configurations with varying
complexity have been selected.
1) Simplified models of only the combustion chamber that include only one
burner (Fig. 5.7, right-hand side). Impedance-boundary conditions are set on the
boundary faces.
2) The impedance boundary conditions are important for the final stability and
the acoustic-pressure level of the system because the impedance at the acoustic
boundary face determines the acoustic energy loss over that surface.7 Hence, to
Can-combustion-chamber
Transition-piece
Simulated ambience Plenum
Burner
Vane-simulation-section
Diffuser-inlet
Exhaust passage
Fig. 5.7 Typical computational domain for finite element acoustic analyses ranging
from diffuser inlet–compressor exit to exhaust.
improve the accuracy of the prediction the computational domain can be extended
as shown in Fig. 5.7 (left-hand side). The computational domain now includes the
incoming flow path upstream of the burner (plenum), the combustion chamber,
turbine vanes or the vane-simulation section, and the exhaust passage.
3) The most complex (and most costly) model covers the whole annular or
can-annular combustion system, including all the burners in the entire engine.
Because the acoustic environment of a test rig and an engine differ, the three-
dimensional acoustic analysis using finite element codes is essential for the evalu-
ation of test-rig results. In addition the codes are used to point out the differences
in acoustic properties between engine and rig.
C. Flame Response
The analysis of the flame response is crucial for the evaluation of the thermoa-
coustic stability of gas-turbine combustion systems because, as shown by Poinsot
and Veynante,7 it determines the source term in the transport equation for acoustic
energy. In general, the flame-response function expresses the heat-release fluctua-
tions induced by acoustic-pressure waves. The instantaneous response of a flame
caused by an acoustic perturbation is given by the pressure-coupled and velocity-
coupled response functions given in Eqs. (5.1) and (5.2), respectively:
q (t)/q̄
F1 (t) = (5.1)
p (t)/ p̄
q (t)/q̄
F2 (t) = (5.2)
u (t)| burner exit face /ū burner exit face
q describes the integral of heat-release fluctuation over the flame surface, defined
as

q (t) = Q (r , t) dV [W ]
heat release zone
p and u denote the acoustic-pressure fluctuation and the acoustic-velocity fluctu-

ation, respectively. Both flame-response functions have been nondimensionalized
by respective mean quantities. F1 expresses the instantaneous relation between
heat-release fluctuation and pressure fluctuations and is valid for all types of flame–
acoustic interactions. F2 expresses the instantaneous relation between heat-release
fluctuations and the acoustic-velocity fluctuation at the burner or nozzle exit.
Most measurements of the flame-response function refer to this function type F2
(e.g., Büchner et al.8 ).
To get more insight into the impact of design changes, the flame response has
to be related to aerodynamic design parameters. Quite a lot of work has been
devoted to the development of models addressing the flame–acoustic interaction.
A comprehensive overview is given by Cho and Lieuwen9 in Chapters 9 and 14.
98 W. KREBS ET AL.
Technical aspects and the impact of design parameter on the flame response are
discussed by Krebs et al.10
In conclusion, the dynamic properties of the flame are mainly represented by
the time lag and different combustion designs can be compared by looking at their
time-lag distribution. The time-lag distribution can be obtained in a postprocessing
step to steady computational fluid dynamics (CFD) analysis.
The approaches based on a steady-state analysis have a main drawback; that
is, the impact of pressure waves on the strain rate and the impact of the flame–
vortex interaction cannot be represented. In addition these approaches assume that
the amplitude of the fluctuation once generated does not change its value while
approaching the flame front. Finally, the steady-state approach assumes that the
flame is stationary; in some cases, the mean flame position shifts when combustion
instability occurs. More insight into the detailed processes can be obtained by
unsteady CFD methods like large-eddy simulation (LES), in which the large-scale
vortices are resolved. Siemens is currently developing unsteady methods to treat
this issue.
D. Full-Stability Analysis
The goal of a full-stability analysis is to predict the excitation or damping
of an acoustic mode. This process involves the representation of the acoustic-
feedback cycle between the flame response and the acoustic environment. The
thermoacoustic stability is the result of the thermoacoustic flame response and
the acoustic properties of all components between compressor and turbine exit. In
contrast to CFD methods the computational domain must be extended to encompass
all components. A solution method that has been extended to meet this requirement
is the transfer-matrix network. This method is explained subsequently, with the
analysis of the annular combustor rig as an example. The cross section of the
annular combustor rig is shown in Fig. 5.8.
Fig. 5.8 Geometry of the annular combustor rig.

Fig. 5.9 Acoustic transfer-matrix network representing the annular test rig.
The airflow enters at the test rig inlet (left), flows through the compressor outlet
diffuser, and is discharged through 24 burners. Combustion takes place in an an-
nular combustor. The hot gases are finally discharged through the exhaust-diffuser
passage. The annular combustor is operated at ambient conditions, and hence
acoustically soft conditions (i.e., low impedance) have been assumed at the rig
inlet and outlet.
The acoustic network representing the test rig is shown in Fig. 5.9. In this ex-
ample the annular combustor rig is represented by two interconnected rings of
one-dimensional duct elements. The outer ring represents the plenum, whereas
the inner ring represents the annular combustor. The length of the duct elements
representing the annulus are selected to cover the average acoustic-passage length.
The average acoustic-passage length is determined as the circumference having
the same azimuthal eigenfrequencies as the annulus of finite width. The eigenfre-
quencies of the combustor are obtained by three-dimensional finite element modal
analysis. This approach is valid because in the annular combustion systems of
the Siemens product family (SGT-1000F, SGT6-4000F, and SGT5-4000F) pure
azimuthal mode shapes are predominant. The elements shown are submodels con-
sisting of further elements representing the actual geometry as shown in Fig. 5.5.
Altogether the whole system may contain more than 1000 different transfer-
matrix elements. The transfer-matrix representation is more-or-less a mathematical
framework in which all the information on acoustic properties of the different
components can be lumped together. The quality of the model depends on the
quality of these elements. In addition to models of annular ducts, Siemens AG has
specially developed models for diffusers and nozzles capable of generating the
100 W. KREBS ET AL.
Table 5.3 Complex eigenfrequencies found for the annular combustor rig
No. Oscillatory frequency, Hz Damping coefficient, 1/s Comments
1 93.4 10.2 Stable

2 169 6.4 Stable
3 196 −3.1 Unstable
4 214 0.1 At instability limit
transfer matrix, even for high-Mach-number flows that may occur in fuel-supply
systems. The network formulation results in a matrix equation described by, for
example, Hubbard and Dowling,11 which can be solved to obtain the complex
eigenfrequencies (ωn = ω f + i ∗ α) of the system.
To determine the stability of the system, the determinant of the transfer ma-
trix is calculated by using the appropriate boundary conditions. According to the
decomposition defined earlier, the amplitude of a pressure oscillation grows if
the imaginary part of the complex eigenfrequency becomes negative. This part
is called the damping coefficient. The complex eigenfrequencies found for the
annular combustor rig operated at nominal conditions are listed in Table 5.3.
Table 5.3 indicates an unstable eigenmode at 196 Hz, which is in agreement with
experiments in which a single unstable mode at about 200 Hz has been found.
V. Application
Because thermoacoustic stability results from the interaction of several different
components, numerous design options are available to increase the thermoacoustic-
stability range. As described in the preceding section, solution methods and design
tools are needed to identify promising design modifications and to investigate
their impact quantitatively. The goal of each of these options is to extend the
operating range of the engine to improve performance or emissions and to extend
the operating life of the engine components.
Several design options are listed in Table 5.4. Most design options considered
involve changes of the burner design and they primarily affect the flame response.
One approach is to reduce the interaction of the thermoacoustic source with the
pressure field by changing the time-lag distribution. Another option is the ap-
plication of different types of resonators to increase the damping of the system
by absorbing acoustic waves amplified in a certain frequency range. Their effi-
ciency greatly depends on the width of the frequency range at which damping
is added. A third approach is to make use of control methods. Active instability
control (AIC) systems achieve stability by perturbing the combustion process at or
near the frequency of the combustion instability (normally 100–200 Hz) to damp
the cycle-to-cycle pressure oscillations that occur on the order of milliseconds.
Hoffmann and Hermann (Chap. 19) describe such a system that was used in
Siemens products SGT6-4000F and SGT5-4000F. Another control strategy is to
use low-bandwidth control (<2 Hz; e.g., industrial fuel-control valves) to make
adjustments to the mean operating conditions (airflow, fuel flow, and fuel distri-
bution). Such methods are called active combustion control or automatic tuning
systems. Both the high-bandwidth and low-bandwidth control approaches require
reliable real-time measurement of the combustion-process oscillations; dynamic
Table 5.4 Design options for extending the safe operation range of
gas-turbine combustion systems
No. Design option Impact on
Burner
1 Change flame-front location by modifiying Flame response
burner exit geomentry (cylindrical burner, outlet)
2 Changing of fuel-concentration profiles Flame response
3 Fuel staging technology Flame response
4 Adaptive fuel-nozzle impedance Flame response
5 Modify heat-release distribution Flame response
Combustor
6 Resonators Damping
Control
7 Active instability control Flame response
8 Active combustion control Flame response
pressure transducers or accelerometers are typically used in industrial applica-

tions. Whether control action is needed for the engine has to be determined on a
case-by-case basis.
A. Mode-Shape Analysis of Annular and Can-Annular

Combustion Systems
As outlined in Sec. IV the three-dimensional analysis of the acoustic-pressure
distribution is essential for the understanding of the acoustic phenomenon in
combustion systems and for the design of passive means to extend the operation
envelope.
In Fig. 5.10 the acoustic-pressure distribution of an annular combustion system
is shown. The modulus of the dynamic pressure is coded in gray scale. Medium
Engine:
24-burner
configuration
Fig. 5.10 Modulus of the dynamic-pressure distribution: first azimuthal mode shape
in annular combustion chamber.
102 W. KREBS ET AL.
Acoustic excitation
at burners
in-phase alternating out-of-phase
Phase ‘Can-Can
- Phase Interaction’!
+
Fig. 5.11 Axial mode shape in can-annular combustion chamber.
grey indicates a pressure antinode, whereas dark grey indicates a pressure node. In
the intermediate range of frequencies, azimuthal mode shapes are predominant in
annular combustion systems of large-scale engines like Siemens products SGT6-
4000F and SGT5-4000F. Azimuthal mode shapes are characterized by a dominant
dependence of the dynamic-pressure amplitude on the azimuthal coordinate as
visualized in Fig. 5.10. The presence of two pressure antinodes and two pressure
nodes indicates a first azimuthal mode corresponding to a frequency of about
100 Hz. The thermoacoustics of can-annular combustion systems is driven by
axially oriented mode shapes as shown in Fig. 5.11. In the engine, basically two
types of mode shapes are possible. On the left-hand side the dynamic pressures of
all single can-combustion chambers are acoustically in phase. On the right-hand
side the pressures in neighboring cans are acoustically out of phase. In this case
the different cans are acoustically connected by the passage in front of the turbine.
The impact of the can–can interactions through the annular manifold upstream
of the turbine inlet on the acoustical behavior of can-annular combustion systems is
shown in Fig. 5.12. In addition to the acoustic modes that are present in a single-can
configuration (compare, solid line), there are further modes in between (compare,
dotted line). These modes are characterized by relatively high-acoustic velocities
in the annular manifold between the can-combustion chambers. To include all
phenomena during the testing phase the exhaust passages of the test rigs have to be
modified accordingly to get the desired mode shape. In addition the analysis reveals
that acoustic phenomena in can-annular combustion systems are distributed over
neighboring cans rather than limited to a single can.
B. Flame Response
A design method that makes use of changing the flame response has been suc-
cessfully applied in the V-Frame. This design approach is sketched in Fig. 5.13.
Mounting cylinders of different lengths on the burner exit results in an elongation
of the flame front.
Fig. 5.12 Can–can interaction in can-annular combustion chambers.
The temperature distributions for the different design configurations are shown
in Fig. 5.14. The zero-velocity contours are also plotted to indicate the size of the
recirculation zones. The burner installed in the SGT-1000F is the so-called hybrid
burner shown in Fig. 5.5. Swirl induces a central-recirculation zone that provides
flame stabilization over the whole gas-turbine operation envelope. In addition,
outer-recirculation zones are formed that also help to stabilize the flame. As typical
Modifications
NBO
CBO short
CBO long
Fig. 5.13 Modification of the burner outlet by cylinders of different lengths (NBO =
normal burner outlet; CBO = cylindrical burner outlet).
104 W. KREBS ET AL.
NBO CBO-Short
CBO-Long
Fig. 5.14 Reacting flowfield for the different burner-exit configurations.
for swirl-stabilized burners, the location of heat release is mainly aerodynamically

determined by the size of the recirculation zones. This size can be modified by the
application of cylindrical burner outlets (CBOs), and, hence, the location of the
flame front, characterized by the steep temperature gradients, is shifted.
The change of the flowfield affects the time-lag distribution, which is shown
in Fig. 5.15. In Fig. 5.15 the time-lag distribution has been normalized by the
residence time of the combustor. The time-lag distribution has been obtained from
a postprocessing CFD procedure in which the time needed by seed particles to
1.00
0.80
Distribution [ - ]
NBO
0.60
Short CBO
0.40 Long CBO
0.20
0.00
0.0000 0.1000 0.2000 0.3000 0.4000
t / t_0 [ - ]
Fig. 5.15 Normalized time-lag distribution for different burner-outlet modifications.

travel between the burner-outlet surface and the flame front has been statistically
evaluated. In this procedure the flame front is characterized by the maximum heat-
release rate encountered by the particle track during its trip through the combustor.
The statistical evaluation results in a time-lag distribution that is generated by the
finite extension of the flame. The time lag corresponding to the modification with
no CBO is the smallest and is characterized by the narrowest distribution. The
longer CBO leads to significantly increased time-lag values and more widespread
distribution. Application of the Rayleigh criterion showed that the increased time
lag improves the thermoacoustic stability; this result was verified in field tests.5
The concept works well because the reacting flowfield is only shifted by the CBO
and the main parameter that has been changed is the time lag.
Other design options for affecting the flame response–time-lag distribution are
changing the fuel-concentration profile at the burner exit, changing the fuel-nozzle
impedance, and changing residence time in the combustor by adjusting the burner
diameter or swirl number. These design options may change other stability para-
meters, mainly arising from the flame–vortex interaction, which is not understood
at presents hence, the simple time-lag concept cannot be used to evaluate these
design changes. Siemens is currently investigating the flame–vortex interaction to
improve the design methodology in this aspect. Another design option widely used
is the implementation of fuel-staging concepts, which allows modification of the
flame response during engine operation in response to combustion instability.
C. Resonators
1. High-Frequency Resonators
Relatively few analytical or numerical analyses of high-frequency dynamics
have been performed so far because of the high-temporal resolution that is needed
to investigate the phenomenon. Feedback cycles supporting the dynamic excitation
have time constants in the range of 0.2 to 0.6 ms, which are the time constants
of the combustion process itself. Hence, the origin of high-frequency dynamics
should be found in the reacting shear layer. Although the origin of the phenomenon
remains unclear, high-frequency dynamics can be well damped by resonators.
With Helmholtz resonators, a cavity is acoustically connected to the inside of the
combustion chamber by an orifice as shown in Fig. 5.16.
The eigenfrequency of a typical Helmholtz resonator configuration is given by

c Sneck
f Helmholtz = (5.3)
2π V (l + l)
The eigenfrequency is determined by the ratio of the area of the orifice of the
resonator tube S and resonator volume V times the length of the resonator tube l.
l expresses the elongation of the resonator tube to take into account radiation
effects. l is computed approximately as 0.85Dtube . The speed of sound is denoted
by c.
To increase the damping performance of high-frequency resonators, the
acoustic-absorption area of the resonators has to be high. In practical configu-
rations of resonators, several resonator tubes are connected in parallel to one large
106 W. KREBS ET AL.
Holes for
entering purge
air
Helmholtz-
Volume V
Combustor
L
Resonator Tube
Dtube
Fig. 5.16 Typical Helmholtz Resonator Configuration.
volume. To maintain a constant resonator temperature the resonators are purged

with compressor discharge air. In Fig. 5.17 the arrangement of resonator boxes
around a combustor basket of Siemens gas-turbine SGT6-5000F is shown. To
damp different high-frequency modes in the range between 2 and 4 kHz, resonator
boxes of different sizes are mounted around the basket. In the practical design of
the resonators, the need for acoustic damping of the high-frequency dynamics must
be balanced with the air-purging requirements and the physical dimensions of the
resonators. Placement of the resonators is critical to their overall performance. The
resonators should be placed near pressure antinodes, which can be measured or pre-
dicted by using three-dimensional analysis techniques described earlier. Because
high-frequency dynamics are multidimensional the axial, azimuthal, and radial
pressure profiles are considered for optimum resonator placement. A good start-
ing point is to place the resonators at the axial location of maximum heat release
because the flame is the source of the acoustic energy; however, in some cases, this
may not be possible because of geometrical constraints. It may also be possible
that another, more optimum location is desired for damping certain modes.
2. Intermediate-Frequency Resonators
Helmholtz resonators may also be used for suppressing intermediate-frequency
dynamics. To design intermediate-frequency Helmholtz resonators for annular
combustors, tests in an annular combustor rig have been performed. The annular
combustor rig is shown in Fig. 5.18.
Row of high frequency resonators for

suppressing High Frequency Dynamics
Fig. 5.17 High-frequency resonators designed to suppress high-frequency dynamics.
A cross section of the rig has already been shown in Fig. 5.8. The annular
combustor rig is equipped with an exit diffuser connected to a large exhaust volume.
When operating the annular combustor rig at atmospheric conditions, a strong
200-Hz combined azimuthal and axial mode is observed. In the tests 14 Helmholtz
resonators were mounted on the outer shell of the combustor to suppress this
mode. The arrangement of resonators is depicted in Fig. 5.19. The nonsymmetric
1.5 m
Fig. 5.18 Annular combustor rig.

108 W. KREBS ET AL.
Fig. 5.19 Combustor test rig with 14 Helmholtz resonators placed circumferentially.
placement of resonators on the circumference was chosen in a way to provide

optimum damping of acoustic waves.
Two types of resonators shown in Fig. 5.20 have been tested in the annular com-
bustor rig. The smaller resonator produced a minor improvement of the stability
limit. The larger resonator resulted in a substantial shift of the stability limit as
shown in Fig. 5.21.
Fig. 5.20 Resonators mounted in the annular combustor rig.

Fig. 5.21 Operation envelope of the annular combustor rig.
In Fig. 5.21 the rms value of the dynamic pressure is plotted vs air–fuel ratio.
The air mass flow through the combustor is held constant at 8 kg/s and the pilot
mass flow is set to 48 g/s. For the baseline case without resonators the fuel mass
flow can increase toward an airflow ratio (AFR) of 35. If the fuel mass flow is
further increased a sudden excitation of the 200-Hz mode occurrs. This mode can
only be damped if the fuel mass flow is considerably reduced toward an AFR of 38.
Fig. 5.22 Modeling of resonators.

110 W. KREBS ET AL.
Fig. 5.23 Eigenfrequencies of test rig with (circles) and without (squares) resonators.
Such a hysteresis has also been observed at different operating points characterized
by different air mass flows and pilot mass flows. With the large resonator, the fuel
can be increased up to an AFR of 30.5 before a sudden excitation occurs. Again
the mode can only be completely damped if the fuel mass flow is decreased toward
an AFR of 38; however, the amplitude is reduced quickly when AFR is increased
above 31.0. The implementation of the resonators results in a significantly wider
oscillation-free operation envelope.
To assess the damping achieved with this configuration before its realization, a
thermoacoustic model was set up using the network method described previously.
The model shown in Fig. 5.22 corresponds to the model of the combustor rig in
Fig. 5.9 equipped with resonators at the locations where they are planned to be
installed in the test rig.12 The transfer-matrix model of the Helmholtz resonators has
been qualified in single-impedance tests. It has been connected between the outer
annular ring of the plenum and the inner ring representing the annular combustor.
Computational analyses have shown that sufficient damping can be achieved with
this configuration. In Fig. 5.23, eigenfrequencies of the test rig without and with
resonators are compared. It can be seen that the originally critical eigenfrequency
is now replaced by a high number of strongly damped eigenfrequencies. It can
therefore be expected that the test rig can be operated at design conditions in
stable mode.
VI. Conclusion
Thermoacoustically induced combustion oscillations are a complex phe-
nomenon arising from the interaction of the flame with the acoustic environ-
ment that is present in all types of combustion systems. The complexity of the
engineering treatment originates from the long wavelengths associated with the
phenomenon, which includes all components downstream of the compressor and

upstream of the turbine. Although this phenomenon complicates the engineering
design methods, both theoretically and experimentally, it also makes a lot of de-
sign approaches possible. Main design modifications affect the burner design and
hence the flame response. The motivation for burner-design modifications is that
the flame response plays an important role as the thermoacoustic source and the
burner is the part that can be replaced most easily in an existing design. The cylin-
ders mounted on the Siemens hybrid burner have extended the flame and hence the
time-lag distribution. Other options are the use of resonators that, general, increase
the damping performance of the combustion system. High-frequency resonators
have already been successfully developed for the can-annular combustion systems
installed in the SGT6-5000F and SGT6-6000G.
To improve thermoacoustic stability in combustion systems, theoretical and
experimental design tools must be developed further. Network tools are used to
understand the interaction of the flame with the combustion environment and to
design test rigs to meet special thermoacoustic-design parameters. The main chal-
lenge is still the understanding of the flame. Although time-lag models have been
used to characterize the dynamic-time constants of the flame, the impact of coher-
ent structures on the dynamic behavior of the flame has not been resolved. Unsteady
CFD is one of the best methods to give more insight in the dynamic interaction
between vortices and the flame front. However, the computational intensity of
unsteady CFD calculations still makes this approach impractical for use as a de-
sign tool in full-scale industrial gas turbines. At present, hybrid methods involving
the use of both one- and three-dimensional analysis tools combined with stability
analysis provide the best approach for thermoacoustic-design optimization.
References
1
Szabo, T. L., “Lumped-Element Transmission-Line Analog of Sound in a Viscous
Medium,” Journal of the Acoustical Society of America, Vol. 45, 1969, pp. 124–130.
2
Munjal, M. L., Acoustics of Ducts and Mufflers with Application to Exhaust and Venti-
lation System Design, Wiley, New York, 1987.
3
Prade, B., Gruschka, U., Hermsmeyer, H., Hoffmann, S., Krebs, W., and Schmitz, U.,
“V64.3A Gas Turbine Natural Gas Burner Development,” American Society of Mechanical
Engineers, Paper GT-2002-30106, 2002.
4
Krüger, U., Hüren, J., Hoffmann, S., Krebs, W., Flohr, P., and Bohn, D., “Prediction and
Measurment of Thermoacoustic Improvements in Gas Turbines with Annular Combustion
Systems,” Journal of Engineering for Gas Turbines and Power, Vol. 123, 2001, pp. 557.
5
Bethke, S., Krebs, W., Flohr P., and Prade, B., “Thermoacoustic Properties of Can
Annular Combustors,” AIAA Paper 2002-2570, May 2002.
6
Morse, P. M., and Ingard, K. U., Theoretical Acoustics, McGraw–Hill, New York, 1968.
7
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, R. T. Edwards,
Flourtown, PA, 2001.
8
Büchnér, H., Lohrmann, M., Zarzalis, N., and Krebs, W., “Flame Transfer Func-
tion Characteristics of Swirl Flames for Gas Turbine Applications,” American Society of
Mechanical Engineers, Paper GT-2003-38113, June 2003.
9
Cho, J. H., and Lieuwen, T., “Laminar Premixed Flame Response to Equivalence Ratio
Oscillations,” Combustion and Flame., Vol. 140, No. 1-2, pp. 116–129, Jan. 2005.
112 W. KREBS ET AL.
10
Krebs, W., Flohr, P., Prade, B., and Hoffmann, S., “Thermoacoustic Stability Chart for
High Intense Gas Turbine Combustion Systems,” Combustion, Science and Technology,
Vol. 174, 2003, pp. 99–128.
11
Hubbard, S., and Dowling, A. P., “Acoustic Instabilities in Premix Burners,” AIAA
Paper 98–2272, 1998.
12
Lepers, J., Krebs, W., Prade, B., Flohr, P., Pollarolo, G., and Ferrante, A. “Investigation
of Thermoacoustic Stability Limits of an Annular Gas Turbine Combustor Test Rig with
and without Helmholtz Resonators,” American Society of Mechanical Engineers, Paper
GT-2005-68246, 2005.
Chapter 6
Characterization and Control of Aeroengine

Combustion Instability: Pratt & Whitney
and NASA Experience
Jeffrey M. Cohen∗ and William Proscia†

Pratt & Whitney, East Hartford, Connecticut
and
John DeLaat‡
NASA Glenn Research Center, Cleveland, Ohio
I. Introduction
A GGRESSIVE goals for increased performance and decreased emissions for

aeroengine gas-turbine combustors have led to the development of combustor
concepts that may operate closer to the boundaries of static and dynamic stability.
The combustor designers’ response to these performance and emissions demands
is often to design more compact, better mixed combustors. Advanced combus-
tors have higher volumetric heat release than legacy combustors and have annular
geometries that allow the possibility of thermoacoustic coupling with longitu-
dinal, tangential, and radial acoustic modes. Furthermore, lean, direct-injection
combustors under development share many features with the lean, premixed com-
bustors used in stationary gas turbines, in which combustion instabilities are widely
observed.
To minimize development costs and time, it is critical to possess the ability
to evaluate the dynamic behavior of a combustor design during the component-
development phase, thereby mitigating the need for design changes during ex-
pensive full-scale engine testing late in the development cycle. Solutions can
be evaluated and developed proactively and more quickly through analyses and
Sections 1–3 and 5–7 are works of the U.S. Government and are not subject to copyright protection
in the United States. Section 4, Copyright
c 2005 by the United Technologies Corporation. Published
by the American Institute of Aeronautics and Astronautics, Inc., with permission.
∗ Aerodynamics Manager, Combustor and Augmentor Technology.
† Fellow, Instability Modeling & Analysis.
‡ Controls Lead, Active Combustion Control Technology.
113
114 J. M. COHEN ET AL.
laboratory-scale experiments, enabling a faster transition of technology into pro-

duction engines.
Solutions for combustion instabilities can be categorized into passive and ac-
tive techniques. Passive solutions act either on the acoustics of the combustor (by
changing resonant frequency or adding damping) or on the unsteady heat release
(by changing the response of the flame to acoustic pressure fluctuations). Active-
control solutions use feedback control to attenuate the instabilities and have typi-
cally employed unsteady-pressure sensors and high-speed fuel flow rate actuation.
Both types of control require an understanding of the root-cause physics behind
the thermoacoustic-coupling process. Whereas passive-solution techniques, such
as screech liners and resonators, are somewhat mature, active-instability-control
techniques are still immature, especially with respect to liquid-fueled aeroengine
application. A step toward advancing the technical maturity of active instability
control is to evaluate the technology at laboratory scale on a realistic problem that
has engine relevance. Because of the complexity of the problem, the uncertainties
that can affect the solutions and the difficulty of making detailed measurements in
an operating engine, it is necessary to have analytical and experimental tools that
capture and scale the relevant dynamics.
Laboratory- or component-scale testing of developmental combustor concepts
is standard practice in the aircraft gas-turbine industry, with use of single-nozzle
flame tubes, single- and multinozzle sectors, and full-annular combustor rigs. The
gas-turbine community has developed an experience base regarding the fidelity
with which such test rigs must replicate engine-design details to characterize the
emissions and operability characteristics of engine combustors. However, an un-
derstanding of how to incorporate dynamic- and acoustic-boundary conditions and
measurement techniques into these types of test rigs needs to be developed.
Most off-the-shelf acoustic analysis tools do not include models for the effects
of mean flow, nonuniform temperature distributions, mass addition, or complex-
boundary conditions, all of which are important in the study of combustion instabil-
ities. This analytical capability is necessary to understand the root cause of the prob-
lem and also to scale solutions developed in the laboratory to engine application.
The primary objectives of this effort were to identify a combustion instability ex-
perienced in an engine environment and to demonstrate that this engine instability
could be replicated in a single-nozzle test rig that provides an engine-traceable test
platform for combustion-instability control research. These steps formed the build-
ing blocks of a process for addressing combustion instabilities early in the design
cycle. The approach described here is limited to instabilities in which the partici-
pating acoustic modes are longitudinal or Helmholtz modes, not transverse modes.
This chapter discusses a process for addressing combustion instabilities early
in the design cycle and presents an example combustion-instability problem, be-
ginning with the analysis of an engine-traceable instability, through the design of
subscale experiments, comparison of their results with results from the engine,
identification of a potential passive method for instability reduction, and a demon-
stration of active-control techniques. The process involves the following steps and
is discussed in detail in the corresponding sections of this chapter.
1) Analysis of dynamic data from the subject engine to determine characteristics

of instability: frequency, amplitude, and sensitivity to changes in hardware con-
figuration and operating conditions (Sec. II).
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 115
2) Acoustic analysis of the engine to determine acoustic modes associated with

instability (Sec. III)
3) Dynamic-response measurements of subcomponents, such as fuel injectors
and air swirlers, to determine criticality/coupling to unsteady heat release (Sec.
IV).
4) Conceptual design of single-nozzle experiment to reproduce the engine’s
acoustic environment and replicate the relevant dynamic processes as determined
in steps 1, 2, and 3 (Sec. V).
5) Acoustic analysis of subscale experiment to determine its fundamental acous-
tic modes to confirm similarity with the acoustic modes observed in the engine
(Sec. V).
6) Test of the finalized laboratory-scale experiment to provide comparison of
data with analyses and engine data (Sec. V).
7) Demonstration of active-control techniques in the subscale experiment to
show feasibility of active control via fuel modulation in a realistic enginelike
environment (Sec. VI).
The demonstration of the utility of this process will lend credence to its a priori
use in the combustor design and development process.
II. Engine Combustion Instability

The problem studied in this effort was a combustion instability that was observed
during the development phase of a high-performance aeroengine that employed
a full-annular combustor with 24 fuel nozzles. The frequency of the instability
varied from about 420 Hz at low-power conditions to about 580 Hz at high-power
conditions, as shown in Fig. 6.1. At a midpower operating point, corresponding to
that used for the analytical phase of this study, the frequency of the instability was
525 Hz. The magnitude of the pressure oscillations resulting from the instability
were sufficient to cause unacceptable vibratory stresses in the turbine component
of the engine.
70
60
50
Elapsed Time, sec
40
Mid-power
point chosen for
30 analysis
20
10
0
0 100 200 300 400 500 600 700 800 900 100
Frequency, Hz
Fig. 6.1 “Waterfall” plot of combustor pressure spectra from an engine test, showing
evolution of the instability during an acceleration event.
Although fast-response combustor pressure data were acquired during the engine
tests, they were acquired at a limited number of locations. For this reason, it was
difficult to draw any significant conclusions about the nature of the instability
purely from the engine data. The analyses described in Sec. III were used to
augment and interpret the engine data. The combination of the engine data, the
acoustic analyses, and the nonreacting swirler–injector characterization provided
the basis for replicating the problem in a single-nozzle combustor rig.
III. Engine Acoustic Analysis

A quasi-one-dimensional unsteady Euler analysis1 was used to predict the bulk
and longitudinal modes of the engine combustor. The one-dimensional Euler equa-
tions were solved with area variation and modeled source terms to account for mass
addition, heat addition, and pressure losses. The resonant acoustic frequencies of
the combustor were determined by solving for the unsteady time-accurate response
of the system and by monitoring the fluctuating pressure at a specified location.
The analysis can be used to examine the sensitivity of the frequency and pres-
sure amplitude to system parameters, such as physical dimensions, temperature
distribution, Mach number, flow rate, and pressure drop.
This analysis is useful for its intended role for the determination of acoustic
modes. The analysis does not attempt to incorporate the physics of the interac-
tion of the fluid mechanics/acoustics with the heat release (effects of flame-shape
variation, local fuel–air variations, etc.) and therefore is not capable of predicting
the absolute magnitude of the pressure response. The approach demonstrated will
show that it is, however, a satisfactory tool for evaluating engine acoustic charac-
teristics and for design of a test rig that replicates the relevant acoustic modes for
combustion-dynamics experiments.
Engine
Rig
Downstream
Upstream Plenum
Plenum Air Swirler
Prediffuser Combustor
Choked
Exit
Fig. 6.2 Cross-sectional area vs axial position for quasi-one-dimensional Euler model.
The solver was first used to compute the steady-state results, and then unsteady
results were obtained by forcing the system. Forcing was accomplished by adding
an unsteady component to the heat release. Two types of forcing signals were
used: 1) broadband white-noise distributed forcing, and 2) swept-sine forcing
over a range of suitable frequencies. The resulting pressure response indicated the
frequency dependence of the combustion system.
Quasi-one-dimensional Euler calculations were conducted for the engine
configuration at an intermediate operating condition: 771◦ F (684 K) and 200 psia
(1.2 MPa). The engine geometry was converted into a one-dimensional description
of area vs axial position, as shown by the dashed line in Fig. 6.2, which shows the
distributions for both the engine and the rig (to be discussed later). The geometry
used included an inlet plenum, the engine prediffuser, diffuser plenum, the cowl
or hood, the swirler, the combustor liner, and turbine vanes. The combustor lies
between x = 0 and 9.25 in. (23.5 cm). Beyond the turbine vane choked exit the
area was expanded rapidly to create a plenum dump. The boundary conditions
used were constant total pressure at the inlet plenum and constant static pressure
at the exit plenum.
The acoustic response of the system was obtained by swept-sine forcing of
the entire heat-release distribution. The fluctuation levels imposed on the heat
release were 10% of the mean. The unsteady pressure amplitude at the x = 3 in.
(7.6 cm) location in the combustor (3 in. downstream of the combustor dump
plane) was used to determine the pressure response. The swept-sine response is
shown in Fig. 6.3 and indicates a resonance at approximately 575 Hz. The width
log(Unsteady Pressure (psi))
Fig. 6.3 Computed power spectrum of combustor pressure for engine configuration
with sine-sweep forcing.
0.02
Unsteady Pressure Amplitude @575 Hz(psi)
0.015
0.01
0.005
−0.005
−0.01
−0.015
−0.02
−10 −5 0 5 10 15 20
X (in.)
Fig. 6.4 Computed pressure mode shape for 575-Hz mode for the engine configura-
tion at evaluation point conditions.
of the amplitude-response peak indicates a large amount of damping. However,

the width of the response also indicates a broad range of frequencies over which
the system may be susceptible to combustion instability.
These results indicate a longitudinal mode in the combustor near 575 Hz, which
is near the observed engine instability frequency of about 525 Hz. The pressure
mode shape for the 575-Hz mode is shown in Fig. 6.4. The mode represents a full-
wave solution to the system equations with zero unsteady pressure specified at
each end. Given the high impedance at each end caused by the high-Mach-number
boundaries, the mode shape can also be interpreted as a half-wave across the
diffuser-combustor domain with closed ends. Note that a pressure node is apparent
at the air swirler–fuel injector (x = 0). The calculated fluctuating pressure in the
diffuser was 180 deg out of phase from the pressure in the combustor. It is believed
that this was the basic acoustic mode that occurred in the engine configuration
instability. Although tangential acoustic modes exist in the full-annular engine
combustor, analysis of the engine data and two-dimensional Euler results have
indicated that they are not associated with the observed instability.
IV. Fuel Injector–Air Swirler Dynamic Response

Because they play such a large role in defining the stoichiometry and fluid me-
chanics of the primary combustion zone, fuel injectors may take part in determining
the level of coupling that takes place. The investigation of the role which the air
swirler from a gas-turbine fuel injector plays in thermoacoustic coupling in a real
gas-turbine combustor is discussed in this section. Three separate studies are de-
scribed and additional details can be found in Refs. 2 and 3. The objectives of these
studies were to identify and isolate the root cause for the thermoacoustic coupling
and to evaluate candidate solution paths to attenuate that coupling mechanism.
In the fuel injector–air swirler under study, air is passed through inlets into
a multipassage swirler. At the exit of the swirler, the highly swirled air flows
sandwich a thin sheet of fuel between them, shearing it into small droplets. The
first study, discussed in Sec. IV.A, investigated the pure acoustic (no throughflow)
of the fuel injector–air swirler to identify potential acoustic resonances in the
frequency range of interest. The second study, discussed in Sec. IV. B, measured
the effects of throughflow on the swirler impedance and unsteady fluid mechanics
to identify potential unsteady fluid mechanic coupling. The third study, described
in Sec. IV.C, used a new measurement technique to assess how the fuel spray was
modulated by acoustic pressure fluctuations.
A. Acoustic-Impedance Measurements
The objectives in making air swirler acoustic-impedance measurements were
2-fold. First, acoustic-impedance measurements will reveal any interesting atten-
uation or amplification characteristics of the air swirlers in the frequency range of
interest, 300 Hz to 600 Hz. Any resonances or antiresonances observed will aid in
interpretation of the combustor test data and guide design of an efficient strategy for
acoustic–fluid dynamic interaction tests. Second, experimental air swirler acoustic-
impedance measurements are needed inputs into combustor acoustic models.
Acoustic impedance is defined as the complex quotient of acoustic pressure di-
vided by acoustic volume velocity at a surface.4 It can be thought of as a frequency-
dependent transfer function that describes the response of the device to an incident
acoustic wave. The acoustic impedance includes resistive effects associated with
pressure losses and reactive effects associated with the inertance and compliance
of the device.
An experimental impedance-tube rig was used to measure the air swirler acoustic
impedances. Important design considerations for the rig were the overall tube
length, tube inner diameter, and the axial locations of the acoustic transducers.5
Overall tube length is suggested at least 5 L/D’s (length/diameters) to ensure that
the waves incident on the duct exit plane are planar. Tube lengths of 0.3 m (1 ft) and
0.91 m (3 ft) were used for these measurements. The duct had roughly the same
cross-sectional shape and diameter as that of the swirler for which impedance was
measured. The two-transducer, transfer-function method was used to measure the
acoustic impedance of the air swirler.5 To validate the measurement technique, the
open-tube acoustic impedance was measured and compared with analytical theory
for a long, open radiating tube.6 The measured impedance was in good agreement
with theory in the band of frequencies from 100 Hz to almost 1000 Hz.
Experimental measurements of air swirler acoustic impedance are shown with
the bounding cases of open-tube and rigid-wall boundary conditions in Fig. 6.5.
The impedance curves of different swirlers were essentially grouped based on the
effective airflow area of the separate devices. These different swirlers represented
a design matrix of swirl angles, effective areas, and flow splits. The frequency
response of the air swirlers is similar to that of an open tube with about 10 dB higher
impedance. The air swirlers did not exhibit significant low- or high-impedance
spikes in the frequency range of interest, indicating the absence of resonances or
antiresonances. As a result, the no-flow acoustics-impedance measurements did
not identify any particular resonances in the frequency range of interest.
60 200
Open tube
Open tube 150 Closed tube
40 Closed tube Swirler
Swirler 100
Magnitude (dB)
Phase (deg)
20 50
0 0
-50
-20
-100
-40 -150
-60 -200
200 400 600 800 1000 200 400 600 800 1000
Frequency (Hz) Frequency (Hz)
Fig. 6.5 Comparison of measured swirler impedance with open and closed tubes.
B. Fluid Mechanic Response Measurements

A similar experimental apparatus was used to test for possible interaction be-
tween air swirler exit velocity and acoustic excitation. Airflow from a 2.7-MPa
(400-psi) source was throttled through a pressure regulator and a choked venturi
into a Ling Electro-Pneumatic driver (model EPT-94B). This device is a high-
capacity, high-response (up to 1 kHz) valve that was driven by a sinusoidal voltage
to supply a pulsating flow into the plenum. The plenum supplied the acoustically
excited airflow to the air swirler mounted in the exit flange with ambient, un-
confined conditions downstream. The plenum inner dimensions were carefully
chosen to place longitudinal and Helmholtz acoustic resonances above 700 Hz.
The plenum exit flange was very thick (12.5 mm/0.5 in.) to prevent acoustic exci-
tation of the flange as a membrane. As a result, the large flange thickness limited
the axial distance from the hot wire to the air swirler exit plane to greater than
9.5 mm (0.375 in.).
A pressure gauge on the plenum allowed for air swirler P/P measurement. An
ac-coupled acoustic transducer mounted in the airflow plenum directly adjacent
to the swirler inlet was used to record acoustic-pressure oscillations inside the
plenum. A hot-film anemometer mounted on a traversing assembly was used to
measure velocity fluctuations at different radial locations along the air swirler exit.
The hot-wire probe was oriented with its length in line with the radial direction
to measure the resultant of the axial and tangential velocity components. Both the
acoustic transducer and hot-wire anemometer measurements were recorded on an
Hewlett–Packard dynamic signal analyzer. The natural frequency of the acoustic
transducer was 13 kHz, and the response of the hot wire was measured to be
20 kHz.
Mean airflow through the rig was measured upstream of the Ling driver by using
a choked venturi with a throat diameter of 5.5 mm (0.215 in.). Choosing a cold test
flow rate to ensure dynamic similarity with the full-scale combustor allowed direct
comparison of test results with engine data. Because the unsteady fluid mechanic
behavior of highly turbulent flow is often described in terms of a nondimensional
frequency, it was appropriate to scale by the Strouhal number:
fD
St = (6.1)
U
where f indicates characteristic frequency, D indicates characteristic length, and

U indicates velocity. The cold acoustic–fluid dynamic interaction tests used the
same full-scale air swirlers (i.e., characteristic length) as the hot-combustor tests,
leaving only velocity U to be scaled to ensure similarity. Adjusting the airflow
to achieve a P/P of 15.5% achieves dynamic similarity between the full-scale
hot-combustor tests and our cold-flow acoustic tests. The Ling driver could not be
operated at lower flow rates because of cooling considerations.
The air swirler exit air velocity to inlet pressure coherence and transfer function
(U/P) was measured as a function of excitation frequency to determine the pres-
ence of any acoustic–fluid dynamic interactions. Discrete tone forcing was first
used to measure the coherence between the two signals and then the swept-sine
technique was used to acquire the transfer-function data. Earlier attempts to use
broadband white-noise forcing proved to have inadequate signal-to-noise ratios.
Coherence may be defined as a measure of linearity between two signals. The
transfer function, because of the assumption of linearity, is only valid for cases in
which the coherence is high. The maximum value of the coherence function is unity,
which indicates an ideal linear relationship between the two signals. The minimum
value of coherence is zero, which indicates no linear relationship between the two
signals. A more detailed discussion of coherence and its importance may be found
in Bendat and Piersol.7
Traversing radially across the exit plane of the swirler and observing the
velocity–pressure coherence, the point of maximum sensitivity to acoustic ex-
citation was at a radial location near the wall between the two swirler passages.
This radial location was also had the highest measured mean velocities and was
directly adjacent to the fuel-filming surface. Figure 6.6 shows the pressure and
velocity spectra for a case in which the coherence at the forcing frequency (500
Hz) was high (0.85).
The swirler flow exhibited strong velocity–pressure coherence over the fre-
quencies from 350 Hz to 550 Hz. This coherence trend indicates a strong linear
relationship between velocity oscillations and pressure oscillations not present in
other swirlers tested. Based on the results of the coherence data, further testing
was conducted with swept-sine techniques.
-40
-42
-44
Velocity (dB)
-46
-48
-50
-52
-54
200 400 600 800 1000
Frequency (Hz)
Fig. 6.6 Spectra of plenum pressure, swirler exit velocity for 500-Hz forcing. Coher-
ence = 0.85.
-10 200
150
-20
100
Magnitude (dB)
-30
Phase (deg)
50
-40 0
-50
-50
-100
-60
-150
-70 -200
100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800
Fig. 6.7 Swirler exit velocity/plenum pressure transfer function using swept-sine
technique.
Swept-sine data acquisition integrated the dynamic-forcing and acquisition pro-

cesses to allow for discrete tone excitation and also for construction of a Bode
plot (magnitude and phase vs. frequency) representing the exit velocity–plenum
pressure relationship. A Hewlett–Packard signal analyzer (model 35665A) was
programmed to force the system in a series of discrete tones, collecting and aver-
aging response data over the frequency range 200–600 Hz, with a resolution of 2
Hz. The results of the swept-sine test are shown in Fig. 6.7. Multiple tests were
performed and demonstrated good repeatability. Note that the swirler exhibited a
high-velocity fluctuation–pressure fluctuation gain over the range of 300–500 Hz.
Directly outside this range, the incident pressure fluctuations cause little or no
velocity fluctuation at this measurement location. Additional testing of the swirler
at an increased pressure drop corresponding to twice the air velocity indicated that
the frequency range over which this sensitivity occurs scales directly with veloc-
ity, establishing the validity of Strouhal number scaling. The shape of the transfer
function indicates that the swirler acts in a manner analogous to a bandpass filter.
These results led to the conclusion that the unsteady fluid mechanics of the swirler
internal airflow were relevant to the instability root cause. The degree to which this
fluid mechanic response could affect the unsteady heat release was then assessed
by using a similar technique to measure the fuel-spray response.
C. Fuel-Spray Response Measurements

An instrument was built to measure the fluctuations of fuel flow through trans-
verse planes of a fuel spray and to determine the level of coupling between an
applied acoustic excitation and those fuel flow fluctuations. The instrument used
was in the same test facility that was used to measure the swirler’s fluid mechanic
response, in which the swirler airflow could be modulated. The fluctuating air
pressure upstream of the injector and fluorescence signal derived from the liquid-
fuel spray were continuously monitored to characterize the degree of correlation
between them.
The instrument was then used to characterize four fuel injectors–air swirlers
from the design matrix. In particular, the study was to ascertain whether injector
response occurred within any preferred frequency ranges in which the motion of
the atomized fuel spray could couple with the induced airflow pulsations caused
Air flow system

Pressure
Venturi regulator
Microphone Acoustic
driver
Spray
nozzle
PMT
Spray
pattern Receiver
Amplifier
Light sheet
Signal
forming
Analyzer
system
Optical Laser
Fiber
Fig. 6.8 Test setup for fluorescence measurements of fuel mass flow fluctuations. The
airflow system includes an acoustic driver that perturbs the flow. The microphone
measures the acoustic forcing of the spray.
by resonant acoustics, thereby providing a feedback mechanism leading to the

formation of combustion instabilities.
The measurement technique takes advantage of the natural ability of typical jet
fuels such as Jet A to fluoresce when excited by visible light in the blue/green
region of the spectrum. The fluorescence is generated from additives and aromatic
constituents.8 An atmospheric-pressure spray rig provided the framework in which
this instrument was used. This facility provided the air and fuel flows necessary to
operate a fuel nozzle, as shown in Fig. 6.8. The Ling acoustic driver was used to
perturb the airflow over a range of frequencies so that the effects on the spray could
be observed. An ac-coupled acoustic transducer located in the plenum chamber
was used to yield an accurate measure of the energy input to the system. The
acoustic-resonance frequencies of the plenum chamber were designed to be far
above the range of interest (>1 kHz). A uniform laser sheet projected through
the spray excited fluorescence proportional to the mass of the fuel within the
sheet. Continuous monitoring of the fluorescence intensity with a single detector
provided a temporal record of the mass and its fluctuations. Correlations of mass
fluctuations with measurements of the acoustic pressure signal provided a means
of examining the level of coupling. The laser sheet could be moved along the spray
axis to determine the spatial nature of the coupling process. A detailed discussion
of the optical technique is contained in Ref. 2.
The optical technique was used to characterize the dynamical signatures of four
fuel injector–swirler combinations. They will be referred as injectors A through D,
respectively, with injector A being the injector tested in the engine. The injectors
utilize the relative motion9 between a low-velocity sheet of liquid fuel and sur-
rounding high-velocity air streams to effectively disrupt and break apart the liquid
sheet into unstable ligaments and large droplets. The liquid fuel (Jet A) is injected
onto a filming surface and is then atomized through the combined influences of
inner and outer airflows having identical swirl directions.
An initial assessment of the technique was done in the form of a calibration.
This entailed measuring the photomultiplier tube (PMT) signal for a variety of
fuel flow rates. The airflow rate was regulated by a choked venturi and adjusted
to yield a pressure drop of 17.2 kPa (2.5 psi) across the air swirler, whereas the
mass flow rate of fuel was varied between 0 and 113 kg/h (250 lbm/h). Both
quantities encompassed the normal (scaled to atmospheric pressure) operating
range of the device. For this series of tests, the laser sheet was located 100 mm (4 in.)
from the exit plane of the injector. The calibration for this technique was roughly
linear.2
The measured PMT signal was nondimensionalized by the incident laser in-
tensity Io . Two data sets are shown with a linear curve fit to demonstrate the
repeatability of the technique. The linearity of the curves is apparent: increasing
the mass flow rate of fuel increased fluorescence intensity and consequently the
PMT signal, as expected.
To quantify the degree of interaction between the pulsed airflow (forcing func-
tion) and the fuel spray (output), the coherence between the acoustic pressure
signal and PMT signal signals was measured for a variety of flow conditions. A
Hewlett–Packard signal analyzer provided an input to drive the valve and recorded
both the acoustic pressure measured in the air-supply plenum and the PMT signal,
representing the spray response. By continuously varying the frequency of the
forcing function, a map of the output’s response was obtained, detailing possible
frequency bands in which interactions between the two signals could be seen. This
measurement, therefore, yielded essential information on the dynamic behavior of
prospective injectors. Figure 6.9 depicts the coherence measurements from injec-
tor A at several axial stations as the driver was forced at a variety of frequencies
between 0 and 900 Hz. The figure shows a location at which the acoustic fluc-
tuations have the greatest effect on the spray. This finding is consistent with the
understanding of the evolution of the atomization process. As the liquid fuel is-
sues from the injector, it is first atomized through the influences of the momentum
flux ratio existing between the high-velocity airflow and the slower moving liquid
flow. After a primary atomization process in which large, unstable ligaments and
droplets are formed, a secondary atomization commences that further reduces the
particles’ size. This process is typically Weber number dependent.10 Only when
1.0
1 "
0.8
2 "
3 "
Coherence
0.6 4 "
5 "
0.4
0.2
0.0
0 300 600 900
Forcing Frequency (Hz)
Fig. 6.9 Coherence measurements for the injector A at several axial stations.
the droplets have attained a unique size can they be more easily influenced by
the surrounding airflow; if they are too large, prevailing forces are insufficient to
accelerate them. This process, in general, relies on the governing Stokes number
to be much less than unity. The Stokes number (St ) is typically defined as the
particle’s response time (τ p ) divided by the eddy-turnover time (τe ) or
τp
St = (6.2)
τe
If particles are too large, they become centrifuged inside turbulent or large-scale
structures so that their response to flowfield changes is negligible. As the particles
are reduced in size with increasing axial distance, creation of an optimal axial
location for forcing should occur. The reason for the drop in droplet response
for greater axial positions (>4 in.) is unclear, however. One possible explanation
could be the attenuation of the acoustic energy over an increasingly larger area
with subsequent downstream locations, thereby diminishing its influence on the
liquid droplets. Another reason could simply be the damping of the spray’s motion
because of spreading and drag effects.
Also apparent is the strong coupling in the 300- to 700-Hz region at all axial
distances. There are even tails on either side of the main peak, hinting at other
frequency bands of interaction. The increasing coherence near-zero frequency is
typical of zero-frequency functions, which have very strong linear relationships,
varying only by a gain factor.3
Figure 6.10 summarizes the coherence measurements for all four injectors taken
at 4 in. from the exit plane. Except for injector D, all the injectors exhibited spray–
acoustic coupling within the frequency range of interest (350–700 Hz). This is
important because it reflects the ability of the droplets to respond to the exter-
nal excitation, thereby allowing for the possibility of a corresponding fluctuating
heat-release trace that could lead to unstable burning. Injector C exhibited the
highest coherence over the frequency range of interest. Between 400 and 650 Hz,
the coherence remained at a fairly constant and high value of about 0.85 for this
swirler. Even at lower frequencies (150–350 Hz), the coherence remained at fairly
1.0
injector A
injector B
0.8
injector C
injector D
Coherence
0.6
0.4
0.2
0.0
0 300 600 900
Forcing Frequency (Hz)
Fig. 6.10 Summary of coherence measurements for all injector configurations, 4 in.
from the exit plane.
substantial values, even matching the highest coherence measurements of injector

A (compare the coherence of injector C at 250 Hz with the maximum coherence
of injector A). The coherence plots also reveal that injector B exhibited the most
peaked coherence map, centering at approximately 550 Hz. It therefore has a
narrower or more selective interaction region. Injector A, conversely, has a more
intermediate region of interest when compared with the injectors C and D. Because
acoustic forcing had a negligible influence on the spray’s behavior with injector
D, this injector would appear to provide the potential for a passive solution to the
observed instability. In fact, engine implementation showed a significant attenua-
tion of the instability amplitude to acceptable levels, and these modified injectors
were incorporated into the production engine design.
Further evaluation of the acoustic coupling was done through measurement of
the spray mass flow–acoustic pressure transfer function, using the swept-sine tech-
nique over the 350- to 650-Hz frequency range. Because the goal of this effort was
to identify a process for replicating the engine-observed instability, injector A con-
tinued as the subject of these investigations. This frequency range was chosen for
injector A, based on the high levels of coherence exhibited there. Figure 6.11 shows
the results of measurements of nozzle A in a plane 100 mm (4 in.) downstream of
the nozzle exit. The plots show the transfer function for the mass fluctuations in
this plane resulting from upstream acoustic pressure perturbations ( p ). The raw
data representing time-dependent mass concentration were collected and stored
by the analyzer in the frequency domain. The time derivative of mass in the mea-
surement plane was derived in the frequency domain to avoid amplifying noise in
the signal. The results show the characteristics often found in bandpass-filtering
devices. The fuel injector acted as a bandpass amplifier, only allowing significant
interaction between the spray and the acoustics over a range of frequencies, peak-
ing at approximately 450 Hz. Evaluation of this characteristic for an injector and
comparison of this frequency range with the dominant acoustic modes of the com-
bustion system will allow designers to address combustion-instability problems
earlier in the development process.
0.10 180
120
0.08
60
Phase (deg.)
Ma gn itu de
0.06
0
0.04
-60
0.02 -120
0.00 -180
350 400 450 500 550 600 650 350 400 450 500 550 600 650
Fig. 6.11 Magnitude and phase of the mass fluctuation/ p transfer function at a lo-
cation 4 in. downstream of nozzle A. The mass fluctution transfer function is derived
from the mass/ p transfer function by multiplying by frequency.
V. Subscale Combustor Experiment

At this time there is no proven methodology for replicating engine-scale combus-
tor dynamics in laboratory-scale rigs. The challenge in designing laboratory-scale
combustion dynamics experiments is to replicate the engine-dynamic environment
in as simple (low-cost) an apparatus as possible. Recent published work suggests
that bulk-mode and longitudinal-mode instabilities can be replicated in single-
nozzle rigs. Cohen et al.11 and Hibshman et al.12 performed active-instability-
control experiments in single-nozzle and sector combustors that reproduced a
bulk-mode instability observed in a lean, premixed industrial combustor. Paschereit
et al.13 have developed a subscale combustor in which the boundary conditions at
the inlet and exit ends can be varied to impose a desired acoustic mode. No relevant
work has been published on replication of tangential modes in multinozzle sectors
or configurations other than full-annular combustors.
To attempt to replicate the engine-observed instability described earlier in this
chapter, a single-nozzle test rig was designed. The test rig design approach incor-
porated the following guidelines.14
A. Use of Full-Scale Fuel Preparation Subcomponents

(Fuel Nozzles, Air Swirlers)
Results from the fuel-injector dynamic-response measurements indicated that
the prototype injector–swirler (swirler A) exhibited an enhanced response near
the 500-Hz instability frequency observed in the engine data, so it was important
to utilize that precise design. Although reduced-size hardware may be of interest
to minimize facility requirements, this approach was not taken because of the
introduction of uncertainties associated with reduced-scale flows.
B. Acoustic Isolation of the Combustor from Facility Air Piping

A venturi was used to choke and meter the inlet airflow. Because the isolation
provided by the sonic throat condition was desired over a range of conditions, the
venturi was designed to be underexpanded, resulting in a normal shock at a distance
Table 6.1 Comparison of engine and rig acoustic features
Feature Engine Rig
Combustor volume per injector, in.3 /cc 13/851 08/770

Combustor length, in./cm 0.5/21.3 0.5/1.3
Shroud volume per injector, in.3 /cc 29.3/119 03.4/695
Nominal shroud height, in./cm 0.2/3.0 0.80/00
Diffuser length 0.7/6.9 0.7/6.9
Prediffuser length, in./cm 0.9/9.9 0.9/9.9
of 1.38 in. (3.5 cm) downstream of the venturi throat. The upstream boundary was
largely established by the normal shock and the sudden expansion of the flow at
the prediffuser dump. The downstream boundary was defined by using a choked
exhaust nozzle at the station occupied by the first turbine inlet vane.
C. Reproduction of the Longitudinal Acoustic Behavior

Along with providing acoustic isolation, the choked exhaust nozzle set the acous-
tic length of the test section. Other critical elements in this regard were the com-
bustor, the diffuser, the prediffuser, and the cowl (hood) which guided the diffuser
air to the fuel nozzle. Table 6.1 shows a comparison of the geometric features of
the annular engine burner with those of the rig. Cross-sectional areas were chosen
to replicate associated volumes and expansion/contraction ratios.
D. Reproduction of the Airflow Distribution, Pressure Drops,

and Flow-Damping Characteristics
The pressure drops and airflow splits used in the engine were duplicated. That
is, the fractions of air used for liner cooling and for primary and dilution air were
reproduced. Designing for equivalent damping is important for achieving similar
instability amplitudes between the test rig and the engine. The resistive damping
of the system was maintained by replicating the system’s pressure drops. Repro-
ducing the airflow splits between the different airflows replicates the distribution
of stoichiometry and heat-release rate within the combustor.
E. Design for Testing at Engine-Operating Conditions

The instability observed in the engine occurred over a range of test conditions. A
single evaluation point was chosen at 200-psia (1.2-MPa) combustor pressure, an
entrance temperature of 771◦ F (684 K), and combustor fuel–air ratio of approxi-
mately 0.03. All analyses were conducted at these conditions. Operating at reduced
conditions with full-scale hardware can change the operating characteristics of the
components (pressure drops, atomization, etc.).
The final consideration was whether to utilize an apparatus with a simple cylin-
drical cross-section burner or an apparatus having a cross-section representative of
1/24 of the 24-nozzle engine burner, that is, a single-sector burner. A circular cross
section was used, because this was the lower-cost, higher-strength approach. The
Fig. 6.12 The combustor test section assembly.
number and size of the combustion and dilution air holes was adjusted to provide
proper penetration of these air jets.
The single-nozzle combustor rig design was established to preserve the axial
lengths and cross-sectional areas of the engine configuration relative to a single
fuel nozzle. The approximately the same area vs axial position distribution was
maintained (Fig. 6.2), but some variation existed because of differences in engine
hardware and the axisymmetric hardware to be used in the single-nozzle combustor.
Airflow splits and pressure losses (swirler, bulkhead, liner, primary and dilution
jets) were also preserved by design. As mentioned, the inlet and exit of the rig were
choked to acoustically isolate the system. The configuration of the test section is
illustrated in Fig. 6.12.
Provisions for high-response pressure transducers and for gas sampling (not
reported herein) were incorporated into the test section. Three transducers, equally
spaced around the circumference, were located in the primary combustion zone.
One transducer was located in the secondary zone, and one was in the dilution
zone.
Unsteady shroud-flow pressure measurements were provided at the location of
the liner primary and dilution holes. Bosses for diffuser unsteady-pressure mea-
surements, upstream of the combustor, were also provided.
Quasi-one-dimensional Euler acoustic analyses were conducted for the baseline
rig configuration at the evaluation-point operating conditions: 771◦ F (684 K), 200
psia (1.2 MPa). The acoustic analyses included inlet and exit plenums upstream
and downstream of the choke points to allow constant total pressure and constant
static pressure to be specified, respectively, as boundary conditions to be applied
to the Euler code domain. Swept-sine forcing over the frequency range from 100
to 800 Hz was applied to the heat release. The resulting power spectrum of the
pressure response is shown in Fig. 6.13, indicating the presence of resonances at
115 Hz and 550 Hz. The level of forcing used in the analysis is arbitrary, and
within the linear-response range, so that no significance should be attached to the
absolute levels of the ordinates in Figs. 6.13 and 6.14.
Further analysis of the mode shapes associated with these resonances revealed
that the low-frequency 115-Hz mode was a first-order longitudinal mode in which
Rig Pressure Response to Sine Sweep

101
log (Unsteady Pressure (psi))
00
10-1
100 200 300 400 500 600 700 800
freq (Hz)
Fig. 6.13 Computed power spectrum of combustor pressure at x = 3.0 in. for the rig
configuration. Quasi-one-dimensional Euler code results for 100 to 800-Hz swept-sine
forcing of heat release.
the diffuser and combustor were in phase. The primary mode of interest was the
550-Hz mode because the observed instability frequency in the engine was 525
Hz. The pressure-mode shape is shown in Fig. 6.14. The 550-Hz mode was essen-
tially a half-wave longitudinal mode considering closed/closed acoustic boundary
conditions from diffuser inlet to combustor exit. A pressure node appeared to occur
at the air swirler–fuel injector location. The pressure in the diffuser was 180 deg
out of phase from the pressure in the combustor. Note that some activity occurred
downstream of the combustor exit, but calculations performed with varying exit
Unsteady Pressure Amplitude @ 575 Hz (psi)
0.08
0.06
0.04
0.02
−0.02
−0.04
−0.06
−0.08
−15 −10 −5 0 5 10 15 20
X (in.)
Fig. 6.14 Computed pressure mode shape for the 550-Hz mode. Rig configuration at
evaluation-point conditions.
0.4
0.35
Unsteady Combustor Pressure Amplitude (psi)
0.3
0.25
0.2
0.15
0.1
0.05
0
0 500 1000 1500 2000 2500
Frequency (Hz)
Fig. 6.15 Measured power spectrum of unsteady combustor pressure at x = 1.9 in.
for evaluation-point operating conditions, showing resonance at 566 Hz, with an am-
plitude of 0.39 psi (0.78 psi p- p).
plenum lengths did not indicate significant changes in the resonant frequency.
Comparison of the engine and rig acoustic analyses (Fig. 6.3 vs. Fig. 6.13 and
Fig. 6.4 vs. Fig. 6.14) showed good agreement between the acoustic response of
the engine and the rig as designed.
Based on the results of these analyses and the stated design principles, the
experimental test rig design was finalized. It was fabricated and installed in a high-
pressure, high-temperature combustion test cell. The operating conditions of the
combustor could be completely described by the following parameters: diffuser
air pressure (P3), diffuser air temperature (T3), and combustor fuel–air ratio (f/a).
Values for each of these parameters were chosen to correspond to three different
engine-operating conditions. These are shown in Table 6.2. The fuel–air ratio
referred to is that estimated at the exit of the combustor and accounted for all of
Table 6.2 Test conditions corresponding to engine operating points
Inlet air pressure, Inlet air Fuel–air

P3, psia/MPa temperature, T3, F/K ratio
70/0.48 500/533 0.016

110/0.76 600/589 0.024
175/1.21 771/684 0.030
Evaluation-point conditions are in bold.

the air flowing into the combustor through the air swirler, primary and dilution
holes, and liner- and bulkhead-cooling passages. It was not possible to vary the
test parameters independently because of the choked, fixed-area combustor exit.
PCB piezoelectric pressure transducers (P/N 124A21) were selected for this
experiment. They were capable of measuring pressure fluctuations at frequencies
between 0.5 Hz and 10 kHz at high-mean pressures. Satisfactory durability was
achieved by use of an integral water-cooled mounting fixture that maintained an ac-
ceptable temperature around the sensor. The liner-pressure sensors communicated
with the combustor through a 0.062-in. (1.6-mm)-diam, 0.83-in. (2.1-cm)-long
sensing tube. Nitrogen was used to purge the tube; the amount of purge flow rate
was negligible. The quarter-wave resonant frequency of the cavity within this tube
was far above the frequency range of interest in this experiment. Analog data were
low-pass filtered at 2 kHz and digitally sampled at 5 kHz by using a simultaneous
sample-hold data-acquisition system.
For the evaluation-point operating conditions, an instability was observed at a
frequency of 566 Hz (Fig. 6.15). The amplitude of this mode at these conditions
was ±0.39 psi (2.7 kPa). The unsteady pressure results presented here are from
the transducer located at 1.9 in. (4.7 cm) downstream of the combustor bulkhead.
The amplitude of the instability increased with increasing fuel–air ratio for fixed
P3 and T3. At higher fuel–air ratios, the overall rms pressure fluctuations were
dominated by this single tone. There was significant noise generated in the 100-
to 300-Hz range, although none of it was particularly coherent.
Figure 6.16 shows the spatial distribution of the unsteady pressure at three
locations within the combustor and one location in the diffuser region upstream.
3.5
Magnitude Phase 150
3
100
Amplitude/ Amplitude PLA1C1
Phase - Phase PLA1C1 (deg)
Bulkhead
2.5
50
2
0
1.5
-50
Combustor
1 Exit
-100
0.5
-150
0
-2 0 2 4 6 8 10
X (in.)
Fig. 6.16 Measured distribution of 566-Hz mode, showing magnitude and phase ref-
erenced to pressure measurement PLA1C1 at x = 1.9 inches downstream of the com-
bustor bulkhead.
Unsteady Combustor Pressure Amplitude (psi) 0.5
0.45
0.4 P3 = 175 psi

P3 = 110 psi
0.35 P3 = 70 psi
0.3
0.25
0.2
0.15
0.1
0.05
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 6.17 Power spectra of unsteady combustor pressure at three operating conditions
corresponding to the conditions in Table 6.2, showing decreasing amplitude of the
566-Hz mode with decreasing power level.
Within the combustor, there was no phase difference between measurements at

different axial stations, and only small differences in amplitude. A significant
566-Hz signal was also apparent upstream of the combustor, which lagged the
combustor pressure by 92 deg in phase and was smaller in amplitude by a factor
of 2. There were no phase or magnitude differences (at 566 Hz) between pressure
measurements at equivalent axial, but differing circumferential stations, consistent
with a longitudinal acoustic mode. The unsteady pressure in the shroud, just outside
the dilution holes, was a factor of 2 smaller than the unsteady pressure in the
combustor and lagged the combustor pressure by approximately 40 deg.
This mode was also observed at the other two, lower-power operating conditions,
although at a smaller amplitudes and lower frequencies, as shown in Fig. 6.17.
The experimental results can be compared with the analytical results by ref-
erencing Fig. 6.13, which shows the predicted pressure spectrum, and Fig. 6.15,
which shows the measured pressure spectrum. Recall that the Euler code model
predicted broad acoustic resonances at about 575 Hz and 115 Hz. In the experi-
ment, a broad instability centered near 570 Hz was observed, and some incoherent
activity was indicated near 100–200 Hz. Thus, the agreement appears to be good.
The mode shape measured in the experiment was of limited spatial resolution and
showed little spatial variation of unsteady-pressure amplitude or phase within the
combustor chamber itself for the 566-Hz mode (Fig. 6.16). This mode shape was
consistent with the mode shape of the 575-Hz mode predicted by the Euler code
(Fig. 6.14). Note both results did indicate a slight decrease in amplitude toward the
upstream end of the combustion chamber. The Euler code prediction indicated the
NASA SNR R004p38 and Engine TX3081.2B

0. 4
Engine
Rig
0. 3
Amplitude (psi)
0. 2
0. 1
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 6.18 Comparison of engine and combustor rig pressure spectra for evaluation-
point operation.
unsteady pressure in the diffuser upstream of the combustor would be 180 deg out
of phase with the combustor pressure. The experimental results indicated a signif-
icant phase shift in the diffuser section, lagging the combustor pressure by about
90 deg at 566 Hz. This discrepancy is likely associated with the one-dimensional
limitations of the model. For example, it is expected that some level of coupling to
the outer-shroud passage would occur, which is also coupled to the combustor via
the air-mixing holes. Therefore, some transition of the phase from in phase with
the combustor outside the mixing holes to out of phase in the diffuser section is
expected in the three-dimensional problem. The result could be a phase relation
in the diffuser section between 0 and 180 deg.
Note that, because the Euler code is essentially an acoustic calculation, it is
fundamentally limited in its ability to calculate the amplitude of the pressure
oscillations without the addition of a combustion–acoustic coupling model. In
calculations for the engine using a constant relative forcing level, the Euler code
indicated that both the frequency and amplitude of the instability should increase
with increasing engine-power level. This trend was validated with engine data and
was also reproduced in the single-nozzle experiment (Figs. 6.1 and 6.17). It is also
encouraging that the damping mechanisms present in the calculations produced a
broad peak at 575 Hz, much like the peak seen in the experiment (Fig. 6.17).
Figure 6.18 shows a comparison between the fluctuating-pressure spectrum in
the engine and the single-nozzle combustor at comparable operating conditions.
Both data sets were acquired over 10 and were processed by using the same
techniques. The frequency of the target mode was reproduced within 12%. The
amplitude of this mode was matched within 3%. The spectral peak was signif-
icantly narrower in the engine data, indicating a more coherent instability. The
single-nozzle combustor also exhibited a higher overall level of noise in the sig-
nal, especially at frequencies below 350 Hz.
Table 6.3 High frequency fuel valve specifications
Maximum mean fuel flow rate 500 lbm/h

Maximum inlet pressure 600 psi
Minimum pressure into injector 300 psi
Maximum modulation flow ±40% of mean flow
Actuator bandwidth Minimum, 600 Hz
Flow media JP-8 jet fuel
VI. Active-Control Demonstration

Active combustion control (ACC), which provides feedback-based control of
the fuel injection, the fuel–air mixing process, and the staging of fuel sources
can provide an alternative approach to achieving acceptable combustor dynamic
behavior, and thus can provide flexibility during the combustor design process. Ac-
tive instability control has been demonstrated on full-scale industrial gas turbines15
(which use gaseous fuel), but has yet to be demonstrated on liquid-fueled aero-
engine combustors. The single-nozzle combustor rig developed under this effort
was used to investigate the feasibility of active instability control using fuel modu-
lation in a realistic aeroengine combustor. This investigation required development
of 1) an actuator capable of modulating the fuel at the desired flow rate and fre-
quency, 2) reduced-order models for control design, and 3) control methods able
to identify and suppress the instability. These technology developments, discussed
in the following sections, culminated in demonstrating active suppression of the
single-nozzle rig instability.
A. Actuator Characterization
To demonstrate instability control, a suitable fuel actuator was necessary. The
specifications for the fuel valve, shown in Table 6.3, were derived from the rig
fuel flow requirements and prior experience. Several fuel actuator concepts were
investigated, and two were chosen for further development. A high-frequency fuel
valve built by the Georgia Institute of Technology was selected for experimental
testing because of the maturity of the concept (Fig. 6.19) The valve included both
a high-frequency flow modulation component and a mean flow control component
in a single device.
To provide a way to conduct steady-state and dynamic characterization of the
capabilities of the fuel valve, a characterization rig was developed and fabricated.
The rig was able to deliver up to 2 gal/min continuous water flow at up to 600 psia.
It was designed to provide an isolated test section for the valve to simulate the
valve/feed-line/injector (VFI) environment encountered in combustor rig testing.
An accumulator at the valve inlet provided isolation from the supply dynamics.
Downstream from the valve, an orifice simulated a fuel injector and a pressurized,
air-filled volume emulated the combustor.
Steady valve flow characterization was conducted first. This procedure consisted
primarily of mapping fluid flow vs valve displacement to quantify the valve mean
flow control authority. Steady valve flow characterization was also used later to op-
timize the valve position to maximize high-frequency fuel-modulation amplitude.
Fuel
Outlet
Auxiliary Fuel
Cooling Inlet
Air
1î
Fig. 6.19 High-frequency fuel valve developed by Georgia Institute of Technology.
The valve exhibited a well-behaved, monotonic increase in flow as the valve open-
ing was increased. Also, once the valve reached a displacement of approximately
0.015 in. it was fully open and was no longer able to modulate the flow. To control
the mean flow and also to modulate the dynamic flow, the valve position had to be
maintained in the range between 0.005 and 0.015 in.
For the dynamic characterization of the valve, dynamic pressure transducers
were placed upstream and downstream of the valve, and downstream of the fuel
injector. Initially, a minimum feed-line length (just long enough to incorporate
the transducer) was used between the valve and the injector. Use of the minimum
feed-line length between the valve and injector allowed direct measurement of the
valve P frequency response while minimizing the interaction with the feed line.
A sinusoidal input signal of ±1 V was sent to the valve, and the pressure drop
across the valve was analyzed with respect to the input signal. Line lengths of 1 ft
and then 2 ft were inserted between the valve and the simulated injector orifice
to simulate the effect of realistic line lengths as would be encountered when lines
were installed on a combustor rig or on an engine.
Figure 6.20 shows the transfer function between valve command and valve
pressure drop. As can be seen from the transfer function, adding line length between
the valve and the fuel injector decreased the resonant frequency of the fuel system.
As line length was increased from 0 to 2 ft, the resonant frequency decreased
toward the 500- to 600-Hz combustor resonance frequency. Having this response
singularity at or near the controller frequency of interest should be avoided because
this proximity of frequency values can frustrate attempts to control the unsteady
fuel mass flow as required for active instability control. Thus, there is a maximum
installation fuel-line length between the fuel valve and the injector above which
the interaction between the fuel system and the combustor instability will become
extremely complicated.
Thus, the valve authority, that is, the level at which the valve is able to perturb
the fuel flow and thus the combustor pressure, was ultimately determined experi-
mentally in the combustor rig. The valve was set to a nominal steady opening based
on the characterization rig results. Open-loop, sinusoidal valve command voltage
0.2
500Hz resonance
0.15
DP23a/input, psi/volt
0.1
0.05
700Hz resonance
0
360
180
-180
Phase, deg
-360
-540
-720 0ft-1V
1ft-1V
-900
2ft-1V
-1080
0 500 1000 1500 2000 2500
Frequency, Hz
Fig. 6.20 Dynamic-valve response showing the transfer function between valve-
commanded voltage (input) and valve Delta-P (DP23) for three different feed-line
lengths.
variations were provided to the valve. The single-nozzle combustor rig was oper-
ated at conditions close to those in Table 6.2 that gave a ∼530-Hz combustion in-
stability. The frequency and amplitude of the valve command voltage were varied,
and the combustor pressure was monitored by using the combustor pressure sensor
1.9 in. downstream of the bulkhead. Representative results are shown in Fig. 6.21.
For a 300-Hz, ±2.5-V (maximum allowed) valve command, the combustor rig
dynamic pressure was shown to have a sharp response to the valve perturbations.
The pressure response was imposed on top of the combustion-instability-pressure
variations. Similar results are shown for a 600-Hz valve command. There was
some initial concern that, even if the valve was able to impose large fuel mass
flow variations, the prefilming features of the fuel injector would reduce actuator
a) 300 Hz 2.5
Valve Command Voltage
3
2 2
Amplitude, volts
Amplitude, volts
1
1.5
0
1
-1
0.5
-2
0 -3
Combustor Pressure
0.4 4
0.3 2
Amplitude, psi
Amplitude, psi
0.2 0
0.1 -2
0 -4
0 200 400 600 800 1000 0 0.01 0.02 0.03 0.04
Frequency, Hz Time, sec
b) 600 Hz 2.5
Valve Command Voltage
3
2 2
Amplitude, volts
Amplitude, volts
1
1.5
0
1
-1
0.5
-2
0 -3
Combustor Pressure
0.4 4
0.3 2
Amplitude, psi
Amplitude, psi
0.2 0
0.1 -2
0 -4
0 200 400 600 800 00 0 0.01 0.02 0.03 0.04
Frequency, Hz Time, sec
Fig. 6.21 Combustor pressure response to commanded valve perturbations shows

open-loop actuator authority.
Instability Pressure
Acoustics
Pressure from +
+
Σ
Fuel Fuel lines, Injector Fuel Modulation Combustor Pressure
Flame
Valve & Combustion
Phase Shift
Controller
Filter
Fig. 6.22 Adaptive sliding phasor-averaged control approach.
authority. However, these tests confirmed that sufficient authority (on the order of
the instability amplitude) was available.
B. Control Methods Development and Demonstration

To achieve closed-loop suppression of the combustion instability, a controller
must sense the combustion pressure oscillations and actuate the fuel at a frequency
and phase that interferes with the instability. In addition, to avoid damage to the
combustor, it is desirable that the controller be able to isolate and mitigate the
instability while it is still small, that is, while it is still on the same order as
the combustor noise. To increase the probability of successful instability suppres-
sion, two alternative control methods were developed. These control methods were
formulated to deal with the large-wideband combustor noise, severe time delay,
and randomness in phase associated with the combustor thermoacoustic pressure
oscillations. The first control method was based on an adaptive, phase-shifting
approach. The Adaptive Sliding Phasor Averaged Control (ASPAC), shown in
Fig. 6.22, sensed the combustor pressure, filtered the signal to capture the in-
stability frequency, calculated the average power in the pressure oscillations, and
adapted the phase of the valve-commanded fuel flow variations to reduce the power
in the pressure oscillations.
The ASPAC method assumes that the overall combustor pressure is the sum of
the instability pressure and the pressure oscillations caused by the fuel modula-
tion (Figs. 6.22 and 6.23). A fast-acting phase-adaptation algorithm initially takes
large phase steps to find the phase region in which power decreases (Fig. 6.23,
Boundary of restricted control region). The phase then slides in smaller steps until
power increases and then reverses direction (Fig. 6.23, Boundary of effective sta-
bility region). The effective stability region shrinks as the instability is suppressed.
However, if there is a persistent increase in power (effective control is lost), then
a new restricted control region is established. By constantly dithering the phase
within the region that causes cancellation, the algorithm rapidly adapted to ran-
domness in the instability pressure, especially that due to background combustor
noise. The controller sample frequency was 10 kHz. The algorithm also provided
Overall combustor pressure

Pressure from
Fuel modulation
Pressure from
instability
Boundary of effective
stability region
Boundary of restricted
control region
Fig. 6.23 ASPAC control approach finds a phase region that provides reduction in
the overall combustor pressure.
a slower, more gradual adaptation of the controller gain. Further details on the
ASPAC method can be found in Refs. 6.16 and 6.17.
The second control method was a model-based approach. The multiscale ex-
tended Kalman (MSEK) approach, like the first method, also sensed combus-
tor pressure. The MSEK method, shown in Fig. 6.24 combined a multi scale
(waveletlike) analysis and an extended Kalman filter observer to predict (model)
the time-delayed states of the thermoacoustic combustion pressure oscillations.
The commanded fuel modulation was calculated from a predictive (damper) ac-
tion based on the predicted states, and an adaptive, tone-suppression action based
on the multiscale estimation of the pressure oscillations and other transient dis-
turbances. The controller attempted to automatically adjust the gain and phase of
these actions to minimize timescale-averaged variances of the combustor pres-
sure. The controller operated at a sample frequency of 5 kHz. Further information
on the MSEK control approach is in Ref. 18. Both control methods were ini-
tially evaluated against reduced-order oscillator models of the combustor pressure
Upstream
injector
sensed
Sensed Time-Scale pressure
combustion Averaged
pressure Pressure
Multi-Scale Variance Fuel
Tones Analysis modulation
command
Parameter Upstream
Tuning Compensation
Phase Drift
Estimation
EK States Damper
Phase-Adjusted Predictor
Reconstruction
Suppression
Fig. 6.24 Multiscale extended Kalman combustion instability control approach.

ACC sim
0.5
0.4
Open Loop
PSI / Hz
0.3
Closed Loop
0.2
0.1
0
0 200 400 600 800 1000
Frequency, Hz
Fig. 6.25 Pressure amplitude spectra showing approximately 60% reduction in in-
stability amplitude predicted for model-based control method.
to verify basic functionality. To provide a better-fidelity validation of controller

performance before rig testing, both controllers were then tested against a quasi-
one-dimensional model of the combustor rig.19 In addition to mass, momentum,
and energy equations, there were also one or more species transport equations with
associated, relatively simple reaction and heat-release equations. The combustor
was approximated by dividing it into a finite number of one-dimensional (constant
area) sectors. The resulting simulation and associated boundary conditions essen-
tially represent a one-dimensional, multiblock technique. This modeling approach
provided a simulation testbed for the control-method evaluation prior to experi-
mental demonstration. Uncontrolled and controlled simulation results are shown
in Fig. 6.25. The simulation results predicted that the controller would be able to
achieve approximately a 60% reduction in the peak unsteady-pressure amplitude.
The high-frequency fuel valve and developed control methods were used to
demonstrate closed-loop instability suppression in the single-nozzle combustor rig.
The rig was operated at the conditions shown in Table 6.2, and exhibited roughly
the instability behavior shown in Fig. 6.15. Combustor pressure was sensed 1.9 in.
(4.8 cm) downstream of the fuel injector. The control algorithms were implemented
on a dSpace real-time processor. The fuel flow was dynamically controlled via the
high-response fuel valve.
For evaluation of each controller, the baseline operating condition was estab-
lished first, and open-loop perturbations were injected to verify actuator health and
0.4 0.4
open-loop
open-loop
Model-Based Control
0.35 0.35 Adaptive Phase-Shift Control
0.3 0.3
Amplitude, psi
0.25
Amplitude, psi
0.25
0.2 0.2
0.15 0.15
142
0.1 0.1
0.05 0.05
0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
Frequency,Hz Frequency,Hz
a) Model-based control method b) Adaptive phase-shifting control method
Fig. 6.26 Combustor pressure amplitude spectra for initial active combustion instability control test showing closed-loop instability
suppression for both control methods.
authority. The closed-loop controller was then engaged. Two sets of tests were
run with both controllers being evaluated during each test. During the first test, a
reduction in instability amplitude was observed for both control methods. However,
for both control methods, low-frequency (<30 Hz) oscillations were seen in the
combustor pressure frequency spectra. It was suspected that these low-frequency
oscillations were caused by interactions between the instability controller and the
valve mean flow control or some other low-frequency phenomena.
For the second set of tests, additional filtering was added to the controllers to
reduce this interaction. The results, shown in Fig. 6.26, demonstrated that this
approach was effective in greatly reducing the low-frequency oscillations. As
can be seen, both control methods were able to identify the instability frequency
and reduce the amplitude by about 30%. This reduction was accomplished with-
out inducing secondary peaks at adjacent frequencies as has been seen in other
combustion-instability-control studies.12,20,21,22 Chapter 23 of this collection in-
cludes a detailed discussion of several factors that will affect the controller’s ability
to attenuate an instability.
VII. Conclusion
In summary, aggressive performance and emission goals for aircraft gas-turbine
combustors are leading to the development of combustors with increased suscep-
tibility to combustion instabilities. In this effort, a process has been developed
that addresses the difficulty in predicting and mitigating these combustion insta-
bilities. This process has been applied to an example problem consisting of an
engine-observed combustion instability.
Acoustic analysis of the engine combustor geometry using quasi-one-
dimensional Euler calculations was conducted to help explain the engine-instability
behavior. This analysis showed a susceptibility of the combustion chamber to
acoustic resonance at the observed 525-Hz instability frequency. Based on the
limited engine-dynamic data available, the analysis results seem to indicate a lon-
gitudinal acoustic mode frequency and mode shape which are believed to be that
exhibited by the engine.
Dynamic-response experiments were conducted with the engine fuel injector–
air swirler. Acoustic impedance tests showed that no specific acoustic parameters
relating to the geometry of this swirler would cause preferential amplification (and
instability coupling) of pressure fluctuations in the 400- to 600-Hz range. Fluid
dynamic forcing tests showed, however, that the air velocity near the filmer of
the swirler was highly sensitive over this range. The swirler exhibited preferen-
tial amplification ranging from 300 to 500 Hz, peaking at approximately 350 Hz.
The frequency of the observed fluid mechanic unsteadiness scaled directly with
velocity, indicating that this unsteadiness may be related to periodic vortex shed-
ding or other phenomena that are velocity sensitive. A novel optical technique
was devised to monitor the fuel flow fluctuations in an acoustically excited spray,
using the fluorescence of additives and aromatic constituents in the fuel. The tech-
nique was applied to four aeroengine fuel injectors and identified strong acoustic
coupling between the pulsed airflow and the fuel spray mass flow in the 350- to
650-Hz range. Modifications to the fuel injector–air swirler design based on the
dynamic spray and fluid mechanic response measurements resulted in significant
attenuation of the instability amplitude to acceptable levels. These modified injec-

tors were incorporated into the production engine design.
A subscale combustor experiment was designed using quasi-one-dimensional
acoustic analyses, information from subcomponent experiments on critical hard-
ware, and simple scaling rules. The acoustic analyses predicted a longitudinal
acoustic resonance at about 550 Hz that was very similar to the mode observed
in engine data and predicted by acoustic analysis of the engine. In the testing
of this experiment, the frequency of the “target” mode was reproduced within
12%. The amplitude of this mode was matched within 3%. The spectral peak was
significantly narrower in the engine data, indicating a more coherent instability.
Finally, active-control techniques using unsteady-combustor-pressure sensors
and high-speed fuel actuation were demonstrated on the laboratory-scale combus-
tor test rig. A high-speed fuel flow actuator capable of setting the mean flow and
perturbing the fuel mass flow about that mean was developed and characterized.
Two control methods, one based on adaptive phase-shifting and one using model-
based control, were developed using reduced-order and physics-based simulations
of the test rig instability behavior. Demonstration of these active control methods
on the test rig achieved amplitude reductions at 530 Hz of about 30%.
Acknowledgments
Numerous outstanding engineers and technicians made very significant contri-
butions to the concept and execution of the projects we have described. At NASA,
they are Clarence Chang, Joseph Saus, Daniel Paxson, George Kopasakis, Dzu Le,
Daniel Vrnak, Kevin Breisacher, and James May. At Pratt and Whitney they are
Jeffery Lovett, Michael Ondas, Saumil Shah, Donald Kendrick, and Saadat Syed.
At United Technologies Research Center they are Brian Wake, Thomas rosfjord,
John McVey, Torger Anderson, Randy Hibshman, Karen Teerlinck, Michael Carey,
Jeffrey Walker, and Walter Borst. At Georgia Institute of Technology they are
Yedidia Neumeier, Eugene Lubarsky, Ben Zinn.
References
1
Wake, B. E., Choi, D., and Hendricks, G. J., “Numerical Investigation of Pre-Mixed
Step-Combustor Instabilities,” AIAA Paper 96-0816, Jan. 1996.
2
Anderson, T. J., Kendrick, D. W., and Cohen, J. M., “Measurement of Spray/Acoustic
Coupling in Gas Turbine Fuel Injectors,” AIAA Paper 98-0718, Jan. 1998.
3
Cohen, J. M., and Hibshman, J. R., “An Experimental Study of Combustor Air Swirler
Acoustic and Fluid Dynamic Sensitivities,” Propulsion Engineering Research Center 9th
Annual Symposium on Propulsion, 1–2 Oct. 1997.
4
Kinsler, L. E., Frey, A. R., Coppens, A. B., and Sanders, J. V., Fundamentals of Acoustics,
3rd ed., Wiley, New York, 1982, pp. 230–243.
5
Munjal, M. L., Acoustics of Ducts and Mufflers with Application to Exhaust and Venti-
lation System Design, Wiley, New York, 1987, pp. 201–207.
6
Beranek, L. L., Acoustics, Acoustical Society of America, Melville, NY, 1993, pp. 123–
128.
7
Bendat, J. S., and Piersol, A. G., Random Data, 2nd ed., Wiley, New York, 1986,
p. 172.
8
Arnold, A., Dinkelacker, F., Heitzmann, T., Monkhouse, P., Schafer, M., Sick, V., and
Wolfrum, J., “DI Diesel Engine Combustion Visualized by Combined Laser Techniques,”
Twenty-fourth Symposium (International) on Combustion, The Combustion Institute, Pitts-
burgh PA, 1992, pp. 1605–1609.
9
Castleman, R. A., “The Mechanism of the Atomisation of Liquids,” Journal of Research
of the National Bureau of Standards, Vol 6, No. 281, 1931, pp. 369–376.
10
Hopfinger, E., and Lasheras, J., “Breakup of a Water Jet in High Velocity Co-flowing
Air,” 6th International Symposium on Liquid Atomization, Spray Systems, 1994, Institute
for Liquid Atomization and Spray Systems, Irvine, CA, pp. 110–117.
11
Cohen, J. M., Rey, N. M., Jacobson, C. A., and Anderson, T. J., “Active Control of
Combustion Instability in a Liquid-Fueled Low-NOx Combustor,” Journal of Engineering
for Gas Turbines and Power, Vol. 121, No. 2, April 1999, pp. 281–284.
12
Hibshman, J. R., Cohen, J. M., Banaszuk, A., Anderson, T. J., and Alholm, H. A., “Active
Control of Combustion Instability in a Liquid-Fueled Sector Combustor,” American Society
of Mechanical Engineers, Paper 99-GT-215, June 1999.
13
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Control of Combustion Driven
Oscillations by Equivalence Ratio Modulations,” American Society of Mechanical Engi-
neers, Paper 99-GT-118, June 1999.
14
Peracchio, A. A., Rosfjord, T., McVey, J., Anderson, T., Banaszuk, A., Cohen, J.,
Hibshman, J., Jacobson, C., Khibnik, A., Proscia, W., and Rey, N., “Active Control for
Marine and Land-Based Aeroderivative Gas Turbine Engines,” Vol. 1, Defense Advanced
Research Projects Agency Final Contractor Rept., United Technologies Research Center
Rept. 98-16, December 1998.
15
Hoffman, S., Weber, G., Judith, H., Herrmann, J., and Orthmann, J., “Application of
Active Combustion Instability Control To Siemens Heavy Duty Gas Turbines,” Symposium
of the AVT Panel on Gas Turbine Engine Combustion, Emissions and Alternative Fuels,
RTO-MP-14, AGARD, NATO Research & Technology Organization, Oct. 1998.
16
Kopasakis, G., and DeLaat, J., “Adaptive Instability Suppression Controls in a Liquid-
Fueled Combustor,” 38th Joint Propulsion Conference and Exhibit, AIAA Paper 2002-4075,
NASA TM-2002-21805, July 2002.
17
Kopasakis, G., “High-Frequency Instability Suppression Controls in a Liquid-Fueled
Combustor,” 39th Joint Propulsion Conference and Exhibit, AIAA Paper 2003-1458, July
2003.
18
Le, D., DeLaat, J., and Chang, C., “Control of Thermo-Acoustic Instabilities: The
Multi-Scale Extended Kalman Approach,” 39th Joint Propulsion Conference and Exhibit,
AIAA Paper 2003–4934, July 2003.
19
Paxson, D., “A Sectored-One-Dimenstional Model for Simulating Combustion Insta-
bilities in Premix Combustors,” 38th Aerospace Sciences Meeting & Exhibit, AIAA Paper
2000-0313, NASA TM-1999-209771, Jan. 2000.
20
Murugappan, S., Acharya, S., Gutmark, E., and Messina, T., “Characteristics and Con-
trol of Combustion Instabilities in a Swirl-Stabilized Spray Combustor,” 35th Joint Propul-
sion Conference and Exhibit, AIAA Paper 99-31259, June 1999.
21
McManus, K. R., Magill, J. C., Miller, M. F., and Allen, M. G., “Closed-Loop System
for Stability Control In Gas Turbine Combustors,” AIAA Paper 97-0463, Jan. 1997.
22
Barooah, P., Anderson, T. J., and Cohen, J. M., “Active Combustion Instability Control
with Spinning Valve Actuator,” American Society of Mechanical Engineers, Paper GT-
2002-30042, June 2002.
Chapter 7
Monitoring of Combustion Instabilities:

Calpine’s Experience
Jesse B. Sewell∗ and Peter A. Sobieski†

Calpine Turbine Maintenance Group, Pasadena, Texas
I. Introduction
T HIS chapter describes the experience with combustion instabilities and, in

particular, combustion-dynamics monitoring at the Calpine Turbine Mainte-
nance Group. Calpine is a non-utility-based electrical power company operating
plants in 21 states and 3 provinces in North America. The Calpine fleet includes
both dry, low NOx (DLN) and conventional systems. The fleet consists of primar-
ily the Siemens–Westinghouse (S/W) 251B12, 501D5, 501D5A, 501FC/FD1/FD2,
and 501G and the General Electric (GE) 6B, 7EA, 7FA, LM2500, LM5000, and
LM6000 engines.
Although research on many of the fundamental aspects of combustion insta-
bilities has increased substantially during the past few years, much of this focus
has been on either design, fundamental analysis, or small-scale-combustor labo-
ratory testing. Published results on combustion-instability experience from actual
engines are rare. Nonetheless, combustion-dynamics monitoring (CDM) has be-
come a standard tool in most power plants for monitoring not only combustion
instabilities themselves, but also for tuning engines for emissions, stability, out-
put, and even heat rate. Many large utilities have dedicated staff, such as Calpine,
for installing, maintaining, monitoring, and tuning engines based on these CDM
systems, which are often monitored remotely from some central location. Section
II describes a typical CDM system used at Calpine.
Continuous CDM has two key benefits that very quickly recoup the large ex-
pense associated with installing and continuously monitoring a system. First, by
directly monitoring the combustion-instability amplitude, corrective action can
be taken in systems that start to oscillate before expensive and destructive part
Copyright 2005
c by the American Institute of Aeronautics and Astronautics, Inc. All rights
reserved.
∗ Turbine Technology Engineer.
† Director of Turbine Maintenance.
147
148 J. B. SEWELL AND P. A. SOBIESKI
failures occur. Cracking of combustion parts and other component failures have
been known to liberate pieces and introduce significant downstream damage. Sec-
tion III describes part damage encountered in Calpine power plants because of
combustion-instability problems. Second, CDM can be used as a general com-
bustor health monitoring tool; part failures, flow blockages, fuel–air distribution
problems, etc., can often be detected from the CDM system by recognizing changes
in the dynamic-spectrum signatures. Section IV describes several case studies in
which the CDM system is used to detect combustor problems in this way.
II. Combustion-Dynamics Monitoring System

Instrumentation for CDM systems takes a variety of configurations that varies
between the OEM’s, engine class, and user. In general, quantitative accuracy of
pressure pulsations CDM data is not as important as, say, temperature or static-
pressure measurements. The reason for this is that original equipment manufactur-
ers (OEMs) do not know a priori what an unacceptable level of pressure amplitude
at a given frequency is based on part-life or failure analyses. Rather, in general,
such information is determined, empirically often by the users. As such, under-
standing of tolerable amplitudes and durations of these amplitudes is determined
for specific systems.
A. Pressure Instrumentation
In general, the instrumentation configuration is driven by the temperature limita-
tions of the transducers. Because of the lack of suitable high-temperature pressure
instrumentation, the sensor is always physically placed away from the flame. This
necessarily introduces some transfer function between the actual pressure in the
combustion chamber and that measured at the transducer. Depending on the stand-
off tube configuration and sensor location, amplitude roll off or attenuation can
become significant.
Either of the following configurations were used for the data discussed in the
following sections. In Siemens–Westinghouse engines, a 3/8-in. waveguide or tube
is mounted flush to the combustion liner, providing acoustic access to a sampling
point in the primary zone of the combustor (Fig. 7.1). The waveguide passes
through a pressure seal in the engine casing at the combustor “top hat,” through
an isolation valve, and to the dynamic-pressure sensor. For this configuration the
dynamic-pressure sensor is located about 12–14 in. from the combustion zone. The
waveguide further extends to an attenuation tube that consists of a 3/4-in. tube filled
with 1/16-in. capillary tubes that dampens the signal and minimizes reflections.
Because of condensation in the tubing, it is periodically purged with dry nitrogen.
Figure 7.2 shows a general description and layout of the system. The sensor is a
fast-response piezoelectric crystal-type dynamic-pressure transducer. The sensor is
hermetically sealed, electrically isolated, and vibration compensated. In multi-can
combustors, one of these sensors has acoustic access to each combustor can.
In GE engines, a 1/4-in. tubing is run from an access point on the combustion
liner wall to an instrumentation/purge panel located outside the turbine enclosure
(Fig. 7.3). The typical distance from the acoustic-access point and the dynamic-
pressure sensor is not more than 27 ft. Damping or attenuation is done by traditional
infinite coils, that consist of 125 ft of 1/4-in. tubing. The system also uses dry
COMBUSTION INSTABILITIES MONITORING: CALPINE EXPERIENCE 149
Waveguide
Acoustic Access Point
Combustor
Top Hat
Fig. 7.1 Acoustic access to primary zone of the combustor for Siemens–Westinghouse
F class engines.
Attenuation Tube
Piezoelectric transducer
Charge
Converter
Nitrogen
Combustor Purge
Top Hat
Analyzer
Fig. 7.2 General Siemens–Westinghouse system layout.

Acoustic Access Point
Fig. 7.3 Acoustic access to primary zone of the combustor for General Electric
engines.
nitrogen to purge any condensation buildup. The sensors are piezoelectric-type

isotrons with a built-in charge amplifier.
B. Data-Acquisition System and Spectral Analysis

Spectral data are recorded and stored by a data-acquisition system or are reduced
and brought into the plant historian database, in which they can readily be analyzed
along with plant-supervisory instrumentation. Combustor dynamics are commonly
analyzed in the frequency domain by applying a fast Fourier transform (FFT) to the
time-domain pressure signal of each combustor. Although the FFT is rigorously
defined, important field issues associated with how the FFT has been set up arise
when comparing amplitudes from engine to engine and when trying to compare
the different OEMs. For example, one manufacturer will look at zero to peak
amplitudes, but others will use peak to peak. In addition, combustor pulsations are
often erratic and noisy and the sampling interval can have a significant impact on
the amplitude in a given frequency bin. For example, if the acquisition time is very
short (0.2 s), the resolution may be as coarse as 10 Hz and the amplitudes will
be exaggerated as compared with longer acquisition times (∼6.0 s), at which the
resolution can be as fine as ∼0.3 Hz.
It is not practical to save continuous frequency and amplitude data of the entire
spectrum for every combustor. A common practice is to save the peak amplitude
observed within each spectral bin (called “peak hold”) over some interval,
commonly 1 min. These data are then compared with frequency-dependent alarm-
threshold levels, which warn of harmful or abnormal conditions. Note that these
alarm thresholds are determined empirically; all that is really known from this type
of data is that large amplitudes are damaging and low amplitudes are less damaging.
III. General Instability Characteristics and Tuning Considerations

Instability characteristics in the Calpine fleet are, in general, grouped into low-,
midrange-, and high-frequency dynamics. Low-frequency dynamics (LFD), often
called rumble, generally occur in the 10- to 50-Hz frequency range. For example,
they occur at 17 Hz in the GE 7FA and at 25 Hz in the S/W 501F. In the industry,
these types of oscillations are referred to as “cold tones,” because their amplitude
increases as the flame temperature decreases. They are often observed at very lean
conditions near blowout. LFD is one of the easiest modes to identify; it is very
audible and may sound like a freight train is running through the plant. Most engines
with LFD signatures will need seasonal tuning each spring and fall. Typically in
the fall as colder weather comes in the engines start running leaner because of
cooler, more dense air. Typical pretuning symptoms are lower engine outlet NOx
and increased LFD. Figure 7.4 depicts typical LFD signatures for lean-running
Fig. 7.4 Example of low-frequency dynamics in a GE 7FA.

engines. A key indicator of a tuning issue in the dynamic spectrum is that LFD is
present in all combustors. If it were a localized issue, such as a part failure with one
combustor, the symptom would be observed in the localized area. In the spring,
when ambient temperatures are on the rise, the opposite symptoms occur.
Midfrequency dynamics (MFD) are generally observed in the 100- to 250-Hz
frequency range. For example, tones occur at 130–150 Hz in the GE 7FA and at
120–140 Hz in the S/W 501F. In the industry, these oscillations are often referred to
as “hot tones,” because their amplitude often increases with firing temperature and
engine power output. For S/W engines the 135- to 145-Hz hot tone is dominant
at higher loads, whereas a 315-Hz mode can become excited depending on the
degree of gas heating and fuel splits. There are situations in which both hot tones
and cold tones can be present at base load depending on the fuel splits. At part
loads of < 85% of base load these engines are relatively stable as long as a high
degree of pilot fuel is contributing in the splits. Both hot tones and cold tones
are rarely present at the same time unless a hardware issue exists. In contrast,
GE7FA engines burn quite a bit leaner throughout the load range. These engines
can see both hot tones and cold tones throughout the load range depending on the
fuel splits. In addition, it is possible to excite multiple dominant frequencies by
adjusting fuel splits.
Finally, high-frequency dynamics (HFD), sometimes called screech, is occa-
sionally observed in industrial gas turbines. Although not very common, HFD is
particularly destructive to engine hardware; parts have been observed to fail in as
little as a few minutes during high-amplitude HFD. HFD at ∼1600 Hz and above
are observed in the S/W 501F and 501G and at ∼260 Hz or above in the GE F class.
Dynamics monitoring has traditionally been used for engine tuning. Tuning DLN
combustors involves changing the amount of fuel delivery between the various
fuel stages and in some cases controlling the amount of air by adjusting the inlet
guide vane (IGV) angles. If the engine has a closed-loop exhaust temperature
control, the IGVs are modulated by the control system to either maintain exhaust
temperature based on a defined curve or to maintain a specified margin lower than
the actual exhaust temperature limit or the so-called firing curve. Open-loop control
modulates the IGVs based on the load in either power output or a percentage of
full load. For either control scheme, the tuner can make adjustments to the IGV
schedule throughout the load range. Care must be taken not to run into a stall or
surge, which are not typically published or provided by some OEMs.
Changes are made on line and are done in small increments. Good practice will
include mapping the engine out to optimize emissions and stability. For example,
at a given load, the tuner will start with one fuel stage and increase the fraction by
enough of a percentage to make a noticeable change in dynamics and emissions.
If the changes improve the current condition, this process is continued. Next, the
original fraction is restored and compared with the original base line. Then, the
fuel fraction of that same stage is decreased until noticeable changes occur in
emissions and dynamics. This procedure will expose the optimum envelope for
that stage and load.
This procedure is done stage by stage, sometimes adjusting airflow throughout
the load range until the stability has been mapped out and an acceptable set of
fuel fractions or splits is obtained. Some balancing or iterations will be required
because of coupling of the stages.
Many combinations of fuel splits for the same load range can meet emissions
and stability requirements. The fuel-split changes are typically used to control
NOx and stability. The IGVs have a significant impact on part-load CO emissions,
whereas a 0.5- to 1-deg change can change CO emissions by as much as 100+ ppm.
Compared with DLN combustors, conventional systems are much simpler to
tune. Conventional systems have only one fuel stage, and water or steam injection is
used for NOx abatement. For water injection, the water is injected at the fuel nozzle
directly into the primary zone. For steam injection, the steam can be introduced
at the nozzle in the primary zone, downstream of the primary zone, or in the
combustor shell before the combustor. They usually run at a steam-to-fuel ratio of
1.2 to 1.5.
In general, it is not possible to tune in anticipation of upcoming ambient changes
such as temperature, humidity, or barometric pressure. In some cases, a unit can
experience extremes in temperature and humidity through various load ranges
several times a day. For example, increases by 50–100% in dynamics amplitude
have been observed as ambient-temperature drops and the engine output increases.
Without continuous monitoring, the only way operators know if there is a tuning
issue, is when dynamics become audible and the engine begins to rumble or trips.4
Other indications of tuning troubles include excessive emissions and unstable
operation such as load swings. In such instances the engine will be load limited
to prevent harmful operation in the range of instability until the engine can be
retuned. Keeping the engine tuned to meet the changes in ambient and special
operations extends parts life and can significantly improve maintenance cycles.2,5
IV. Detrimental Impacts of Combustion Dynamics

Combustion instabilities have been responsible for damaging a variety of hard-
ware, including fuel nozzles, combustor cans, and transition pieces. In extreme
cases, a part may be liberated into the gas path and ultimately through the turbine
section. Although current knowledge about acoustic instability and it’s relation to
combustion hot-gas-path part life is limited, it is known that parts must be designed
such that the natural mechanical frequencies have a design margin offset to the
combustion frequencies that is not close to instability frequencies.
Cold tones occur axially in the combustion system and have an impact on the
turbine section and combustion parts. Excessive cold-tone operation can show up
on the hardware as increased distress on the wear surfaces of combustion parts
and fretting and damage to the exhaust end of the turbine and heat-recovery steam
generator or stack. The amplitude allowable for the cold tones is typically less than
for hot tones.
Hot tones are also detrimental to combustion hardware if allowed to run at
excessive amplitudes for extended periods; however, our experience has been that
they are less destructive than cold tones or screech. It is difficult to say exactly which
parts are being affected by either hot or cold tones without knowledge of material
characteristics, natural frequencies, etc. Most combustion hardware that exhibits
distress caused by combustion dynamics or pulsations associated with instabilities
fails in low-cycle fatigue (LCF) because of either thermal or mechanical stress, and
then the crack propagates in high-cycle fatigue (HCF) caused by a combination
of acoustic, mechanical, or thermal contributions. Crack initiation is usually at
Material removed from area around dilution

holes with several crack propagations
Crack initiation area

around dilution holes
Cross Flame Tube assembly
completely broken off
Fig. 7.5 Photograph of combustor damage caused by high-frequency dynamics.
an area of high-stress concentration caused by material, geometry, welds, or even

cooling holes.
Screech tones can damage hardware very rapidly. For this reason it is difficult
to try to tune out of such situations. One of the reasons that screech is so unique
is that the part can fail in HCF simply because of cycles to failure. If the part is
subject to 2000-Hz dynamics at 1.0 psi or greater, it takes 500 s to reach a mil-
lion cycles. The most destructive manifestations of screech usually occur when a
particular part goes into resonance, driven by the acoustic pulsations. The follow-
ing photographs show some of the results of HFD to S/W combustion hardware.
In Fig. 7.5, damage around several dilution holes on the forward section of the
combustion can is evident, as well as the missing cross-flame-tube assembly that
has broken completely off. Under normal operation these sections of the can are
able to withstand a lifetime of thermal and mechanical stress loading. However, in
HFD parts fail quickly. Small parts with welds or connected by weld or other areas
of stress concentration are especially vulnerable. A second example is shown in
Fig. 7.6, where a crack in the combustor basket is clearly evident.
V. CDM for Combustor Health Monitoring: Case Studies

This section describes several examples in which analysis of pressure oscil-
lations is used to detect problems. One of the key empirical insights aiding this
analysis is that there should be uniformity in combustor dynamics from one can
to the next. Although spectral characteristics of the engine vary with fuel splits
and load, these signatures should be fairly uniform from basket to basket. This
insight is used to identify problems. A local event such as a part failure or flow
disturbance will usually be isolated to the specific combustor and the adjoining
Crack propagation on plate fin

inside of combustor
Fig. 7.6 Photograph of combustor damage caused by high-frequency dynamics.
positions. In contrast, a tuning issue will be visible in most, if not all, combustors
and can be further isolated by using plant supervisory data.
However, it is not always possible to determine which part has failed until after
an inspection. By using combustor-dynamics data along with plant supervisory
data, the objective for operations is to quickly determine one of the following; run
the engine normally, shut it down, or operate at reduced load until parts, manpower,
and equipment can be brought to the site.
A. Flow Obstructions
In the situation described here, an engine experienced LFD at about 25 Hz as it
approached base load, which would periodically exceed the alarm level. Figure 7.7
shows the combustor exhibiting LFD (0.837 psi at 23.75 Hz). The solid lines de-
pict the amplitude threshold value programmed in the software. Alarm threshold
in the low-frequency zone (10–50 Hz) was set for 0.5 psi (0 peak). Note that there
is actually a higher amplitude MFD in the 120- and 140-Hz range. However, as
can be seen, this mode’s amplitude is below the alarm threshold. Examination
of the data from all positions indicated that the problem was only experienced
in a single combustor can; the amplitude in other cans always remained below
the threshold value. Further investigation of the incidents with the plant historian
showed at one of the same instants that the dynamics spiked, a flashback thermo-
couple in the same combustor increased in temperature by 20 to 60◦ F, as shown
in Fig. 7.8. This figure shows combustor temperatures and pressure amplitude as
the engine is loaded, holding base load, and subsequently unloaded on operator
control.
The flashback thermocouple (Tfb ) is located in a high-velocity region of the
combustor, downstream of the premixers, and before the primary zone of the
combustor, which is in the same plane as the combustor-dynamics access point.
Alarm
Threshold
Upper line is peak hold
Lower line is live trace
Fig. 7.7 Dynamic spectrum with 23.75-Hz spike in alarm, amplitude is pounds per
square inch (0 peak).
Fig. 7.8 Energy correlation to low-level flashback as engine is loaded.

Fig. 7.9 Borescope pictures of combustor premixer showing flow obstructions in swirl
vanes.
This temperature rise by itself would not have triggered an alarm, because typical
alarm limit set points for these engines are about 200◦ F. Also the indication has to
stay above threshold for 30 s before the engine will auto unload. In this example,
there were no changes in emissions, exhaust, or blade-path temperatures. Without
the combustor dynamics alarm, there would have been no control indication that
something was wrong.
This signature has been seen several times in our fleet and is an indication
of either lean blowout or a flow disturbance in the combustor. Findings from
inspections include lock wire and other debris lodged in the swirlers, broken swirler
pins, and holes burned in the sides of the combustors. Two examples are shown
in Fig. 7.9. In one case a linkage assembly for a combustor bypass valve had
been installed one spline off, roughly 15% more open than the other valves. In
this particular case the airflow in the subject combustor was reduced, slowing
down the air velocity, and the conditions were favorable for true flashback. Note
that an increased-temperature indication is not always a real flashback. In the
preceding case, it was caused by a localized disturbance or recirculation caused
by an obstruction in the airflow at the premixers.
B. Pilot-Nozzle Cracking
In this example, the first indication that the operator received was an alarm
in the 100- to-500-Hz band on combustor 16. This alarm indication can be seen
in Fig. 7.10, which plots the pressure spectra for combustors 11–16. The alarm
indication showed a distinct spike at 156 Hz, which is slightly outside the nor-
mal 125- to 135-Hz frequency range of the system. Looking closer, there is
some 156-Hz contribution in combustor 15 and, although not shown, in com-
bustor 1 (which is adjacent to position 16). However, the 156 Hz is not visible in
combustors 11 through 14. The significance of this detail is that combustors are
acoustically coupled by means of cross-flame tubes and noise can be heard and
detected in the adjoining baskets. This observation is a very important in any part-
failure signature, rather than instrumentation or tuning, to distinguish a localized
event.
156 Hz component also seen

in adjacent combustor
Fig. 7.10 Initial pilot-nozzle failure spectrum without peak hold, amplitude in
position 16 is above threshold, and the adjacent combustor exhibits the same
frequency.
A review of engine data indicated that there was a 3- to 5-ppm increase in

NOx emissions and a gradual 45-deg increase in blade-path variance, defined as
the difference between the highest thermocouple reading and the overall average.
However, blade-path variance was still below alarm levels and would have gone
undetected until further damage was caused. Because of the CDM alarm and
subsequent engine-data analysis, combustors 1, 16, and 15 were borescoped, which
revealed that a pilot nozzle was broken in combustor 16 (Fig. 7.11). Additional
fuel was introduced into the combustor through a crack on the weld at the flange.
Had the condition been allowed to continue, it is possible that the broken part
could go downstream and cause catastrophic damage. Figure 7.12 shows an engine
borescope picture in which the pilot nozzle was in fact liberated into the gas stream
and, in this case, became lodged in the row 1 vanes, which is the stationary set of
airfoils just upstream of rotating turbine blades.
Cold Side Hot Side
Fillet Weld Crack Location

Butt Weld Crack Location
Fig. 7.11 Location of pilot nozzle cracked at the bolt flange.
The nozzle failed on the hot side at which the flange bolts onto the engine.
This area is subject to a cantilever motion and stresses. The crack initiated at the
base of the hot-side cantilever. Some studies have shown that the failure mode
was at 125 Hz, which is very close to the dominant operational range of this
combustor. Such pilot-nozzle failures are caused by high-vibratory motion during
engine operation. They have occurred on two separate styles of pilot nozzles, one
that has a fillet weld at hot-side–cold-side flange interface and one that has a butt
weld at the hot-side–cold-side flange interface. Both designs have had similar
failures, which were caused for the same reason.
Fig. 7.12 Borescope of a pilot nozzle lodged between row 1 turbine vanes.
Total outage time for this case was about 12 h. The pilot nozzle was replaced
and the engine was returned to service.
C. Transition-Piece Failure
In this combustor, the typical dominant frequency is about 135 to 145 Hz, which
can be seen in all the combustors (Fig. 7.13). Note, however, an additional spike at
225 Hz in combustors 4, 5, and 6. This frequency has higher amplitudes in position
5 than the adjoining combustors 4 and 6, but it is clearly more obvious than the
pilot-nozzle-failure example.
Typical spectrum distribution,

135 to 145 Hz
225 Hz contribution from

combustor 5
Fig. 7.13 Early indication of transition piece failure (combustor 5). Lines indicate
peak hold and live trace: Combustors 1 through 4. (Continued)
225 Hz signature of
combustor with cracked
transition piece
225 Hz contribution from

combustor 5
Fig. 7.13 (Continued) Early indication of transition piece failure (combustor 5). Lines
indicate peak hold and live trace: Combustors 5 through 8.
At this point, it is difficult to say how big the crack in the transition piece was. As
time progressed, this tone grew in amplitude, even when the load was reduced from
100% power. Shortly after this spectral snapshot the engine tripped on high blade-
path variance, this time because of a very-low-temperature reading. As the crack
propagates on the transition panel, it allows compressor discharge air to bypass the
head end of the combustor, thus cooling the exit of the damaged combustor and
showing up in the exhaust as a cold thermocouple or high-blade-path variance.
Figure 7.14 shows the as-found condition of the transition. A typical transition
failure of this type is initiated by low-cycle fatigue followed by high-cycle-fatigue
propagation. The combination of high steady-state stresses coupled with high-
dynamic stresses can cause quick and damaging failures.
Fig. 7.14 As-found condition of damaged transition piece.
References
1
and Power, Vol. 123, 2001, pp. 182–189.
2
Stuttaford, P., Martling, V., Green, A., and Lieuwen, T., “Combustion Noise Measure-
ment System for Low Emissions Combustor Performance Optimization and Health Moni-
toring,” American Society of Mechanical Engineers, Paper GT2003-38255, 2003.
3
Richards, G., and Straub, D., “Passive Control of Combustion Dynamics in Stationary
Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 795–810.
4
Mongia, C., Held, T., Hsiao, G., and Pandalai, R., “Challenges and Progress in Con-
trolling Dynamics in Gas Turbine Combustors,” Journal of Propulsion and Power, Vol. 19,
No. 5, 2003, pp. 822–829.
5
Hobson, D., Fackrell, J., and Hewitt, G., “Combustion Instabilities in Industrial Gas
Turbines-Measurements on Operating Plant and Thermoacoustic Modeling,” Journal of
Engineering for Gas Turbines and Power, Vol. 122, 2000, pp. 420–428.
6
McManus, K., and Lieuwen, T., “That Elusive Hum,” Mechanical Engineering Power,
June 2002.
Chapter 8
Monitoring Combustion Instabilities:

E.ON UK’s Experience
Catherine J. Goy,∗ Stuart R. James,∗ and Suzanne Rea∗

E.ON UK, Nottingham, England, United Kingdom
I. Introduction
P OWER Technology, the central engineering and scientific consultancy of

E.ON UK, supports the operation of generating plants for a wide range of
customers. Its condition-monitoring facility provides a daily engine-health screen-
ing service to many powerplants, including gas turbines. An important element of
the screening activity is focused on combustion dynamics because of the critical
impact that combustion dynamics can have on machine reliability and availability.
This chapter describes the equipment and methods used for monitoring combustion
dynamics, including several case studies that highlight the value of such systems.
II. Why Monitor Combustion Dynamics?

Operators of a gas turbine plant aim to achieve the most cost-effective utilization
possible. To meet increasingly stringent emissions legislation, manufacturers of
gas turbines have adopted a lean premixed approach to combustion.1 All modern
gas turbines fitted with these dry, low-emission (DLE) combustion systems are
susceptible to combustion dynamics.2−6 If left unchecked, the resulting pressure
pulsations can cause catastrophic damage to combustion hardware and to the down-
stream hot-section components. Combustion failures can, therefore, be extremely
costly, not only because of the cost of replacement hardware, but also in terms
of loss in electricity generation. Unplanned shutdowns incur additional financial
penalties in a dynamic energy-trading market because the shortfall in generation
must be brought in from other generators.
Coupled with the inherent sensitivity of DLE machines to combustion dynamics,
the increased machine flexibility required in today’s market increases the likeli-
hood of encountering problems related to dynamics.7 It is, therefore, beneficial to
Copyright c 2005 by E.ON UK. Published by the American Institute of Aeronautics and Astronau-
tics, Inc., with permission.
∗ Power Technology.
163
164 C. J. GOY ET AL.
Fig. 8.1 On-line combustion dynamics monitoring system schematic.
continuously monitor the health of the gas turbine and its combustion system to
obtain the earliest possible indication of deterioration or failure. Early detection
of abnormal operation enables commercially appropriate action to be planned,
whether this involves a machine inspection or the reoptimization of the combus-
tion system.
If just one major combustion failure can be prevented by regular health mon-
itoring, the resulting financial savings will far exceed the costs of installing the
required equipment.
III. Description of the On-Line Combustion-Monitoring System

Even when a combustion-dynamics monitoring system has not been provided
by the manufacturer, it is usually possible to install the necessary equipment. The
installation may basically involve the connection of a data-acquisition system to
the buffered voltage outputs from existing dynamic pressure transducers. If the
sensors are not available, then a full-system installation is required,8 as shown
schematically in Fig. 8.1.
A complete on-line combustion-dynamics health-monitoring and screening sys-
tem includes the following:
1) Dynamic pressure transducers. Sensors record the fluctuating pressures

within the combustion system to give a direct measurement of the combustion dy-
namics. The transducer ideally is positioned as close as possible to the combustion
space to minimize signal attenuation. It must, therefore, have high temperature
capability. A typical installation of a dynamic pressure transducer is shown in
Fig. 8.2.
2) Signal conditioners. If a piezo-electric sensor is used, the measured fluctuat-
ing pressure is converted to an electrical charge in the sensing head; however, this
signal may be transmitted only a short distance. Signal conditioners convert the
charge into a voltage to enable more distant signal transmission.
3) Signal processors. The voltage signal is further processed before it enters
a data-acquisition system that acquires raw time-varying data and performs fast
Fourier transform and averaging operations before sending the resulting data to a
database.
4) Data reduction. If the dominant dynamic modes of the combustion system
are known, the data can be further reduced by recording peak levels only. This
significantly reduces the storage requirement of the data-acquisition system.
MONITORING COMBUSTION INSTABILITIES 165
Fig. 8.2 Installed dynamic pressure transducer.
5) Remote data analysis. Data can be transferred by network or modem connec-

tion from the power station to feed into the integrated engine-health monitoring
software package developed in-house.
6) Automated data screening. Combustion dynamics information is merged with
plant-operating data, enabling analysis of dynamic response as a function of the
machine’s operating regime. Monitoring systems may also provide complementary
data on vibration, pyrometry, performance, and debris (Fig. 8.3). The integration of
data from all available sources improves the understanding of machine operation.
This improved understanding results in additional operator confidence when as-
sessing whether the gas turbine is operating normally. Analysis can be automated
when sufficient data characterizing the machine’s behavior have been acquired.
IV. Benefits of Combustion-Dynamics Monitoring

On-line combustion-dynamics monitoring has been carried out for the past
three years and, already, significant benefits have been realized. The number of
combustion failures on monitored machines has significantly declined. As a result,
the level of confidence in the systems has greatly improved. Scheduled borescope
inspections can now be more focused because the specific combustors and regions
of concern can be targeted, thus, reducing machine downtime. Because unsched-
uled shutdowns are particularly costly to generating companies, close, continuous
monitoring can be used to allow a deteriorating machine to be run until a commer-
cially suitable shutdown time. A load restriction can often be applied and additional
Power Station
Power Technology
Specialist Advice
On Line
Combustion
Dynamic data
On Line
Vibration
Data exported by
Wide Area
On Line Network
EDMS Specialist PC
Evaluation
and Optimisation
Plant data On Line
Performance
On Line
Pyrometry
EDMS – Engine Distress Monitoring System
Fig. 8.3 Integration of combustion dynamics with other monitoring systems.
alarms can be configured to provide greater protection during the period of oper-
ation until the shutdown can be arranged.
Combustion components were inspected during overhauls and detailed infor-
mation about their condition has been recorded. On the basis of these inspections,
correlations have been developed between individual modes of instability and dam-
age to specific combustion components. This information has been fed back into
the daily screening activity to focus on the most damaging modes and to fine-tune
alarm criteria.
V. Case Studies
Once a machine’s operating behavior is characterized, deviations from the nor-
mal dynamic response can quickly be identified. This characterization must take
place over an extended period to take account of variations caused by ambient
conditions and operating regimes. Subsequently discussed are examples of this
sensitivity in a heavy industrial gas turbine with can-annular combustion system
that is operating in a combined cycle application for power generation.
Combustion-dynamics monitoring also enables investigations of complete fail-
ures to be undertaken. The examples given here illustrate how this experience has
been used to avoid the recurrence of failure modes.
VI. Impact of Ambient Conditions on Dynamic Response

An understanding of the likely response of the combustion system to changing
ambient conditions is invaluable when optimizing the machine at any given time of
year. For example, Fig. 8.4 shows the response of two different dynamic combus-
tion modes as a function of gas turbine load and ambient temperature. Both repre-
sent axial modes of instability within the combustor and its air-delivery plenum.
Of the two modes, the lower-frequency mode (solid line) is essentially insensitive
12.00 400
Low f level Mid f level O Ambient T X Unit Load
Dynamic Pressure / Temperature
X XX X X XX X X X X X X X X X XX X 350
X X XXX XX O X X OX X XX
10.00
X X X X X X O X X XO X O
X X O O O OO X
XX X X O O O XX X X X
X XX O 300
X XX O O O X
X X
8.00
O O OO
O
Unit Load
250
O O O
O O
6.00 O 200
O O
O
150
4.00
O O
Mid frequency dynamics 100
2.00 dominate at lower loads

50
0.00 0
00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 03:00 06:00 09:00
Time
Fig. 8.4 Combustion dynamics as a function of load and ambient temperature.
to the range of conditions during the 33 low periods shown. However, the higher-
frequency mode (dashed line) is excited during part-load operation, and especially
at times of low ambient temperature. The amplitude of the peak instability occur-
ring between 0300 and 0600 hrs on the first day, as shown in Fig. 8.4, is more than
three times greater than the peak level recorded between 0300 and 0600 hrs on the
second day. The operating load is identical in the two cases; the only difference is
that the ambient temperature is significantly lower on the first day. The difference
in response of the two dynamic modes is a result of the impact of varying ambient
conditions on the coupling between the premixed flame and the combustion cham-
ber acoustics, as has been investigated elsewhere.9 This sensitivity must, therefore,
be kept in mind if combustion optimization is carried out on a warm day because
recorded dynamics levels would be minimal throughout an engine’s operating
range under warm ambient conditions. A safety margin must be maintained from
damaging levels of all dynamic modes and over all anticipated ambient conditions
in which the machine is expected to operate. Continuous combustion-dynamics
monitoring throughout the year provides the necessary understanding of the re-
sponse of each individual instability mode to a change in ambient temperature,
such that the combustion system can be optimized for dynamically quiet operation
at all times.
VII. Impact of Operating Regime on Dynamic Response

Increased machine flexibility is required in today’s dynamic market, and it is
important to understand the impact that different modes of operation may have on
the structural integrity of the combustion system. Figure 8.5 shows a combustor’s
dynamic response to a gas turbine operating in a frequency-responsive mode.
This describes the transient mode of machine operation in which the gas turbine
is continually responding, by either over- or underfiring to counteract fluctuations
in grid frequency. The response is achieved by continually varying the gas-turbine
Load Max Dynamics FSMode

FS Mode
230 3.5
210 3.0
Dynamic Pressure
190 2.5
GT Load
170 2.0
150 1.5
130 1.0
110 0.5
90 0.0
6-Oct AM 6-Oct PM 7-Oct AM 7-Oct PM 8-Oct AM 8-Oct PM

FS mode = 1 ACTIVE
FS mode = 0 INACTIVE
Fig. 8.5 Combustion dynamics as a function of load and frequency response.
load to modify the engine speed; therefore, the operation of the machine is never
steady. When operating in this much more transient, frequency-sensitive (FS)
mode (described as “FS mode = 1 ACTIVE” in Fig. 8.5), the dynamics levels
increase significantly, then gradually decay each time the machine returns to
steady-state operation.
VIII. Combustion Liner Failure

When damage (e.g., a crack in a combustion liner) occurs, changes also occur
in the dynamic signature. The growth of a crack results in compressor-delivery
air being routed through the wall of the liner, rather than through the premixers,
causing changes in the fuel–air distribution and mixing within the combustor and,
therefore, an alteration in the dynamic response. In many cases, signal power levels
increase at the onset of damage, although in some circumstances, a failure can have
a damping effect on the dynamic signature of the combustor. Moreover, changes
in the airflow distribution will alter temperatures within the combustor, affecting
the dominant frequencies and amplitudes related to that combustor. Tracking these
changes through continuous monitoring enables early detection of a growing crack
and potential catastrophic failure. When damage does occur, monitoring can pro-
vide forewarning of the failure. This is dramatically shown in a major failure of the
combustion liner shown in Fig. 8.6. A selection of the data collected during this
period is illustrated in Figs. 8.7 and 8.8. In this particular example, combustion
dynamics levels doubled the week before the failure. It is believed that the crack
had been developing during the week and opened up around midnight, as marked
by the sudden step change in dynamics levels at this time.
During this time, the combustor-related exhaust gas temperature dropped, indi-
cating that the effective area of the damaged combustor had increased as a result
of the presence of a crack. Finally, at 1400 hrs, the nitrous oxide levels increased
Fig. 8.6 Damaged combustion liner (view looking downstream, burners removed).
Dynamic pressure
Hot tone level NOx 140
6
120
Dynamic Pressure
5
100
4
NOx
80
3
60
2
40
1 20
0 0
1-Dec 2-Dec 3-Dec 4-Dec 5-Dec 6-Dec 7-Dec 8-Dec 9-Dec 10-Dec
Fig. 8.7 NOx and dynamics levels in the week leading up to the combustion liner
failure. (Note that the periodic spikes of NOx emissions correspond to monitoring
equipment calibrations.)
625 250
Can 8 Exhaust Temperature GT Load

620 200
Temperature
GT Load
615 150
610 100
605 50
600 0
08 Dec 08 Dec 09 Dec 09 Dec 09 Dec 09 Dec

12:00 18:00 00:00 06:00 12:00 18:00
Fig. 8.8 Combustor-related exhaust temperature during the combustion liner failure.
from about 20 mg/m3 to 180 mg/m3 . This change led to the decision to shut down
the unit. When the relevant liner was removed from the engine, it was found that a
crack had initially developed at a weld at the aft end, propagated around the weld,
then moved toward the forward end to form a flap of material that had become
partially detached (Fig. 8.6).
The ability to confirm, via reference to exhaust gas temperature data, the com-
bustor in which components are deteriorating is a valuable check. Mapping of
temperature variations back to the individual combustor at the root of the problem
Fig. 8.9 Failed burner assembly.

Failure
(High Frequency Mode)
Dynamic Pressure
0.00
0 50 100 150
Days
Fig. 8.10 Dynamics data from failed combustor in the 6 months before failure. (The
high-frequency, radial acoustic mode is shown here.)
helps to reduce repair downtime. Further indications of liner crack growth can be
found in changes in emissions. Low NOx depends on good mixing and distribu-
tion of fuel and air, both of which can be affected by liner damage, resulting in
significant variations in NOx emissions.
IX. Burner Assembly Failure

A very different catastrophic failure has been encountered on two burner
assemblies. The resulting damage to the worst affected component is shown in
Fig. 8.9. Once again, combustion dynamics data were a very useful part of the
failure investigation. Figure 8.10 shows the data from one of the failed combustors
during the six months leading to the failure, with a data interval of 20 min. There
were sporadic cases of high-frequency dynamics (caused by a radial instability
mode within the combustor) during the four to six month period before the failure
occurred. This was raised as a concern; however no data were available at the time
to correlate this dynamic mode to specific hardware damage. Not shown in Fig. 8.10
are the amplitudes of other combustion modes. They have been omitted for clarity
because the amplitudes of the lower-frequency dynamics were within acceptable
limits.
Figure 8.11 shows the dynamics data for the month leading to the failure. As
before, to present a clear picture, only the dynamics of interest are plotted. During
the month before the failure, no incidence of high-frequency dynamics occurred.
However, the levels of lower-frequency dynamics suddenly decreased at the time
of the failure, these levels are shown in Fig. 8.11. The reason for this decrease
can be clearly seen once the damage to the combustor is examined. An undam-
aged burner assembly is shown in Fig. 8.12 as compared with a failed component
(Fig. 8.9). Failure of the burner nozzles caused the loss of premixer tubes, resulting
in the reversion of a premixed flame to a diffusion flame. Because diffusion
flames tend to exhibit lower levels of combustion instability, the sudden transition
Step Decrease
(Lower Frequency Mode) .
In Dynamics At
Dynamic Pressure
Time Of Failure
0.00
125 135 145 155 165

Days
Fig. 8.11 Dynamics data from affected combustor in the month before failure.
Fig. 8.12 New burner nozzle assembly.

NOx EGT Spread Increase

In EGT
Increase
Spread
In NOx
Levels
EGT Spread
NOx
0 0
125 135 145 155 165

Days
Fig. 8.13 NOx levels and EGT spread in the six months before failure.
resulted in reduced lower-frequency dynamics. Although not shown, the levels of

other dynamic modes also fell.
The formation of a diffusion flame would be expected to result in an increase in
NOx emissions. Figure 8.13 shows that this was the case, although the increase is
quite small. This is the result of only one of the 18 combustors reverting to a diffu-
sion flame. Therefore, the high NOx emission arising from the failed combustor is
masked by the low emissions from the other 17 combustors. Because the dynamics
are measured within each combustor, it was easier to pinpoint the failed combustor
by using these measurements.
The exhaust gas temperature (EGT) spread is also shown in Fig. 8.13. The EGT
spread is the difference between the highest and lowest readings from the thermo-
couples that are situated around the circumference of the gas-turbine exhaust. The
step increase of the EGT spread at the time of the failure was caused by a decrease
in the temperature recorded by thermocouples related to the failed combustor. The
decrease in temperature was caused by an increase in the combustion airflow to
the failed burner assembly, which itself was a result of the increase in the effec-
tive area caused by the loss of material at the time of the failure. Although the
overall temperature from the failed combustor decreased at the time of the failure,
this decrease in overall temperature did not cause a decrease in the levels of NOx
produced from this combustor because of the presence of a diffusion flame that
resulted in higher local peak temperatures. Most of the excess combustion air now
passing through would not have taken part in the combustion process, rather it
would have passed around the diffusion flame.
Inspection of the affected components of the failed combustor revealed that
several welds attaching the premix tubes to the burner nozzle assemblies had failed.
The weld failures occurred on the two combustors that had exhibited the highest
levels of high-frequency dynamics. On the basis of this evidence, the investigation
team concluded that the failure of the premix tubes was a direct result of the
undesirable levels of high-frequency dynamics in the combustors.
Fig. 8.14 Damage to a burner assembly caused by high-frequency dynamics.
In another instance, the continuous monitoring system detected abnormally high

levels of high-frequency dynamics in a combustor. A borescope inspection revealed
that a fist-sized section of the burner assembly was detached, raising the real risk
of debris damage to the turbine (Fig. 8.14).
As a result of this investigation, upon occurrence of high-frequency dynam-
ics above a predetermined level, operators examine the combustion components
of their gas turbines for damage and take any necessary remedial action. This
investigation also concluded that a different combustion optimization philosophy
should be adopted to avoid areas of high-frequency dynamics.
Knowing that this particular high-frequency dynamic mode is extremely dam-
aging to the premixing region of the combustion system has prompted increased
interest in its occurrence during continuous monitoring. Subsequent experience has
shown that the high-frequency mode becomes dominant at low ambient tempera-
ture and during part-load operation (Fig. 8.15). In the example shown in Fig. 8.15,
Gas Turbine Load
Ambient Temperature
Dynamic Pressure
Time
Fig. 8.15 Impact of ambient temperature and gas turbine load on high-frequency
dynamics.
on detection of the high-frequency instability, the operator was immediately in-

structed to restrict the gas-turbine load to a dynamically quiet operating range and
to arrange a combustion system reoptimization as soon as possible. Consequently,
the machine operator was able to run the gas turbine without incident until its next
scheduled inspection.
X. Conclusion
On-line combustion-dynamics monitoring offers significant commercial advan-
tages to power generators by providing a tool with which greater gas-turbine
reliability and availability can be achieved. The principal purpose of this activity
is to make operators aware of occasions when their gas turbines are operating
in a potentially damaging way. This aim has been realized. Although the cost of
installing monitoring equipment may be high, the potential savings resulting from
the increased flexibility of operation and the avoidance of catastrophic failure far
outweigh the cost. As a result, combustion monitoring has become a valuable
component of the integrated health-screening activity.
In addition, dynamics data have been used retrospectively to aid in failure in-
vestigations. An understanding of the effects of specific combustion modes on
individual components has also been developed. This is continuously revisited to
provide the most up-to-date understanding possible.
References
1
Lefebvre, A. H., “Gas Turbine Combustion,” Taylor & Francis, Washington, DC, 1998.
2
Lieuwen, T., and Yang, V., “Combustion Instability in LPP Combustors,” Journal of
Propulsion and Power, Vol. 19, No. 5, 2003.
3
Willis, J. D., and Moran, A. J., “Industrial RB211 DLE Gas Turbine Combustion Up-
date,” American Society of Mechanical Engineers, Paper 2000-GT-109, May 2000.
4
Scarinci, T., and Halpin, J. L., “Industrial Trent Combustor–Combustion Noise Char-
acteristics,” Journal of Engineering of Gas Turbines and Power, Vol. 122, April 2000,
pp. 280–286.
5
Berenbrink, P., and Hoffmann, S., “Suppression of Dynamic Combustion Instabilities
by Passive and Active Means,” American Society of Mechanical Engineers, Paper 2000-
GT-079, May 2000.
6
Pandalai, R., and Mongia, H., “Combustion Instability Characteristics of Industrial En-
gine Dry Low Emissions Combustion Systems,” AIAA Joint Propulsion Conference, AIAA,
Cleveland, OH, Paper AIAA 98-3379, July 1998.
7
Colechin, M., Rea, S., Goy, C., and James, S., “On-line Combustion Monitoring on Dry
Low NOx Industrial Gas Turbines,” Institution of Mechanical Engineers Seminar, London,
Jan. 2003.
8
Rea, S., James, S., Goy, C., and Colechin, M., “On-line Combustion Monitoring on Dry
Low NOx Industrial Gas Turbines,” Measurement Science and Technology, Vol. 14, 2003,
pp. 1123–1130.
9
Janus, M., Richards, G., Yip, M., and Robey, E., “Effects of Ambient Conditions and
Fuel Composition on Combustion Stability,” American Society of Mechanical Engineers,
Paper 97-GT-266, June 1997.
III. Fundamental Processes and Mechanisms
Chapter 9
Combustion Instability Mechanisms

in Premixed Combustors
Sébastien Ducruix,∗ Thierry Schuller,† Daniel Durox,‡

and Sébastien Candel§
CNRS and Ecole Centrale Paris, Châtenay-Malabry, France
I. Introduction
C OMBUSTION instabilities constitute a central problem in many fields of ap-

plication from aerospace propulsion, gas turbines operating in the premixed
mode to domestic boilers and radiant heaters. Instabilities that result from resonant
interactions lead to oscillations of the flow, inducing many undesirable effects:
large-amplitude structural vibrations, increased heat fluxes at the system walls,
flashback, and flame blowoff. In some extreme cases, the outcome is a spectacular
failure. Much of the recent work in this field has relied on detailed experimentation
with advanced optical diagnostics and on numerical simulation tools. In general,
the objective of this work is to reveal the instability scenario and develop predictive
models for combustion-dynamic phenomena. Schematically, a driving process
generates perturbations of the flow, and a feedback process couples these perturba-
tions to the driving mechanism and produces the resonant interaction that may lead
to oscillations. The feedback (or coupling) process relates the downstream flow
to the upstream region where the perturbations are initiated. As a consequence,
acoustic-wave propagation is usually responsible for the feedback path. This
coupling process may also involve convective modes, like entropy waves, which
are associated with temperature fluctuations generated by the combustion process.
Vorticity convected by the flow may be part of the coupling process as well. When
such fluctuations in entropy or vorticity reach a nozzle on the downstream end of
the system, they are reflected in the form of upstream-propagating pressure waves.
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights
reserved.
∗ Research Scientist, Laboratoire EM2C.
† Assistant Professor, ECP, Laboratoire EM2C.
‡ Senior Research Engineer, Laboratoire EM2C.
§ Professor, ECP and Institut Universitaire de France, Laboratoire EM2C, Fellow AIAA.
179
180 S. DUCRUIX ET AL.
Acoustics
Atomization/ Flame
vaporization/ wall
Upstream mixing interactions Down-
dynamics stream
dynamics
Feed line Injection Stabilization Heat
dynamics Entropy Exhaust
release waves impedance
Impedance conditions
conditions Mixing Flame/
Organized
vortex vortex
structures interactions
Acoustics
Fig. 9.1 Basic interactions leading to combustion instabilities (from Ref. 7).
A variety of complex physical processes may then be involved in the develop-

ment of instabilities, depending on the system characteristics, operating conditions,
etc. Figure 9.1 synthesizes some of the interactions, that can participate in the
process. Extensive experimental and theoretical work has been performed to iden-
tify the fundamental mechanisms and devise analytical models. Some early ob-
servations are in a classical study by Mallard and Le Châtelier.1 An often-quoted
paper by Rayleigh2 establishes a criterion that oscillations are sustained when
heat release and pressure fluctuations are in phase. This criterion may be used to
investigate an unstable situation, but it does not allow predictions of combustion
instabilities. In many situations, neither the driving path, which leads from heat
release to acoustic fluctuations, nor the coupling path, which leads from acoustic
to heat-release fluctuations, are known. They are both crucial to predict the ampli-
fication or damping of an initial perturbation traveling in the system. Moreover,
the phase (or time delay) between acoustics and combustion plays a key role in
combustion instabilities, and this parameter is usually difficult to predict or even
measure in practical situations.
The different mechanisms susceptible to coupling involve time lags, because
reactants introduced in the chamber at one instant are converted into burnt gases at
a later time. Systems with delays are more readily unstable. This is easily shown
by considering a second-order model featuring a linear damping (second term)
and a restoring force with a delay (third term):
d2 x dx
+ 2ζ ω0 + ω02 x(t − τ ) = 0 (9.1)
dt 2 dt
Expanding Eq. (9.1) in a Taylor series to first order yields
d2 x dx
2
+ ω0 (2ζ − ω0 τ ) + ω02 x(t) = 0 (9.2)
dt dt
The damping coefficient is negative if ω0 τ > 2ζ . If the delay τ is long enough
with respect to the period T = 2π/ω0 , the amplitude of any perturbation will grow
exponentially. More generally, combustion instability occurs when the natural
COMBUSTION INSTABILITY MECHANISMS 181
resonant time of the flow configuration is commensurate with the characteristic

time of the combustion process.
It is thus important to understand the elementary processes of interaction be-
tween combustion and waves or flow perturbations (acoustics, convective modes,
injection inhomogeneities, etc.), which may become driving or coupling processes
under unstable conditions. No attempt will be made in what follows to describe all
the processes involved in combustion instabilities, because these processes have
already been reviewed extensively.3–8 Here, we will examine the aspects that typ-
ify what occurs in gas-turbine combustors. This chapter specifically focuses on
gaseous-fueled, premixed systems and uses simple and well-controlled situations,
which can be examined in detail to analyze the elementary processes.
One additional complication is that interactions in practical systems take place in
a complex configuration, and that the flow is, in most cases, turbulent and swirling.
Various groups are now making a large-scale effort to develop numerical tools for
combustion dynamics in such structures, based in particular on Large-Eddy Simu-
lation (LES) (see, for example, recent computations in Refs. 9–15). The numerical
tools of combustion dynamics are covered by Yang in this book and by reviews
in Refs. 7 and 16. Elementary processes like those described subsequently should
be carefully taken into account in comprehensive simulation tools; some of the
experiments reviewed in this chapter could clearly be validation cases for these
numerical tools. Ideas developed in simple cases can be transposed to the more
complex turbulent cases by noting that, in many circumstances, the occurrence of
instability is intimately related to large-scale motion or to organized convective
modes. Then, the random turbulent fluctuations corresponding to fine-grain turbu-
lence act as a noisy background to the unstable oscillation. When considering the
unstable process, one can focus on the organized motion, which is well illustrated
in laminar experiments. Other chapters in this book, like those by Lieuwen or
Dowling, provide further information on the dynamics of turbulent combustors of
the type used in gas-turbine systems (see also Refs. 17–19).
Some of many possible interactions that need to be examined are especially
relevant because they directly cause fluctuations in heat release or generate pres-
sure perturbations. According to Rayleigh’s criterion, these mechanisms can be
of great importance in the development of combustion instabilities. The follow-
ing processes will be considered in this chapter: 1) flame–vortex interactions;
2) acoustic–flame coupling; 3) interactions of perturbed flames with boundaries;
4) mutual flame annihilations; 5) flame response to incident composition inhomo-
geneities; and 6) unsteady strain rate effects. These processes, illustrated schemat-
ically in Fig. 9.2, only correspond to a few of the blocks in Fig. 9.1. Many other
interactions deserve attention and have already been surveyed in previous articles
and in other parts of this book. For each of these elementary processes, a driving
or coupling path is proposed as an example, relating heat release to acoustic vari-
ables (pressure, velocity) in the first case or leading from acoustic variables to heat
release in the other case. These links are illustrated by simple calculations and/or
data from well-controlled experiments.
As mentioned, characteristic times and delays associated with the elementary
processes are closely related to combustion instabilities. For example, convective
processes often induce the longest time lag in the system and are therefore central
in the analysis of the problem (see, for example, Refs. 20 or 21). The convective
F+O F+O Vortex

Flame ε(t)
F+O
Flame
a) b) P c) P
P
F+O Flame
P Plate
Flame Flame
d) P
Flame
Equivalence f) g)
ratio e) F+O Plane
perturbation acoustic
Plane Plane waves
acoustic acoustic
waves waves
Fig. 9.2 Elementary processes: a) Unsteady strained diffusion flame, b) flame roll up
in a vortex, c) premixed flame/vortex interaction, d) equivalence ratio perturbation
interacting with a premixed flame, e) acoustically modulated conical flame, f) acous-
tically modulated V-flame, and g) perturbed flame interacting with a plate (adapted
from Ref. 7).
process is exemplified in Fig. 9.3 (from Ref. 20), which gives an illustration of
Rayleigh’s criterion in a vortex-driven instability. Characteristic time delays consti-
tute a generic feature of combustion instability, and are introduced in the following
illustrations.
It is convenient to begin with a wave equation for reacting flows (Sec. II). This
wave equation for the pressure fluctuations in the system features an unsteady
heat-release source term. At this point, the problem is not completely solved,
because no simple expression exists that relates the heat-release fluctuations to
the acoustic variables (pressure, velocity). A classical representation based on the
(n − τ ) model is introduced, showing once again the importance of characteristic
time delays. This development indicates that an accurate description of the driving
path linking heat release to pressure fluctuations is crucial.
Fig. 9.3 Illustration of Rayleigh’s criterion (from Ref. 20).

Heat release is then considered as a pressure source in Sec. III, in which three
different situations are analyzed. The first is that of vortex-driven fluctuations, in
which vortices interact with a flame producing a heat-release pulse. This mech-
anism has been studied in many laminar and turbulent configurations, because it
constitutes a powerful driving process. In the second case studied in this section,
self-sustained oscillations of a flame impinging on a plate produce an intense radi-
ation of sound. This elementary process typifies heat-release fluctuations resulting
from flame–wall interactions. One may also infer from this example that similar
processes may take place as a result of mutual flame interactions in the core of
the flow. This third process effectively produces rapid changes of flame surface
area and correspondingly large fluctuations in heat release. These processes are
demonstrated by well-controlled model-scale experiments, but these mechanisms
are generic and probably drive many of the instabilities observed in larger-scale
combustors.
Section IV deals with heat-release fluctuations driven by waves or flow pertur-
bations. Three situations are envisaged. In the first situation, a conical flame is
modulated by acoustic waves. It is shown that the response of the flame may be
represented by a transfer function, which can be used to describe the stability map
of the burner. Experimental measurements of this transfer function are compared
with analytical estimates and numerical results. In the second situation, inhomo-
geneities formed in the upstream flow impinge on a flame producing a fluctuation
in heat release. The time delay between injection and combustion is the key pa-
rameter in the process and it defines conditions of oscillation. In the third situation,
heat-release perturbations result from a time-variable strain rate. A low-pass filter
behavior of the flame is found in this configuration. These three examples typify
interactions that may take place in practical systems. Many other cases are treated
in this book and in the references listed at the end of the chapter.
II. Acoustics for Reacting Flows

Combustion instabilities can be analyzed by starting from a wave equation
that relates the pressure field and source terms associated with heat release and
turbulence. This equation is briefly derived to highlight one of the relations that
exist between acoustics and combustion. More elaborate theoretical descriptions
of instabilities may be developed in various other ways, as exemplified in this book
or in the literature (see Ref. 22 for a review).
A. Role of Heat-Release Fluctuations

The following analysis provides a simplified framework for theoretical inves-
tigation of combustion oscillations; its intent is limited. A low-speed (low Mach
number) reactive flow is assumed, because this is the case in most combustors,
to minimize head losses. Aerodynamic sources of sound are neglected. A more
complete description of sound sources in reactive flows may be found in Ref. 23.
Our objective is to underline the role of heat-release fluctuations and to demon-
strate that the rate of change of these fluctuations acts as a source, driving pressure
waves in the system. Starting from the balance equations for a chemically reacting
mixture of N species24 and using various simplifications, one can derive a wave
equation for the logarithm of the pressure16 :

c2 d 1 d
∇ · ∇ln p − ln p = ∇ · (ρ −1 ∇ · τ )
γ dt γ dt

d 1 N N

− ∇ · λ∇T + − h k ẇk − ρYk c pk vk · ∇T
D
dt ρc p T k=1 k=1
d2
− (ln R) − ∇v : ∇v (9.3)
dt 2
where c designates the speed of sound; ρ, p, T , Yk , v, and vkD are the density,
pressure, temperature, species mass fractions, velocity, and diffusion velocity, re-
spectively. , c pk , γ , λ, and R designate the viscous dissipation function, specific
heats, specific heat ratio, heat conductivity, and gas constant. h k and ẇk are, re-
spectively, the specific enthalpies and rates of reaction.
In expressions similar to Eq. (9.3), the splitting of terms between the left- and
right-hand sides is somewhat arbitrary, because some of the terms in the right-
hand side describe features of the propagation of sound in the medium and should
then be included in the left-hand side. This point is discussed by Doak25 in the
context of aerodynamic sound and by Kotake26 in a study of combustion noise.
Nevertheless, it is useful to regard the terms appearing in the right-hand side of
Eq. (9.3) as the source terms generating the pressure waves in the reactive mixture.
In a turbulent reacting mixture, an order-of-magnitude analysis indicates that, in
low-speed combustors, the dominant source terms are associated with the chemical
heat-release fluctuations.26 Neglecting all other terms, one obtains

c2 d 1 d d 1 N
∇ · ∇ln p − ln p = dt ρc T h k ẇk (9.4)
γ dt γ dt p k=1
Considering low-speed reactive flows, the convective term in the material derivative
may be neglected d/dt ∼ ∂/∂t. Assuming, in addition, that the specific heat ratio
is constant, Eq. (9.4) becomes

∂ 2
∂ 1 N
∇ · c2 ∇ln p − 2 ln p = h k ẇk (9.5)
∂t ∂t ρcv T k=1
This equation is not linearized, and it can be used to describe finite amplitude waves.
However, in many circumstances, the wave amplitude is relatively weak, and lin-
earization is appropriate. The pressure is then expressed as a sum of a mean and
fluctuating components: p = p0 + p1 with p1 / p0 1. Then, ln p p1 / p0 and
Eq. (9.5) becomes

p1 ∂ 2 p1 ∂ 1 N
∇·c ∇2
− 2 = h k ẇk (9.6)
p0 ∂t p0 ∂t ρcv T k=1
In practical continuous-combustion devices, the mean pressure does not change

by more than a few percent, the spatial derivatives of p0 may be neglected, and,
hence, Eq. (9.6) may be written as

∂2 ∂ N
∇ · c ∇ p1 − 2 p1 =
2
(γ − 1) h k ẇk (9.7)
∂t ∂t k=1
In addition to Eq. (9.7), an expression is needed for the acoustic velocity. This
expression can be obtained by linearizing the momentum equation and neglecting
the viscous stresses. This yields
∂v1 1
= − ∇ p1 (9.8)
∂t ρ0
where v1 represents the velocity fluctuations. Equations (9.7) and (9.8) describe
the propagation and generation of small perturbations in the reactive mixture. As
already mentioned, the problem is not completely solved, because a third relation
between heat release ( h k ẇk ) and acoustic fluctuations ( p1 , v1 ) is necessary to
close the system. This last relation can be deduced from experiments, theories,
or simulations. Section II.B gives an example of an analytical model using the
so-called (n − τ ) model.
Considering again the source term corresponding to the nonsteady heat release,
one may assume for simplicity that the chemical change occurs by a single-step
reaction. Then, if
h ◦f designates the change of formation enthalpy per unit mass
of the mixture, and if ẇ represents the rate of reaction, the chemical source term
becomes (∂/∂t)(γ − 1)(−
h ◦f )ẇ. In most cases, the only time dependence in this
expression is a result of the nonsteady rate of reaction, and, as a consequence, the
acoustic source term associated with chemical reaction may be written in the form:
∂ Q 1m
(γ − 1) (9.9)
∂t
where Q 1m represents the nonsteady rate of heat release per unit mass of mixture.
The wave equation (9.7) and the source term (9.9) indicate that the pressure field
is driven by the nonsteady release of heat. A coupled motion can take place if this
last quantity is influenced by acoustic variables, pressure, or velocity.
B. Case of a Compact Flame in a Duct

To get an understanding of the relation between heat-release fluctuations and
acoustic perturbations, it is instructive to consider the flow of a combustible mix-
ture through a long duct (an acoustic resonator), with a flame stabilized at the
axial location x = a, as shown in Fig. 9.4 (adapted from Ref. 4). In this develop-
ment, the following assumptions are made: 1) Acoustic wave frequencies are low
compared with the duct-cutoff frequency, and the perturbed motion corresponds
to plane waves propagating in the axial direction. 2) The flame thickness is small
compared with the acoustic wavelength, so that the region of heat release may be
Fig. 9.4 Sketch of the model compact flame geometry. The flame zone is thin
compared to the acoustic wavelength. Arrows A, B, C and D indicate acoustic
waves propagating in the system. Flame is assumed to be located at x = a (adapted
from Ref. 4).
approximated by a thin sheet located at x = a. The portion of the duct upstream

of the flame holder is denoted as region 1, with a fresh gas density ρ f and sound
speed c f . Region 2 corresponds to the downstream side of the flame holder, with
a burnt-gas density ρb and sound speed cb . The acoustic velocity v is easily ex-
pressed in terms of upstream- and downstream-propagating waves. Combustion
acts as a velocity source term because of the strong dilatation associated with heat
release. This effect may be quantified by integrating the wave equation (9.7) over
a thin control volume containing the flame. This leads to (see Ref. 16 for details)
Q 1a
vb (a+ , t) − v f (a− , t) = (γ − 1) (9.10)
ρ f c2f
where Q 1a represents the instantaneous heat-release rate per unit area. When the
flame is compact, the nonsteady release of heat determines the jump in acoustic
velocities.
The determination of Q 1a as a function of the perturbed motion is by no means
trivial. One has to relate the time-varying flow variables and the dynamic response
of the flame. One approach27–29 uses a time-lag hypothesis to express Q 1a in terms
of the time-delayed upstream velocity perturbation,
Q 1a
(γ − 1) = nv f (a− , t − τ ) (9.11)
ρ f c2f
where n is an interaction index, and τ represents a time lag. The heat-release term
is modeled as a function of an acoustic-wave variable alone. This approach is
clearly a simplified representation of more complex processes involving the flow,
turbulence dynamics and large-scale motions, flame interactions with neighboring
flames and walls, heat transfer at the boundaries, etc. Some of these processes are
described subsequently.
The value of the time lag τ relative to the frequency often defines ranges of
instability.30, 31 This value is recognized from the early work on rocket engine
instability29 (see also the review in Ref. 22) and from many recent studies. One
possible use of expressions like Eq. (9.11) is reduced modeling of active control.
As shown, for example, in Ref. 4, this modeling yields simple time-lag conditions
for instability development and control. In what follows, time lags will be analyzed
in various laboratory-scale situations.
III. Heat Release as a Pressure Source

It is worth examining the elementary processes in which heat release acts as a
pressure source term. As already mentioned, this is meant to be an illustration of
more complex gas-turbine combustion dynamics. Vortex structures drive various
types of combustion instabilities. In many premixed systems, the ignition and
delayed combustion of these structures constitute the mechanism that feeds energy
into the oscillation. This mechanism is analyzed and illustrations are given. Section
III.B is devoted to the interaction of a flame with a wall. This mechanism may not
be of major importance for gas-turbine combustors, but it serves to show that
rapid changes in flame surface area can induce heat-release fluctuations, which in
turn may feed energy into the pressure field. This process, illustrated with flame–
wall interactions, may also result from mutual interactions between neighboring
flame elements in the core of the flow, which may certainly arise in gas-turbine
combustors. Mutual interactions will be briefly analyzed in the third subsection.
A. Flame–Vortex Interactions
Flame–vortex interactions have been observed in many unstable combustion
systems. Two distinct mechanisms are usually involved. In the first, the flame
area is rapidly changing because of vortex roll up.20, 32 In the second, the vortex
interacts with a wall or another structure, which induces a sudden ignition of fresh
material.21
Vortex roll up often controls the mixing of fresh gases into the burning regions.
This roll up determines the nonsteady rate of conversion of reactants in the flow
and the amplitude of the pressure pulse resulting from the vortex burnout. When
the flame is rolled up, the surface area increases rapidly. The growth is limited
by flame shortening, which results from interactions of neighboring elements, and
consumption of the reactants entrained by the vortex. Such rapid variations of
flame surface correspond to the first mechanism. Flame–vortex dynamics have
been studied extensively (see Ref. 33 for a review). Much of the experimental
work has concerned toroidal or pairs of counter-rotating vortices running into a
traveling premixed flame34 or an established strained diffusion flame,35 which do
not quite correspond to situations of interest in combustion instability.
Observations of combustion oscillations indicate that vortex roll up takes place
while the flame develops. The vortex entrains fresh materials and hot products and
ignites at a later time, producing a pulse, which feeds energy in one of the resonant
modes of the combustor.36, 37 This process is more difficult to study experimentally
and is less well documented. Interactions between adjacent reactive vortices may
also take place, leading to formation of fine-grain turbulence.
Many studies have also focused on the natural instabilities of wakes and jets.38–41
Indeed, the vortical structures that are involved in the flame–vortex interactions
are often naturally generated and shed at dump planes. When one of the natural
frequencies of the jet matches one of the acoustic resonance frequencies, that
is, when the characteristic times are close, the coupling between acoustics and
combustion is made easier. An alternative way to study these situations is to force
the flow by using driver units or pistons and to analyze the response sensitivity of
the jet or wake to this forcing.42–46 The conversion of energy between vortices and
acoustics is also of great importance and has been studied in nonreactive47, 48 and
reactive situations.33
One example of self-sustained oscillations controlled by vortices is reported in
Ref. 36. A multiple-inlet combustor is fed with a mixture of air and propane, and it
features a dump plane (Fig. 9.5). The low-frequency instability observed in this case
Fig. 9.5 Geometry of the multiple-flame-holder dump combustor studied in Ref. 36.
Spark-schlieren photograph of the central jet for the 530 Hz unstable regime.
is acoustically coupled and occurs at one eigenfrequency of the system. The flame
visualization of Fig. 9.5 clearly shows that the largest-amplitude oscillations are
vortex driven. The following processes are involved: 1) A vortex is shed at the dump
plane when the velocity perturbation is maximum (v1 −→ Ω1 in a driving path,
where Ω1 represents vorticity fluctuations). 2) The vortex is convected, accelerated,
and entrains hot gases from its surroundings. A combustion pulse is produced when
two adjacent vortices interact, creating a large amount of small-scale turbulence
and flame surface area. 3) The sudden heat release constitutes a source that feeds
energy into the perturbed acoustic motion. Self-sustained oscillations can only
occur when the processes are correctly phased, that is, when the convective time
lag is in a suitable range with respect to the period of the motion as shown in Fig. 9.6.
The self-sustained oscillations of a laminar V-flame interacting with vortices are
analyzed in Refs. 20, 36, and 49.
Collisions of reacting vortices with boundaries are less well covered but are
often observed in premixed devices.21, 50, 51 A mechanism of this type is featured
in Ref. 50. A vortex is shed from a single inlet into a dump combustor (Fig. 9.7,
from Ref. 51). The vortices are synchronized by one of the longitudinal modes of
the system. Figure 9.7 shows a typical vortex-shedding event and the heat-release
distribution at a later time. Reducing the height of the combustor enhances the
interaction between the vortex and the lateral boundary, which produces longer
axial burning regions and augments the overall straining of the vortex. Fast burning
of the fresh reactants entrained by the structure takes place when the vortex collides
with the wall. The general process of flame interactions with boundaries is detailed
in the next section, because it can constitute a source of heat-release fluctuations
even in the absence of vortex shedding.
In the first situation described in this section, the mechanism involves flow
perturbations producing vorticity, which results in rapid changes of flame area,
inducing a heat-release pulse. In the second case, the vorticity directly causes a
volumetric expansion, leading to the heat-release pulse. This may be represented
globally by the following expression:
1 −→ Q 1 −→ p1
Many other studies21, 37 have revealed the key role of vortex structures. Coming
back to the theoretical expressions of Sec. II, one clearly sees that a model giving Q 1
as a function of 1 would make it possible to solve the complete set of equations.
Because the process involves convection, ignition, and combustion delays, one
may try an (n − τ ) formulation. This, however, requires further analysis of the
elementary steps, leading from vortex shedding to vortex burning.
B. Interactions of Flames with Boundaries

Interactions of flames with solid walls constitute a source of heat-release fluctu-
ations.52–54 Under certain conditions, such interactions can lead to self-sustained
oscillations, which are briefly described in what follows. These experiments in-
dicate that heat-release fluctuations of large amplitude can be induced by rapid
changes of flame area and that these fluctuations generate an intense sound field.
Fig. 9.6 Two-dimensional combustion tunnel facility studied in Ref. 20. Sequence of
phase-locked schlieren photographs of the combustor during an unstable mode oper-
ation. The trace shown is the pressure record of one instability cycle. The photographs
are taken at the respective phases marked on the trace.
Fig. 9.7 Schlieren photograph of a vortical structure entering a dump plane combus-
tor and chemiluminescence image representing the heat release rate distribution at a
later time during the instability cycle (from Ref. 51).
In the driving path, surface-area fluctuations produce nonsteady heat release, which
induces acoustic pressure radiation. It may be represented schematically by
A1 −→ Q 1 −→ p1
This sequence may be used to model the instability mechanism. It requires an

analytical description of the interactions between the flame and the wall and an
expression of the noise generated by the flame. Eventually, the determination of
the associated time lags defines regions of instability.
In experiments reported in Ref. 52, a laminar premixed flame, anchored on a
cylindrical burner, impinges on a horizontal plate, and a driver unit modulates the
upstream flow. This forcing generates perturbations in flame-surface area and heat
release (coupling path, typically ( p1 , v1 ) −→ Q 1 ). The driving path (typically,
Q 1 −→ ( p1 , v1 )) is also easy to characterize, because the sound produced by the
thermocouple cooled plate

LDV
M3
PM
zone 2
CH*
filter
zone 1
mixture of
gases
M0
loudspeaker
Fig. 9.8 Schematic view of the experimental setup used to study interactions of a
perturbed flame with a cooled wall. This configuration radiates an intense acoustic
field (from Ref. 53).
system is 10–20 dB higher than that emitted by a free flame submitted to the
same modulation, without the plate. The interaction of the flame with the plate
leads to rapid changes of the flame surface, which constitutes a major source of
sound in this situation. This well-controlled experiment typifies more complicated
situations in which the flame spreads in a chamber (as in a gas-turbine combustor)
and can produce pressure oscillations when impinging on the walls or on adjacent
flame sheets. When the phase is suitable, and when the gain exceeds the losses,
the oscillation may reach large amplitudes leading to instability.
It is also possible to observe self-sustained oscillations of a flame impinging
on a plate. The experimental setup is similar to that used in Ref. 52 but the driver
unit is removed (Fig. 9.8).53 A 10-mm-thick water-cooled plate, which can move
vertically, is placed above the cylindrical burner. An oscillation develops naturally
if both the driving and coupling paths are present in the system. For certain plate-
to-burner distances, intense emission of sound is observed. Figure 9.9 shows the
steady flame (a), when no sound emission is observed, and a complete cycle of
oscillation (b–e), when the instability is triggered. The flame front is undulated
by the perturbation, which is convected from the burner rim to the plate. These
visualizations are close to those obtained in the external modulation case described
previously. The sound emitted features many harmonics, with a fundamental fre-
quency at about 200 Hz.
The burner behaves like a Helmholtz resonator with a resonance frequency of
200 Hz. The resonant behavior of the system may be described analytically by
combining a model for the flame interaction with the plate and a representation of
the burner acoustics. The acoustic velocity v1 and pressure p1 at the burner exit
may be related by a second-order equation53 :
d2 v1 dv1 d p1
M 2
+R + kv1 = −S1 (9.12)
dt dt dt
Fig. 9.9 Different views of a flame interacting with a wall: a) Steady state; b–e)
Instantaneous images of the flame during an instability cycle (from Ref. 53).
3 3
CH*
v1 (m/s) - I(CH*) - Mic
d(I(CH*))/dt (arb. unit)

2 2
LDV 1
1
Micro M 3 d(I(CH*))/dt
0 0
-1 -1
0 5 10 15 20
Time (ms)
Fig. 9.10 Self-sustained oscillations of a flame interacting with a plate. Simultaneous

measurements of the velocity v1 at the burner outlet, of the CH∗ emission and of the
pressure p∞ signals (from Ref. 53).
where R is the system damping, and k is the stiffness of the gas volume acting
as a restoring force on the effective mass of air M. According to Eq. (9.12), the
resonator is driven by external pressure fluctuations p1 at the burner outlet. To
pursue the analysis, it is necessary to model the driving process appearing in the
right-hand side of Eq. (9.12).
The formulation is based on the following considerations. First, the source term
on the right-hand side of Eq. (9.12) originates from rapid changes of the flame
surface and subsequent noise radiation. Noise is generated when large portions of
the flame collapse because of interaction with and quenching by the plate. The
pressure field radiated by a compact source of nonsteady heat release takes the
form55–57

ρ∞ ρ f dQ
p∞ (r, t) = −1 (9.13)
4πr ρb dt t−τa
In this equation, ρ∞ , ρ f , ρb are the densities in the far-field air, the fresh gas, and
the burned gas, respectively; τa is the time required by sound propagation over
a distance r from the sources to the detector. In gaseous premixed flames, the
far-field radiated pressure p∞ can be related to the time-retarded rate of change of
the flame-surface area A:

ρ∞ ρ f dA
p∞ (r, t) = − 1 SL (9.14)
4πr ρb dt t−τa
where SL is the laminar burning velocity. The fast rate of extinction of the flame
area at the cold boundary induces a significant acoustic pressure radiation, which
shown in Fig. 9.10, where p∞ is measured by a microphone. The time derivative
of the heat-release signal is shown at the bottom of the graph. This signal nearly
coincides with the pressure signal detected by the microphone.
Next, it is important to relate the flame-surface area A to the velocity perturba-

tions at the burner outlet. This coupling (or feedback) mechanism logically belongs
in the next section, but it is envisaged here to complete the stability analysis of
Eq. (9.12). Fluctuations of the flame-surface area A are induced by velocity pertur-
bations at the burner exhaust. Flame perturbations, caused by velocity fluctuations
v1 at the burner outlet, are convected along the flame front toward the plate, which
can be modeled by
A(t) = n [v1 ]t−τc (9.15)
where n characterizes the coupling between the surface fluctuations and the veloc-
ity perturbations, and τc is the time required by convection from the burner lip to
the plate. This (n − τ )-like formulation is supported by the detailed experiments
carried out in Ref. 52. Expressions (9.12) and (9.15) may be combined yielding a
second-order equation for the velocity fluctuations:
2
d2 v1 dv1 d v1
+ 2δ + ω0 v1 = −N
2
(9.16)
dt 2 dt dt 2 t−τ
where N is a normalized combustion–acoustics interaction factor, and τ = τa + τc

is a global time delay. It is shown in Ref. 53 that this model correctly retrieves the
phase relations between the various signals and reproduces the shift in frequency
observed in the experiments, when the burner-to-plate distance is varied. This
demonstration confirms the existence of a mechanism whereby interactions of the
flame with the wall produce high rates of surface changes, which in turn generate
an intense pressure field. The stability map of the system can be determined by
considering the time delays involved and the detailed balance between gain and
losses in the process.
C. Mutual Flame Annihilation

The rapid consumption of reactants trapped between two adjacent flames may
also produce a heat-release pulse and the subsequent emission of pressure waves.58
If this interaction is properly phased with respect to an acoustic eigenmode, it may
drive the unstable motion. It is illustrated in another laminar experiment,59 in
which a central rod is placed in the burner and the flame is anchored on the burner
rim and on the central rod. The flame takes an “M” shape (one may also speak
of a “fountain” flame in this case, as shown in Fig. 9.11). This configuration is
well suited to interaction studies of adjacent flame-front elements, which may take
place between neighboring branches of the “M” shape.59, 60
The case presented corresponds to an equivalence ratio = 1.04, a mixture
flow velocity v̄ = 1.89 m s−1 , a modulation level fixed to vrms = 0.5 m s−1 , and
a modulation frequency f = 150 Hz. The description of the flame motion over a
cycle of excitation starts as in the flame–plate interaction. A velocity perturbation
originates at the burner lips and produces a deformation of the flame front at the
base of the burner (Fig. 9.11a). The perturbation mostly affects the outer branch of
the “M” flame. It is then convected by the mean flow toward the top of the flame
a) b)
c) d)
Fig. 9.11 Visualizations of the flame-flame interaction. Four different instants

of a cycle (clockwise time sequence). Φ = 1.04, v̄ = 1.89 m s−1 , f = 150 Hz, v = 0.5
ms−1 (from Ref. 59).
(Fig. 9.11b). As the deformation travels along the flame front, the two branches
of the “M” are stretched in the vertical direction and get closer (Fig. 9.11c), up
to an instant in the cycle where the flame-surface area is maximum, and two
flame elements interact (Fig. 9.11d). The outcome of this mutual annihilation de-
pends on the spatial position of the first interaction. In some cases, pockets of
fresh reactants may be trapped in a torus, but in other cases this will not oc-
cur.59 For some operating conditions not shown here, up to two flame tori can be
produced.
During interaction of these flame elements, the shape of the reactive front under-
goes a strong alteration. As in the flame–plate situation, after the mutual interaction,
the flame quickly retrieves its initial shape at the beginning of the following cycle
(Fig. 9.11a). In this cycle, the short phase of flame-surface destruction produces a
faster rate of change of the flame-surface area than the longer phase of flame-surface
production by stretch. The same mechanism operates as in the flame–plate
interaction, except that flame-surface destruction is produced by mutual anni-

hilation of neighboring front elements and not by thermal losses. The path is also
similar:
A1 −→ Q 1 −→ p1
The overall sound-pressure level is considerably enhanced. The pressure spectrum

is quite similar to that associated with the flame–plate interaction, with many
harmonics of the fundamental frequency indicating that the pressure signal is
periodic, but that the wave shape is nonlinear with a rich harmonic content. These
energetic harmonics indicate that the physical process, which is at the origin of
the noise, involves a rapid change of the rate of heat release.
Because mutual flame annihilation is believed to control and limit flame-surface
area in turbulent combustion, the previous findings suggest that this mechanism
could also be an important source of noise in turbulent combustors.
IV. Heat-Release Fluctuations Driven by Waves

Unsteady fluctuations in pressure, temperature, strain rate, induced curvature,
and chemical composition directly influence the rate of reaction in the flame. Of
course, pressure, temperature, or composition have a direct effect on the kinetics
of the system, but these usually produce weak effects. Pressure and temperature ef-
fects are considered, for example, by McIntosh61, 62 and Edwards et al.,63 whereas
Park et al.64 deals with the response of a distributed reaction zone to incident
waves. However, unsteady changes in the rates of conversion in the local flame
elements or in the available flame-surface area are probably more relevant. These
unsteady changes are illustrated here by considering heat-release fluctuations in-
duced by various perturbations; acoustic waves, equivalence ratio inhomogeneities,
and unsteady strain rates are successively discussed. These perturbations are not
the only possible sources of coupling (or feedback), which may drive heat-release
fluctuations, but they are most significant.
A. Modulated Conical Flames

The coupling may be represented schematically by
p1 −→ v1 −→ A1 −→ Q 1
This path may be investigated by modulating an initially stable flame by acous-

tic waves. As mentioned in Sec. III, the forcing technique is used to study the
presence of a coupling mechanism. If the geometry is simple enough, one may
determine the flame response to incident perturbations. Early investigations of this
type were proposed, for example by Markstein,65 Blackshear,66 and De Sœte.67
If the process remains in the linear regime, one may define a transfer function
between the incident velocity fluctuations and the nonsteady heat release, which
will depend on the burner geometry, operating parameters, and steady-state flame
configuration.
Transfer functions proposed in Refs. 68–71 indicated that the flame usually
behaves like a low-pass filter, providing a qualitative representation of the flame
response. Laminar conical flames are considered by Blackshear,66 De Sœte,67 and,
more recently, Baillot et al.72 Further theoretical efforts by Fleifil et al.73 and a
combination of theoretical analysis and detailed measurements by Ducruix et al.74
have advanced the status of this basic problem. More recent work by Schuller et al.75
provides additional clues on flame response in the high-frequency range. Dowling76
uses a similar approach to derive a model for the low-frequency nonlinear response
of a ducted V-flame in a geometry close to that considered in an earlier work by
Marble and Candel.77
The modulation of a conical flame is now considered in further detail. Our
objective is to describe the unsteady rate of heat release as a function of acoustic
variables. A laminar premixed flame is anchored on a cylindrical burner, and it is
submitted to acoustic waves generated by a loudspeaker placed at the bottom of the
burner. The flame response is driven by the acoustic velocity, and the aim is to find
the transfer function between heat-release fluctuations and velocity modulations:
Q 1 (ω)/Q 0
F(ω) = (9.17)
v1 (ω)/v0
where ω is the angular frequency of the modulation. In the linear range, the transfer
function is a good representation of the relation leading from acoustic variables to
heat-release fluctuations. Nonlinear effects will not be considered here, but they
are examined in many references.76, 78, 79 The modulus of F gives the amplitude of
heat-release fluctuations as a function of velocity modulations, whereas its phase
characterizes the time lag existing between velocity and heat-release fluctuations.
A complete analysis of this problem can be found in Refs. 74 and 80. Selected
results are highlighted subsequently.
The burner consists of a converging nozzle, which is water cooled, and a 120-
mm-long cylindrical tube, placed upstream from the nozzle and containing various
grids and honeycombs to produce a laminar flow. The conical flame is stabilized on
a 22-mm-diam burner rim. A driver unit placed at the base of the burner generates
perturbations, which wrinkle the flame front. The shape of the perturbed flame
depends on the frequency and amplitude of modulation. The typical flame shapes
displayed in Fig. 9.12 are visualized with a four-color schlieren technique. The
use of modern diagnostic techniques [particle-imaging velocimetry (PIV), instan-
taneous visualizations using intensified cameras, etc.] has provided new informa-
tion concerning the geometry of the flame front, the local and global heat-release
rates,74 and the velocity field at the burner exhaust and in the flowfield.80 This has
allowed direct measurements of the flame-transfer function defined by Eq. (9.17).
These measurements can be compared with theoretical and numerical predictions.
An analytical transfer function can be derived by decomposing the flow in mean
and perturbed components. The geometry of the problem is sketched in Fig. 9.13.
A G equation is used to describe the flame position:
∂G
+ v · ∇G = −S D |∇G| (9.18)
∂t
40 40
30 30
y (mm)
y (mm)
20 20
10 10
0 0
-20 -10 0 10 20 -20 -10 0 10 20
x (mm) x (mm)
Fig. 9.12 Methane air conical flame modulated by longitudinal acoustic perturba-
tions. fe = 150.5 Hz, ω∗ 28, v̄ = 1.44 ms−1 , v /v̄ = 0.13, Φ = 1.05. Top: schlieren
images for two different instants. Bottom: corresponding numerical simulations
(adapted from Ref. 80).
y
v
burnt
u gases
αo
ds
n r fresh
L gases
ηo (r) η (r,t)
r
-R (a) (b) R
Fig. 9.13 Geometry of a) the conical flame in the steady situation and b) in the
perturbed case; from Ref. 74.
where v = (u, v) is the velocity vector, and S D is the flame-displacement speed.

In what follows, S D is assumed to be a constant and equal to the laminar burning
velocity SL .
The G variable increases from the fresh mixture to the burnt gases, and one
contour G = G 0 represents the flame. In the simplest velocity-perturbation model,
the radial component u 1 is supposed to be negligible, compared with the vertical
component v1 , which is assumed to be uniform and sinusoidal: v1 = v1 cos ωt.
This corresponds to a bulk motion of the fresh stream. Figure 9.13 shows that G
may be replaced by η − y, where η designates the flame position. Substituting η =
η0 + η1 in Eq. (9.18), where η0 represents the steady flame shape, and considering
small perturbations η1 (Fig. 9.13b), one may expand the resulting equation to the
first order and obtain
∂η1 ∂η1
= SL cos α0 + v1 (9.19)
∂t ∂r
where α0 denotes the half-angle of the steady flame cone.

The heat-release fluctuations may be evaluated from the flame-surface varia-
tions:
R
A1 = 2π cos α0 η1 dr (9.20)
0
Heat-release fluctuations Q 1 are directly related to the fluctuations of flame-surface

area: Q 1 = ρ f SL q A1 , where ρ f is the unburnt-gas density, and q designates the
heat release per unit mass of mixture. Some calculations yield the following ex-
pression for the relative heat-release fluctuations74 :
Q1 v1 2
= [(1 − cos ω∗ ) cos (ωt) + (ω∗ − sin ω∗ ) sin (ωt)] (9.21)
Q0 v0 ω∗2
and the transfer function is easily deduced therefrom. The resulting expression
depends on a reduced frequency ω∗ = ω R/(SL cos α0 ), where R is the burner
radius. Expression (9.21) may be used as a source term in wave equation (9.7),
providing a complete dynamical description of a system featuring an initially
conical flame.
The analytical flame response to acoustic modulations obtained in this way relies
on many simplifying assumptions. It was assumed that the perturbed velocity is
axial and uniform. Data obtained with PIV80 show that this assumption may be
acceptable for weakly wrinkled flames, with a small velocity radial component, that
is, in the low-frequency range (ω∗ < 2). In this case, the flame responds as if it were
globally stretched and compressed by the modulation while keeping an essentially
conical shape. In contrast, these assumptions are too strong for larger frequencies
to correctly represent the acoustic–flame interactions. In that range, the velocity
field convects structures with important gradients and a radial component exists
near the burner exhaust, which clearly shows that simplified low-order models
have a limited range of validity.
1.0
(Qrms/Q)/(vrms/v)
0.8
0.6
0.4
0.2
0
v=0.97m/s
v=1.22m/s
phase difference (rad)
v=1.70m/s
model A
4 model B
0
1 10 20 30
ω*
Fig. 9.14 Comparisons between calculations (solid line), analytical results (dashed
and solid line) and measurements (symbols) for the transfer function of a conical flame
(from Ref. 80).
An alternative model represents the convective nature of the perturbed mo-

tion and the related phase differences. This alternative model is developed in
Ref. 80, where a revised formulation of the velocity-modulation incident on the
flame is proposed. This formulation is combined with G equation (9.18), and a
level-set approach is then used in the numerical integration of this equation. Typical
results of calculations shown in Fig. 9.12 are very close to the experimental flame
shapes. In Fig. 9.12, the perturbation velocity is axial and uniform in model A (bulk
perturbation model) and convected in the axial direction in model B (convective
perturbation model). The experimental and model A transfer-function amplitudes
essentially agree, but this is not the case for the phase (Fig. 9.14). The experimen-
tal phase increases with frequency, whereas the theoretical phase corresponding
to expression (9.21) tends to π/2. With model B for the velocity perturbation, the
results are notably improved.
It is also possible to derive a new expression for the transfer function by making
use of an earlier analysis of perturbed oblique flames.81 Schuller et al.75 show
that this function depends on two parameters ω∗ and SL /v̄, and one obtains an
improved agreement with the experimental data. The phase of the transfer function
shifts from a purely convective behavior for elongated flames to a saturated value
for flat flames.
B. Flame Response to Composition Inhomogeneities

Experiments and theoretical analysis indicate that certain types of instabilities in
lean premixed combustors may be driven by perturbations in the fuel–air ratio.82–85
This situation is illustrated in this section by assuming that pressure oscillations
in the combustor interact with the fuel-supply line and change the fuel flow rate,
as proposed by Lieuwen and Zinn.82
A positive pressure excursion produces a decrease of the fuel supply at a later
instant, which causes a negative perturbation in the equivalence ratio φ1 , which is
then convected by the flow to the flame zone. The interaction may also take place
with the air supply, which will also affect the equivalence ratio. The two types
of interactions will produce a heat-release perturbation, which, if properly phased
with the pressure, may feed energy in the resonant acoustic mode involved in the
process. This interaction can be represented schematically by
p1 −→ φ1 −→ convection −→ Q 1
This mechanism is illustrated in Fig. 9.15. In a first step, a pressure oscillation arises
in the system, which will modify the fuel flow rate and change the equivalence
a)
b)
c)
d)
e)
f)
Fig. 9.15 Instability driven by equivalence ratio perturbations. Time traces of pres-
sures, equivalence ratios, and heat release in the flame (from Ref. 82).
ratio. Three time delays define the process. The first τi corresponds to a phase shift
between the pressure at the injector and fuel mass flow rate ṁ F1 . Oscillations in this
flow rate induce fluctuations in the equivalence ratio φ1 . An inhomogenous mixture
is then convected to the reaction zone with a delay τconv . The response of the flame
to the impinging fluctuations φ1 comes after a combustion delay τchem . Oscillations
will be sustained by this process if the pressure and heat-release fluctuations are
in phase (Rayleigh’s criterion), which is the case if the total delay is such that
T
τi + τconv + τchem = (2n − 1) (9.22)
2
where T is the period of the combustion instability, and n is an integer (n > 0).
In many cases, the dominant delay is associated with convection, and the last
condition becomes τconv (2n − 1)T /2.
One fundamental aspect of this process is the response of the flame to incom-
ing equivalence ratio perturbations.86 Another aspect that will also influence this
mechanism is the level of mixing taking place between the injector and the flame.87
If this mixing is efficient, the initial level of fluctuations will be diminished to a
great extent by reducing the fluctuation in heat release. Effects of inhomogeneities
are also examined in more detail in other parts of this book (see also Ref. 8).
C. Unsteady Strain Rate Effects

An unsteady strain-rate field can be induced by the resonant acoustic motion
acting on the flow. This field may change the rate of heat release in two major
ways. The first way consists of perturbations in the flame-surface area. To analyze
this first possibility, let us consider a model equation for the flame-surface density:
d
= − β 2 (9.23)
dt
Balance equations of this type are extensively used in turbulent combustion. In

Eq. (9.23), the first term on the right-hand side represents production of surface den-
sity by strain rate, and the second term describes mutual annihilation of flame sur-
face density (flame shortening). At equilibrium, d0 /dt = 0 and 0 0 − β02 = 0.
A sinusoidal perturbation of the strain rate = 0 + 1 cos ωt is now assumed,
which produces a perturbation in surface density: = 0 + 1 . Injecting this
expression in the balance equation (9.23) and, retaining first-order terms only, one
obtains7
d1
+ 0 1 = (1 cos ωt)0 (9.24)
dt
The response in terms of 1 is that of a low-pass filter. The steady-state solution

takes the general form:
1 1
= 2 (0 cos ωt + ω sin ωt) (9.25)
0 0 + ω 2
In the low-frequency limit, ω 0 , the relative perturbation of flame-surface den-

sity is in phase with the strain rate:
1 1
= cos ωt (9.26)
0 0
In the high-frequency limit, ω 0 , the relative perturbation of flame-surface

density is in quadrature with the strain rate, and it decreases with frequency:
1 1
= sin ωt (9.27)
0 ω
This mechanism applies equally well to premixed and nonpremixed flames. This
type of interaction modulates the flame-surface density and can be represented
schematically by
p1 −→ v1 −→ flow −→ 1 −→ A1
The second type of interaction involves a direct effect on the reaction rate per unit
flame surface and is represented by
p1 −→ v1 −→ flow −→ 1 −→ ω̇1
This second effect is effective in the nonpremixed case, because the reaction rate
is directly related to the species gradients at the flame, which are fixed by the strain
rate. In the premixed case, the consumption rate is weakly influenced by the strain
rate, except near extinction conditions.
The flame response to strain rate has been extensively studied in turbulent com-
bustion.88, 89 Other studies deal with the response of flames to external strain-rate
modulations. The problem is envisaged experimentally90 and often treated by di-
rect calculations using time-dependent solutions of strained flames with complex
chemistry.91, 92 Analytical expressions of the flame response have also been de-
termined by using asymptotics.93 It was found that flames behave like low-pass
filters, when the perturbed strain-rate fluctuations do not exceed the extinction
value. Considering nonpremixed flames and assuming the infinitely fast chemistry
limit, the flame-transfer function is defined in the frequency domain as the ratio of
the relative reaction rate modulation to the relative strain-rate perturbation7 :

ṁ(ω) − ṁ 0 (ω) − 0
F(ω) = (9.28)
ṁ 0 0
This transfer function has the form of a low-pass filter:
1 1
F(ω) = (9.29)
2 1 + i(ω/20 )
a) b)
c) d)
e) f)
Fig. 9.16 DNS computations of the mutual flame annihilation as a limitation mecha-
nism for flame surface production. The different lines represent the peak consumption
rates of CH4 , O2 , H2 and CO at a) 0.61t f , b) 0.72t f , c) 0.75t f , d) 0.78t f , e) 0.81t f , and
f) 1.1t f , where t f is the flame time (from Ref. 96).
The effect of unsteady strain on premixed flames cannot be described in such simple
terms. Numerical calculations by Im and Chen94 indicate that the reponse of the
flame to modulated strain rates takes the form of cycles around the steady-state
line. The size of the cycle diminishes as the frequency increases.
The flame-surface area is augmented when the strain rate acting on the reactive
elements is lower than the extinction value. Conversely, the flame area is limited by
a mechanism of mutual interactions of adjacent reactive elements. This mechanism
has been identified as a fundamental process reducing the flame-surface area (see,
for example, Echekki et al.95 or Chen et al.96 and Fig. 9.16). This mechanism may
also influence the dynamics of turbulent flames. The mutual interaction of strained
flames is now well understood in cases in which the flame elements tend to propa-
gate away from each other. The case of strained elements approaching each other
and leading to a shortening of the flame is less easy to study experimentally. The
possibility of having synchronized interactions leading to instability is not gener-
ally considered. This type of coupling process has been observed experimentally
at least by Schuller et al.59 The corresponding driving process is briefly evoked in
Sec. III.
Table 9.1 Summary of driving processes examined in this chapter
Effect Induced
Initial on flow field
perturbation and flame Main result Consequences changes
Flame–vortex Vorticity Flame roll up

interaction generation
Flame– Flow Flame Changes of Heat-release Pressure-
boundary perturbation wrinkling flame-surface fluctuations wave
interaction area radiations
Mutual flame Flame Flame front
interaction wrinkling annihilation
V. Conclusion
The development of predictive methods for combustion instabilities is an impor-
tant technological objective. This prediction is now essential to the development
of advanced combustors for gas turbines. Considerable progress has been made
in this direction. Experiments and detailed analysis have generated a wealth of
information on the basic processes involved. This chapter illustrates some of these
processes and focuses on the driving and coupling relations that exist between
heat-release fluctuations and acoustic variables. Tables 9.1 and 9.2 summarize the
different paths examined in this review. By using well-controlled experiments, it
is shown that rapid changes of the flame surface generate an intense radiation of
sound. In practical situations, there are many possible mechanisms that may pro-
duce or destroy flame surface at a fast rate, such as flame–wall interactions and
collisions between adjacent flames or between neighboring flow structures like
vortices or reactant jets. These processes may feed energy into a resonant mode if
Table 9.2 Summary of coupling processes examined in this chaptera
Effect
Initial on flow
perturbation and flame Main result Consequences
Flame response Flow Flame surface

to upstream modulation wrinkling
modulation
Flame response Acoustic Injection Equivalence- Unsteady heat
to composition wave motion perturbation ratio release
inhomogeneities fluctuations
Flame response Flow Fluctuations of
to strain rate modulation strain rate field
a When interacting with the proper phase lag, driving and coupling mechanisms can lead to combustion
instabilities.
they are properly phased with respect to the pressure. Fast changes in flame-surface
area constitute an important driving process of combustion instabilities.
The coupling (or feedback) mechanism between the pressure field and the com-
bustion process may take many different forms. It is illustrated here with a set of
experiments with laminar flames, but previous experiments on turbulent ducted
configurations have indicated that premixed flames are quite susceptible to such
modulations. The flame is highly wrinkled by the external field, giving rise to
surface and heat-release fluctuations. In simple cases, it is possible to define a
transfer function between the relative velocity and heat-release fluctuations. Com-
parisons between analytical models, numerical simulations, and experiments are
reviewed. It is shown that simple filter models do not provide a suitable description
of the phase when the modulation frequency is high and that more refined methods
must be used to get a better description of this quantity. In practical systems, the
flame may also be modulated by many other means. Equivalence ratio perturba-
tions caused by the differential response of the injection system may also induce
heat-release fluctuations when these perturbations convected downstream reach
the flame, which has been identified as a possible driving process for some types
of gas-turbine instabilities. Flame modulation may result from the field of variable
strain rate, which can be induced by the nonsteady motion in the combustor. The
variable strain rate can produce or diminish the flame-surface area, and modify the
local rate of reaction per unit surface. If the fluctuations are suitably phased, they
will feed energy back into the acoustic motion.
Although the information accumulated over many years of research is quite sub-
stantial, additional fundamental experiments and intermediate scale investigations
are still needed. Further modeling, with a focus on coupling and driving processes
is required together with detailed simulations. The results gathered recently could
be used to check numerical tools and validate simulations of combustion dynamics.
The interactions examined in this chapter only portray some of the mechanisms in-
volved in the more complex dynamics of gas-turbine combustors, a subject covered
in further detail in this book.
References
1
Mallard, E., and Le Châtelier, H., “Recherches Expérimentales et Théoriques sur la
Combustion de Mélanges Gazeux Explosifs,” Annales des Mines, Paris Series, Vol. 8,
1883, pp. 274–377.
2
Lord Rayleigh, “The Explanation of Certain Acoustic Phenomena,” Nature, Vol. 18,
1878, pp. 319–321.
3
Candel, S., “Combustion Instabilities Coupled by Pressure Waves and Their Active Con-
trol,” Proceedings of the Combustion Institute, Vol. 24, The Combustion Inst., Pittsburgh,
PA, 1992, pp. 1277–1296.
4
McManus, K., Poinsot, T., and Candel, S., “A Review of Active Control of Combustion
Instabilities,” Progress in Energy and Combustion Science, Vol. 19, No. 1, 1993, pp. 1–29.
5
Yang, V., and Anderson, W. E. (eds.), Liquid Rocket Engines Combustion Instabilities,
Vol. 169, Progress in Astronautics and Aeronautics, AIAA, Reston, VA, 1995.
6
De Luca, D., Price, E. W., and Summerfield, M. (eds.), Nonsteady Burning and Com-
bustion Stability of Solid Propellants, Vol. 143, Progress in Astronautics and Aeronautics,
AIAA, Reston, VA, 1992.
7
Candel, S., “Combustion Dynamics and Control: Progress and Challenges,” Proceed-
ings of the Combustion Institute, Vol. 29, The Combustion Inst., Pittsburgh, PA, 2002,
pp. 1–28.
8
Lieuwen, T., and McManus, K. (eds.), “Combustion Dynamics in Lean-Premixed Pre-
vaporized (LPP) Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5, 2003,
pp. 721–829.
9
Poinsot, T. (ed.), “Large Eddy Simulation of Reacting Flows,” Flow Turbulence and
Combustion, Vol. 65, No. 2, 2000, pp. 111–244.
10
Desjardins, P. E., and Frankel, S. H., “Two-dimensional Large Eddy Simulation of Soot
Formation in the Near-field of a Strongly Radiating Nonpremixed Acetylene-Air Turbulent
Jet Flame,” Combustion and Flame, Vol. 119, No. 1–2, 1999, pp. 121–133.
11
Kim, W. W., Menon, S., and Mongia, H. C., “Large Eddy Simulation of a Gas Turbine
Combustor Flow,” Combustion Science and Technology, Vol. 143, No. 1–6, 1999, pp. 25–62.
12
Fureby, C., “A Computational Study of Combustion Instabilities due to Vortex Shed-
ding,” Proceedings of the Combustion Institute, Vol. 28, The Combustion Inst., Pittsburgh,
PA, 2000, pp. 783–791.
13
Pitsch, H., and Duchamp de Lageneste, L., “Large-Eddy Simulation of Premixed Tur-
bulent Combustion Using a Level-Set Approach,” Proceedings of the Combustion Institute,
Vol. 29, The Combustion Inst., Pittsburgh, PA, 2002, pp. 2001–2008.
14
Huang, Y., Sung, H.-G., Hsieh, S.-Y., and Yang, V., “Large-Eddy Simulation of Com-
bustion Dynamics of Lean-Premixed Swirl-Stabilized Combustor,” Journal of Propulsion
15
Selle, L., Lartigue, G., Poinsot, T., Koch, R., Schildmacher, K.-U., Krebs, W., Prade,
B., Kaufmann, P., and Veynante, D., “Compressible Large Eddy Simulation of Turbulent
Combustion in Complex Geometry on Unstructured Meshes,” Combustion and Flame,
Vol. 137, No. 3, 2004, pp. 489–505.
16
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, Edwards,
Philadelphia, 2001, 473 pp.
17
Ducruix, S., Schuller, T., Durox, D., and Candel, S., “Combustion Dynamics and Insta-
bilities: Elementary Coupling and Driving Mechanisms,” Journal of Propulsion and Power,
Vol. 19, No. 5, 2003, pp. 722–734.
18
Lieuwen, T., “Modeling Premixed Combustion-Acoustic Wave Interactions: A Re-
19
Dowling, A. P., and Stow, S. R., “Modal Analysis of Gas Turbine Combustor Acoustics,”
Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 751–764.
20
Yu, K. H., Trouvé, A., and Daily, J. W., “Low-Frequency Pressure Oscillations in
a Model Ramjet Combustor,” Journal of Fluid Mechanics, Vol. 232, Nov. 1991, pp. 47–72.
21
Smith, D. A., and Zukoski, E. E., “Combustion Instability Sustained by Unsteady Vortex
Combustion,” AIAA-SAE-ASME-ASEE Twenty-First Joint Propulsion Conference, AIAA
Paper 85-1248, July 1985.
22
Culick, F. E. C., and Yang, V., “Overview of Combustion Instabilities in Liquid-
Propellant Rocket Engines,” Liquid Rocket Engines Combustion Instability, Vol. 169,
Progress in Astronautics and Aeronautics, AIAA, Reston, VA, 1995, pp. 3–37.
23
Strahle, W. C., “Duality, Dilatation, Diffusion, and Dissipation in Reacting Turbulent
Flows,” Proceedings of the Combustion Institute, Vol. 19, The Combustion Inst., Pittsburgh,
PA, 1982, pp. 337–347.
24
Williams, F. A., Combustion Theory, Benjamin Cummings, Menlo Park, CA, 1985.
25
Doak, P. E., “Fundamentals of Aerodynamic Sound Theory and Flow Duct Acoustics,”
Journal of Sound and Vibration, Vol. 28, No. 3, 1973, pp. 527–561.
26
Kotake, S., “On Combustion Noise Related to Chemical Reactions,” Journal of Sound
and Vibration, Vol. 42, No. 3, 1975, pp. 399–410.
27
Crocco, L., “Aspects of Combustion Instability in Liquid Propellant Rocket Motors,”
Journal of the Aeronautical Research Society, Vols. 21 and 22, 1952.
28
Tsien, H. S., “Servo-stabilization of Combustion in Rocket Motors,” American Rocket
Society Journal, Vol. 22, 1952, pp. 256–263.
29
Crocco, L., and Cheng, S. L., Theory of Combustion Instability in Liquid Propel-
lant Rocket Motors, Agardograph No. 8, Butterworths Science Publication, Butterworths,
London, 1956.
30
Lang, W., Poinsot, T., and Candel, S., “Active Control of Combustion Instability,”
Combustion and Flame, Vol. 70, No. 3, 1987, pp. 281–289.
31
Gulati, A., and Mani, R., “Active Control of Unsteady Combustion-Induced Oscilla-
tions,” Journal of Propulsion and Power, Vol. 8, No. 5, 1992, pp. 1109–1115.
32
Hedge, U. G., Reuter, D., and Zinn, B. T., “Sound Generation by Ducted Flames,” AIAA
Journal, Vol. 26, No. 5, 1988, pp. 532–537.
33
Renard, P. H., Thévenin, D., Rolon, J. C., and Candel, S., “Dynamics of Flame-Vortex
Interactions,” Progress in Energy and Combustion Science, Vol. 26, No. 3, 2000, pp. 225–
282.
34
Sinibaldi, J. O., Mueller, C. J., and Driscoll, J. F., “Local Flame Propagation Speeds
Along Wrinkled, Unsteady, Stretched Flames,” Proceedings of the Combustion Institute,
35
Renard, P. H., Rolon, J. C., Thévenin, D., and Candel, S., “Investigation of Heat Release,
Extinction and Time Evolution of the Flame Surface, for a Nonpremixed Flame Interacting
with a Vortex,” Combustion and Flame, Vol. 117, No. 1, 1999, pp. 189–205.
36
Poinsot, T., Trouvé, A., Veynante, D., Candel, S., and Esposito, E., “Vortex-Driven
April 1987, pp. 265–292.
37
Schadow, K. C., Gutmark, E. J., Parr, T. P., Parr, D. M., Wilson, K. J., and Crump,
J. E., “Large-Scale Coherent Structures as Drivers of Combustion Instability,” Combustion
Science and Technology, Vol. 64, No. 4–6, 1989, pp. 167–186.
38
Crow, S. C., and Champagne, F. H., “Orderly Structure in Jet Turbulence,” Journal of
Fluid Mechanics, Vol. 48, No. 3, 1971, pp. 547–591.
39
Ho, C. M., and Nosseir, N. S., “Dynamics of an Impinging Jet. Part 1: The Feedback
Phenomenon,” Journal of Fluid Mechanics, Vol. 105, April 1981, pp. 119–142.
40
Ho, C.-M., and Huerre, P., “Perturbed Free Shear Layers,” Annual Review of Fluid
Mechanics, Vol. 16, 1984, pp. 365–424.
41
Kaiktsis, L., Karniadakis, G. E., and Orszag, S., “Unsteadiness and Convective Instabil-
ities in Two-Dimensional Flow over a Backward-Facing Step,” Journal of Fluid Mechanics,
Vol. 321, August 1996, pp. 157–187.
42
Hussain, A. K., and Zaman, K. B., “Vortex Pairing in a Circular Jet under Controlled
Excitation. Part 1. General Jet Response,” Journal of Fluid Mechanics, Vol. 101, Dec. 1980,
pp. 449–491.
43
Hussain, A. K., and Zaman, K. B., “Vortex Pairing in a Circular Jet under Controlled
Excitation. Part 2. Coherent Structure Dynamics,” Journal of Fluid Mechanics, Vol. 101,
Dec. 1980, pp. 493–544.
44
Oster, D., and Wygnanski, I., “The Forced Mixing Layer Between Parallel Streams,”
Journal of Fluid Mechanics, Vol. 123, Oct. 1982, pp. 91–130.
45
Gaster, M., Kit, E., and Wygnanski, I., “Large-Scale Structures in a Forced Turbulent
Mixing Layer,” Journal of Fluid Mechanics, Vol. 150, Jan. 1985, pp. 23–39.
46
Ghoniem, A. F., and Ng, K. K., “Numerical Study of the Dynamics of a Forced Shear
Layer,” Physics of Fluids, Vol. 30, No. 3, 1987, pp. 706–723.
47
Mitchell, B. E., Lele, S. K., and Moin, P., “Direct Computation of the Sound Generated
by Vortex Pairing in an Axisymmetric Jet,” Journal of Fluid Mechanics, Vol. 383, March
1999, pp. 113–142.
48
Bogey, C., Bailly, C., and Juvé, D., “Numerical Simulation of Sound Generated by
Vortex Pairing in a Mixing Layer,” AIAA Journal, Vol. 38, No. 12, 1999, pp. 2210–2218.
49
Durox, D., Schuller, T., and Candel, S., “Combustion Dynamics of Inverted Conical
Flames,” Proceedings of the Combustion Institute, Vol. 30, The Combustion Inst., Pittsburgh,
PA, 2004, pp. 1717–1724.
50
Kendrick, D. W., Zsak, T. W., and Zukoski, E. E., “An Experimental and Numerical
Investigation of Premixed Combustion in a Vortex in a Laboratory Dump Combustor,”
Unsteady Combustion, NATO ASI Series, Vol. 306, Kluwer, Dordrecht 1996, pp. 33–69.
51
Zsak, T. W., “An Investigation of the Reacting Vortex Structures Associated with Pulse
Combustion,” Ph.D. Dissertation, California Inst. of Technology, Pasadena, CA, April
1993.
52
Schuller, T., Durox, D., and Candel, S., “Dynamics of and Noise Radiated by a Perturbed
Impinging Premixed Jet Flame,” Combustion and Flame, Vol. 128, No. 1–2, 2002, pp. 88–
110.
53
Durox, D., Schuller, T., and Candel, S., “Self-induced Instability of a Premixed Jet
Flame Impinging on a Plate,” Proceedings of the Combustion Institute, Vol. 29, The Com-
bustion Inst., Pittsburgh, PA, 2002, pp. 69–75.
54
Schäfer, O., Koch, R., and Wittig, S., “Measurement of the Periodic Flow of an En-
closed Lean Premixed Prevaporized Stagnation Flame,” Tenth International Symposium on
Applications of Laser Techniques to Fluid Mechanics, Lisbon, 2000.
55
Bragg, S. L., “Combustion Noise,” Journal of the Institute of Fuel, Vol. 36, 1963,
pp. 12–16.
56
Abugov, D. I., and Obrezkov, O. I., “Acoustic Noise in Turbulent Flames,” Combustion,
Explosions and Shock Waves, Vol. 14, 1978, pp. 606–612.
57
Clavin, P., and Siggia, E., “Turbulent Premixed Flames and Sound Generation,” Com-
bustion Science and Technology, Vol. 78, No. 1–3, 1991, pp. 147–155.
58
Kidin, N., Librovich, V., Roberts, J., and Vuillermoz, M., “On Sound Sources in Tur-
bulent Combustion,” Dynamics of Flames and Reactive Systems, Vol. 95, Progress in As-
tronautics and Aeronautics, AIAA, New York, 1984, pp. 343–355.
59
Schuller, T., Durox, D., and Candel, S., “Self-induced Combustion Oscillations of
Laminar Premixed Flames Stabilized on Annular Burners,” Combustion and Flame,
Vol. 135, No. 4, 2003, pp. 525–538.
60
Candel, S., Durox, D., and Schuller, T., “Flame Interactions as a Source of Noise and
Combustion Instabilities,” AIAA Paper 2004-2928, AIAA/CEAS Aeroacoustics Confer-
ence, Manchester, England, U.K., 2004.
61
McIntosh, A. C., “On Flame Resonance in Tubes,” Combustion Science and Technology,
Vol. 69, No. 4–6, 1990, pp. 147–152.
62
McIntosh, A. C., “The Linearised Response of the Mass Burning Rate of a Premixed
Flame to Rapid Pressure Changes,” Combustion Science and Technology, Vol. 91, No. 4–6,
1993, pp. 329–346.
63
Edwards, N. R., McIntosh, A. C., and Brindley, J., “The Development of Pressure
Induced Instabilities in Premixed Flames,” Combustion Science and Technology, Vol. 99,
No. 1–3, 1996, pp. 373–386.
64
Park, S., Annaswamy, A. M., and Ghoniem, A. F., “Heat Release Dynamics Modeling
of Kinetically Controlled Burning,” Combustion and Flame, Vol. 128, No. 3, 2002, pp. 217–
231.
65
Markstein, G. H., Non Steady Flame Propagation, Pergamon Press, Elmsford, NY,
1964.
66
Blackshear, P. L., “Driving Standing Waves by Heat Addition,” Proceedings of the
Combustion Institute, Vol. 4, The Combustion Inst., Pittsburgh, PA, 1953, pp. 553–566.
67
De Sœte, G., “Etude des Flammes Vibrantes. Application à la Combustion Turbulente,”
Revue de l’Institut Français du Pétrole et Annales des Combustibles Liquides, Vol. 19, No. 6,
1964, pp. 766–785.
68
Merk, H. J., “An Analysis of Unstable Combustion of Premixed Gases,” Proceedings
of the Combustion Institute, Vol. 6, The Combustion Inst., Pittsburgh, PA, 1956, pp. 500–
512.
69
Becker, R., and Günther, R., “The Transfer Function of Premixed Turbulent Jet Flames,”
Proceedings of the Combustion Institute, Vol. 13, The Combustion Inst., Pittsburgh, PA,
1971, pp. 517–526.
70
Baade, P. K., “Design Criteria and Modes for Preventing Combustion Oscillations,”
ASHRAE Transactions, Vol. 1, 1978, pp. 449–465.
71
Matsui, Y., “An Experimental Study on Pyro-acoustic Amplification of Premixed
Laminar Flames,” Combustion and Flame, Vol. 43, No. 2, 1981, pp. 199–209.
72
Baillot, F., Durox, D., and Prud’homme, R., “Experimental and Theoretical Study of
a Premixed Vibrating Flame,” Combustion and Flame, Vol. 88, No. 2, 1992, pp. 149–
168.
73
Fleifil, M., Annaswamy, A. M., Ghoneim, Z. A., and Ghoniem, A. F., “Response of a
Laminar Premixed Flame to Flow Oscillations: A Kinematic Model and Thermoacoustic
Instability Results,” Combustion and Flame, Vol. 106, No. 4, 1996, pp. 487–510.
74
Ducruix, S., Durox, D., and Candel, S., “Theoretical and Experimental Determinations
of the Transfer Function of a Laminar Premixed Flame,” Proceedings of the Combustion
Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 765–773.
75
Schuller, T., Durox, D., and Candel, S., “A Unified Model for the Prediction of Flame
Transfer Functions: Comparison Between Conical and V-flame Dynamics,” Combustion
and Flame, Vol. 134, No. 1–2, 2003, pp. 21–34.
76
Dowling, A. P., “A Kinematic Model of a Ducted Flame,” Journal of Fluid Mechanics,
Vol. 394, September 1999, pp. 51--72.
77
Marble, F. E., and Candel, S., “An Analytical Study of the Non-steady Behavior of
Large Combustors,” Proceedings of the Combustion Institute, Vol. 17, The Combustion
Inst., Pittsburgh, PA, 1978, pp. 761–769.
78
Bourehla, A., and Baillot, F., “Appearance and Stability of a Laminar Conical Premixed
Flame Subjected to an Acoustic Perturbation,” Combustion and Flame, Vol. 114, No. 3–4,
1998, pp. 303–318.
79
Lieuwen, T., and Neumeier, Y., “Nonlinear Pressure-Heat Release Transfer Function
Measurements in a Premixed Combustor,” Proceedings of the Combustion Institute, Vol. 29,
The Combustion Inst., Pittsburgh, PA, 2002, pp. 99–105.
80
Schuller, T., Ducruix, S., Durox, D., and Candel S., “Modeling Tools for the Prediction
of Premixed Flame Transfer Functions,” Proceedings of the Combustion Institute, Vol. 29,
81
Boyer, L., and Quinard, J., “On the Dynamics of Anchored Flames,” Combustion and
Flame, Vol. 82, No. 1, 1990, pp. 51–65.
82
Lieuwen, T., and Zinn, B. T., “The Role of Equivalence Ratio Oscillations in Driving
Combustion Instabilities in Low NOx Gas Turbines,” Proceedings of the Combustion Insti-
tute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 1809–1816.
83
Lee, J. G., Kwanwoo, K., and Santavicca, D. A., “Measurement of Equivalence Ratio
Fluctuation and Its Effect on Heat Release during Unstable Combustion,” Proceedings of
the Combustion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 415–421.
84
Hathout, J. P., Fleifil, M., Annaswamy, A. M., and Ghoniem, A. F., “Heat-Release Actu-
ation for Control of Mixture-Inhomogeneity-Driven Combustion Instability,” Proceedings
730.
85
86
Marzouk, Y. M., Ghoniem, A. F., and Najm, H. N., “Dynamic Response of Strained
Premixed Flames to Equivalence Ratio Gradients,” Proceedings of the Combustion Institute,
87
Lieuwen, T., Neumeier, Y., and Zinn, B. T., “The Role of Unmixedness and Chemical
Kinetics in Driving Combustion Instabilities in Lean Premixed Combustors,” Combustion
88
Law, C. K., “Dynamics of Stretched Flames,” Proceedings of the Combustion Institute,
89
Peters, N., Turbulent Combustion, Cambridge University Press, Cambridge, U.K., 2000.
90
Welle, E. J., Roberts, W. L., Decroix, M. E., Carter, C. D., and Donbar, J. M., “Simulta-
neous Particle-Imaging Velocimetry and OH Planar Laser Induced Fluorescence Measure-
ments in an Unsteady Counterflow Propane Diffusion Flame,” Proceedings of the Combus-
tion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 2021–2027.
91
Darabiha, N., “Transient Behaviour of Laminar Counter Flow Hydrogen-Air Flames
with Complex Chemistry,” Combustion Science and Technology, Vol. 86, No. 1–6, 1992,
pp. 163–181.
92
Egolfopoulos, F. N., and Campbell, C. S., “Unsteady Counterflowing Strained Diffusion
Flames: Diffusion-Limited Frequency Response,” Journal of Fluid Mechanics, Vol. 318,
July 1996, pp. 1–29.
93
Joulin, G., “On the Response of Premixed Flames to Time-Dependent Stretch and
Curvature,” Combustion Science and Technology, Vol. 97, No. 1–3, 1994, pp. 219–229.
94
Im, H. G., and Chen, J. H., “Effects of Flow Transients on the Burning Velocity of
Laminar Hydrogen-Air Premixed Flames,” Proceedings of the Combustion Institute, Vol. 28,
The Combustion Institute, Pittsburgh, 2000, pp. 1833–1840.
95
Echekki, T., Chen, J., and Gran, I., “The Mechanism of Mutual Annihilation of
Stoichiometric Premixed Methane-Air Flames,” Proceedings of the Combustion Institute,
96
Chen, J., Echekki, T., and Kollman, W., “The Mechanism of Two-Dimensional Pocket
Formation in Lean Premixed Methane-Air Flames with Implication to Turbulent Combus-
tion,” Combustion and Flame, Vol. 116, No. 1–2, 1999, pp. 15–48.
Chapter 10
Flow and Flame Dynamics of Lean Premixed

Swirl Injectors
Ying Huang,∗ Shanwu Wang,† and Vigor Yang‡

Pennsylvania State University, University Park, Pennsylvania
I. Introduction
F UEL injection and mixing are critical to achieving efficient and clean com-
bustion in modern gas-turbine engines, whether they are powered by gaseous
or liquid fuels. For gaseous fuels, the major concern is to obtain an optimal level of
mixing between air, fuel, and combustion products in the combustion zone. When
liquid fuels are employed, they must be atomized into small droplets and then dis-
tributed in an airstream before entering the combustion zone.1 Most gas-turbine
injectors employ swirl configurations that produce central toroidal recirculation
zones (CTRZs) to provide the dominant flame-stabilization mechanism. Flows in
this region are generally associated with high shear rates and strong turbulence
intensities resulting from vortex breakdown.
Many experimental studies have been conducted to investigate the flow and flame
dynamics of swirl injectors. An overview of the use of these injectors in liquid-
propellant rocket engines was recently published by Bazarov et al.2 For fuel in-
jectors typical of gas-turbine engines, excellent descriptions and their applications
in modern dry low-emission (DLE) combustors were given by Lefebrve.3 Wang
et al.4, 5 conducted an experimental study of a 3 × scale model of the CFM56 coaxial
swirl cup. The droplet dynamics were characterized by means of phase Doppler in-
terferometry. Jeng and colleagues6–8 examined the counter-rotating flow structures
produced by the CFM56 swirl cup. The effects of air temperature, fluid property,
and the equivalence ratio on spray characteristics were studied under both nonre-
acting and reacting conditions. Cowell and Smith9 tested a liquid-fueled injector
in a bench-scale can combustor to evaluate critical design and operating parame-
ters for emission characteristics. Both axial and radial swirlers were explored. The
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and
Astronautics, Inc., with permission.
∗ Postdoctoral Research Associate. Member AIAA.
† Research Associate. Member AIAA.
‡ Distinguished Professor. Fellow AIAA.
213
214 Y. HUANG ET AL.
size and shape of the recirculation zone associated with each swirler were found
to exert a strong influence on combustor performance. Snyder et al.10 carried out
an experimental program to develop a liquid-fueled injector with tangential entry.
The combustion performance was evaluated in a high-pressure, single-nozzle test
facility. A similar injector configuration was later explored by Cohen et al.11 in
the development of an active combustion control system at realistic engine (i.e.,
FT-8) operating conditions.
For laboratory research swirl injectors, a considerable amount of experimental
work was conducted. Richards et al.12, 13 investigated the flame dynamics of a pre-
mixed fuel injector using natural gas. A simple time-lag model was proposed to
characterize experimentally observed combustion oscillations. Broda et al.,14 Lee
et al.,15 Venkataraman et al.,16 and Lee et al.17, 18 studied the combustion dynamics
of gaseous-fueled single-element swirl injectors. The effects of the equivalence
ratio, inlet velocity, temperature, fuel distribution, swirl number, and the pres-
ence of a centerbody recess on combustion stability were examined. Mordaunt
et al.19 conducted a series of experiments to investigate combustion dynamics
in a single-element injector using various fuels, including gaseous ethylene and
three different liquid hydrocarbon fuels: n-heptane, JP-8, and a coal-based fuel.
Cohen and Rosfjord,20, 21 Chin et al.,22, 23 and Hardalupas et al.24, 25 measured the
spray characteristics downstream of swirl injectors to improve the understanding
of the liquid fuel atomization process. Presser et al.26, 27 examined the aerodynamic
characteristics of a swirling spray flame using a pressure-jet atomizer. The effects
of swirl on droplet transport, as well as the interactions between droplets and air
flowfield, were studied under both nonreacting and reacting conditions. Paschereit,
et al.,28, 29 Acharya et al.,30, 31 Murugappan et al.,32 Lee et al.,33 Richards, et al.,34, 35
and Zinn and colleagues36–39 investigated the characteristics and control of com-
bustion instabilities in swirl-stabilized combustors. Bernier et al.40 analyzed the
combustion dynamics in a liquid-fueled premixed prevaporized burner using co-
and counter-rotating swirl injectors. Li and Gutmark41 examined the effects of
swirler orientation and exhaust-nozzle geometry on the flow and flame character-
istics of a dual-fueled multiple swirler combustion system.
Extensive efforts were applied to numerical studies of gas turbine combustion.
Brewster et al.42 conducted a comprehensive review of numerical simulations for
stationary gas turbines that were based on the Reynolds-averaged Navier–Stokes
(RANS) equations. The RANS simulation may be appropriate for time-mean tur-
bulent flow properties, but its validity for unsteady flow evolution has yet to be
established, especially for problems involving such complicated configurations
as swirl injectors. Recent advances in large-eddy simulations (LES) have shown
promise for studying the dynamics of swirl injectors. The technique computes ex-
plicitly the contributions of large energy-carrying structures to mass, momentum,
and energy transfer in the flowfield, with the effects of unresolved small-scale tur-
bulence modeled either analytically or empirically. Although the RANS method
remains the main workhorse for combustor design analysis, LES is considered to be
the next-generation analytical design tool for gas turbine combustors. In an effort
to improve the understanding of the flow and flame phenomena within the CFM56
aero-engine swirl cup, a series of numerical investigations were performed by GE
Aircraft Engines using both RANS and LES techniques.43–51 The LES method was
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 215
also implemented to facilitate the combustor designs at Rolls-Royce52 and Pratt

& Whitney.53, 54
The literature on LES studies of lean-premixed combustion with gas-fueled
swirl injectors was reviewed by Huang, et al.55 in 2003. A number of studies have
appeared since then. Stone and Menon56, 57 used LES modeling to investigate a
swirl-stabilized combustor flow. The effects of swirl and equivalence ratio on flame
dynamics were studied. Pierce and Moin58 conducted a numerical simulation of
a coaxial jet combustor. A flamelet/progress-variable approach was developed to
treat nonpremixed turbulent combustion. Selle et al.59 treated the full burner of a
premixed gas-turbine engine using LES for both nonreacting and reacting cases.
A strong precessing vortex core was observed for nonreacting flows. This vortex,
however, disappears when combustion occurs. Grinstein et al.60 simulated the
flowfield in a gaseous swirl combustor, with emphasis on the effects of combustor
confinement on the flow and flame evolution. Sommerer et al.61 conducted an LES
study of the flashback and blowoff in a lean partially premixed swirl burner. Wang
et al.62, 63 examined the vortical flow dynamics in swirl injectors with radial entry
under conditions with and without external excitations. Various flow instability
mechanisms, such as the Kelvin–Helmholtz, helical, and centrifugal instabilities,
as well as their interactions, were investigated in detail. Huang and Yang64–66
investigated the influences of inlet flow conditions on the combustion dynamics in
a lean-premixed swirl-stabilized combustor. The flame bifurcation phenomenon
and stability boundary were investigated as a function of the burner operating
conditions.
As compared with gaseous fuels, liquid spray combustion involves an additional
array of intricacies,67, 68 such as atomization, droplet dispersion and evaporation,
mixing, and combustion. All these processes must be considered for accurate pre-
diction, but most of them have not yet been fully understood and well modeled
in the highly turbulent environments typical of gas turbine combustors. Very few
LES studies of liquid-fueled swirl-injectors have been reported.69−72 Sankaran and
Menon70 performed an LES study of swirling spray combustion. The configuration
consisted of an inlet section with a central injection cone and a cylindrical dump
combustor, similar to that of a dual-annular counter-rotating swirl (DACRS) in-
jector for aero engines. A dilute-spray approximation, which neglected the droplet
breakup and coalescence processes, was employed. Apte et al.71 simulated particle-
laden, swirling flows in a coaxial-jet combustor, with emphasis on the particle
dispersion characteristics. The results compared favorably with the experimen-
tal data in mean velocity fields, turbulence properties, and particle distributions.
Wang et al.72 recently conducted an LES of spray-field dynamics in cross flows.
Liquid-fuel-jet breakup and droplet transport were considered.
This chapter provides an overview of various dominant processes associated
with lean-premixed swirl injectors. Emphasis is placed on the detailed flow evo-
lution and flame dynamics of gaseous systems. The situation with liquid spray
combustion dynamics will be covered in the chapter by Menon73 (Chap. 11). The
remainder of this chapter is organized as follows. In Sec. II, cold flow character-
istics of three different types of swirl injectors, including both axial- and radial-
entry configurations, are explored. The effects of single versus multiple swirlers
and co-rotating vs counter-rotating arrangements are examined systematically. In
216 Y. HUANG ET AL.
inlet dump chamber
air 101.6 152.4
25.4
unit: mm 600
Fig. 10.1 Schematic of a dump chamber with a co-axial swirler.
Sec. III, the flame dynamics of an axial-entry swirl injector operating over a wide
range of flow conditions is studied. Finally, a summary is given in Sec. IV.
II. Cold Flow Characteristics of Swirl Injectors

This section deals with the cold flow characteristics of three different kinds of
swirl injectors, including both axial- and radial-entry configurations that are rep-
resentative of contemporary gas-turbine injectors. Much of the discussion given
herein is based on the results obtained from large-eddy simulations.62, 63, 74, 75 Ex-
perimental observations are also presented, where appropriate, to provide a com-
prehensive understanding of the injector flow dynamics under various operating
conditions. The formulation accommodates the Favre-filtered conservation equa-
tions in three dimensions and is solved numerically by using a density-based,
finite-volume methodology. The code is further equipped with a multiblock do-
main decomposition feature to facilitate parallel processing in a distributed com-
puting environment using the Message Passing Interface (MPI) library. A detailed
description of the numerical approach is given by Wang.76
A. Flow Dynamics of Coaxial Swirl Injector

Figure 10.1 shows a coaxial swirler along with a dump chamber of concern.
Favaloro et al.77 conducted an experimental study on this configuration, in which a
swirler with 12 circular inlet guide vanes is located 50.8 mm upstream of the dump
plane. The leading edge of each blade is designed to be tangential to the incoming
flow and perpendicular to the centerline of the chamber. The chamber consists of a
plexiglass pipe measuring 152.4 mm in diameter and 1850 mm in length. The inlet
temperature and pressure are 300 K and 1 atm, respectively. The Reynolds number
is 1.25 × 105 based on the inlet diameter. The centerline velocity in the inlet pipe,
U = 19.2 m/s, and the height of the backward-facing step, H = 25.4 mm, are
used as the reference quantities to normalize the flow properties. The detailed
configuration of the experimental rig can be found in Ref. 77.
Two different swirl numbers (S = 0.3 and S = 0.5), defined below as the ratio
of the axial flux of the angular momentum to the product of the axial momentum
flux and a characteristic radius, are considered:
Rn Rn
S= ū w̄r 2 dr/ Rn ū 2r dr (10.1)
Rh Rh
S = 0.3
0 5 10 x/H
S = 0.5
0 5 10 x/H
Fig. 10.2 Streamlines based on mean axial and radial velocity components; swirl
numbers S = 0.3 and 0.5.
where Rh and Rn are the radii of the centerbody and the inlet duct, respectively. If
we assume that the axial and azimuthal velocities are uniform and that the vanes
are thin, the swirl number can be written as

2 1 − (Rh /Rn )3
S= tan ϕ (10.2)
3 1 − (Rh /Rn )2
where ϕ is the swirler vane angle. The present discussion focuses on the high
swirl-number case of S = 0.5 because of the occurrence of vortex breakdown.
The situation with S = 0.3, in which vortex breakdown is not observed, is also
treated for comparison.
1. Vortical Flow Evolution

Figure 10.2 shows the streamlines of the time-mean flowfield based on the
axial and radial velocities. Both the primary and secondary separation bubbles
are observed in the downstream region of the backward-facing step. The length
of the corner recirculation zone (CRZ) is shorter for the high swirl-number case
(S = 0.5) because of the stronger expansion of the main flow resulting from the
higher centrifugal force. A small separation bubble exists behind the centerbody,
and the flow rapidly merges along the centerline.
A simplified momentum equation indicates that a radial pressure gradient is
produced by the centrifugal force arising from the swirling effect:
∂p ρUθ2
= (10.3)
∂r r
The pressure tends to be minimized in regions where strong swirling motions
occur, that is, in the wake of the centerbody. As the flow expands and the azimuthal
velocity decays with the axial distance, the pressure is recovered in the downstream
218 Y. HUANG ET AL.
Fig. 10.3 Time evolution of streamlines based on mean axial and radial velocity
components spatially averaged in the azimuthal direction (time increment of 0.6 ms),
swirl number S = 0.5.
region. A positive pressure gradient is consequently generated along the axial axis,
which may lead to the formation of a recirculation zone in a high swirling flow, a
phenomenon commonly termed vortex breakdown. In the present configuration and
flow condition, vortex breakdown occurs only at high swirl numbers, as evidenced
in Fig. 10.2. A CTRZ is formed in the central region for S = 0.5, reaching from
x/H ≈ 0.36 to 7.8.
The temporal evolution of the flowfield is examined to explore the phenomenon
of vortex breakdown. Figure 10.3 shows the instantaneous streamlines on an x-
r plane, spatially averaged in the azimuthal direction, at various times during a
typical flow evolution period. The time increment between the snapshots is 0.6 ms,
and t = 0 corresponds to the instant at which data collection begins after the flow
reaches its stationary state. At t = 37.7 ms, a new vortical bubble is generated in
front of a braid of vortical bubbles. These bubbles then coalesce at t = 39.0 ms.
The bubble located in the downstream side of the vortical braid is separated into
two structures at t = 39.6 ms; one stays at basically the same location, and the
other is convected downstream and finally disappears because of turbulent dif-
fusion and viscous dissipation at t = 41.4 ms. During this period, the coalesced
vortical bubble separates, and another new bubble appears in the upstream region at
t = 42.0 ms. These snapshots exhibit a very complicated vortex evolution in the
central region.
Figure 10.4 shows snapshots of the axial velocity fields for the swirl numbers of
0.3 and 0.5. Shear layers are produced at the trailing edges of the centerbody and
the backward-facing step because of the Kelvin–Helmholtz instabilities in both
the axial and azimuthal directions. The flow evolution in the azimuthal direction
is presented on the two crosssections at x/H = 0.36 and 1.54. The shear layer
originating from the backward-facing step remains almost symmetric immediately
downstream of the dump plane (x/H = 0.36). It then rolls up and forms large
Fig. 10.4 Instantaneous axial velocity fields on x−r plane and two cross-sections,
x/H = 0.36 and 1.54. Swirl numbers S = 0.3 and 0.5. Contour levels between –12
and 33 m/s with increment of 3 m/s. Solid lines represent positive values and dashed
lines represent negative values.
asymmetric structures at x/H = 1.54, because of the strong shear force in the
azimuthal direction.
The large velocity difference in the azimuthal direction at a high swirl number
(S = 0.5) significantly increases the strength of the shear layer, especially around
the boundary of the CTRZ. As a consequence of flow reversal, the effective flow
passage area in the chamber is reduced, which increases the axial velocity differ-
ence and further enhances the shear layer in the axial direction. The large-scale
structures are eventually dissipated by turbulent diffusion and viscous damping
when the flow convects downstream.
Figure 10.5 shows snapshots of the isosurfaces of vorticity magnitude at || =
1.5U /H (i.e., 1133 1/s) for S = 0.3 and 0.5. Helical vortex tubes develop from
Fig. 10.5 Snapshots of isosurfaces of vorticity magnitude at 1.5 × U/H. Dark lines
represent streamlines; swirl numbers S = 0.3 and 0.5.
220 Y. HUANG ET AL.
Fig. 10.6 Instantaneous fluctuating pressure field on x−r plane and cross-sections
at x/H = 1.94, 5.87, 9.81, and 13.75. Contour levels between –600 and 600 Pa with
increment of 50 Pa. Solid lines represent positive values and dashed lines represent
negative values; swirl number S = 0.5.
the inlet and travel in a direction opposite to the main swirling flow, although the
whole structure follows the motion of the main flow. The swirl number plays an
important role in dictating the flow evolution and its underlying mechanisms. The
helical structure at S = 0.5 arises from the vortex breakdown and expands in the
downstream region. The situation is, however, different for a low swirl number, in
which the helical structure of the vortex tube shrinks in the downstream region.
This phenomenon may be attributed to the precession of the vortex core around the
centerline. The resultant intermittent occurrence of vortex breakdown causes the
helical structure issuing from the centerbody to vanish rapidly as the flow evolves
downstream. A high swirl number apparently helps maintain flow coherence and
leads to strong flow reversal.
2. Vortico-Acoustic Interaction
The strong vortical motion in the chamber often produces acoustic waves prop-
agating throughout the entire field. The shear layers, however, are susceptible to
acoustic excitations if such disturbances occur at appropriate locations and frequen-
cies. A feedback loop can thus be established, depending on the mutual coupling
between the vortical and acoustic fields. To explore vortico-acoustic interaction,
the fluctuating pressure field is obtained by subtracting the longtime averaged
pressure from its instantaneous quantity.
Figure 10.6 shows snapshots of the fluctuating pressure field on several cross
sections for the high swirl-number case with S = 0.5. Considerable pressure fluc-
tuations take place immediately downstream of the centerbody, where strong vor-
ticity is present. The entire field exhibits a wide range of length scales, with
broadband turbulent motion in the upstream region. This motion develops into
large-scale coherent acoustic motion in the downstream region. To help identify
the wave characteristics, especially those associated with longitudinal waves, the
complicated three-dimensional field shown in Fig. 10.6 is spatially averaged in
Fig. 10.7 Time evolution of fluctuating pressure field spatially averaged in the
azimuthal direction. Contour levels between –600 and 600 Pa with increment of
50 Pa (time increment of 0.3 ms). Solid lines represent positive values and dashed
lines represent negative values; swirl number S = 0.5.
the azimuthal direction. Figure 10.7 presents the time sequence of the resultant
quasi-two-dimensional fields. The data are further reduced by spatial averaging
over each cross section along the axial axis. The averaged quasi-one-dimensional
fields shown in Fig. 10.8 indicate that a negative pressure peak, followed by a
positive one, forms periodically and travels downstream at the speed of sound.
Two negative pressure peaks are observed in the chamber at t = 35.9 ms. Because
the evolution pattern of these two fluctuations is almost identical, the frequency
of the pressure wave is approximately 655 Hz, based on the distance between the
two pressure peaks at t = 35.9 ms and the wave propagation speed.
The frequency spectra of pressure fluctuations are obtained to quantitatively
characterize the acoustic flow evolution. Figure 10.9 shows the result at y/H =
0.06 and z/H = 0.07 with three different axial positions, x/H = 0.56, 2.19, and
7.20, which are located in the upstream, center, and downstream regions, re-
spectively, of the central recirculation zone. The most dominant mode is 1380
Hz, and other characteristic frequencies include 660, 2040, and 3420 Hz. The
222 Y. HUANG ET AL.
50 t = 34.4 ms t = 35.6 ms
0
p' , Pa
–50
–100
50 t = 34.7 ms t = 35.9 ms
0
p' , Pa
–50
–100
50 t = 35.0 ms t = 36.2 ms
0
p' , Pa
–50
–100
50 t = 35.3 ms t = 36.5 ms
0
p' , Pa
–50
–100
0 5 10 15 20 0 5 10 15 20
x/H x/H
Fig. 10.8 Time evolution of fluctuating pressure field spatially averaged over r –θ
cross section; swirl number S = 0.5.
a) b) c)
60
p' , Pa
40
20
0
0 2000 4000 6000 0 2000 4000 6000 0 2000 4000 6000
frequency, Hz frequency, Hz frequency, Hz
Fig. 10.9 Frequency spectra of pressure fluctuations at different axial locations near
the centerline: a) x/H = 0.38; b) x/H = 2.19; and c) x/H = 7.20, y/H = 0.06, and
z/H = 0.07. Swirl number S = 0.5.
corresponding wave amplitudes decrease in the downstream region because of

dissipation and dispersion effects. These frequencies represent various acoustic
modes in the chamber and are determined by the mean flow properties and chamber
geometry.
A simple acoustic modal analysis, without accounting for the mean flow effects,
indicates that the eigenfrequencies of the first tangential (1T) modes in the dump
chamber and the inlet duct are f c = 1350 Hz and f i = 2000 Hz, respectively. They
are almost identical to the observed harmonics shown in Fig. 10.9. The analysis
assumes the speed of sound to be 340 m/s for air at ambient conditions, and the radii
of the inlet duct and the chamber are taken to be 50.8 and 76.2 mm, respectively.
The two transverse acoustic modes interact with each other through nonlinear
gasdynamics to generate a subharmonic and a superharmonic with frequencies of
f S1 = f i − f c = 650 Hz and f S2 = f i + f c = 3350 Hz, respectively. The former
propagates in the form of a traveling longitudinal wave, as shown in Fig. 10.8.
To identify the mechanisms of acoustic wave generation and its relationship with
the shear-layer evolution, the vortical flow dynamics need to be further explored. It
is well established that the shear layer originating from the centerbody is sensitive
to external forcing,78−80 such as acoustic motion in the chamber. Thus, we may
employ shear-layer instability theories to help explain the mutual coupling between
the vortical and acoustic motions. Following common practice,79 the Strouhal
number, St, is defined as
fSδ 1
St = and Ū = (U1 + U2 ) (10.4)
Ū 2
where δ is the initial momentum thickness of the shear layer, and U1 and U2
are the freestream velocities on the two sides of the shear layer. The mean averaged
axial velocity Ū is approximately 10 m/s near the trailing edge of the centerbody.
The most unstable mode of an unforced planar shear layer occurs at St ≈ 0.044–
0.048 for turbulent flows.78 The momentum thickness δ is roughly one-fourth of
the vorticity thickness,79 which can be calculated from the axial velocity profile in
the radial direction. On the basis of Eq. (10.4), the frequency of the most unstable
mode of shear-layer instability, f S0 , is approximately on the order of 103 Hz near
the downstream region of the centerbody.
For the case with a swirl number of 0.5, the most prevalent acoustic mode
shown in Fig. 10.9 has a frequency of 1380 Hz, which is consistent with the
frequency of the most unstable shear-instability mode. Consequently, the shear
layer originating from the centerbody can be easily locked to the first-tangential
mode of the acoustic oscillation in the chamber. At this response frequency, f SR ,
the shear layer rolls up into discrete vortices and reinforces the acoustic oscillation
in the chamber. A feedback loop between the acoustic fluctuation and shear-layer
instability is thus established and leads to a large excursion of flow motions. When
the discrete vortices are convected downstream, they pair with the adjacent ones
to form larger structures with a characteristic frequency of f SR /2 = 690 Hz. This
subharmonic frequency is also close to the frequency of the longitudinal acoustic
mode in the chamber, 655 Hz, as determined from the reduced one-dimensional
pressure profiles shown in Fig. 10.8.
The effects of swirl number on the acoustic field were examined by considering a
low swirl number of S = 0.3. The dominant frequency over the entire field becomes
224 Y. HUANG ET AL.
3900 Hz, which corresponds to the mixed first tangential (1T) and first radial (1R)
acoustic mode in the chamber. A proper orthogonal decomposition (POD)81 anal-
ysis of the fluctuating pressure flowfield also confirms the prevalence of the mixed
1T/1R acoustic mode. As previously mentioned, the two shear layers originating
from the trailing edges of the centerbody and the backward-facing step may exert
significant influence on the oscillatory flow characteristics. Their specific effects
depend on the swirl number and the chamber geometry. At the high swirl number of
S = 0.5, the large vortical structure associated with the central recirculating flow
overshadows the shear layer originating from the corner region and dominates the
flow development in the chamber. The resultant acoustic wave thus has a character-
istic frequency of 1380 Hz, which matches the frequency of the centerbody shear-
layer instability. At a low swirl number of S = 0.3, no vortex breakdown occurs and
the importance of the flowfield on the downstream side of the centerbody in excit-
ing acoustic oscillations diminishes. In contrast, the shear layer in the corner region
plays a crucial role in dictating the acoustic flow evolution, whose characteristic fre-
quency of 3900 Hz matches that of the mixed 1T/1R acoustic mode in the chamber.
In short, the dominant acoustic mode in the chamber is sensitive to unsteady
vorticity evolution, which in turn strongly depends on the swirl number.
B. Flow Dynamics of Radial-Entry Swirl Injector

This section examines the flow dynamics in an air-blast swirl injector with
radial entry. This type of injector has been widely used in contemporary gas-
turbine engines because its atomization performance is superior to that of pressure
injectors. The model considered herein consists of a mixing duct and a fuel nozzle
located coaxially at the head end,82 as shown schematically in Fig. 10.10. The
S1
S2 S3
1st guide vane

2nd guide vane
y
fuel
x nozzle
S2 S3
S1
Fig. 10.10 Schematic of gas turbine swirl injector with radial entry.
mixing duct includes a center cylindrical passage and two annular passages, which
are spaced radially outward from the axial axis. Three radial-entry swirlers—
denoted S1 , S2 , and S3 and counter-rotating with each other—are located at the
entrance. This injector is also referred to as a high-shear nozzle/swirler (HSNS)
and has four major advantages. First, it reduces smoke by introducing high swirl
from the first passage. Second, the middle swirler is implemented to generate
strong shear layers in both the axial and azimuthal directions and to reduce the
overall swirl angle. Consequently, the fuel–air mixing is improved. The inclusion
of the second passage makes it easy to control the initial swirl number of the flow.
Third, a stronger CTRZ is generated, which increases the relight stability. Finally,
the relight stability and the total flow rate can be decoupled by shifting the airflow
through the third passage.82
The mixing duct in the injector, shown in Fig. 10.10, has a diameter of D0 =
32 mm at the exit. Two different sets of swirl vanes are explored herein.62, 63 The
low swirl-number (LSN) case has swirl vane angles of S1 = 30 deg, S2 = −45 deg,
and S3 = 50 deg; and the high swirl-number (HSN) case has S1 = 45 deg, S2 =
−60 deg, and S3 = 70 deg. The corresponding swirl numbers are 0.35 and 0.49,
respectively, based on the flow properties at the injector exit. The baseline flow
condition includes an ambient pressure of 1 atm, an inlet temperature of 293 K,
and a mass flow rate of 0.077 kg/s. The Reynolds number that is based on the
diameter and the bulk axial velocity at the exit is 2 × 105 .
1. Vortical Flow Evolution

Figure 10.11 shows snapshots of the vorticity magnitude fields on two cross
sections for both the high and low swirl numbers. The flow evolution exhibits
several distinct features, as follows. First, when the flow travels downstream of
the centerbody, the strong swirling motion and its associated centrifugal force
produce large radial pressure gradients, which then induce a low-pressure core
around the centerline. As the flow expands and the azimuthal velocity decays
with the axial distance, the pressure is recovered. A positive pressure gradient is
consequently generated in the axial direction and leads to the formation of a central
recirculating flow, a phenomenon commonly referred to as vortex breakdown or
vortex burst. The resultant flow detachment from the rim of the centerbody gives
rise to a vorticity layer, which subsequently rolls, tilts, stretches, and breaks up into
small eddies. These small vorticity bulbs interact and merge with the surrounding
flow structures while being convected downstream. The entire process is highly
unsteady and involves a wide range of length and time scales.
Second, because of the opposition of the swirler vane angles, two counter-
rotating flows with different velocities in the streamwise and azimuthal directions
merge at the trailing edges of the guide vanes. Vortices are generated in the shear-
layer regions and shed downstream sequentially because of the Kelvin–Helmholtz
instabilities. In comparison with the vortex-breakdown-induced central recirculat-
ing flow, the flow structures associated with the periodic vortex shedding in the
outer region are small and well organized. The shear-layer instability, along with
the helical and centrifugal instabilities, induces large asymmetric structures on the
transverse plane. Finally, the aforementioned flow structures in various parts of the
injector and their underlying mechanisms interact and compete with one another.
226 Y. HUANG ET AL.
Fig. 10.11 Snapshots of vorticity magnitude contours: a) low swirl number and
b) high swirl number.
When the swirl number changes, the dominant instability mode may switch cor-
respondingly. A detailed analysis of these phenomena is given in the following
sections.
a. Vortex Breakdown. Much insight into the vortex breakdown in the core
flow region can be obtained from the isosurfaces of the azimuthal velocity shown in
Fig. 10.12. In the low swirl-number case, a stable bubble type of vortex breakdown
is clearly observed in the downstream region of the centerbody, whereas a much
more complex structure prevails at the high swirl number. The streamlines of
the mean flowfields given in Fig. 10.13 quantitatively reveal the formation of a
central toroidal recirculation zone in this region. As the swirl number increases,
the size of the recirculation zone accordingly becomes greater. The stagnation
point of the vortex breakdown moves upstream for an equilibrium position and
finally reaches the centerbody. The local flow development depends on the relative
magnitudes of the downward momentum inertia of the incoming flow and the
outward flow motion arising from the centrifugal force. Although the downward
momentum inertia remains almost the same because of the fixed inlet mass flow
rate employed, the weaker centrifugal force in the low swirl-number case causes the
a)
θ
−
b)
θ
−
Fig. 10.12 Instantaneous isosurfaces of azimuthal velocities at uθ = 10 and 50 m/s:

a) low swirl number and b) high swirl number.
228 Y. HUANG ET AL.
Fig. 10.13 Streamlines of mean flowfields for swirl numbers of S = 0.35 and 0.49.
incoming flow to penetrate all the way to the core region, as evidenced in Fig. 10.13.
The ensuing flow structure bears a close resemblance to a tornado near the ground
where a large accumulation of vorticity in the center region takes place, a kind of
collapse of the swirling flow.83
The temporal evolution of the flowfield permits insight into the vortex break-
down phenomenon. Figure 10.14 shows instantaneous streamlines on a longitudi-
nal plane, spatially averaged in the azimuthal direction, at various times during a
typical flow evolution period for the high swirl-number case. Uneven time inter-
vals between frames were chosen to show the important phases of the oscillation.
Obviously, the spatially averaged flow structures are more distinguishable than
Fig. 10.14 Close-up views of streamlines downstream of centerbody for high swirl-
number case of S = 0.49. Flowfields spatially averaged in azimuthal direction. The
time interval between pictures is not constant.
a)
b)
Fig. 10.15 Snapshots of azimuthal velocity fields on four transverse cross sections,
contour levels between –70 and 120 m/s with increment of 10 m/s. Solid lines represent
positive values and dashed lines represent negative values: a) low swirl number and
b) high swirl number.
those of the original three-dimensional flowfield, which is too complex to allow an

effective analysis. Two large vortices exist in the region downstream of the center-
body, and they evolve in two different forms. First, between 14.45 and 14.85 ms,
a small vortex separates from its parent structure, travels downstream, and even-
tually coalesces with the large vortex located in the downstream region. In a later
stage, between 15.25 and 15.85 ms, a small vortex is generated in front of the
array of vortices; and the large vortex, which is normally anchored at the cen-
terbody, is detached, causing a switch in the flow topology. The instantaneous
flow pattern at 15.85 mm is considerably different from its time-mean counterpart
and bears a close resemblance to the situation for the low swirl-number case, in
which a strong wall jet exists in the wake of the centerbody and the incoming flow
can penetrate deeply into the core region. The temporal variation in the vortical
structure affects the injector characteristics through its influence on the effective
flow-passage area.
b. Outer Shear-Layer Instability. Vortex shedding arising from the Kelvin–
Helmholtz instabilities in both the axial and azimuthal directions takes place at the
trailing edges of the guide vanes. The flow evolution in the azimuthal direction,
as shown in Fig. 10.15, clearly indicates the existence of an outer shear layer
because of the counter-rotating flows through the first and second passages and
a center recirculating flow induced by the vortex breakdown. For the low swirl-
number case, the azimuthal velocity remains almost uniform up to x = 11 mm
in spite of the small-scale turbulence embedded in the inlet flow. Large organized
structures then develop under the effect of the Kelvin–Helmholtz instability when
the incoming streams merge. The situation becomes more obvious for the high
230 Y. HUANG ET AL.
swirl-number case. The center recirculating flow even intersects the outer shear
layer, causing a complex flowfield near the injector exit.
The dominant frequency of the vortex shedding because of the Kelvin–
Helmholtz instability in the streamwise direction can be estimated by using Eq.
(10.4). In the present configuration, the mean velocity, Ū , is 50 m/s, and the mo-
mentum thickness of the shear layer, θ , is around 0.2 mm for both swirl numbers.
The frequency of the most unstable mode, f n , is estimated to be 1 × 104 Hz.
This value is comparable with the numerically calculated instability frequency
of 13,000 Hz, further demonstrating that the outer shear flow dynamics is dic-
tated by the Kelvin–Helmholtz instability in the streamwise direction in the low
swirl-number case.
The situation is vastly different in the high swirl-number case. As a result of the
strong shear force and the associated Kelvin–Helmholtz, helical, and centrifugal
instabilities in the azimuthal direction, the flow becomes highly disordered soon
after the incoming streams merge in the region downstream of the guide vanes.
The interaction between the outer shear layer and the central toroidal recirculating
flow also contributes to the eddy breakup and mixing processes.
c. Interaction and Competition of Instability Modes. As previously men-
tioned, three major flow mechanisms (i.e., vortex breakdown, Kelvin–Helmholtz
instability, and helical instability) exist and interact with one another within the
injector. The specific type of coupling depends on the swirl number and can be
classified in two categories. First, the outer shear layer may interact with the large
disorganized structures arising from evolution of the central recirculating flow
when the swirl number exceeds a threshold value, as evidenced in Fig. 10.11. The
interaction usually increases with increasing swirl number and varies within each
flow evolution period. The vortex shedding tends to be more organized when the
center recirculation zone shrinks and less organized when it grows. The turbulent
kinetic energy in the central recirculation zone and in the wake of the guide vanes
is much greater than that in the rest of the domain because of vigorous vortical
motions in these regions. The two shear layers are distinctly separate in the low
swirl-number case but merge in the high swirl-number case. Because liquid fuel is
delivered into the injector from the centerbody, the high turbulence intensity in this
region can significantly enhance the atomization of the injected fuel. At the same
time, the strong shear stress in the downstream region of the second guide vane
promotes rapid mixing between the air and the fuel impinging and accumulating
on the second guide vane.
In the second type of flow coupling, the instability waves in the axial and az-
imuthal directions in the outer shear layer compete with each other. In the low
swirl-number case, the streamwise instability dominates the shear-layer evolution;
therefore, the billow structures and subsequent hairpin vortices prevail in the flow-
field. In the high swirl-number case, the development of the billows is suppressed
and flow structures are severely distorted by the azimuthal flow instabilities.
Several other competing mechanisms may also exist in the flowfield, such as the
one involving the Kelvin–Helmholtz and centrifugal instabilities. Swirling flows
usually result in an unstable radial stratification, thereby leading to centrifugal
instability,83 which is enhanced by a higher azimuthal velocity gradient and which
further influences the streamwise Kelvin–Helmholtz instability in the outer shear
layer.
2. Spectral Characteristics
The injector dynamics involve an array of intricate flow processes characterized
by a wide range of time and length scales. Quantitative information can be obtained
by using spectral and proper-orthogonal-decomposition analyses for the low swirl-
number case, the dominant frequency of pressure oscillation along the main flow
passage is 13,000 Hz, corresponding to the most amplified mode of the shear-
layer instability downstream of the first guide vane. In the outer region of the
central recirculation zone, the prevalent frequency of 5783 Hz corresponds to the
precession of the vortex core (PVC). The phenomenon is confirmed by visual
inspection of the flow evolution data.
The situation is qualitatively different for the high swirl-number case, as shown
in Fig. 10.16. As a consequence of the strong interactions between the outer shear
layer and the central recirculation zone, the spectral content of the flowfield be-
comes very rich and is characterized by several different frequencies in various
regions. A low-frequency mode around 500 Hz dominates the flow oscillations
Fig. 10.16 Frequency spectra of pressure oscillations along main flow passage; high
swirl flow (S = 0.49).
232 Y. HUANG ET AL.
p: −4 −3 −2 −1 0 1 2 3 4
Fig. 10.17 POD mode shape of pressure field showing existence of precessing vortex;
f = 4.0 kHz.
near the inlet (probes 1-1 and 1-2), whereas high-frequency modes around 4000 Hz
prevail in the downstream region (probe 1-4). The former may be attributed to the
flow displacement effect of the central recirculation zone. The occurrence of the
4000 Hz oscillation at the injector exit can be explained by considering the flow
development along the boundary of the central recirculation zone in Fig. 10.17,
which shows the three-dimensional POD mode shape corresponding to the fre-
quency of 4000 Hz. The existence of precessing vortex motion in the outer region
of the central recirculating flow is revealed. Figure 10.18 shows the characteris-
tic frequencies in different regions of the flowfield. The prevalence of distinct
frequencies in different regions suggests that the flow instability mechanisms
vary in different regions, a phenomenon consistent with Martin and Meiburg’s
expectation.84
3. Injector Response to External Excitation

Most previous studies on gas-turbine combustion instabilities focused on
thermal–acoustic interactions in the chamber. The dynamic behavior of an in-
jector was loosely modeled with an acoustic admittance function at the injector
exit; the specific value of this function was treated as an empirical coefficient.
Very limited effort was applied to investigate the injector internal flow evolution
and its response to external forcing. This section examines the response of the
swirl injector by exciting the system at discrete sinusoidal frequencies.63 Periodic
oscillations of the mass flow rate ṁ are enforced at the injector entrance, similar
to the experiment conducted by Cohen and Hibshman,85
ṁ = ṁ 0 [1 + α sin(2π f F t)] (10.5)
where ṁ 0 and f F denote the mean mass flow rate and the forcing frequency,
respectively. The amplitude of the oscillation, α, is fixed at 10%. The forcing
fsh ~
ear 13k fshear~13k Hz
Hz
fpassage~1.2-1.8k Hz
fPVC ~ 4.0k Hz
recirculation zone
facoustic~11k Hz
fvortical~1.7k Hz
Fig. 10.18 Characteristic frequencies in the injector at high swirl number.
frequency covers a range from 400 through 13,000 Hz, commensurate with the
broadband nature of the injector flow dynamics. Only the higher swirl-number
case with S1 = 45 deg, S2 = −60 deg, and S3 = 70 deg is considered herein.
The vortical and acoustic fields in the injector can be globally characterized
by two frequencies, f v and f a , measuring the convective and acoustic motions,
respectively. The former can be estimated by the mean flow residence time and
has a value of 1.7 kHz. The latter is obtained on the basis of the time required
for a downstream acoustic wave to travel through the injector and has a value of
11 kHz. The phase difference of the traveling acoustic wave between the entrance
and the exit of the injector, θ , is
θ ≈ 2π L/l F = 2π f F / f a (10.6)
where L is the length of the main flow passage and l F is the acoustic wavelength
at the forcing frequency.
a. Instantaneous Flow Structures. Figure 10.19 shows snapshots of the fluc-
tuating vorticity magnitude fields, | |, obtained by subtracting the longtime aver-
aged quantity from its instantaneous value, at various forcing frequencies. When
the frequency is higher than f v , well-defined vortical structures are observed in
the forward section of the injector. These waves, generated by the flow oscilla-
tions at the entrance, are convected downstream with the local flow velocity. The
wavelength is inversely proportional to the forcing frequency and shortens in the
234 Y. HUANG ET AL.
Fig. 10.19 Snapshots of fluctuating vorticity magnitude field on a longitudinal cross

section under conditions with and without forcing. Contour levels between 103 and
105 1/s with exponential distribution.
middle region of the injector because of the flow turning effect, that is, the flow
direction turns in this region and the velocity component perpendicular to the
wave front decreases. The intensive turbulent fluctuations downstream of the cen-
terbody overshadow the organized vortical waves, which are eventually damped
out by turbulent diffusion and viscous dissipation. When the forcing frequency is
less than f v , it is difficult to clearly observe organized vortical waves inside the in-
jector because of the long vortical wavelengths associated with the low-frequency
oscillations.
Figure 10.20 shows snapshots of the fluctuating velocity and pressure fields
under external forcing with a frequency of 13 kHz. This case was chosen be-
cause of the presence of a well-established vortical wave, which helps identify
the disturbance propagation mechanisms. The vortical wave is mainly aligned
with the fluctuating azimuthal velocity, whereas the acoustic wave is most closely
related to the pressure oscillation. The imposed excitation at the injector entrance
can be decomposed into two components in the azimuthal and radial directions.
The former generates a vortical wave because of the shear stress resulting from
the flow oscillation in the azimuthal direction, and its dynamics are governed by the
conservation of angular momentum. The latter produces an irrotational, traveling
acoustic wave and can be characterized by the pressure and streamwise-velocity
fluctuations through mass conservation.
Fig. 10.20 Snapshots of velocity and pressure fluctuations at forcing frequency of

fF = 13 kHz. Velocity contour levels between –49 and 49 m/s with increment of
0.2 m/s in square root of velocity magnitude; pressure contour levels between –
6 and 3 kPa with increment of 0.1 kPa.
To further clarify the wave-propagation mechanisms, the fluctuating velocity

components in various regions of the injector are investigated. Figure 10.21 shows
the temporal variations of the velocity fluctuations in the streamwise and azimuthal
directions at three different locations along the streamline originating from the
middle point of the entrance. These measurement points are all in the forward
section of the injector; and the corresponding distances from the entrance are 0,
5.9, and 12.2 mm. Both the streamwise and azimuthal velocity fluctuations in-
crease when the fluid particles travel downstream, because of the conservation
of mass and angular momentum, respectively. The traces are smoothed by fil-
tering out the background turbulence. Of particular interest is the propagation
of the streamwise disturbance in the form of an acoustic wave with its phase
speed equal to the local acoustic-wave propagation speed. The flow disturbance
in the azimuthal direction, however, travels in the form of a convective–vortical
wave, with its phase speed equal to the local flow velocity. The large disparity
between the two phase speeds indicates that the streamwise disturbance arrives
in the downstream region much earlier than its azimuthal counterpart. This phe-
nomenon of decomposed oscillations is analogous to wave propagation during an
earthquake: the vertical oscillation is always detected earlier than the horizontal
counterpart at the surface because of the higher propagation speed of the vertical
oscillation.
The decoupling between the two velocity oscillations is significant in that the
development of the oscillating flowfields in different spatial directions may be
distinct. Because the fluctuations have the same frequency but different speeds, the
236 Y. HUANG ET AL.
azimuthal velocity streamwise velocity

15
10
Distance, mm
5
vortical wave acoustic wave
~40 m/s ~400 m/s
0
35.4 35.6 35.8 36 36.2

Time, ms
Fig. 10.21 Fluctuations of streamwise and azimuthal velocities at three different

locations along the streamline originating from the middle point of the entrance.
vortical wavelength is smaller than its acoustic counterpart by almost one order
of magnitude. Considering the injector dimension and forcing frequency under
consideration, the wavelength of the organized vortical motion is closer to the
large scales in various regions of the injector and are less than the characteristic
length of the main flow passage. Since interactions between flow motions with
similar scales are generally stronger than those with highly disparate scales, a
vortical wave with a frequency higher than f v (i.e., wavelength less than the flow-
passage length) exerts more significant influence on the energy transfer process
involving different scales.
The impressed periodic forcing provides an additional channel to transfer energy
between the mean and turbulent flowfields through organized motions.63, 66, 86, 87
This energy redistribution process is manifested by the presence of vorticity pock-
ets in the flowfield, in which the fluctuating vorticity is greater than a prespecified
threshold value, ||T . Figure 10.19 shows that those pockets with intensive vor-
ticity fluctuation are enhanced at low forcing frequencies (e.g., 500 and 1500 Hz),
but suppressed at high forcing frequencies (e.g., 4000 and 13,000 Hz).
Figure 10.22 shows the evolution of the instantaneous axial velocity field, which
is spatially averaged in the azimuthal direction, within one cycle of oscillation at
a forcing frequency of 1500 Hz. Also included is the time trace of the mass flow
rate at the injector exit, obtained by filtering out turbulent fluctuations. When the
mass flow rate achieves the maximum at t f F = 38.43, a ring structure with strong
positive velocity appears in the downstream region of the second guide vane, where
the mean axial-flow velocity also reaches its maximum. The ring structure then
sheds downstream while the mass flow rate decreases. A new one is produced
when the mass flow rate increases in a new cycle. The strong flow oscillation in
this region can potentially influence the atomization process of the liquid film
accumulated on the surface of the second guide vane. This evolution pattern,
however, cannot be observed at the other forcing frequencies. The discrepancy may
be attributed to the fact that 1500 Hz is closer to the characteristic frequency of flow
Fig. 10.22 Time evolution of axial velocity field within one cycle of oscillation with
forcing frequency of 1500 Hz, spatially averaged in azimuthal direction. Contour levels
between –50 and 100 m/s with increment of 6 m/s. Solid lines represent positive values
and dashed lines represent negative values.
238 Y. HUANG ET AL.
free oscillation fF = 1500 Hz
fF = 4000 Hz fF = 13,000 Hz
Fig. 10.23 Effect of forcing frequency on longtime averaged azimuthal velocity field.
Contour levels between –90 and 150 m/s with increment of 10 m/s. Solid lines represent
positive values and dashed lines represent negative values.
convection, f v , than the others studied under this flow configuration. The flow
tends to resonate with the external excitation at this frequency in the streamwise
direction.
b. Mean Flow Properties. Figure 10.23 shows the longtime averaged az-
imuthal velocity fields at f F = 1500, 4000, and 13,000 Hz. No discernible differ-
ence is observed between the flows with and without external excitations except in
the region where the counter-rotating flows through the S1 and S2 swirlers merge
at f F = 13, 000 Hz. The mixing region can be characterized by the line of zero
azimuthal velocity, which shrinks by almost half at this forcing frequency. The
impressed oscillation resonates with the local shear-layer instability (i.e., 13,000
Hz) when the two frequencies match each other and even causes the reversal of the
azimuthal flow direction. The effect of external forcing on flow development can
be further examined in Fig. 10.24, which shows snapshots of the isosurfaces of
the azimuthal velocities at u θ = −2 and 2 m/s in the azimuthal phase space (θ =
0 to 360 deg). The flowfield exhibits a helical structure that originates from the
trailing edge of the first guide vane under conditions without external forcing. The
coherent structure, however, is destroyed by the impressed axisymmetric distur-
bance at the injector entrance and breaks into small bulbs. As discussed previously,
in addition to the external forcing, two mechanisms contribute to this phenomenon:
the Kelvin–Helmholtz instability in the azimuthal direction and centrifugal insta-
bility. Both of them strongly depend on the swirl number. The ensuing enhancement
of flow instability in the azimuthal direction considerably enhances local turbulent
mixing.
a)
b)
Fig. 10.24 Snapshots of isosurfaces of azimuthal velocities at uθ = −2 and 2 m/s in

azimuthal phase space (θ = 0 to 360 deg) in shear layer originating from trailing edge
of first guide vane: a) top view and b) bottom view.
240 Y. HUANG ET AL.
It should be noted that in an operational injector, liquid fuel injected from the
centerbody impinges onto the inner surface of the second guide vane and forms a
liquid film, which is then atomized to a spray of fine droplets by the local shear
flow near the rim of the second guide vane. The potential influence of external
forcing on the breakup of the liquid film appears in the two conflicting areas.
On the one hand, the strong fluctuation in the azimuthal direction promotes the
development of an instability wave on the fuel filming surface and the subsequent
atomization process.88 On the other hand, as shown in Fig. 10.23, the external
forcing may significantly modify the mean azimuthal velocity field near the fuel
filming surface, especially when the forcing frequency approaches the shear-layer
characteristic frequency (i.e., 13 kHz). The flow near the downstream part of
the second guide vane even changes its direction from counter-rotating to co-
rotating with the flows in the main and the third (S3 ) passages. This qualitative
switch of flow pattern represents an undesired feature from the perspective of fuel
atomization.23
In spite of the modification of the flowfield between the first and second guide
vanes at f F = 13 kHz, the distribution of the turbulent kinetic energy appears to be
insensitive to external forcing in the bulk of the flowfield. This may be attributed
to the weakness of the excitation as compared with the intrinsic high-intensity
flow motion. The kinetic energy of the periodic motion is considerably smaller
than that of the turbulent motion at the injector outlet. The broadband nature of
the injector flow also discourages the modulation of the mean flow by a single-
harmonic excitation unless the forcing resonates with the local flow structure at
appropriate frequencies.89
c. Acoustic Admittance at Injector Exit. The global response of the injector
can be described by the acoustic admittance at the exit. The information obtained
can be effectively used to serve as the upstream boundary condition for analyzing
the unsteady flow motion in a combustion chamber.90 The admittance function,
also the reciprocal of the impedance function, measures the velocity fluctuation in
response to incident pressure fluctuation. Following common practice, the acoustic
admittance function, Ad , is defined as
û a /ā
Ad ( f ) = (10.7)
p̂a /γ p̄
where p̄ and ā denote the mean pressure and the speed of sound, respectively.
The overhat ( )a represents the Fourier component of the oscillation at the forcing
frequency. Because the background noise in the free-forcing case is too strong
to obtain meaningful results, an external excitation is required to determine the
acoustic admittance at the frequency of concern.
Figure 10.25 shows the radial distributions of the admittance functions at the
injector exit for four different forcing frequencies: 500, 900, 1500, and 4000Hz.
The maximum response occurs at 500 Hz, especially near the rim of the second
guide vane. Excitations at 500, 900, and 1500Hz exhibit the same trend; and the
admittances achieve their maxima when the outer boundary r = R0 is approached.
This outcome may be attributed to the relatively low pressure oscillation and high
velocity fluctuations near the upper boundary. In this region, the pressure response
1.0
r/R0
0.5 fF (Hz)
500
900
1500
4000
0.0
0 5 10
Magnitude of Ad
1.0
fF (Hz)
500
900
r/R0
0.5 1500
4000
0.0
−1.0π −0.5π 0.0 0.5π 1.0π
Phase of Ad , radian
Fig. 10.25 Radial distributions of acoustic admittance function at injector exit for
different forcing frequencies.
at 500 Hz forcing is less than 300 Pa, which is smaller than its counterparts at other
excitation frequencies (> 1000 Pa). When the oscillation is impressed at 4000 Hz,
the velocity response in the outer region (0.8 < r/R0 < 1.0) becomes very small.
Since the liquid film breaks up in the trailing edge of the second guide vane,
the flow response in this region plays an important role in dictating the dynamic
behavior of the liquid fuel.85 A small pressure oscillation at 500 Hz may result in
a large velocity fluctuation, which consequently exerts a strong influence on spray
formation at that location.
The phase distribution of the admittance function indicates a lag between
the velocity and pressure fluctuations of around 90 deg in the main flow passage
242 Y. HUANG ET AL.
(0.3 < r/R0 < 0.8). The situation is consistent with the behavior of a simple trav-
eling acoustic wave without much influence from shear layers. The phase behavior
for the 4000 Hz case exhibits a trend distinct from that of the other cases, especially
in the central recirculation zone. A major factor contributing to this phenomenon
is the proximity of this forcing frequency to the characteristic frequency of the
central recirculating flow. The imposed axisymmetric excitation in the streamwise
and azimuthal directions does not promote the evolution of the precessing vortex
along the boundary of the central recirculation zone. The pressure and velocity
coupling at 4000 Hz differs from that at other frequencies because of the phase
difference between the oscillations induced by external forcing and intrinsic flow
instabilities.
d. Mass Transfer Function. Another important measure of the injectors dy-
namic response is the transfer function of the total mass flow rate between the
injector entrance and exit, defined as
ṁˆ aex

m ( f ) = (10.8)
ṁˆ in
a
Here ṁˆ a is the Fourier component of the mass flow rate at the forcing frequency,
which is obtained by integrating the mass flux over the entire surface of concern.
Figure 10.26 shows the magnitude and phase of
m as a function of the forcing
frequency. The magnitude reaches its maximum at f F = 1500 Hz, as expected
from the previous results. A large disparity in the fluctuation of the mass flow rate
between the entrance and exit is clearly found. At first glance, this observation
seems to violate the law of mass conservation for such an acoustically compact
2
Magnitude of Πm
0
0
Phase of Πm , radian
−π
−2π
0 5000 10000
Forcing frequency, Hz
Fig. 10.26 Effect of forcing frequency on transfer function of total mass flow rate.
injector. The forcing frequency is much lower than the acoustic characteristic fre-
quency of the injector, f a . Under this condition, the flowfield in the injector can
be treated as incompressible, and the instantaneous total mass flow rate at the en-
trance and exit should be identical. To explore the underlying physical mechanisms
responsible for the phenomenon shown in Fig. 10.26, and to ensure numerical ac-
curacy, the time-averaged mass flow rates at the injector entrance and exit are
calculated. The result confirms the conservation of the overall mass flow rate for
all the forcing frequencies considered herein. The 1500 Hz forcing indeed excites
the flowfield at the expense of suppressing fluctuations at other frequencies. The
mass-flow transfer function for the 4000 Hz forcing is less than unity. In addition to
channeling mass flow among different Fourier components, flow compressibility
takes effect at high-frequency forcing, allowing temporarily for a relatively large
mass variation inside the injector. In short, the forcing frequency affects not only
the spatial distribution of the mass flux fluctuation but also the temporal variation
of the overall mass flow rate.
The phase shift in Fig. 10.26 exhibits a linear distribution with the forcing
frequency because of compressibility effects. This phenomenon can be examined
by using the acoustic characteristic frequency, f a , and the phase difference, θ ,
in Eq. (10.6). The good agreement between the analytic estimation, Eq.(10.6),
and the numerical result further verifies that the oscillation of the mass flow rate
propagates in the form of an acoustic wave.
C. Flow Dynamics of Axial-Entry Swirl Injector

This section deals with the flow dynamics of an axial-entry injector, the CFM56
swirl injector of GE Aircraft Engines, as shown in Fig. 10.27. It is a prefilming
airblast injector, and it has been implemented in aero gas-turbine engines because
of its high combustion efficiency, broad lean blowout (LBO) limit at low power,
and low NOx and smoke emissions during high-power operations.43
The injector consists of eight counterclockwise elliptical primary-jet inlets, ten
clockwise secondary vanes, a venturi, and a flare. The fuel nozzle is located at the
center of the primary-jet plane, where air swirlers give rise to a strong swirling
flowfield around the fuel nozzle outlet. As a result, a region with high-intensity
turbulence and strong shear stress is established in the vicinity of the fuel nozzle,
and a finely atomized spray is produced in this region. The fuel droplets carried by
the primary counterclockwise air stream mix with the counter-rotating secondary
flow to further promote rapid fuel–air mixing. At the same time, part of the liquid
fuel injected from the centerbody impinges onto the inner surface of the venturi
and forms a liquid film, which is then atomized to a spray of fine droplets by
the local shear flow near the entrance of the secondary swirlers. In an effort to
optimize the combustor design, characterization of the flowfield inside the swirl
injector becomes critical.43, 75
The injector flow was studied in a 3 × 3 dump chamber by using laser-Doppler
velocimetry (LDV). The inlet air was supplied with 4.0% pressure drop across the
injector at 1 atm and 291 K. The diameter, D0 = 27 mm, and mean axial velocity,
U0 = 30 m/s, at the downstream side of the secondary swirl vanes were employed
as the reference length and velocity, respectively. The corresponding Reynolds
number is 5.4 × 104 .
Next Page
244 Y. HUANG ET AL.
Fig. 10.27 Schematic of CFM56 gas-turbine swirl cup assembly with radial jet entry.
1. Mean Flow Properties

Figure 10.28 shows the time-mean axial velocity field, in which a central recir-
culation zone is clearly observed. Owing to the strong swirling flow induced by
the primary swirlers, steep pressure gradients are established in the radial direc-
tion to balance the centrifugal force, and they lead to a low-pressure core around
the centerline. The pressure is then recovered in the downstream region as the
flow expands and the azimuthal velocity decays with the axial distance. Conse-
quently, a positive (adverse) pressure gradient forms along the axial axis, and flow
reversal occurs. The recirculating flow originates from the middle of the venturi at
x = 0.15D0 and extends into the chamber at x = 4.89D0 . The calculated length
of the recirculation zone closely matches the experimental measurement.43
The central recirculating flow affects the injector performance in two areas.
First, the flow provides a low-speed region with high turbulence intensity, which
promotes fuel–air mixing and stabilizes combustion with reduced emissions. Sec-
ond, the recirculation zone exerts a blockage effect and reduces the effective flow
passage area in the injector, consequently increasing the flow velocity in the outer
Previous Page
Fig. 10.28 Contours of normalized mean axial velocity, counter levels between
−0.6 and 2.2, with increment of 0.2. Solid lines represent positive values and dashed
lines represent negative values.
region. The resultant strong shear stress near the wall, especially in the vicinity of
the venturi surface, facilitates atomization of the injected fuel. It is worth noting
that in an operational injector, liquid fuel injected from the nozzle impinges onto
the venturi surface and forms a thin film, which needs to be atomized to a spray
of fine droplets by the local shear flow. The flowfield near the injector wall plays a
crucial role in dictating the liquid-sheet-breakup and droplet-formation processes.
The turbulent kinetic energy field shown in Fig. 10.29 exhibits three different re-
gions with high turbulence intensity. Region 1, formed by the merging of the eight
swirl jets, includes the head-end region near the centerline, where the liquid fuel is
discharged. The vigorous flow motion promotes liquid breakup and atomization.
Region 2 covers the stagnation point of the central recirculating flow. A precessing
vortex core is also observed originating from this region, and it is addressed in
detail in the following section. Both the vortex precession and flow recirculation
enhance local flow oscillations. Region 3 consists of the field surrounding the cen-
tral recirculation zone in the main flow passage. Large coherent structures evolve,
especially when the primary flow merges with the counter-rotating flow through
the secondary swirlers. The strong radial motion accelerates the development of
the Kelvin–Helmholtz instability in the liquid film issuing from the trailing edge
of the venturi. At the same time, the shear force associated with the counter-rotating
flow in the azimuthal direction also enhances the atomization process. Thus, the
flow structure in the present injector provides an effective capability to atomize
the liquid film.
2. Unsteady Flow Evolution: Precessing Vortex

One of the most important flow characteristics of the present injector is the
precession of the recirculating flow around the axial axis at a well-defined fre-
quency. This phenomenon is often observed when vortex breakdown occurs in a
high-Reynolds-number flow, such as turbulent swirling flows in cyclone chambers
and combustion devices.91,92 Figure 10.30 shows a snapshot of an isobaric surface
246 Y. HUANG ET AL.
Fig. 10.29 Distribution of normalized turbulent kinetic energy.
Fig. 10.30 Isobaric surface at p = 99, 500 Pa.

a) b)
c) d)
Fig. 10.31 Time evolution of streamlines and pressure field (time increment of
0.1 ms); the thick dark line indicates the contour of zero axial velocity.
with p = 99.5 kPa, indicating the existence of a PVC structure. The low-pressure
core initially aligns with the axial axis in the region downstream of the fuel nozzle.
It is then driven away from the centerline at the stagnation point of the recirculat-
ing flow and extends downstream spirally against the direction of the main flow
rotation, although the whole structure follows the main flow. Figure 10.31 shows
snapshots of the instantaneous streamlines and pressure fields on a longitudinal
plane at various times. The clustered streamlines indicate that large vortices around
the low-pressure core are pushed outward. Furthermore, the induced low-pressure
core is located outside the region defined by zero axial velocity. The situation is
consistent with the observation of Syred and Beer,93 in that the PVC is located
between the zero axial-velocity and dividing surfaces in the outer region of the
recirculation zone. When the vortex core rotates in the injector, large structures
are peeled off from the spiral core periodically and are convected downstream by
the local flow.
Novak and Sarpkaya94 found that the direction of the spiral winding changed
randomly in their experimental study of turbulent swirling flows at high Reynolds
numbers. Lucca-Negro and O’Doherty92 reviewed the twist direction for the spiral
form of vortex breakdown in detail but drew no definite conclusions. Although the
248 Y. HUANG ET AL.
Fig. 10.32 First two POD modes of normalized fluctuating pressure field.
vortex precession at low Reynolds numbers is not identical to the spiral form of
vortex breakdown, the precessing process can be considered as periodical spiral
motion in the conical form of turbulent vortex breakdown observed by Novak and
Sarpkaya.94
Figure 10.32 shows the first two POD mode shapes of the fluctuating pressure
field, which account for more than 30% of the total energy of the field. The mode
shapes and associated time-varying coefficients demonstrate the presence of spiral
rotating motion (i.e., vortex breakdown) at a frequency of 1266 Hz. This result
further confirms that vortex precession is a dominant mechanism in the cold flow
evolution.
3. Effects of Swirler Orientation on Flow Development

The influence of the inlet flow direction on the injector dynamics is explored
by reversing the orientation of the secondary swirl vanes. Thus, both the primary
and secondary swirlers generate counterclockwise swirling flows. Figure 10.33
shows the radial distributions of three velocity components in the injector. The
mean velocity field in the co-rotating case bears a close resemblance to its coun-
terpart in the counter-rotating case, except that the former possesses a much larger
recirculation zone because of the stronger swirling motion. It is worth noting that
a large recirculation zone tends to reduce the effective flow passage in the injector
and consequently weakens the effects of the secondary swirling flow on the flow
development within the venturi. As a result, the injector flow evolution is mainly
driven by flow through the primary swirlers. The difference of the flow pattern in
the streamwise and spanwise directions between the two cases is modest. In the
flare and chamber, the flow disparity is limited, except for a noticeable difference
in the azimuthal velocity field and a deficiency of the axial velocity along the
centerline in the co-rotating case because of its higher swirl strength.
U(Ux)
1
y/D0
−1
1 2 3 4 5 6
x/D0
V(Ur)
1
y/D0
−1
1 2 3 4 5 6
x/D0
W(Uθ)
1
y/D0
−1
1 2 3 4 5 6
x/D0
Fig. 10.33 Radial distributions of mean-flow axial, radial, and azimuthal velocity
components (symbol: co-rotation; line: counter-rotation).
250 Y. HUANG ET AL.
In addition to the central recirculation zone, another major difference between

the two configurations lies in the distributions of turbulent kinetic energy and shear
stress near the exit of the venturi. At that location in the counter-rotating case, strong
shear stress is observed not only near the boundary of the central recirculation zone
as in the co-rotating case, but also near the tip of the prefilming surface. The co-
rotating configuration also exhibits the appearance of PVC, which rotates and
twists in the same direction as its counterparts in the counter-rotating case but at a
slightly lower frequency of 1100 Hz. This common characteristic indicates that the
key mechanism dictating the flow evolution in the two configurations is identical.
The observation is further corroborated by the fact that the origin of PVC is located
within the venturi, in which the flow motion is mainly controlled by the swirling
flow through the primary swirler.
In conclusion, the counter-rotating arrangement appears to be more desirable
than its co-rotating counterpart for this particular injector design for several rea-
sons. First, the co-rotating configuration produces a large recirculation zone, which
is more susceptible to flame oscillation.94 Second, the strong shear layer and high-
intensity turbulence near the trailing edge of the venturi in the counter-rotating
case promote the development of the Kelvin–Helmholtz instability in the liquid
film and subsequently the formation of fine droplets, as suggested by Chin and his
coworkers23 in their experiments. Third, the counter-rotating flow accelerates the
pressure recovery in the downstream region and leads to a higher adverse pres-
sure gradient along the centerline. The process further enhances turbulent motion
downstream of the fuel nozzle, and facilitates the breakup of the liquid fuel. There-
fore, the counter-rotating design is expected to produce finer droplets and a much
more stable flame.
III. Flame Dynamics of Axial-Entry Swirl Injector

This section treats the flame dynamics of an axial-entry swirl injector typically
employed in industrial gas turbine combustors. The physical model of concern
consists of a single swirler injector, an axisymmetric chamber, and a choked nozzle,
as shown in Fig. 10.34, simulating the experimental setup described in Refs. 14
and 95. Figure 10.35 shows a schematic of the top and cross-section views of
the flat vane swirler. Natural gas is injected radially from the centerbody through
10 holes immediately downstream of the swirler vanes. The fuel–air mixture is
swirl combustion choked

injector chamber exit
45 mm
235 mm
38 mm
Fig. 10.34 Schematic of a model swirl-stabilized gas-turbine combustor (after
Broda et al., 1998, Ref. 14).
top view cross section view A – A'

A Dn = 20.3 mm; Dh = 9.53 mm
ϕ
Dn Dh
c
A'
Fig. 10.35 Schematic of top and cross-section views of a flat vane swirler
(after Seo, 1999, Ref. 95).
assumed to be perfectly premixed before entering the combustor. The chamber

measures 45 mm in diameter and 235 mm in length. The choked nozzle at the exit
prevents any downstream disturbances from traveling upstream and maintains the
desired chamber pressure. A choked venturi is also installed at the inlet entrance to
acoustically isolate the test section from the air supply line. The baseline condition
includes an equivalence ratio of 0.57 and a chamber pressure of 0.46 MPa. The
mass flow rates of the natural gas and air are 1.7 and 50.7 g/s, respectively. The
inlet flow velocity of 86.6 m/s gives rise to a Reynolds number of 35,000 on
the basis of the height of the inlet annulus.
A broad range of equivalence ratios and inlet air temperatures was considered
experimentally.14, 15, 95 Figure 10.36 shows the stability maps as functions of inlet
air temperature, equivalence ratio, and chamber pressure. Instabilities occur only
when the inlet air temperature is greater than a threshold value Tin∗ around 660 K
and the equivalence ratio falls into the range between 0.5 and 0.7. Figure 10.37
shows typical photographic images of a stable flame and an unstable flame with
an equivalence ratio of 0.6. As the inlet temperature increases and exceeds the
threshold value Tin∗ , the flame structure transforms from a stable state to an unstable
one, and the amplitude of pressure oscillation increases and reaches a limit cycle.
This kind of bifurcation phenomenon in flame structure, as well as oscillatory
flame dynamics, is discussed systematically in this section.
The basis of the analysis is the LES technique developed specifically for in-
vestigating lean premixed (LPM) combustion instabilities in swirl-stabilized com-
bustors.55, 64, 65 The formulation employs the Favre-filtered conservation equations
in three dimensions and an appropriate subgrid-scale (SGS) model. A level-set
flamelet library approach, which has been successfully applied to simulate pre-
mixed turbulent combustion,55 is used. In this approach, the filtered flame surface
evolution is modeled using a level-set G-equation, where G is defined as a dis-
tance function outside the flame front. Thermophysical properties are obtained by
using a presumed probability density function (PDF) along with a laminar flamelet
library.
A. Stable Flame Dynamics

Stable flame evolution was first obtained for an inlet mixture temperature of
600 K (below the threshold value Tin∗ for the onset of combustion oscillation). The
inlet swirl number defined by Eq. (10.2) is 0.76. The flame bifurcation phenomenon
252 Y. HUANG ET AL.
a) 15.0
p'rms /Pc
p'rms /Pc (%)

7.5
0.0
600 620 640 660 680 700
inlet air temperature, T in (K)
15.0
(b) 15.0
Pc = 0.286 MPa
12.5 Pc = 0.463 MPa
Pc = 0.638 MPa
10.0
p'rms/Pc (%)
7.5
7.5
(%)
c
/P p'
rms
5.0
2.5
0.0
0.0
0.50
0.50 0.55 0.60
0.60 0.65 0.7
0.70 0.75 0.80
0.80
equivalence
Overall ratio,
Equivalence φφο
Ratio,
Fig. 10.36 Stability maps as function of inlet air temperature and equivalence ratio:
a) Pc = 0.45 MPa, φ = 0.573; b) Tin = 669 K (after Seo, 1999, Ref. 95).
was then investigated by increasing the inlet temperature from 600 to 660 K. The
mean temperature contours and pseudo streamlines on the x–r plane, shown in
Fig. 10.38, are based on the mean axial and radial velocity components for a stable
flame. A CRTZ is established in the wake of the centerbody under the effects
of the swirling flow. The CTRZ, a form of vortex breakdown, serves as a flame
stabilization region, where hot products mix with the incoming mixture of air and
fuel. In addition, as a result of the sudden increase in combustor area, a CRZ is
formed downstream of the backward-facing step.
The calculated pressure and velocity fields exhibit small-amplitude fluctuations,
with a dominant harmonic mode at 3214 Hz, corresponding to the frequency of the
vortex shedding from the centerbody. Figure 10.39 presents the flame evolution
and vortex shedding process in the upstream region of the chamber over one cycle
of oscillation. The pressure and velocity are measured at the middle point of the
T in = 570 K
T in = 660 K
20
15
10
p' (kPa)
−5
−10
−15
−20
0 10 20
time (ms)
Fig. 10.37 Top: photographic images of stable and unstable flames; bottom: pressure-
time trace, Pc = 0.483 MPa, φ = 0.573 (after Seo, 1999).
254 Y. HUANG ET AL.
Fig. 10.38 Mean temperature contours and streamlines of stable flame for S = 0.76.
θ
t
θ
t
θ
t
θ
t
140 470
u p
u (m/s)
p (kpa)
120 460
100 450
0.30 0.50 0.70

t (s)
Fig. 10.39 Stable flame evolution over one cycle of oscillation (3214 Hz): temperature
contours and streamlines for S = 0.76.
inlet annulus exit. The phase angle θ is referenced with respect to the acous-
tic velocity at the interface between the inlet and combustor. The entire process
is dictated by the temporal evolution and spatial distribution of the flame front,
which moves back and forth under the influences of the vortical motion (indicated
by the concentrated streamlines) in the chamber. A new vortex begins to shed
from the center body at θ = 90 deg, accompanied by a higher local flow velocity.
As the vortex moves downstream (θ = 180 deg −270 deg), it distorts the flame
front or even produces a separated flame pocket. At the same time, the higher-
speed mixture pushes the flame downstream. When the vortex moves away from
the flame (θ = 360 deg) and dissipates into small-scale structures, the flame front
propagates upstream (since the higher-speed mixture is convected downstream)
and interacts with another incoming vortex. During this process, a new vortex
appears at the corner of the centerbody and the cycle repeats.
B. Bifurcation of Flame Structure

The inlet temperature has enormous effects on the flame dynamics in the system.
On the one hand, when the inlet temperature increases, for a fixed mass flow rate,
the flow velocity also increases and pushes the flame downstream. On the other
hand, the increased inlet temperature leads to an increase in the flame speed and
consequently causes the flame to propagate upstream. In addition, flashback may
occur near the wall because of the small local flow velocity. The combined effects
of flow acceleration, flame-speed enhancement, and flashback determine the final
form of the flame structure.
In the present study, as the inlet temperature increases from 600 to 660 K,
flame bifurcation takes place. The flame originally anchored in the center re-
circulation zone penetrates into the corner recirculation zone and flashes back.
Consequently, the flame is stabilized by both the corner-recirculating zone and the
center-recirculating flow and forms a compact enveloped configuration. The flame
flaps dynamically and drives flow oscillations through its influence on unsteady
heat release. At the same time, the pressure oscillation increases and reaches an-
other limit cycle with a much larger amplitude. The entire bifurcation process
can be divided into three stages: high-temperature-mixture filling process, flame-
trapping process, and vortex-flashback process, as shown in Fig. 10.40, where
t = 0 ms denotes the time at which the inlet mixture temperature starts to increase
from 600 to 660 K.
Figures 10.40a–10.40c show the high-temperature-mixture filling process. As
the inlet mixture temperature increases, the flow speed increases because of
the decreased density for a fixed mass flow rate. As a result, the original low-
temperature mixture is pushed downstream toward the flame. Although a flash-
back phenomenon is observed near the wall, the high-temperature mixture has not
reached the flame front near the wall and the flame speed remains unchanged at
this stage.
Figures 10.40d and 10.40e show the flame-trapping process. Once the high-
temperature mixture reaches the flame front, with the help of the increased flame
speed, the near-wall flashback overshadows the flow acceleration effects. Conse-
quently, the flame front penetrates into the corner recirculation zone and is trapped
by the local vortical motion.
256 Y. HUANG ET AL.
a)
b)
c)
d)
e)
f)
g)
h)
Fig. 10.40 Transition from stable to unstable flame with increased inlet temperature
from 600 to 660 K.
In the vortex-flashback process, as shown in Figs. 10.40f–10.40h, the flame
propagates upstream under the influence of the vortical motion. A counterclock-
wise rotating vortex originally shed from the edge of the backward-facing step
approaches the flame front in the corner recirculation zone and then pushes it
toward the dump plane. At the same time, a small flame pocket is produced
and separated from the main stream. After this vortex is convected downstream
and passes through the flame, another vortex approaches and interacts with the
flame. The process continues, and eventually the fresh reactants in the corner
recirculation zone are completely burnt. The flame is stabilized by both the corner-
recirculating flow and the center-recirculating flow, and its overall length is sub-
stantially reduced. This situation renders the combustor more prone to instabil-
ities according to the Rayleigh criterion,96 since considerable heat is released
within a short distance close to the chamber head end (i.e., the acoustic antinode
point).
Once the flame becomes unstable (oscillatory) when the inlet flow temperature
exceeds the critical value Tin∗ , it becomes difficult to reestablish stable operation
unless the inlet temperature is reduced to a level significantly lower than Tin∗ .
This phenomenon is commonly referred to as hysteresis and has been experi-
mentally observed by many researchers.14, 64 The occurrence of hysteresis under
the current circumstance may be explained as follows. During unstable com-
bustion, the corner recirculation zone is filled with high-temperature products,
and the chamber wall in this region is heated to reach the local flame tempera-
ture. To recover stable operation, the cold flow needs not only to extinguish the
flame stabilized by the corner recirculating flow through entrainment or flame
liftoff but also to offset the effects of the high-temperature wall, which tends
to increase the local gas temperature and inhibit extinction and near-wall flash-
back. Consequently, a much lower inlet temperature is required to regain stable
operation.
In light of the preceding observations, we conclude that the flashback phe-
nomenon dictates the flame bifurcation process. Flashback in premixed combus-
tion has been the subject of a number of experimental, analytical, and numerical
studies in the past. Its occurrence is usually attributed to two mechanisms. The
first involves flame propagation in the boundary layer along a solid wall, where
the local velocity diminishes toward the surface. The second mechanism is asso-
ciated with flow reversal, which is usually caused by vortical motions or acoustic
oscillations. Both mechanisms are observed in the present case.
For lean-premixed combustion, the laminar flame speed SL increases with an
increase in the equivalence ratio φ. Thus, increases in the equivalence ratio and in-
let temperature exert similar effects on the flame evolution. However, the chemical
reaction rate and heat release are much more sensitive to variations in the equiv-
alence ratio under lean conditions than under stoichiometric conditions. More-
over, near the lean blowout limit, perturbations in the equivalence ratio φ can
cause periodic extinction of the flame. As a result, the equivalence ratio oscil-
lation under lean conditions is prone to inducing flow oscillation97 and subse-
quently increases turbulent velocity fluctuation v . This outcome suggests that a
lean premixed turbulent flame is more susceptible to flashback, since the turbu-
lent flame speed ST increases not only with the laminar flame speed SL , but also
with turbulent velocity fluctuation v .98 The result helps explain why the tran-
sition from a stable to an unstable state, as described in Refs. 14 and 95, only
occurs when the equivalence ratio falls in the range between 0.5 and 0.7. Since
the flame bifurcation is largely determined by the flashback phenomenon in the
corner recirculation zone in the present case, one effective way to avoid its oc-
currence is to inject cold flow into that region. This procedure suppresses the
upstream propagation of local flame and consequently leads to a much more stable
system.
258 Y. HUANG ET AL.
C. Oscillatory Flame Dynamics

When the inlet flow temperature exceeds the threshold value, unstable (oscilla-
tory) flames are clearly observed and are accompanied by large excursions of flow
oscillations with frequencies corresponding to the various characteristic dimen-
sions of the chamber. Three different swirl numbers (S = 0.44, 0.76, and 1.10) are
investigated for oscillatory flame dynamics. For S = 0.44, three dominant modes
exist at the frequencies of 1761, 10,367, and 17,618 Hz, corresponding to the first
longitudinal (1L), first tangential (1T), and second tangential (2T) modes of acous-
tic motions in the chamber, respectively. For S = 0.76, the frequencies of these
three modes slightly shift to 1795, 10,970, and 17,356 Hz because of the change
in the temperature field. For S = 1.10, the longitudinal wave disappears, and the
frequencies of the 1T and 2T modes change to 10,795 and 18,133 Hz.
1. Mean Flow Structures

The mean flow properties are obtained by taking the longtime average of the
instantaneous quantities. In spite of significant flow motions in the azimuthal di-
rection, the mean flowfield remains perfectly axisymmetric. Figure 10.41 shows
the streamline patterns and mean temperature fields on the x–r plane. Three dis-
tinct recirculation zones are observed in the low swirl-number case with S = 0.44,
including a wake recirculation zone (WRZ) behind the centerbody, a corner recircu-
lation zone attributable to the sudden enlargement of the combustor configuration,
and a central toroidal recirculation zone (CTRZ) resulting from vortex breakdown.
The WRZ, however, disappears at the high swirl number of S = 1.10. The overall
flow development with respect to the inlet swirl number can be described as fol-
lows. If there is no swirl, only the wake and corner recirculation zones exist. As the
swirl number increases and exceeds a critical value, vortex breakdown takes place
and leads to the formation of a central recirculation zone. As the swirl number
further increases, the central recirculation zone moves upstream and merges with
the wake recirculation zone. The corner recirculation zone is also reduced. Simi-
lar results were reported by Chao99 in his experimental study of the recirculation
structure in a coannular swirl combustor. The mean temperature fields clearly ex-
hibit enveloped flames anchored at the rim of the centerbody and the corner of the
backward-facing step. The flame is much more compact for the high swirl-number
case with S = 1.10, which is caused by the enhanced flame speed resulting from
the increased turbulence intensity, as we will subsequently show.
Figure 10.42 shows the radial distributions of the mean velocity components
and turbulent kinetic energy at various axial locations for S = 0.44 and S = 1.10,
where r = 0 corresponds to the centerline of the chamber. The negative axial ve-
locities in the central and corner regions indicate the existence of recirculation
zones. The incoming flows from the inlet annulus spread outward from the cham-
ber centerline under the effect of the centrifugal force, producing positive radial
velocities in the main flow passage. The stronger the swirl strength is, the faster the
main flow moves toward the wall. As a result, the size of the corner recirculation
zone is considerably reduced at the high swirl number of S = 1.10. The mean
azimuthal velocity fields suggest that the flow motion in the central region bears a
close resemblance to solid-body rotation. The distributions of the turbulent kinetic
energy indicate that a high turbulence-intensity region develops downstream of
Fig. 10.41 Mean temperature fields and streamline patterns for S = 0.44, 0.76,
and 1.10.
the centerbody and the backward-facing step, where large velocity fluctuations
are produced because of the strong turbulent mixing in the shear layers between
the incoming flow and the recirculation flows. The evolution of turbulent kinetic
energy is governed by the following equation:

D u i2 /2 /Dt = ∂ − u j p/ρ0 − u i2 u j /2 + 2vu i ei j /∂ x j
− u i u j ∂Ui /∂ x j − 2vei j ei j (10.9)
where ei j = (∂u i /∂ x j + ∂u j /∂ xi )/2. The existence of the steep velocity gradi-

ent ∂Ui /∂ x j because of the strong swirling and reverse flow in the high swirl-
number case, facilitates the production of turbulent kinetic energy, −u i u j ∂Ui /∂ x j .
260 Y. HUANG ET AL.
x = 60 mm
x = 40 mm
x = 80 mm
x = 25 mm
x = 30 mm
x = 50 mm
2
r (cm)
1
0
0 100 0 100 0 100 0 100 0 100 0 100
u x (m/s)
x = 25 mm
x = 40 mm
x = 80 mm
x = 60 mm
x = 50 mm
x = 30 mm
2
r (cm)
0
0 100 0 100 0 100 0 100 0 100 0 100
u r (m/s)
x = 40 mm
x = 30 mm
x = 50 mm
x = 80 mm
x = 60 mm
x = 25 mm
2
r (cm)
0
0 100 0 100 0 100 0 100 0 100 0 100
u θ (m/s)
x = 80 mm
x = 30 mm
x = 60 mm
x = 40 mm
x = 25 mm
x = 50 mm
2
r (cm)
0
0 50 0 50 0  50 0 50 0 50 0 50
√ k (m/s)
Fig. 10.42 Radial distributions of mean velocity components and turbulent kinetic
energy at various axial locations for S = 0.44 and 1.10.
Consequently, much stronger turbulent kinetic energy is observed for the high
swirl-number case with S = 1.10.
2. Instantaneous Flowfield
Vorticity is of concern because of its dominant influence in determining the
flow entrainment in the reaction zone and the subsequent flame evolution. Fig-
ure 10.43 shows snapshots of the vorticity magnitude field on an x–r plane. For
the low swirl-number case with S = 0.44, large vortical structures, arising from
the shear layers downstream of the dump plane and centerbody, are convected
downstream and then dissipated into small-scale eddies. The same phenomenon is
observed for the high swirl-number case, in which well-organized vortices are shed
Fig. 10.43 Snapshots of vorticity magnitude field on x–r and x–θ planes for S = 0.44
and 1.10.
from the edge of the backward-facing step. The vortex motions downstream of
the centerbody, however, become quite disordered, because of the presence of
the strong central recirculating flow. In both cases, the vortex shedding frequen-
cies are close to that of the first tangential mode of the acoustic wave in the
chamber.
Figure 10.44 shows snapshots of the isovorticity surface at ω =75,000 1/s.
The flowfield in the region r > 2 cm is blanked to provide a clear picture of the
vortex structures. For the low swirl-number case with S = 0.44, a vortex spiral
evolves from the shear layer originating at the backward-facing step, because of
the Kelvin–Helmholtz instabilities in both the axial and azimuthal directions. This
vortical structure gyrates around the centerline and persists for several turns before
breaking up into small fragments. For the high swirl-number case with S =1.10, a
spiral vortex structure can also be observed. The structure, however, is much more
complex because of the high centrifugal force. It spreads outward rapidly and soon
breaks up into small-scale structures.
The evolution of these spiral vortex structures can be regarded as a kind of
vortex shedding process with well-defined frequencies, as previously described.
262 Y. HUANG ET AL.
Fig. 10.44 Snapshots of isovorticity surface at ω = 75, 000 s−1 (left: r > 0.02 m is
blanked, right: r > 0.01 m is blanked) for S = 0.44 and 1.10.
One may conjecture55 that the vortical motions in the shear layers resonate with
acoustic oscillations in the chamber. In the present configuration, two shear layers
exist downstream of the rear-facing step and the centerbody. The axial momen-
tum thickness, θ0 , of each shear layer is estimated to be around 0.1 mm for the
low swirl-number case based on the calculated mean velocity distribution. For
the high swirl-number case, the momentum thickness of the inner shear layer
(≈0.25 mm) differs from its outer counterpart (≈0.05 mm) by a factor of five, be-
cause of its stronger swirl strength. A linear stability analysis100 has been carried
out to provide more insight into the shear-layer instability phenomena for annular
swirling flows in an open atmosphere. The geometric parameters were selected
to match the current physical model. The effects of momentum thickness, swirl
strength, and density and velocity ratios were studied systematically with different
azimuthal wave numbers. The predicted most-amplified frequencies are different
for the two swirl numbers considered herein, mainly because of the disparity of
the axial momentum thickness between the two cases. In addition, the predicted
values are much higher than the vortex-shedding frequency, which corresponds to
the frequency of the first tangential (1T) mode of acoustic motion in the present
chamber. This observation indicates that the acoustic oscillation acts as a forced
excitation to the system. The shear layers respond to the excitation by locking their
shedding frequencies close to the forcing frequency.
3. Evolution of Flame Surface and Heat Release

To understand the mutual coupling between the flame dynamics and flow oscil-
lation, the total heat release and flame surface area were analyzed in the frequency
domain. The overall heat release in the chamber can be obtained from
Q̇ = ρu h 0f ST A (10.10)
where ρu is the unburnt gas density, ST is the subgrid turbulent flame speed, h 0f
is the heat of reaction, and A is the total filtered flame surface area.
Figure 10.45 shows the power spectral densities of the total filtered flame
surface-area and heat-release fluctuations. At S = 0.44, a dominant mode exists
400 S = 0.44 flame surface area 400 S = 1.10 flame surface area
Amplitude, mm2/Hz
Amplitude, mm2/Hz
1761
300 300
200 3320 200
100 100
11,712
0 0
0 10 20 0 10 20
8 S = 0.44 heat release 8 S = 1.10 heat release

Amplitude, kJ/s/Hz
Amplitude, kJ/s/Hz
1761
6 6
4 3320 4
11,712
2 20,532 2
0 0
0 10 20 0 10 20
Frequency, kHz Frequency, kHz
Fig. 10.45 Power spectral densities of total flame surface-area and heat-release fluc-
tuations for S = 0.44 and 1.10.
264 Y. HUANG ET AL.
Fig. 10.46 Temporal evolution of temperature field over one cycle of 1L mode of
oscillation for S = 0.44.
at 1761 Hz in the flame-surface oscillation, which corresponds to the 1L acoustic

mode of the combustor. A higher harmonic at 3320 Hz is also found, approximately
twice the frequency of the 1L mode. Although transverse acoustic motions includ-
ing the 1T and 2T modes are observed, the flame surface-area oscillations do not
exhibit such a high-frequency behavior. At S = 1.10, a small peak at 11,712 Hz
near the 1T acoustic mode is present, but no corresponding 1L mode oscillation
is found. The frequency content of the total heat-release fluctuations bears a close
resemblance to that of flame surface-area variations. A small spike near the fre-
quency of 20,532 Hz, however, is observed for S = 0.44, which arises from the
fluctuations in the subgrid turbulent flame speed ST .55 In light of the preceding
observations, one can conclude that low-frequency acoustic perturbations exert a
strong influence on the fluctuations of the total flame surface area and heat release.
In contrast, high-frequency acoustic oscillations travel through the flame zone
without significantly affecting the flame surface area and heat-release variations,
although they may impose a significant impact on the local flame propagation. The
results qualitatively agree well with the prediction from a companion analytical
analysis of flame response.101 The calculated mean flame surface area and the root
mean square of the fluctuating quantity for the high swirl-number case are much
smaller than those of the low swirl-number case. However, owing to the increased
turbulence intensity and the ensuing enhancement of the flame speed in the high
swirl-number case, the mean heat-release rate and the associated fluctuation are
very close in these two cases.
4. Acoustic and Flame Interaction

Figure 10.46 presents the temporal evolution of the temperature field in the
upstream section of the chamber over one cycle of the 1L mode of acoustic oscil-
lation for S = 0.44. The phase angle θ is referenced to the 1L acoustic pressure
at the chamber head end. The entire process is dictated by the cold-flow entrain-
ment into and mixing with hot gases in the vortical structures in the flame zone.
Figure 10.47 shows the time histories of the pressure immediately downstream of
the dump plane (top), the total flame surface area (middle), and the rate of heat
release (bottom). These signals involve a wide range of frequencies corresponding
to turbulent-flow and acoustic oscillations. The extracted 1L oscillations (denoted
by the thick black lines) of these quantities are also plotted for clarity. The flame
Fig. 10.47 Time histories of pressure immediately downstream of the dump plane
(top), flame surface area (middle), and heat release rate (bottom) for S = 0.44; the
thick black lines represent the extracted 1L oscillations.
266 Y. HUANG ET AL.
pressure
250
500 4000
flame surface area, mm2

200
heat release, kJ
400 flame surface area
p, kpa
3000
300 total heat release 150

2000
200
100
23 24 25
Time, ms
Fig. 10.48 Time histories of pressure immediately downstream of the edge of the
centerbody (top), flame surface-area (middle), and heat release rate (bottom) for
S = 1.10.
surface-area variation can be elucidated by considering its interaction with the local
oscillatory flowfield. It lags behind the pressure oscillation by 76 deg. During the
period from θ = −166 deg (t = 24.09 ms) to 14 deg (t = 24.38 ms), a relatively
lower pressure field exists near the dump plane, facilitating the delivery of the
fresh reactants into the chamber. Intensive heat release then occurs after a short
fluid-mixing and chemical-induction time. The resultant flow expansion pushes
the flame outward and causes the flame surface area to increase from a trough
to a crest. Unburned mixture fragments may be shattered away from the main
stream and may generate local hot spots when convected downstream. During the
period from θ = 14 deg (t = 24.38 ms) to 194 deg (t = 24.66 ms), the relatively
higher pressure near the dump plane prevents the fresh reactants from traveling
downstream into the chamber. The flame zone is thus reduced and becomes a lit-
tle more compact. The same process then repeats for another cycle of oscillation.
Figure 10.47 also indicates that the heat-release and flame-surface-area fluctuations
are nearly in phase. The former only lags behind the flame surface-area oscilla-
tion by 4 degrees. For the high swirl-number case with S = 1.10, no obvious 1L
oscillation can be observed, as shown in Fig. 10.48.
Figure 10.49 presents the temporal evolution of the temperature field in the
upstream section of the chamber on the x–r plane over one cycle of the 1T mode
of acoustic oscillation. As the swirl number increases, the flame anchored by
the center recirculating flow may propagate upstream periodically and cause flame
flashback. Two mechanisms, as previously stated, have been identified for the
occurrence of flame flashback. In this case, the flashback is closely linked to
the strong reverse flow in the center recirculation zone. The swirl strength is so
strong that it sometimes causes the center recirculating flow to enter the inlet annu-
lus. As a consequence, the flame attached to the centerbody travels upstream and
flashback occurs.
Fig. 10.49 Temporal evolution of temperature field on an x–r plane over one cycle of
1T mode of oscillation for S = 0.44 and 1.10.
5. Vortex and Flame Interaction

Figure 10.50 shows the instantaneous vorticity field at various times within one
cycle of the 1T acoustic oscillation for S = 0.76. The thick black line indicates the
flame front. Well-organized vortices are shed from the edge of the backward-facing
step. The process, however, becomes much more complex in the downstream of the
268 Y. HUANG ET AL.
Fig. 10.50 Vortex and flame front interaction over one cycle of 1T mode of oscillation
for the case S = 0.76.
centerbody because of the existence of a toroidal recirculating flow. New vortices

are produced at the tip of the backward-facing step at θ = 72 deg and bulge the
flame front. They continue to distort the flame or even produce separated flame
pockets when traveling downstream. Finally, these vortices move out of the flame
region and dissipate into small-scale structures. Another set of vortices appears at
θ = 360 deg at the dump plane, and the cycle repeats.
To further examine the preceding process, the temporal evolution of the vortic-
ity, temperature, and heat-release distributions within one cycle of the 1T mode
of acoustic oscillation are plotted for S = 0.76, as shown in Fig. 10.51. The vor-
tex shedding process is clearly visualized in the evolution of the vortex spiral,
which gyrates around the chamber centerline and propagates downstream. The
wavelike structure on the flame surface possesses a characteristic frequency cor-
responding to the 1T acoustic wave. Because the vortex shedding affects the
shapes of the flame front, it also changes the heat-release distribution. As a re-
sult, the acoustic motion in the chamber is closely coupled with the heat-release
fluctuation.
Fig. 10.51 Temporal evolution of isovorticity surface at ω =75,000 s −1 (r > 0.02 m

is blanked), isothermal surface at T = 1700K and normalized heat release contour
over one cycle of 1T mode of oscillation for S = 0.76.
IV. Conclusion
In this chapter, the flow and flame dynamics of several different types of swirl-
injectors for contemporary gas turbine engines were studied. Both axial- and radial-
entry configurations were explored over a wide range of flow conditions. The
effects of single vs multiple swirlers and co-rotating vs counter-rotating arrange-
ments were also examined systematically. Most of the results presented here were
270 Y. HUANG ET AL.
obtained from large eddy simulations of detailed flow structures under conditions
with and without external forcing. Where appropriate, experimental observations
are also given for the sake of completeness.
For cold-flow characterization of injector dynamics, various fundamental mech-
anisms dictating the flow evolution in swirl injectors were identified. These include
vortex breakdown; precession of vortex core; and Kelvin–Helmholtz, helical and
centrifugal instabilities. The injector response to external forcing was studied in
terms of the acoustic admittance and mass transfer functions over a broad range of
frequencies. Low-frequency excitations generally promote flow fluctuations, but
the trend is reversed for high-frequency excitations. The fluctuation of the mass
flow rate of a given frequency component at the injector exit may reach a magnitude
substantially greater than that at the entrance when the forcing resonates with the
injector flow. Results of this kind can be effectively used in analyzing combustion
instabilities in gas-turbine engines. The influence of swirler orientation on injector
flow development was also examined by considering both counter- and co-rotating
inflow conditions. The counter-rotating configuration is more desirable because
of its reduced central recirculation zone and stronger shear stress and turbulence
intensity in regions where liquid fuel atomization occurs.
The second part of the chapter deals with the flame dynamics of swirl injectors.
Detailed flow structures and flame evolution were investigated under various oper-
ating conditions. The inlet air temperature and equivalence ratio were found to be
the two key parameters determining the stability characteristics of a lean-premixed
swirl-stabilized combustor. A slight increase in the inlet airflow temperature across
the stability boundary leads to a sudden increase in the chamber flow oscillation. Fi-
nally, the underlying mechanisms responsible for driving combustion oscillations
and the transition from a stable to an unstable flame were identified.
Acknowledgments
The work reported in this chapter was sponsored in part by the Office of Naval
Research under Grant No. N00014-96-1-0405, in part by the NASA Glenn Re-
search Center under Grant NAG 3-2151, and in part by the Air Force Office of
Scientific Research under Grant No. F49620-99-0290. The support and encourage-
ment provided by Gabriel Roy and Kevin Breisacher are gratefully acknowledged.
References
1
Lefebvre, A. H., “The Role of Fuel Preparation in Low-emission Combustion,” Journal
of Engineering for Gas Turbines and Power, Vol. 117, 1995, pp. 617–654.
2
Bazarov, V., Yang, V., and Puri, P., “Design and Dynamics of Jet and Swirl Injectors,”
Chapter 2, Liquid Rocket Thrust Chambers: Aspects of Modeling, Analysis, and Design,
edited by V., Yang, M. Habiballah, J. Hulka, and M., Popp, Progress in Astronautics and
Aeronautics, Vol. 200, AIAA, Reston VA, 2004, pp. 19–103.
3
Lefebvre, A. H., Gas Turbine Combustion, Taylor & Francis, Philadelphia, 1998.
4
Wang, H., McDonell, V. G., Sowa, W. A., and Samuelsen, S., “Experimental Study of
a Model Gas Turbine Combustor Swirl Cup, Part I: Two-Phase Characterization,” Journal
of Propulsion and Power, Vol. 10, No. 4, 1994, pp. 441–445.
5
Wang, H., McDonell, V. G., Sowa, W. A., and Samuelsen, S., “Experimental Study
of a Model Gas Turbine Combustor Swirl Cup, Part II: Droplet Dynamics,” Journal of
6
Fu, Y., Cai, J., Elkady, M., Jeng, S. M., and Mongia, H. C., “Fuel and Equivalence Ratio
Effects on Spray Combustion of a Counter-Rotating Swirler,” AIAA Paper 2005-0354,
2005.
7
Jeng, S. M., Flohre, N. M., and Mongia, H. C., “Air Temperature Effects on Non-Reacting
Spray Characteristics Issued from a Counter-Rotating Swirler,” AIAA Paper 2005-0355,
2005.
8
Jeng, S. M., Flohre, N. M., and Mongia, H. C., “Fluid Property Effects on Non-Reacting
Spray Characteristics Issued from a Counter-Rotating Swirler,” AIAA Paper 2005-0356,
2005.
9
Cowell, L. H., and Smith, K. O., “Development of a Liquid-Fueled, Lean-Premixed Gas
Turbine Combustor,” Journal of Engineering for Gas Turbines and Power, Vol. 115, 1993,
pp. 554–562.
10
Snyder, T. S., Rosfjord, T. J., McVey, J. B., and Chiappetta, L. M., “Comparison of
Liquid/Air Mixing and NOx Emission for a Tangential Entry Nozzle,” American Society
of Mechanical Engineers, Paper 94-GT-283, 1994.
11
for Gas Turbines and Power, Vol. 121, 1999, pp. 281–284.
12
Richards, G. A., Gemmen, R. S., and Yip, M. J., “A Test Device for Premixed Gas
Turbine Combustion Oscillations,” Journal of Engineering for Gas Turbines and Power,
Vol. 120, 1998, pp. 294–302.
13
Richards, G. A., and Janus, M. C., “Characterization of Oscillations During Premixed
Gas Turbine Combustion,” Journal of Engineering for Gas Turbines and Power, Vol. 119,
1997, pp. 776–782.
14
Study of Combustion Dynamics of a Premixed Swirl Injector,” Proceedings of the Com-
bustion Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1849–1856.
15
Lee, S. Y., Seo, S., Broda, J. C., Pal, S., and Santoro, R. J., “An Experimental Estimation
of Mean Reaction Rate and Flame Structure During Combustion Instability in a Lean
Premixed Gas Turbine Combustor,” Proceedings of the Combustion Institute, Pittsburgh,
PA, Vol. 28, 2000, pp. 775–782.
16
Venkataraman, K. K., Preston, L. H., Simons, D. W., Lee, B. J., Lee, J. G., and
Santavicca, D. A., “Mechanism of Combustion Instability in a Lean Premixed Dump Com-
bustor,” Journal of Propulsion and Power, Vol. 15, No. 6, 1999, pp. 909–918.
17
Lee, J. G., Kim, K., and Santavicca, D. A., “Measurement of Equivalence Ratio Fluc-
tuation and Its Effect on Heat Release During Unstable Combustion,” Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 415–421.
18
Lee, J. G., and Santavicca, D. A., “Experimental Diagnostics for the Study of Combus-
tion Instabilities in Lean-Premixed Combustors,” Journal of Propulsion and Power, Vol.
19, No. 5, 2003, pp. 735–749.
19
Mordaunt, C., Brossard, C., Lee, S. Y., and Santoro, R. J., “Combustion Instability
Studies in a High-Pressure Lean-Premixed Model Combustor Under Liquid Fuel Opera-
tion,” Proceedings of International Joint Power Generation Conference, JPG C2001/FACT-
19089, June 2001.
272 Y. HUANG ET AL.
20
Cohen, J. M., and Rosfjord, T. J., “Spray Pattern at High Pressure,” Journal of Propul-
sion and Power, Vol. 7, No. 4, 1991, pp. 1481–1487.
21
Cohen, J. M., and Rosfjord, T. J., “Influence on the Sprays Formed by High-Shear
Fuel-Nozzle/Swirler Assemblies,” Journal of Propulsion and Power, Vol. 9, No. 1, 1993,
pp. 16–27.
22
Chin, J. S., Rizk, N. K., Razdan, M. K., “Study on Hybrid Airblast Atomization,”
Journal of Propulsion and Power, Vol. 15, No. 2, 1999, pp. 241–247.
23
Chin, J., Rizk, N., and Razdan, M., “Effect of Inner and Outer Airflow Characteristics on
High Liquid Pressure Prefilming Airblast Atomization,” Journal of Propulsion and Power,
Vol. 16, 2000, pp. 297–301.
24
Hardalupas, Y., and Whitelaw, J. H., “Interaction Between Sprays from Multiple Coax-
ial Airblast Atomizers,” Journal of Engineering for Gas Turbines and Power, Vol. 118,
1996, pp. 762–771.
25
Hardalupas, Y., Pantelides, K., and Whitelaw, J. H., “Particle Tracking Velocimetry in
Swirl-Stabilized Burners,” Experiments in Fluids, [Suppl.], Vol. 29, 2000, pp. S220–S226.
26
Presser, C., Gupta, A. K., and Semerjian, H. G., “Aerodynamic Characteristics of
Swirling Spray Flame: Pressure-Jet Atomizer,” Combustion and Flame, Vol. 92, 1993,
pp. 25–44.
27
Presser, C., Gupta, A. K., Semerjian, H. G., and Avedisian, C. T., “Droplet Transport in
a Swirl-Stabilized Spray Flames,” Journal of Propulsion and Power, Vol. 10, No. 5, 1994,
pp. 631–638.
28
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Control of Thermoacoustic In-
stabilities and Emissions in an Industrial-Type Gas Turbine Combustor,” Proceedings of
the Combustion Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1817–1824.
29
Paschereit, C. O., and Gutmark, E., “Proportional Control of Combustion Instabilities
in a Simulated Gas Turbine Combustor,” Journal of Propulsion and Power, Vol. 18, 2002,
pp. 1298–1304.
30
Acharya, S., Murugappan, S., O’Donnel, M., and Gutmark, E. J., “Characteristics and
Control of Combustion Instabilities in a Swirl-Stabilized Spray Combustor,” Journal of
31
Uhm, J. H., Acharya, S., “Control of Combustion Instability with a High-Momentum
Air-Jet,” Combustion and Flame, Vol. 139, 2004, pp. 106–125.
32
Murugappan, S., Acharya, S., Allgood, D. C., Park, S., Anaswamy, A. M., and Ghoniem,
A. F., “Optimal Control of a Swirl-Stabilized Spray Combustion Using System Identification
Approach,” Combustion Science and Technology, Vol. 175, 2003, pp. 55–81.
33
Lee, J. G., Kim, K., Santavicca, D. A., “Effect of Injection Location on the Effective-
ness of an Active Control System Using Secondary Fuel Injection,” Proceedings of the
34
Richards, G. A., Janus, M., and Robey, E. H., “Control of Flame Oscillation with
Equivalence Ratio Modulation,” Journal of Propulsion and Power, Vol. 15, No. 2, 1999,
pp. 232–240.
35
Richards, G. A., Straub, D. L., and Robey, E. H., “Passive Control of Combustion
Dynamics in Stationary Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5,
2003, pp. 795–809.
36
Sattinger, S. S., Neumeier Y., Nabi, A., Zinn, B. T., Amos, D. J., and Darling,
D. D., “Sub-Scale Demonstration of the Active Feedback Control of Gas-Turbine Com-
bustion Instabilities,” Journal of Engineering for Gas Turbines and Power, Vol. 122, 2000,
pp. 262–268.
37
Johnson, C. E., Neumeier, Y., Neumaier, M., Zinn, B. T., Darling, D. D., and Sattinger,
S. S., “Demonstration of Active Control of Combustion Instabilities on a Full-Scale Gas
Turbine Combustor,” ASME 2001-GT-519, 2001.
38
Conrad, T., Bibik, A., Shcherbik, D., Lubarsky, E., and Zinn, B. T., “Slow Control of
Combustion Instabilities by Fuel Spray Modification Using Smart Fuel Injector,” AIAA
Paper 2004-1034, 2004.
39
Conrad, T., Bibik, A., Shcherbik, D., Lubarsky, E., and Zinn, B. T., “Control of In-
stabilities in Liquid Fueled Combustor by Modification of the Reaction Zone Using Smart
Fuel Injector,” AIAA Paper 2004-4029, 2004.
40
Bernier, D., Lacas, F., and Candel, S., “Instability Mechanisms in a Premixed Pre-
vaporized Combustor,” Journal of Propulsion and Power, Vol. 20, No. 4, 2004, pp. 648–
656.
41
Li, G., and Gutmark, E. J., “Effect of Exhaust Nozzle Geometry on Combustor Flow
Field and Combustion Characteristics,” Proceedings of the Combustion Institute, Pittsburgh,
PA, Vol. 30, 2004.
42
Brewster, B. S., Cannon, S. M., Farmer, J. R., and Meng, F., “Modeling of Lean Pre-
mixed Combustion in Stationary Gas Turbines,” Progress in Energy and Combustion Sci-
ence, Vol. 25, 1999, pp. 353–385.
43
Mongia, H. C., Al-Roub, M., Danis, A., Elliott-Lewis, D., Jeng, S. M., John, A.,
McDonell, V. G., Samuelsen, G. S., and Vise, S., “Swirl Cup Modeling, Part I,” AIAA
Paper 2001-3576, 2001.
44
Hsiao, G., Mongia, H. C., Vij, A., “Swirl Cup Modeling Part II: Inlet Boundary Con-
ditions,” AIAA Paper 2001-1350, 2001.
45
Hsiao, G., and Mongia, H. C., “Swirl Cup Modeling Part III: Grid Independent Solution
with Different Turbulence Models,” AIAA Paper 2001-1349, 2001.
46
Cai, J., Fu, Y., Jeng, S. and Mongia, H. C., “Swirl Cup Modeling Part IV: Effect of
Confinement on Flow Characteristics,” AIAA Paper 2003-0486, 2003.
47
Wang, S., Yang, V., and Mongia, H. C., “Swirl Cup Modeling Part V: Large Eddy
Simulation of Cold Flow,” AIAA Paper 2001-0485, 2003.
48
Giridharan, M. G., Mongia, H. C., and Singh, G., “Swirl Cup Modeling Part VI: Dilution
Jet Modeling,” AIAA Paper 2003-1203, 2003.
49
Stevens, E. J., Held, T. J., and Mongia, H. C., “Swirl Cup Modeling Part VII: Pre-
mixed Laminar Flamelet Model Validation and Simulation of a Single-Cup Combustor
with Gaseous N-Heptane,” AIAA Paper 2003-0488, 2003.
50
Giridharan, M. G., Mongia, H. C., and Jeng, S. M., “Swirl Cup Modeling Part VIII:
Spary Combustion in CFM56 Single Cup Flame Tube,” AIAA Paper 2003-0319, 2003.
51
Jeng, S. M., Flohre, N. M., and Mongia, H. C., “Swirl Cup Modeling Part IX: Atom-
ization,” AIAA Paper 2004-0137, 2004.
52
James, S., Zhu, J., and Anand, M. S., “Large-Eddy Simulation of Gas Turbine Com-
bustors,” AIAA Paper 2004-0552, 2004.
53
Kim, W. W., and Syed, S., “Large-Eddy Simulation Needs for Gas Turbine Combustor
Design,” AIAA Paper 2004-0331, 2004.
54
Kim, W. W., Lienau, J. J., Van Slooten, P. R., Colket, M. B., Malecki, R. E., and Syed, S.,
“Towards Modeling Lean Blow Out in Gas Turbine Flameholder Applications,” American
Society of Mechanical Engineers, 2004-GT-53967, 2004.
55
Huang, Y., Sung, H. G., Hsieh, S. Y. and Yang, V., “Large Eddy Simulation of Com-
bustion Dynamics of Lean-Premixed Swirl-Stabilized Combustor,” Journal of Propulsion
and Power, Vol. 19, 2003, pp. 782–794.
274 Y. HUANG ET AL.
56
Stone, C., and Menon, S., “Swirl Control of Combustion Instabilities in a Gas Tur-
bine Combustor,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 29, 2002,
pp. 155–160.
57
Stone, C., and Menon, S., “Open Loop Control Combustion Instabilities in a Model
Gas Turbine Combustor,” Journal of Turbulence, Vol. 4, 020, 2003.
58
Pierce, C. D., and Moin, P., “Progress-Variable Approach for Large-Eddy Simulation
of Non-Premixed Turbulent Combustion,” Journal of Fluid Mechanics, Vol. 504, 2004,
pp. 73–97.
59
Selle, L., Lartigue, G., Poinsot, T., Koch, R., Schildmacher, K. U., Kerbs, W., Kaufmann,
P., and Veynante, D., “Compressible Large Eddy Simulation of Turbulent Combustion in
Complex Geometry on Unstructured Meshes,” Combustion and Flame, Vol. 137, 2004,
pp. 489–505.
60
Grinstein, F. F., and Fureby, C., “LES Studies of the Flow in a Swirl Combustor,”
Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 30, 2005, pp. 1791–1798.
61
Sommerer, Y., Galley, D., Poinsot, T., Ducruix, S. Lacas, F., and Veynante, D., “Large
Eddy Simulation and Experimental Study of Flashback and Blow-Off in a Lean Partially
Premixed Swirled Burner,” Journal of Turbulence, Vol. 5, 037, 2004.
62
Wang, S. W., Hsieh, S. Y., and Yang, V., “Unsteady Flow Evolution in Swirl Injector
with Radial Entry, Part 1: Stationary Conditions,” Physics of Fluids, Vol. 17, No. 4, 2005,
p. 045106.
63
Wang, S. W., and Yang, V., “Unsteady Flow Evolution in Swirl Injector with Radial
Entry, Part 2: External Excitations,” Physics of Fluids, Vol. 17, No. 4, 2005, p. 045107.
64
Huang, Y., and Yang, V., “Bifurcation of Flame Structure in a Lean-Premixed Swirl-
Stabilized Combustor: Transition from Stable to Unstable Flame,” Combustion and Flame,
Vol. 136, 2004, pp. 383–389.
65
Huang, Y., and Yang, V., “Effect of Swirl on Combustion Dynamics in a Lean-Premixed
Swirl-Stabilized Combustor,” Proceedings of the Combustion Institute, Pittsburgh, PA,
Vol. 30, 2005, pp. 1775–1782.
66
Huang, Y., Wang, S., and Yang, V., “A Systematic Analysis of Combustion Dynamics
in a Lean-Premixed Swirl-Stabilized Combustor,” AIAA Journal (accepted for publication),
2005.
67
Oefelein, J. C., and Yang, V., “Simulation of High-Pressure Spray Field Dynamics,”
Chapter 11, Recent Advances in Spray Combustion, Vol. 2, edited by K. K. Kuo Progress
in Astronautics and Aeronautics, Vol. 171, AIAA, Reston, VA, 1996, pp. 263–304.
68
Chiu, H. H., and Oefelein, J. C., “Modeling Liquid-Propellant Spray Combustion Pro-
cesses,” Chapter 6, Liquid Rocket Thrust Chambers: Aspects of Modeling, Analysis, and
Design, Vol. 200, edited by V. Yang, M., Habiballah, J., Hulka, and M., Popp, Progress in
Astronautics and Aeronautics, AIAA, Reston, VA, 2004, pp. 251–293.
69
Caraeni, D., Bergstrom, C., and Fuchs, L., “Modeling of Liquid Fuel Injection, Evapo-
ration and Mixing in a Gas Turbine Burner Using Large Eddy Simulation,” Flow, Turbulence
and Combustion, Vol. 65, 2000, pp. 223–244.
70
Sankaran, V., and Menon, S., “LES of Spray Combustion in Swirling Flows,” Journal
of Turbulence, Vol. 3, 011, 2001.
71
Apte, S. V., Mahesh, K., Moin P., and Oefelein, J. C., “Large-Eddy Simulation of
Swirling Particle-Laden Flows in a Coaxial-Jet Combustor,” International Journal of Mul-
tiphase Flow, Vol. 29, 2003, pp. 1311–1333.
72
Wang, S., Yang, V. and Koo, J. Y., “Large-Eddy Simulation of Spray-Field Dynamics
in Cross Flows,” AIAA Paper 2005-0729, 2005.
73
Menon, S., “Acoustic-Vortex-Flame Interactions in Gas Turbines,” Combustion Insta-
bilities in Gas Turbine Engines: Operational Experience, Fundamental Mechanisms, and
Modeling, edited by T. Lieuwen and V. Yang, Progress in Astronautics and Aeronautics,
AIAA, Reston, VA, 2005, pp. 277–314.
74
Lu, X. Y., Wang, S., Sung, H. G., Hsieh, S. Y., and Yang, V., “Large Eddy Simulation
of Turbulent Swirling Flows Injected into a Dump Chamber,” Journal of Fluid Mechanics,
Vol. 527, 2005, pp. 171–195.
75
Wang, S., Yang, V., Hsiao G., Hsieh, S. Y., and Mongia, H., “Large Eddy Simulation of
Gas-Turbine Swirl Injector Flow Dynamics,” to be submitted to Journal of Fluid Mechanics,
2005.
76
Wang, S., “Vortical Flow Dynamics and Acoustic Response of Gas-Turbine Swirl-
Stabilized Injectors,” Ph.D. Thesis, Pennsylvania State Univ., University Park, PA, 2002.
77
Favaloro, S. C., Nejad, A. S., Ahmed, S. A., Vanka, S. P., and Miller, T. J., “An Exper-
imental and Computational Investigation of Isothermal Swirling Flow in an Axisymmetric
Dump Combustor,” AIAA Paper 89-0620, 1989.
78
Wu, J. Z., Lu, X. Y., Denny, A. G., Fan, M., and Wu, J. M., “Post-Stall Flow Control
on an Airfoil by Local Unsteady Forcing,” Journal of Fluid Mechanics, Vol. 371, 1998,
pp. 21–58.
79
Mechanics, Vol. 16, 1984, pp. 365–424.
80
Panda, J., and McLaughlin, D. K., “Experiments on the Instabilities of a Swirling Jet,”
Physics of Fluids, Vol. 6, 1994, pp. 263–276.
81
Berkooz, G., Holmes P., and Lumley J. L., “The Proper Orthogonal Decomposition
in the Analysis of Turbulent Flows,” Annual Review of Fluid Mechanics, Vol. 25, 1993,
pp. 539–575.
82
Graves, C. B., “Outer Shear Layer Swirl Mixer for a combustor,” U.S. Patent 5-603-211,
1997.
83
Shtern, V., and Hussain, F., “Collapse, Symmetry, Breaking, and Hysteresis in Swirling
Flows,” Annual Review of Fluid Mechanics, Vol. 31, 1999, pp. 537–566.
84
Martin, J. E., and Meiburg, E., “Nonlinear Axisymmetric and Three-Dimensional Vor-
ticity Dynamics in a Swirling Jet Model,” Physics of Fluids, Vol. 8, 1996, pp. 1917–1928.
85
Cohen, J., and Hibshman, J., “An Experimental Study of Combustor Air Swirler Acous-
tic and Fluid Dynamic Sensitivities,” Proceedings of Propulsion Engineering Research
Center, Pennsylvania State Univ., University Park, PA, 1997.
86
Hussain, A., and Reynolds, W., “The Mechanics of Organized Wave in Turbulent Shear
Flow,” Journal of Fluid Mechanics, Vol. 41, 1970, pp. 241–258.
87
Huang, Y., “Modeling and Simulation of Combustion Dynamics in Lean-Premixed
Swirl-Stabilized Gas Turbine Engines,” Ph.D. Thesis, Dept. of Mechanical Engineering,
Pennsylvania State Univ., University Park, PA, 2003.
88
Lasheras, J., and Hopfinger, E., “Liquid Jet Instability and Atomization in a Coaxial
Gas Stream,” Annual Review of Fluid Mechanics, Vol. 32, 2000, pp. 275–308.
89
Brereton, G., Reynolds, W., and Jayaraman, R., “Response of a Turbulent Boundary
Layer to Sinusoidal Free-Stream Unsteadiness,” Journal of Fluid Mechanics, Vol. 221,
1990, pp. 131–159.
90
Propellant Rocket Engines,” Chapter 1, Liquid Rocket Engine Combustion Instability, edited
by V. Yang and W. E. Anderson, Progress in Astronautics and Aeronautics, Vol. 169, AIAA,
Washington, DC, 1995, pp. 3–37.
276 Y. HUANG ET AL.
91
Gupta, A. K., Lilley, D. G., and Syred, N., Swirl Flows, Abacus Press, London, 1984.
92
Lucca-Negro, O., and O’Doherty, T., “Vortex Breakdown: A Review,” Progress in
Energy and Combustion Science, Vol. 27, 2001, pp. 431–481.
93
Syred, N., and Beer, J., “The Damping of Precessing Vortex Cores by Combustion in
Swirl Generators,” Astronautica Acta, Vol. 17, 1972, pp. 783–801.
94
Novak, F., and Sarpkaya, T., “Turbulent Vortex Breakdown at High Reynolds Number,”
AIAA Journal, Vol. 38, No. 5, 2000, pp. 287–296.
95
Seo, S., “Parametric Study of Lean Premixed Combustion Instability in a Pressured
Model Gas Turbine Combustor,” Ph.D. Thesis, Dept. of Mechanical Engineering, Pennsyl-
vania State Univ., University Park, PA, 1999.
96
Rayleigh, J. W. S., The Theory of Sound, Vol. 2, Dover, New York, 1945.
97
Lieuwen, T., Torres, H., Johnson, C. and Zinn, B. T., “A Mechanism of Combustion
Instability in Lean-Premixed Gas-Turbine Combustor,” Journal of Engineering for Gas
Turbine and Power, Vol. 123, 2001, pp. 182–189.
98
Lipatnikov, A. N., and Chomiak, J., “Turbulent Flame Speed and Thickness: Phe-
nomenology, Evaluation and Application in Multi-Dimensional Simulations,” Progress in
Energy and Combustion Science, Vol. 28, 2000, pp. 1–74.
99
Chao, Y. C., Leu, J. H., and Huang, Y.F., “Downstream Boundary Effects on the Spectral
Characteristics of a Swirling Flowfield,” Experiments in Fluids, Vol. 10, 1991, pp. 341–348.
100
Liu, T., Huang, Y., and Yang, V., “Linear Stability Analysis of Annular Swirling Jets,”
Physics of Fluids (submitted for publication).
101
You, D. N., Huang, Y., and Yang, V., “A Generalized Model of Acoustic Response
of Turbulent Premixed Flame and Its Application to Gas-Turbine Combustion Instability
Analysis,” Combustion Science and Technology, Vol. 177, No. 5–6, 2005, pp. 1109–1150.
Chapter 11
Acoustic-Vortex-Flame Interactions in Gas Turbines
Suresh Menon∗
I. Introduction
M OST practical combustion systems, such as gas-turbine engines, internal

combustion engines, ramjets, and rocket motors are confined systems in
which operational design and size and weight constraints define the scale of the
device. Confined combustion systems can have dynamical features that are not
apparent in unconfined systems. For example, many of these devices have choked
outflow, and passage of vortical structures or hot spots through choked nozzles
can result in acoustic wave generation that can propagate upstream and interact
with the incoming flow and the flame zone. Geometric features such as acoustic
liner cavities, secondary injectors and complex ducts can all affect coupling among
vortex flow, acoustic motion, and unsteady heat release in these devices.
Gas-turbine combustors, which are the focus of discussion in this chapter, have
other unique features. The inlet to the combustor typically contains a complex
swirl-vane structure that induces a swirl to the hot air from the compressor. The
airflow may be split into multiple streams and each stream swirled independently
in either the counterdirection or the codirection.1 Fuel (liquid or gas) is injected
before, through or after these swirl vanes, and fuel–air mixing occurs in a highly tur-
bulent, swirling flow. Many propulsion gas-turbine combustors have both primary
and secondary combustion zones, and fuel is also split between them depending
on operation conditions.1,2 Additional complexity is introduced by the fact that
many fuel injectors are present in operational engines. Thus, interaction between
multiple injectors can introduce an additional layer of complexity in the dynamics
that makes analyzing and interpreting data very difficult.
In all of the previously noted phenomena, three physical mechanisms interact
in a highly nonlinear and unsteady manner. These three mechanisms are acoustic
fluctuations, vortex motion, and unsteady combustion heat release. Earlier studies3
∗ Professor, School of Aerospace Engineering.
277
278 S. MENON
characterized these three mechanisms as acoustic, vorticity, and entropy waves,

although only the acoustic field behaves as a wave, whereas both vorticity and
entropy “waves” are convected at the local flow velocity. In combustion systems,
entropy fluctuations can be attributed to unsteady flame propagation.
In this chapter, we discuss acoustic-vortex-flame (AVF) interaction in gas turbine
combustors. AVF interaction occurs in many devices, such as liquid-propellant and
solid-propellant rocket motors,4−6 ramjet engines,7−9 and dump combustors,10 but
these devices are not covered in this discussion, except to highlight observations
that are relevant for gas-turbine operations. Here, we discuss how vortex motion
in a gas-turbine combustor interacts with unsteady combustion heat release in a
confined domain and how this interaction can result in the excitation or enhance-
ment of acoustic disturbances. Chapters 10 and 13 have addressed specific mod-
eling and analysis issues related to AVF and combustion dynamics in gas-turbine
engines; therefore, these issues will not be repeated here. It is noted that acoustic-
vortex-flame interactions have been studied extensively in the past11−15 by using
laboratory scale devices. However, except for rare cases, most of these studies
have focused on low Reynolds (Re) number flows under atmospheric pressure
conditions. Here, we focus on premixed and liquid-fueled gas turbines that oper-
ate at the high pressures that are characteristic of real operating conditions. Earlier
experimental16−19 and numerical20−25 studies have addressed the dynamics in such
combustors, and some of the highlights of these studies are subsequently discussed.
This chapter is organized as follows. In the next section, we discuss briefly the
various length and timescales in a typical gas turbine engine and how these scales
define the AVF interaction process. This discussion is followed by a theoretical
analysis of the equations governing AVF interactions to identify key terms that
explicitly show the contributions of each of these modes. Finally, modes of AVF
interactions and their sensitivity to various system parameters are discussed in the
last section by using results from simulations.
II. Length and Time Scales

The interaction between acoustic waves, vortex motion, and unsteady flame
motion involves a wide range of time and length scales, and this range depends not
only on the actual sources (e.g., compressibility, shear layer separation and rollup,
or heat release) but also the geometry. In confined domains, boundary reflections
can introduce other scales into the interaction process11,26,27 that must be taken
into consideration.
If these modes are to interact, there has to be some overlap between their re-
spective time and length scales. Therefore, it is instructive to make some order-
of-magnitude estimates of these time and length scales for a typical gas turbine
combustor. From geometric data,1,20 the characteristic length scale of the com-
bustor in gas turbines is in the 0.1–1 m range with representative conditions (e.g.,
combustor pressure of 12 atm; inlet air temperature of 670 K; inlet bulk axial ve-
locity, U of 100 m/s; and inlet diameter, D of 0.05 m), the inlet Reynolds number
is estimated to be approximately 500,000. If we assume that the integral length
scale l ≈ D, then the integral length scale is in the range 0.01–0.1 m. The integral
scale represents the characteristic energy-containing eddies that play a major role
in energy and scalar transport in shear flows.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 279
For the preceding length scales, the turbulent Reynolds number, Rel = u l/ν,
where u is the turbulence intensity, is estimated to be in the 102 –104 range. The
higher value reflects the high level of turbulence in the regions of high shear (i.e.,
shear layer) in gas turbine engines. On using inertial range scaling, l/η ≈ Rel 3/4 ,
the Kolmogorov scale η can be estimated to be in the range 10−4 –10−5 m. Thus,
fluid dynamic length scales that are characteristic of vortex motion range from
10−5 –10−1 m. This is a four-order-of-magnitude (O(4)) range in length scales of
interest.
For reacting flows, additional length scales have to be considered. For example,
in two-phase systems, droplets are in the 10−4 –10−6 m range, whereas molecular
mixing and combustion occur in the 10−8 –10−9 m range. The latter estimate is
based on the observation that the typical CH molecule size is around 1.09 Angstrom
(10−10 m) and that the reaction-zone thickness, δ R R is at least 10–100 CH molecules
wide. This figure is just an estimate, since mixing and flame regions vary over
a wide range. Furthermore, the reaction-zone thickness is substantially smaller
than the effective flame thickness, δ f , especially in premixed systems where the
flame thickness can be substantially larger, particularly in the thin-reaction-zone
regime.21,28−30 Nevertheless, it is clear that species vaporization (in the case of
liquid fuel), mixing, and combustion occur in a range from 10−4 –10−9 m, an O(5)
range of scales.
Acoustic timescales can also be estimated on the basis of the range of frequencies
known to be excited in gas turbine engines. Typical frequencies are in the range
of 100–1000 Hz, and under standard conditions the wavelength range is 0.03–
0.3 m31,32 (under “hot” conditions, these values will be even higher). Thus, the
typical acoustic length scales of interest are in the 10−2 –100 m range.
Thus, there is an O(7–9) range of scale that has to interact in a turbulent reacting
flow under realistic conditions. These interactions have to occur in a time-accurate
manner (in nature, there is no other way!); and hence, the range of timescales is
equally large.
These estimates suggest that although a significant disparity exists between
the characteristic length scales in which vortex motion, acoustic fluctuations, and
heat release each dominate, there are also some regions of overlap. For example,
coupling between energy-containing eddies in a turbulent shear flow with acoustic
wavelengths in the 100–10,000 Hz range is feasible; however, at the other extreme,
eddies in the inertial-dissipation range are more likely to interact with unsteady
heat release at the molecular level.
The preceding argument is a rather simplistic view, since in reality, eddies of all
scales coexist and interact in a highly nonlinear manner in a turbulent flow. Regard-
less, these estimates suggest that the mechanism that couples acoustic fluctuations
with unsteady heat release in a turbulent flow is the dynamic range of scales inher-
ent in high-Re turbulent motion. In laminar flows, acoustic-vortex interactions can
still occur32−34 because, in addition to length scales, the timescales of interaction
are of the same order. For example, the flame-response timescale, τ f = δ f /SL ,
where δ f and SL are the laminar flame thickness and speed, respectively, for a
premixed system, is in the range of 10−3 –10−2 s, which is of the same order as
acoustic timescales for frequencies in the 100–1000 Hz range. Thus, it is possible
for an acoustic field to couple with heat release even in the absence of a turbulent
cascade of length scales.
280 S. MENON
This wide range of scales offers a serious challenge to both experimentalists and
modelers. Experimental diagnostic tools and simulation models both have to be
refined well enough to capture this wide range of scales accurately. That is easier
said than done.
The preceding discussion used order-of-magnitude estimates of time and length
scales independent of the problem of interest and the nature of the physics associ-
ated with these scales. As discussed in previous studies,28 the characteristic length
scales of interest can also be defined on the basis of the physics. For example, in
premixed combustion, the well-known Borghi diagram35 has been used in the past
to identify the type of flames (and hence, the associated length scales) in premixed
systems. Other diagrams have also been proposed on the basis of experimental and
numerical results in premixed and nonpremixed systems.31,36 These diagrams can
also be used to identify characteristic time and length scales of interest.
III. Theoretical Considerations

One way to analyze the nature of the three-way coupling is by considering
the various sources and sinks in the flow for each mode. Thus, shear flow (of
any kind) is the source of vorticity generation and convection. In gas-turbine
combustors, both wall-bounded and free shear flow occur. The free shear flow
is of primary interest, since the flow from the inlet duct forms a complex three-
dimensional swirling shear flow in the combustor and contains large-scale coher-
ent structures that undergo growth and eventual breakdown into fine-scale three-
dimensional turbulence further downstream. Since fuel is introduced into the air
in the inlet, these swirling vortical structures can consist of a partially or fully
premixed fuel–air mixture, and in the case of spray systems, can also contain fuel
droplets. Furthermore, in reacting flows, the flame is located in the region in which
these vortices are forming, and so flame–vortex coupling is intrinsic in this type
of flow.
Combustion-related unsteady heat release is the source of “entropy” mode,
which is typically characterized by fluctuations in temperature. Combustion oc-
curs in a compact region because the flame is stabilized in a region upstream of
a recirculation bubble that is created by the swirl in the incoming flow. Thus, en-
tropy perturbations occur in a compact but highly three-dimensional and transient
region.
Compressibility (i.e., density variation resulting in pressure fluctuations) is the
source of acoustic wave motion. Because of the many ducts and passages in a
gas turbine, there are many possible acoustic modes in the combustor. The entire
region from the compressor exit to the turbine entrance boundary can play a role
in acoustic wave motion. [See Chap. 13]. In general, longitudinal acoustic modes
are known to be very important in AVF interactions. However, because of the
geometric nature of the combustor and the swirling flame structure, radial and
circumferential acoustic modes can also exist.16,27 These acoustic modes can be
driven by periodic vortex shedding in dump combustors,7,37 especially since flames
can be entrained or modulated by the large-scale vortical structures.23
To investigate this three-way coupling from a theoretical point of view, it is
necessary to start with the conservation equations for the unsteady, compressible
reacting flow. Although low-Mach-number equations of motion that eliminate

acoustic waves from the governing equations have been employed for many low-
speed reacting flow studies,38 these equations are inappropriate to describe AVF
interactions. Even when only very low-frequency acoustic modes are of inter-
est, the scaling analysis discussed in the previous section shows that because of
the overlap in both time and length scales, the full compressible system needs
to be considered, especially in a confined domain. Combustion instability is
known to be the result of a coupling among acoustics, vortex motion, and un-
steady heat release; and this coupling can only be captured in a compressible
formulation.
In the following section, we discuss the relevant equations of motion to highlight
the terms that explicitly show how each of these modes appears in the equations.
All these equations are well known and reported in the cited literature; therefore,
some of the details regarding their derivation are avoided for brevity.
A. Governing Equations
For the purpose of the current discussion, we assume no external forces and
an inviscid, non–heat conducting, multispecies, nondiffusive reacting mixture.
The more general viscous equations and the linearization of the inviscid form for
acoustic analysis are reported in Chapter 13. The governing equations are
1 Dρ
∇ ·v= − (11.1)
ρ Dt
Dv ∇p
=− (11.2)
Dt ρ
De p
= − (∇ · v) (11.3)
Dt ρ
DYk ω˙k
= , k = 1, . . . N (11.4)
Dt ρ
Here, the substantial derivative is D/Dt = ∂/∂t + v · ∇, where v is the velocity

vector, ρ is the mixture density, p is the pressure, e is the internal energy per unit
mass defined as e = k ek Yk , where ek is the kth species internal energy (sum
of the translational, rotational, and heat of formation), Yk is the kth species mass
fraction, and ω˙k is the kth species production/destruction term. These equations
are usually closed by the equation of state for a perfect gas p = ρ RT , where T
is the temperature and where R = k Rk Yk is the mixture gas constant. Here, Rk
is the the kth species gas constant defined as Rk = R/Mk , where Mk is the kth
species molecular weight.
It is instructive to rewrite these equations in a form that highlights the underlying
nature of the three-way interaction under discussion here. There are many ways to
write these equations,31 and addressing all these alternative approaches is not the
goal here. We focus here on one specific formulation. By using the definition of
internal energy for a multicomponent mixture in terms of entropy, we can rewrite
282 S. MENON
Eq. (11.3) as39 (details are avoided for brevity)
N
DS 1 µk
=− ω˙k (11.5)
Dt ρT k=1 Mk
Here, S is the entropy and µk is the chemical potential.39 This equation shows
that entropy is convected because of fluid motion and can be generated by heat
release.
The mass conservation equation, Eq. (11.1) can be rewritten in terms of pres-
sure to identify the acoustic mode. Many forms of pressure (or wave) equations
have been derived in the past, including an equation that is often called the wave
equation.31 Here, a form that can be obtained by combining Eqs. (11.1), (11.5),
and the state equation is39
1 Dp 1 N
∇ ·v=− − σk ω˙k (11.6)
γ p Dt ρ k=1
where σk = h k /(C p T ) for an ideal gas mixture with constant average molecular
weight and where C p is the specific heat at constant pressure for a calorically
perfect gas mixture.
Equation (11.6) shows that the volumetric dilatation (∇ · v) can be affected by
pressure-wave motion and heat release. Additional manipulation of this equation
can result in a generalized equation for the logarithmic of the pressure31,40,41 that
can be used to model the acoustic wave motion.
By taking the curl of the momentum equation, Eq. (11.2), one can derive an
equation for vorticity, Ω:
DΩ ∇ρ × ∇ p
= (Ω · ∇)v − Ω(∇ · v) + (11.7)
Dt ρ2
For analysis purposes, Eq. (11.7) is often combined with Eq. (11.1) to obtain

D Ω Ω 1
= · ∇v + ∇ρ × ∇ p (11.8)
Dt ρ ρ ρ3
Both forms of the vorticity equation show important features that are relevant
to the current discussion. The first term on the right-hand-side of Eq. (11.7) and
Eq. (11.8) is the well-known vortex-stretching term. The other terms are unique
to compressible and reacting flow. The quantity Ω(∇ · v) is the thermal expansion
term, which is nonzero only in compressible flow and in reacting flow. In Eq. (11.8),
this term is absorbed into the convective term on the left-hand side. Therefore, the
left-hand side of Eq. (11.8) contains convection by both the hydrodynamic and
the acoustic velocity. As shown subsequently, this form can be used to identify
how acoustic and hydrodynamic disturbances can interact and contribute to vortex
motion.
The last term, ∇ρ × ∇ p/ρ 3 , often called the baroclinic torque, is a consequence
of the relative orientation of the pressure and density gradients. This term can be
significant in both compressible and reacting flows. For example, baroclinic torque
has been shown to affect flame wrinkling indirectly because of the changes induced
in the vorticity field.42 Pressure fluctuations normal to the density gradient caused
by the flame also contribute to vorticity generation. In gas turbine combustors,
the flame (and hence, the density gradient) is highly three-dimensional; and there-
fore, both longitudinal and transverse acoustic waves in the combustor can cause
changes to the local vortical field. Furthermore, this term can cause enhancement
or suppression of vorticity in the combustor depending on local conditions. This
behavior is discussed in the final section.
Equations (11.5), (11.6), and (11.7) (and Eq. (11.4) for species conservation) can
be considered to be an alternative set of governing equations in terms of entropy,
pressure, and vorticity instead of the conventional conservation equations. Except
for neglecting transport properties (e.g., dissipative effects) and external forces (and
these assumptions can be easily relaxed), these equations are exact, nonlinear, and
define the flowfield.
Although these equations are informative, their solution is not easy because of
the nonlinearity. These equations are, however, useful for linearized analysis and
for interpreting results. [See Chaps. 13 and 10]. Additionally, the direct numerical
simulation (DNS) or large-eddy simulation (LES) database can be used to extract
some of these terms explicitly.
B. Interpretation Using Field Decomposition

The earliest and classical mode decomposition carried out by Chu and
Kovaznay3 is still a valid starting point for understanding AVF interactions. More
details of this decomposition and its use in linearized analysis are given in
Chapter 13, and therefore, are not repeated here. Any flow variable
(x, t) can be
decomposed as

=
p +
+
S (11.9)
where the subscripts p, , and S, respectively, indicate the acoustic component,

the vortex component, and the entropy component (often identified with hot spots).
As shown elsewhere3 by using a linearized analysis, all three modes not only can
exist independently, but they can also interact and produce one another, especially
in a confined domain. It has been noted that vorticity and entropy modes can exist
even in the absence of pressure fluctuations but cannot exist if there is no mean
flow; and to the first approximation, weak vorticity fluctuations do not generate
pressure or entropy fluctuations of the same order.
Stability analysis11,27 of the linearized one-dimensional conservation equations
using the preceding decomposition has been quite successful in identifying the
typical frequencies of oscillation in combustors and in determining the ones that
will grow exponentially over time. In most of these studies, the linearized equations
of motion are solved subject to appropriate boundary conditions, jump conditions
across the flame, and by assuming that the initial perturbation is made up of the sum
of acoustic, entropy, and vorticity disturbances at a characteristic frequency, which
284 S. MENON
is determined as a part of the solution. For more-complex geometries, transfer

functions are used to account for geometric changes on the disturbances (see
Chap. 13).
In some of the analyses,11 more-complex boundary conditions have been ex-
plored to identify the nature of acoustic–vortex coupling. For example, in dump
combustors, because of the mean flow motion, a mixed-mode coupling between
acoustic waves and vortex motion, which is triggered in part by the coupling at the
boundaries,9 can exist. It was been shown that the large-scale vortices shed from the
rearward-facing step at the dump plane propagate downstream (at a characteristic
velocity of the order of the mean axial velocity) and impinge on the downstream
diffuser wall or interact with the choked nozzle condition. This interaction creates
a backward-propagating acoustic wave (traveling upstream with a velocity u − c)
that can interact with the shear layer at the step, thereby triggering the formation
of the next large-scale vortex and completing the feedback loop. Depending on
the scale of the combustor, the characteristic frequency for this coupling can be a
combination of both acoustic and vortex modes.
Modeling these acoustic–vortex mode requires that the boundary conditions
reflect this coupled nature. For example, at the backward-facing step where the
inlet boundary layer separates and forms a vortical free shear layer, to model the
triggering of the vortex mode by the acoustic fluctuation velocity u , the boundary
condition for the vortex mode can be = βu , where β is a transfer function
that has to be specified. To model backward-moving acoustic fluctuation (iden-
tified as the unsteady part of the dilatation field = ∇ · v) generated by vortex
impingement at, say, x = xn , a condition such as = −α∂δ(x − xn )/∂ x has
been employed in the past.11 Here, δ(x − xn ) is the Dirac delta function, and α
is another (complex) transfer function that has to be determined for a particular
problem (either from DNS/LES or from experimental data). Previous studies of
ramjet-type dump combustors have shown that the computed frequency predicted
by this type of coupled-mode approach agrees well with observations.
Such a coupled mode analysis for gas-turbine combustors has not yet been
carried out, especially when unsteady heat release is included. Some (drastic) sim-
plifications are needed to achieve a tractable formulation, since the complexity of
gas-turbine geometry, the swirl effects, the three-dimensional variation in mean
flow velocity and temperature, and the proper boundary conditions between mul-
tiple duct coupling and flow passages are difficult to implement in a simplified
one-dimensional formulation. Nevertheless, some valuable insight into the insta-
bility mechanism and nature of coupling has been (and can be) obtained in such
studies, as shown elsewhere in this book.
A decomposition suggested by Hussain43 could be used to obtain further insight
into AVF interactions, especially in highly turbulent flow containing large-scale
coherent structures (CSs). The flame structure can be substantially modified by
these CSs. Because of their deterministic nature and quasi-periodic motion, CSs
can contribute significantly to noise production in combustors. Previous studies
have shown that these structures play a major role in combustion instability in
dump combustors.
In this decomposition, any flow variable
is split as

(x, t) =
(x)
¯ +
(x,
˜ t; τ ) +
(x,
ˆ t) (11.10)
Here,
¯ is the mean value,
˜ is an unsteady “coherent” component (over a char-

acteristic time period τ ), and
ˆ is an unsteady “incoherent” contribution. This de-
composition is similar to the classical Reynolds decomposition, except that, in the
latter approach,
=
˜ +
ˆ represents all the unsteady fluctuations in the flow.
It has been shown43 that by using Eq. (11.10) in the governing equations, one
can derive governing equations for each type of motion. Analysis of the various
terms in these equations can be carried out to determine how the mean, the coherent
motion, and the incoherent motion interact in a turbulent medium.
It is feasible (although cumbersome) to use Eq. (11.10) in Eq. (11.9) to further
decompose AVF interactions into explicit contributions from mean flow, coherent
motion, and incoherent motion. It would then be feasible to identify the terms
that control, for example, acoustic field excitation by coherent structure transport,
and so on. Previous studies44 have shown that the DNS database can be post-
processed to obtain CS information in shear flows. Experimental data have also
been postprocessed to extract CS information by using a technique similar to this
triple decomposition.45 Similar analysis using the LES database can be carried
out to investigate the importance of large-scale coherent structures in gas-turbine
combustors.
C. Sources and Sinks in the Field Equations

The acoustic field can be characterized in two ways: 1) acoustics without any
flow and 2) acoustics with flow. In the former case, the generation and the motion of
waves because of small perturbation on top of a stagnant or a steady mean flow can
be studied by using linearized analysis. In the latter case, especially for gas-turbine
combustors, the flow is more complex and consists of three-dimensional swirling
shear layers, boundary layer separation, and variation in temperature caused by
unsteady heat release. In this case, the acoustic field can be excited or sustained
not only by boundary conditions (these conditions may involve unsteady sources
or sinks) but also by the very nature of the flow in the combustor. Here, we identify
some terms in the nonlinear governing equations that explicitly show sources and
sinks for each of these modes. Some of these terms can be extracted from numerical
simulation data (DNS or LES) to understand the physics of AVF interactions.
We consider a form of the acoustic equation that is obtained by taking the
divergence of the momentum equation, Eq. (11.2):

∂ 1
(∇ · v) + v · ∇(∇ · v) + ∇v : ∇v = −∇ · ∇p (11.11)
∂t ρ
and combine with Eq. (11.1), to obtain

D 1 Dρ 1 1
− ∇ 2 p = ∇v : ∇v − 2 ∇ρ · ∇ p (11.12)
Dt ρ Dt ρ ρ
If we consider the equation of state of the form ρ = ρ( p, S), then

Dρ ∂ρ Dp ∂ρ DS
= + (11.13)
Dt ∂ p S Dt ∂ S p Dt
286 S. MENON
On using Eq. (11.13) in Eq. (11.12) and noting that (∂ρ/∂ p) S = a S 2 is the
isentropic (frozen) acoustic speed, we obtain

D 1 Dp 1 2 1 D 1 ∂ρ DS
− ∇ p = ∇v : ∇v − 2 ∇ρ · ∇ p −
Dt ρa S 2 Dt ρ ρ Dt ρ ∂ S p Dt
(11.14)
Some observations for this equation include the following: 1) no linearization

is carried out at this stage, 2) nonlinear acoustics are present on both sides of
the equation, 3) the first term on the right-hand side (RHS) is the generation of
acoustics by hydrodynamic disturbances and some nonlinear acoustics, and 4) the
last term on the RHS is the generation of acoustics by entropy disturbances. This
last term can be replaced by the heat-release term if Eq. (11.5) is used. This term
contributes to combustion noise and instability, especially if its fluctuation is in
phase with the pressure fluctuation.
The term ∇v : ∇v is present even in nonreacting flow. It can be further inter-
preted (by using tensor notation with velocity qi ) as

∂q j ∂qi ∂2 ∂ ∂q j ∂ ∂qi
= qi q j − qi − qj (11.15)
∂ xi ∂ x j ∂ xi ∂ x j ∂ xi ∂x j ∂ xi ∂ x j
The first term in the RHS of this equation is the classical Lighthill’s Reynolds
stress sound-generation term.31 All other terms on the RHS are nonzero only for
reacting and compressible flow.
Thus, the acoustic equation, Eq. (11.14) contains explicit terms for hydrodynam-
ics and combustion heat release that can either enhance or suppress the pressure
disturbance. The analysis of these terms can shed insight into AVF interactions.
Such an analysis can only be done by using numerical simulation data and sug-
gests a possible avenue by which the DNS or LES database could be used to help
interpret experimental observations, which are likely to be limited in spatial and
temporal resolution.
The sources and sinks in the vorticity equation can also be analyzed. If we
consider that the velocity field is decomposed into a hydrodynamic and an acoustic
component, as v = v + v p , then the left-hand side of Eq. (11.8) can be written
as

D D D D 1 1
= + + −
Dt ρ Dt ρ0 Dt p ρ Dt ρ ρ0
(11.16)
The implication of this decomposition is that the first term on the RHS is the
Orr–Sommerfield operator (in the linearized limit). The second term can be rewrit-
ten as

D
≈ vp · ∇ (11.17)
Dt p ρ ρ
and represents the transport of vorticity by the acoustic disturbance; it acts as a

source for the Orr–Sommerfield operator. Physically, it shows that if the frequency
of the acoustic disturbance coincides with the frequency of the most unstable
hydrodynamic mode (as determined by the linearized stability analysis), then this
source can drive the instability to large-amplitude disturbance.
It can be seen that acoustic disturbances can interact with hydrodynamic distur-
bances in the shear layer, and under certain conditions can enhance the instability,
even in the linear limit. Therefore, the frequency content and the spatial distribution
of the source term v p · ∇(/ρ), especially in the region of shear layers, should
be analyzed to understand the physics of acoustic-vortex interactions in turbulent
flows.
Sources and sinks for acoustics and the vorticity field also exist in the flame zone.
Temperature rise because of combustion in the flame region increases viscosity; and
viscosity, in turn, leads to dissipation of small-scale turbulent structures. However,
unsteady flame motion in an acoustic field can result in turbulence enhancement or
suppression by the baroclinic torque term (∇ρ × ∇ p/ρ 2 ) in the vorticity equation.
Flame motion also can contribute to the pressure field by the heat-release source
term in the acoustic equation, Eq. (11.14).
IV. Factors Affecting AVF Interactions

AVF interactions in gas turbines occur for a variety of reasons; however, some
key system parameters or operating characteristics are very important. Here, we
discuss some of these parameters.
A. Swirl
Swirl is a key element in all gas-turbine engines and is used to create a region
of high entrainment and mixing for the fuel–air mixture. Swirl also provides an
efficient mechanism to stabilize the flame in a compact region without requiring
a physical flame holder. In general, the extent of swirl is typically defined by the
swirl number Si , which is the ratio of the axial flux of angular momentum to the
product of the inlet radius and axial flux of axial momentum. Other parameters—
such as inlet swirl-vane geometry, Reynolds number, confinement geometry, and
inlet velocity (both mean and fluctuation) profiles—can all affect swirl effects.
With and without confinement, the nature of the flow downstream changes
significantly as a function of the swirl number Si . As long Si is below a critical
value, typically 0.6 for dump combustors,46 the shear layer from the inlet separates
from the dump plane and rolls up into vortices that eventually coalesce into large-
scale coherent vortex structures. These structures propagate downstream at a phase
velocity, which is of the order of the mean velocity, and maintains coherence for
some distance before breaking down into more irregular, three-dimensional vortical
structures.
Results from classical linear stability theory47,48 can be used to obtain insight
into the vortex motion for low swirl numbers. For example, stability analysis sug-
gests that the characteristic Strouhal number (Stθ = f θ θ/U ) for the most unstable
mode in a shear layer is 0.032. Here, θ is the shear layer momentum thickness, U
is the characteristic inflow mean velocity, and f θ is the characteristic frequency
288 S. MENON
of the mode. An order-of-magnitude estimate for a typical gas-turbine combustor

using U = 100 m/s and θ = 0.1 mm gives f θ = 32 kHz. However, such a high-
frequency mode of instability is very difficult to resolve within a typical “noisy”
combustor, so it is difficult to know whether this classical instability mode exists
within a real combustor.
This initial instability mode causes the vorticity in the shear layer to roll up into
vortices that can undergo multiple pairing and merging processes until the final
large-scale structure is formed at the end of the potential core.43 In practical devices,
since the inflow is highly turbulent and swirling, the rollup or merging process
can be quite abrupt and can be unlike the previously described classical process.
Regardless of how the initial process begins, the final large-scale structure is seen to
exist within a short distance from the shear layer separation point at the inlet dump
plane. Further downstream these large-scale structures break down into smaller,
three-dimensional irregular structures typical of three-dimensional turbulence. The
characteristic frequency at each step of the pairing and merging process decreases
continuously, and the frequency of the final structure is called the jet-preferred
mode. Past experimental studies suggest that the typical Strouhal number for this
mode, St = f D/U , is in the 0.1–1 range.49 Frequencies in this jet-preferred mode
are in the 0.1–1 KHz range, and can be easily resolved in measurements. Note that
these frequencies are also well within the range of acoustic frequencies in the
gas-turbine combustor; and therefore, coupling between the large-scale structures
and acoustic modes can easily occur.
In reacting flows, these large structures, which are shed from the dump plane
in a periodic manner, can interact with the flame. Periodic vortex shedding at a
characteristic frequency is seen in both experiments and numerical simulation.10,50
In dump combustors, it has been shown that these structures can modulate the flame,
so that the flame moves with these structures and the coupling is only broken when
these structures break down farther downstream. If this frequency is the same as
one of the key acoustic modes in the combustor, it is feasible that the vortex–
flame motion can add energy to the pressure fluctuation leading to combustion
instability.
In the following, we briefly describe some characteristic results for premixed
combustion obtained in a General Electric (GE) LM6000 combustor single-
element injector test rig.23,50 This configuration consists of a swirling premixed
mixture entering from a single inlet pipe into a dump combustor. The premixer
itself is not modeled in this study, and the inlet conditions and profiles just down-
stream of the premixer are prescribed on the basis of data from an earlier study
at GE. The inlet temperature is 673 K, and the combustor pressure is 11.8 atm.
The bulk-flow Reynolds number that is based on the inflow velocity and the inlet
diameter is around 527,000.
Simulations were conducted by using a finite-volume code that is second-order
accurate in time and fourth-order in space. Subgrid closure for the momentum and
energy LES filtered equations is carried out by using a transport model for the
subgrid kinetic energy. A localized dynamic closure20,21 is employed to obtain the
coefficients of the subgrid model locally in space and time without requiring any
averaging or smoothing. The premixed methane flame at an equivalence ratio of
0.52 is simulated by using a thin-flame model with a dynamic turbulent flame speed
model. A grid of 181 × 73 × 81 is used, with clustering in regions of high shear.
Fig. 11.1 Flame and vortex structures in the combustor for inlet swirl number
of Si = 0.56.
Previous studies20,21 have compared the LES predictions against measured data
for a similar single-element rig, and very good agreement was obtained. Here,
we discuss more qualitative features extracted from these simulations to highlight
the nature of AVF interactions. Figure 11.1 shows the typical instantaneous flame–
vortex structure for a low-swirl, (Si = 0.56) case. Here, the reference swirl number
Si is defined at the inlet boundary (it drops to 0.42 at the dump plane). The flame in
this figure is represented by an isoscalar surface, and the coherent structure is shown
as the λ2 isosurface.51 It can be seen that low swirl results in large-scale, coherent
ring-like structures that are similar to structures seen in a nonswirling forced jet
shear layer. These structures undergo rotation and exhibit azimuthal perturbation
that rotates and grows as the structures move downstream. The flame moves in
phase with these structures until the vortical structures break down because of
instability. At this stage, the flame decouples from the vortex ring and retreats
toward the inlet. It is then pulled forward by the next large-scale structure. Thus,
the vortex-flame structure undergoes a periodic pulsation.
The shape and pulsation of the flame in the low-swirl case is reminiscent of the
“tulip” flames seen in acoustic coupled laminar flames.40 Here, although the con-
ditions are highly turbulent, the modulation of the flame by the coherent structures
seems to result in the observed shape.
In Chapter 10 additional flame-vortex interaction effects are discussed, except
that in their case, the combustor consists of an annular swirling flow in the inlet with
a centerbody. Although some differences are attributable to the geometric nature
of the two devices, significant similarities suggest that many observed features are
fundamental components of swirling combustion.
When the swirl is increased beyond the critical value, a major physical change in
the flow is observed. Around the critical value, the adverse axial pressure gradient
290 S. MENON
caused by swirl exceeds the forward momentum force of the inflow, and the flow
reverses. This reversal typically occurs first in the centerline region for a single
swirling jet inlet. In the coaxial swirl combustor or in combustors with a centerbody,
the location of flow reversal is off-center but still axially located in the region where
the swirling flow enters the combustor.
High swirl in the inflow results in the formation of a vortex breakdown bubble
(VBB), or a recirculation bubble. This VBB can be a single bubble at the centerline
or a toroidal structure depending on the inlet conditions, the inlet, and the dump
combustor design. In multielement, multiswirler sector rigs, the shape of VBBs can
be even more complex and can consist of both core recirculation zones and offset
toroidal structures.52,53 Depending on the geometry and the operating condition,
the leading edge of the VBB can move into the inlet as well.
In any event, the VBB acts as an aerodynamic blockage that inhibits the classical
growth and rollup–pairing processes in the shear layer. The initial rollup–merging
process and the formation of large-scale structures still occur. However, since the
shear layer is diverted radially away from the centerline because of the presence
of the VBB, the large-scale structures undergo rapid stretching in both the radial
and azimuthal directions. This stretching causes these structures to break down
quickly into three-dimensional relatively small-scale, but still vortically coher-
ent, structures.54 This breakdown leads to the formation of more axially oriented
streamwise vortices that form the precessing vortex core (PVC) that rotates about
the central VBB46,55 while the shear flow propagates downstream.
The effect of high swirl and of the formation of the recirculation bubble on
the flame structure is dramatic. Figure 11.2 shows an instantaneous flame–vortex
Fig. 11.2 Flame and vortex structures in the combustor for inlet swirl number of
Si = 1.12.
structure for a high-inlet-swirl (Si = 1.12, which reduces to around 0.74 at the
dump plane) case for the same combustor shown previously (both simulations
employed identical conditions except for an increase in Si ). With an increase in
swirl, the vortical ring structures break down quickly, and the coupling between
the flame and the vortex is broken. The result is a compact flame that stabilizes
very close to the dump plane and upstream of the VBB.
Figures 11.3a and 11.3b show representative particle paths in the combustor
for the low and high swirl cases, respectively. The streamlines are computed by
using the time-averaged velocity field and therefore do not represent the actual
instantaneous motion of the fluid element in the flow. Nevertheless, these particle
paths provide some insight into the complex flow in the combustor. For each case,
two seed particles are injected at the inlet; one near the centerline and the other
in the boundary layer on the inlet wall. In Fig. 11.3a (low-swirl case), the particle
injected at the centerline is propelled almost without interruption through the entire
combustor. The particle injected near the inlet wall shows a much stronger effect
of the swirl. Inside the combustor, the fluid particle initially slows down and then
moves upstream on entering the separated flow region at the base of the dump plane.
Figure 11.3b shows that for the high-swirl case, the particle injected along the
centerline can get entrained into the VBB and undergoes multiple revolutions in
it before being ejected. The outer particle does not enter the VBB; rather it is
quickly accelerated around the VBB toward the rear of the recirculation zone.
From there, the particle velocity rapidly decreases, and a slow rotating motion is
initiated before the particle is finally ejected into the main flow and accelerated
out of the combustor. An analysis of the instantaneous particle paths indicates that
particles in the shear layer tend to be present more in the PVC than in the VBB.
The flow inside the VBB contains primarily hot burned products, is highly
turbulent with strong three-dimensional variation of turbulent kinetic energy, and
with significant anisotropy in the Reynolds stress components.46 Studies with a
central fuel jet surrounded by a swirling coaxial air56 have shown that the flame
essentially surrounds the toroidal VBB. Furthermore, they showed that the VBB
size, the mixing rate, and the flame length depend on the degree of recirculation and
the ratio of the momentum of the fuel jet to the momentum inside the recirculation
region. These results suggest that the VBB acts as a large eddy that is not just a
passive structure in the flame-stabilization process.
Stability analysis of highly swirling jets, with and without confinement, have
shown that many modes of instability can occur in swirling shear layers. Studies57
in an unconfined swirling water jet in a low-to-moderate Re (∼ 1200) showed
that, in addition to the classical VBB, a conical vortex sheet structure is formed.
Both these structures became asymmetric with an increase in Reynolds number,
suggesting a spiral mode of instability.
In much higher Re swirling free jet and wake flows,54,58 more-complex insta-
bilities were found. In particular, it was shown that both axisymmetric and helical
instability waves exist in an St = 0.75–1.5 range. The vortex structure around the
VBB was less coherent and irregular for this case; but with even weak acoustic
excitation, the coherence and periodicity of the structures improved dramatically.
This observation is an important one for gas-turbine combustors, since confinement
results in the excitation of the acoustic modes in the combustor. Thus, it is possible
that the observed coherence of large-scale structures in these combustors may be
292 S. MENON
a)
b)
Fig. 11.3 Particle paths in the combustor as a function of inlet swirl number:
a) low swirl, Si = 0.56 and b) high swirl, Si = 1.12.
due to forcing by even small-amplitude acoustic fluctuations in the combustor (of

course, the frequency of the acoustic fluctuation must be in the jet-preferred mode
for this to occur).
Figures 11.4a and 11.4b show the center plane time-averaged axial velocity
contours for the low- and the high-swirl cases, respectively. The recirculation
regions near the base of the step and in the combustor are identified by a black
a)
b)
Fig. 11.4 Axial velocity contours and flame structure as a function of swirl number.
The recirculation zones in the combustor are identified by black solid lines: a) low
swirl, Si = 0.56 and b) high swirl, Si = 1.12.
294 S. MENON
solid line. The typical flame structure is also identified is an isocontour. In the
low-swirl case, there is no VBB. However, for the high-swirl case, a large VBB is
clearly visible in the combustor. The dump-plane base recirculation bubble (BRB)
is also visible in both these figures. For low swirl, the BRB is relatively large;
whereas for high swirl, the BRB is very small (but with more intense flow motion)
and trapped up in the step corner.
In gas-turbine combustors, the confinement geometry can introduce new features
that can affect shear flow and flame stabilization. In the configuration discussed
previously, the rearward-facing step at the dump plane creates a base recircu-
lation bubble (BRB) containing hot combustion products that provides another
mechanism for flame reignition and stabilization. Flame stabilization using the
rearward-facing step is well known and is employed in dump combustors as in a
ramjet in the absence of swirl. The current results suggest that the size of the BRB
depends on both the swirl intensity and the shape of the device. Low swirl creates a
well-defined BRB, since no VBB is in the core of the flow [Fig. 11.4a]. However,
when the swirl is very high, the VBB is very close to the dump plane, and the
lateral divergence of the shear layer compresses this BRB into a small region that
is very close to the step corner [Fig. 11.4a]. Thus, it appears that the BRB may not
play a major role in flame stabilization in the high-swirl case.
The preceding observations are only valid in a dump combustor with a single
inlet. In combustors with coaxial flow or with centerbodies,53,59 multiple BRBs can
form. However, they are much smaller than the primary VBB and may only play a
secondary role in flame stabilization. In most operational gas turbine combustors,
the VBB is the primary aerodynamic flame-stabilizing mechanism.
The behavior of the PVC is also significantly altered by combustion heat release
and by the system parameters (e.g., swirl number, axial velocity and geometry).
Previous studies46,60 suggest that the type of combustion (i.e., premixed or non-
premixed) can affect the PVC frequency and its intensity. Geometrical and system
features, such as secondary air injection for cooling and nonaxial injection of
fuel, can also lead to significant changes to the PVC and to the flame-stabilization
process.
The unsteady motion of the VBB, the PVC, and the flame structure all can lead to
unsteady pressure fluctuation in the combustor. As previously noted, in-phase fluc-
tuations can lead to instability. However, under certain conditions, since an increase
in swirl can actually lead to rapid breakdown of the coherent vortices in the shear
layer, it can contribute to a de-coupling of the vortex motion from acoustic fluctu-
ations and unsteady heat release. As a result, instability may actually be averted.
The stability of the combustion process can be determined by evaluating the
Rayleigh criterion. This criterion says that when unsteady heat release is in phase
with the unsteady pressure fluctuation, the heat release adds energy to the oscilla-
tion. This energy addition can lead to combustion instability if it exceed the losses
(caused by viscous dissipation and by outflow) from the system. In spatially evolv-
ing unsteady reacting flow, as in a combustor, the phase between unsteady heat
release and unsteady pressure can vary locally both in space and in time. Thus,
in some regions in the combustor, the instability could be suppressed whereas
in other locations it could be enhanced. To quantify
the state of combustion, a
volume-averaged Rayleigh parameter R(t) = V p q dV can be defined. Here,
the integral is over the entire combustor domain and q and p are the unsteady
a)
6
Unstable ( prms = 2)
4
2
*
p' and R
-2
-4
Rayleigh parameter
Pressure fluctuation
-6
1.0 1.2 1.4 1.6 1.8 2.0
Time (ms)
b)
6
4 Stable ( prms = 1)
2
*
p' and R
-2
-4
Rayleigh parameter
Pressure fluctuation
-6
1.0 1.2 1.4 1.6 1.8 2.0
Time (ms)
Fig. 11.5 Rayleigh parameter and pressure fluctuation in the combustor: a) low swirl,
Si = 0.56 and b) high swirl, Si = 1.12.
heat-release and pressure, respectively. Positive values of R(t) correspond to net

energy addition to the acoustic modes, (i.e., instability growth), whereas negative
values signify energy loss or attenuation.
A time sequence of R ∗ = R(t)/R, where R is the time average of R(t), is shown
in Fig. 11.5a for the low-swirl case. Also shown is the pressure fluctuation (at
the base of the step) for the same time period. Here, p represents the nondimen-
sional pressure fluctuation expressed as a percentage of the mean pressure. For the
low-swirl case, amplification (i.e., positive R ∗ ) is seen to occur in phase with the
pressure signal recorded near the dump plane. During the amplification phase, a
new vortex ring is shed at the base of the flame and the pressure near the flame
296 S. MENON
zone is higher than the mean. Because of this high-pressure zone, the axial velocity
is low and the flame is able to propagate further upstream, consuming more fuel
along the way. Under this condition, unsteady heat release occurs in phase with
pressure fluctuation, leading to R ∗ > 0.
For the-high swirl case, Fig. 11.5b, the flame is very compact and the vortical
structures are no longer coherent rings, as in the low-swirl case. Flow is accelerated
around the VBB; and near the dump plane, heat release is no longer coupled to the
pressure fluctuation. As a result, the Rayleigh parameter and pressure fluctuation
are no longer in phase and combustion is more stable. An estimate for the pressure
root-mean-square fluctuation intensity shows that pr ms is decreased by nearly 100
percent when the combustion process becomes more stable.
In the preceding discussion, the effect of swirl was discussed primarily for
premixed combustion. However, swirl is also very important for all gas-turbine
engines used for propulsion where liquid fuel is employed. The fuel-injection
system creates droplets over a wide range of sizes by the atomization process.
Droplet transport, vaporization, fuel–air mixing and combustion in the combustor
depend in part on how these droplets are entrained and dispersed within the swirling
air stream. These issues are discussed in the following two sections.
B. Droplet-Vortex Interactions
Before discussing spray dispersion and combustion in gas-turbine combus-
tors, summarizing observations from fundamental studies of droplet dispersion
in shear layers is worthwhile. Previous studies22,61 using direct numerical simula-
tions (DNSs) of droplet-laden temporal mixing layers provide insight into droplet–
vortex interaction, and some results are discussed subsequently.
Simulations of two-phase flows are carried out by using the previously noted
finite-volume code with a Eulerian–Lagrangian approach. In this method, the gas
phase is simulated by using the usual finite-volume approach, whereas the particle
motion is simulated by using a Lagrangian scheme. Full two-way coupling is
included in this approach. Droplet vaporization, gaseous fuel mixing with air, and
subsequent combustion can also be simulated with this solver, as reported in the
following section.
Here, we show some results for droplet motion in a temporal mixing layer. The
test conditions used here are identical to those used in an earlier DNS study using
a pseudospectral DNS code.61 A cubic domain is discretized by using a uniform
grid of 64 × 64 × 64. A temporal mixing layer, initialized by a tangent hyperbolic
mean profile and perturbed by the first two most unstable two-dimensional modes
is simulated here. The computational domain is chosen such that one vortex pairing
can be simulated.
We study the effect of Stokes number St0 on particle dispersion. Here, St0 is
defined as St0 = τd /τ f , the ratio of a particle response time τd and a characteristic
flow time τ f . Here, τd = (ρd D 2 )/18µ is the particle response time, ρd is the
particle density, D is the particle diameter, and µ is the molecular viscosity of
the gas phase. The characteristic flow time is τ f = L/U0 where L and U0 are the
characteristic length and velocity scale for the flow.
As previously defined, St0 1 implies that the particle can easily respond to
the changes in flow, and therefore the particle and the flow velocity will reach
an equilibrium. However, for St0 1 particles will not have time to respond to
the flow velocity change and particle motion will not be significantly affected.
For a given flow time, variation in St0 is directly related to the particle diameter.
In the following, we compare the behavior of particles with St0 of 0.1, 1, 4, and
100 in the mixing layer. The initial condition for the droplets is such that they are
uniformly distributed throughout the domain and are in dynamic equilibrium with
the gas phase. One particle per cell is placed at the start of the simulation. For
each simulation, droplets of same size are used (fixed St0 ) and no vaporization is
included here.
Figure 11.6 shows the distribution of droplets at a nondimensional time of
T = 28. At very small St0 , droplets are dragged along with the fluid and can
a) b)
c) d)
Fig. 11.6 Droplet dispersion in a mixing layer as a function of Stokes number:

a) St = 0.1, b) St0 = 1.0, c) St0 = 4.0, and d) St0 = 100.
298 S. MENON
therefore even be convected into the core of the spanwise vortical structures. As
the droplet Stokes number approaches unity, droplets begin to accumulate near the
circumference of the large vortical structures, which leads to an increase in lateral
droplet dispersion with droplets marking the lateral boundaries of the rolled-up
vortical structures. With a further increase in Stokes number, droplet response time
further increases and the influence of flow on the droplets decreases. Some droplets
are then observed to pass through the core of these structures and accumulate in
the braid regions between the large vortical structures. This trend is in very good
agreement with observations in the earlier DNS study.61
Dispersion behavior of droplets of St0 = 4 is shown in Fig. 11.7. Droplet dis-
tribution is shown at two nondimensional times. Figure 11.8 shows the span-
wise vorticity and the droplet distribution at the same instants. Droplets tend
to accumulate around the circumference of the large vortical structures that are
formed by rollup. The primary spanwise vortical structure is essentially devoid of
droplets because the strong vorticity present at the core of the spanwise vorticity
centrifuges the droplets away from the center. These droplets accumulate near
the high-strain regions of the flow, such as the braid regions between the larger
spanwise rollers. These results agree with past observations61−63 that droplets
tend to accumulate in regions of low vorticity and high strain rate. Such prefer-
ential concentration of droplets can have serious implications in liquid-fueled
combustors, where preferential accumulation may lead to large spatial varia-
tion in the mixture-equivalence ratio, which in turn, may lead to incomplete
combustion.
Figure 11.9 shows the root mean square (RMS) of droplet number per cell
(Nrms ) over the whole field. This number can be used to characterize the overall
accumulation tendency of the droplets in the flowfield. The Nrms is defined as61

Nc
N 2
Nrms = i
(11.18)
i=1
N C
where Nc is the total number of computational cells and Ni is the number of

droplets in the ith cell. Nrms increases with time, indicating the correlation between
the growth of the mixing layer and the dispersion of the droplets. Droplets with
a Stokes number of the order of unity (St0 = 1−4) have higher Nrms than lighter
(St0 1) and heavier droplets (St0 1). As previously noted, lighter droplets
have a strong tendency to follow the carrier (gas) phase, and hence they become
more dispersed. Heavier droplets, however, follow their own inertia and are not
dispersed much by the flow. However, droplets with a Stokes number of the order
of unity (St0 = 1−4) are preferentially dispersed by the flow, and hence their
distribution becomes nonhomogeneous even though their initial distribution is
homogeneous. The agreement with the spectral DNS data61 is very good over the
whole range of St0 .
Another way to quantify the dispersion of droplets with different St0 is to
evaluate a dispersion function in the transverse (y) direction for the droplets
initially distributed in the midplane in this direction. The dispersion function is
a)
b)
Fig. 11.7 Droplet dispersion in mixing layer for St0 = 4: a) T = 12 and b) T = 20.
300 S. MENON
a) b)
Fig. 11.8 Droplet distribution and spanwise vorticity in mixing layer for St0 = 4:
a) T = 12 and b) T = 20.
defined as61

1 Np
D y (t) = [Yi (t) − Ym (t)]2 (11.19)
N p i=1
where N p is the total number of droplets, Ym (t) is the mean value of the droplet
displacement in the vertical direction at time t, and Yi (t) is the displacement of an
ith droplet in the vertical direction at time t.
T = 24 (Present DNS )
RMS of particle numbers per cell
T = 28
T = 36
T = 24 (Spectral-DNS)
T = 28
2 T = 36
1.5
0.01 0.1 1 10 100 1000

Stokes number
Fig. 11.9 Droplet concentration fluctuation intensity per cell as a function of
Stokes number.
4
St O = 0.1
St O = 1
St O = 4
3 St O = 10
St O = 100
Dy (t)
0
0 10 20 30
Nondimensional time
Fig. 11.10 Stokes number effect on transverse droplet dispersion.
Figure 11.10 shows this dispersion function as a function of time for the droplets
for various St0 . At very small times, droplets with St0 = 0.1 show more dispersion.
However, at later times, droplets with St0 of order unity show more dispersion.
Droplets with very large Stokes numbers have much lower dispersion compared
with lighter droplets.
Vortex–droplet coupling has also been observed in liquid-fueled com-
bustors.22,64 Figure 11.11 shows a GE Dual-Annular Counter-Rotating Swirling
Fig. 11.11 GE-DACRS gas-turbine combustor configuration.

302 S. MENON
(GE-DACRS) combustor that was used in a recent study to understand droplet dis-
persion in swirling flow. The computational grid used 141 × 75 × 81 grid points
in the axial, radial, and azimuthal directions, respectively, with clustering in high-
shear regions. A large number of particles, typically more than 100,000 com-
putational parcels are tracked. Each of these parcels contains an average of 150
individual droplets. Thus, an average of 15 million droplets are present in the
computation. A monodisperse distribution with spherical droplets of size 40 µm
(which is approximately the Sauter mean diameter of the droplets in DACRS)
is introduced in the inlet through the central shaft at half-angle of 10 deg to the
flow. For the simulated case, the inlet pressure was 13.8 atm and the bulk Re was
260,000.
Figures 11.12a and 11.12b show a low swirl (Si = 0.75) case and a high swirl
(Si = 1.5) case, respectively. Because on the shape of the inlet, the actual swirl
numbers at the dump plane are approximately 0.5 and 0.8, respectively, attributable
to flow relaxation and decay in the inlet. The vortical structures undergo helical
instability and break down rapidly with an increase in swirl, as seen in the pre-
mixed combustor. Closer examination shows that sheetlike structures associated
with the spanwise vorticity z form tubular rings, whereas structures associated
with the streamwise vorticity x form tubular braidlike connections between the
rings. However, farther downstream as the rings break down, more randomly ori-
ented tubular structures appear. Analysis described elsewhere22 has shown that the
magnitude of the strain field controls this breakdown.
With an increase in swirl, droplet dispersion increases significantly. As observed
in temporal mixing layers, droplets tend to collect in the low-vorticity, high-strain
regions; and their dispersion is significantly enhanced once the large-scale ring
structures have lost their coherence.
C. Droplet–Vortex–Flame Interactions
In spray combustion systems, additional complexity can be created by the in-
teraction between droplets and flame. Droplet vaporization is accompanied by
fuel–air mixing and occurs in a swirling flow. However, in most situations, mix-
ing is not perfect everywhere; and in general, premixed, partially premixed, and
nonpremixed regions can coexist in the mixing region. Thus, the flame structure
can be quite complex in these regions. Both isolated flames around droplets and
group combustion of droplets (where droplets are clustered close together so that
only the region at the edge of the cloud contains vaporized fuel) have been seen in
spray combustion simulations.22
These observations suggest that the location and the structure of the unsteady
heat-release zone in the combustor can vary significantly depending on the ability of
the inlet swirl to mix and then redistribute the mixed fluid. Very little experimental
data on operational gas turbines describe how the flame structure actually correlates
with the vortical structures and with the acoustic fluctuation as the operational point
is systematically changed.
However, some insight into the flame structure has been possible from sim-
ulation data. It appears that both spray and premixed combustion systems have
physical characteristics that are remarkably similar.22,65 These results also agree
a)
b)
Fig. 11.12 Spray dispersion and vortex structure in the GE-DACRS combustor:
a) low swirl, Si = 0.75 and b) high swirl, Si = 1.5.
304 S. MENON
X
Z
Fig. 11.13 Schematic of the GE-1 combustor two-cup sector rig.
with classical DNS studies in simpler flows at relatively low Re, suggesting some
sort of universality in these features. The most likely strain state is appears to be
axisymmetric extension; and the vorticity tends to align itself with the intermediate
strain rate, whereas the scalar gradient aligns with the most compressive strain rate.
The magnitude of these alignments is found to decrease in the presence of droplets
and with heat release or an increase in swirl. Analysis shows that both tubelike
and sheetlike structures can coexist in the combustor and that their relative abun-
dance (or lack thereof) is a function of spatial location in the combustor and swirl
number. For example, tubelike structures are more likely in regions with intense
vorticity gradients, whereas regions of increased scalar gradients form sheetlike
structures that in turn wrap around the tubular vortical structures. The scalar gra-
dients in these sheets are amplified by the interaction between the strain rate and
the vorticity fields, thus increasing mixing and reaction in these regions.
As a final example of spray combustion, Fig. 11.13 shows a sector rig consisting
of two cups, each with multiple annular swirlers that surround the primary fuel
injector.52 This combustor is hereafter identified as GE-1.53 Although the two cups
are identical, the sector shape is not. Therefore, the regions downstream of the two
cups are not the same, and significant three-dimensional interaction between the
two cups occurs. Fuel (kerosene) is injected from the primary injector in the center
of each cup and from 20 injectors placed equally far apart on the rim surrounding
the primary cups. The total fuel flow rate is split equally between the two cups.
Spray is injected by using parcels to represent groups of droplets with similar
properties. Typically, an average of approximately 350,000 droplet parcels (with
10–15 droplets per group) are present when the simulation reaches the stationary
state. We consider full-power operation in which the combustor operates at around

24 atm with preheated inlet air at around 900 K. The bulk Re for these conditions
is around 2.2 × 106 . A grid of around 6.9 million grid points is used for this
simulation, with clustering to resolve the many regions of high shear.
To simulate scalar transport and combustion, a subgrid mixing model based on
the linear-eddy mixing (LEM) model is employed for these simulations.53,66−70 In
this approach, the scalar field is simulated within each LES by using the LEM model
in a one-dimensional grid embedded inside each LES cell. This one-dimensional
grid is aligned along the scalar gradient, and the reaction-diffusion processes
are simulated on this grid in a locally exact manner. A three-step, eight-species
kerosene-air mechanism (which includes CO and NO) is used for the chemical ki-
netics. Droplet dispersion and vaporization are included in the subgrid approach.
Details are given in the cited references and are therefore avoided here.
Figure 11.14a shows the mean product CO2 mass fraction contours in the center
x–y plane, along with a typical instantaneous spray pattern. The spray is injected
from both the primary cup and from the outer swirlers, but only the outer swirler
spray is visible here. Most of the droplets quickly vaporize. The combustion regions
from the two cups merge relatively quickly, since the fuel is injected from multiple
locations; and fuel–air mixing therefore occurs more uniformly.
Closer observation shows that the flame structure is actually partially lifted. This
outcome can be seen in Fig. 11.14b, which shows the mean kerosene reaction-rate
contours extracted from the simulation data. Analysis of the results indicates that
very close to the primary injectors, the local strain rate is very high, and mixing
between the vaporized fuel and air is not fully complete. However, farther down-
stream, the strain effects drop off rather quickly, and ignition of the mixed fuel–air
mixture can take place. It is observed that the flame is consistently lifted away
from the dump plane only for the lower cup. The shape of the combustor, which
pushes the flow upward from the lower cup region, contributes partly to this effect.
The reaction-rate contours in this plane are highly wrinkled. Closer examination
shows that both thin and distributed regions of heat release occur in the combustor.
The reaction rate structure is not continuous, since some local regions show very
low reaction rates. Comparing the reaction rate distribution with the CO2 contours
shows that the region inside the reaction-rate contours consists of both completely
burned and partially burned regions coexisting side by side. Full three-dimensional
visualization of these fields shows significant three-dimensional variation in the
flame structure and in the burned regions.
Figures 11.15a and 11.15b show the mean axial velocity contours and the stream-
line pattern, respectively. It can be seen that the vortex breakdown bubbles for the
two cups are very different, primarily because of the combustor shape. The VBB is
much more pronounced and well established for the upper cup. The VBB extends
into the primary inlets in both cups and exhibits a complex three-dimensional
unsteady shape (not shown). In the regions with local rearward-facing step and
centerbody type of geometry at the dump plane, smaller local regions of BRB are
seen. Comparison with the reaction-rate figure shows that the spray flame in this
combustor is anchored at multiple locations by both the BRB and the VBB in this
configuration.
Figure 11.15b shows the streamline pattern and temperature isocontours (1300
K) in the center x–y plane. The streamlines from both the cups quickly interact, and
306 S. MENON
a)
b)
Fig. 11.14 Spray and reaction rate contours in the center x–y plane: a) spray and
CO2 mass fraction and b) reaction rate.
a)
b)
Fig. 11.15 Axial velocity contours and streamline pattern in the center x–y plane:
a) axial velocity contours and b) streamlines and temperature contours.
308 S. MENON
Fig. 11.16 Instantaneous z−component of the baroclinic torque in the center x–y
plane.
the burned regions are trapped between the swirling outer streamlines. Analysis
shows that the classical PVC from each cup persists only for a short region around
the VBB. Farther downstream, the structures from both cups break down and
merge into a single stream with very little local coherence. This result is probably
more realistic (and part of the design strategy), since the outflow from this sector
is actually more uniform than in a single injector case.
To understand how vortex–flame interactions occur in this device and how this
process affects the acoustic fluctuations, additional postprocessing of the results is
required. Analysis of the time-evolving flow data can also be used to understand
how AVF interactions occur. For example, Fig. 11.16 shows a center-plane view
of the instantaneous z component of the previously discussed baroclinic torque
term. Both positive and negative values of this quantity are present in the flow. As
noted, positive values indicate enhancement of vorticity, whereas negative values
indicate suppression. Overall, the pattern closely follows the flame front, but this
quantity is nonzero in the regions where the droplets are vaporizing and mixing as
well.
Figures 11.17a and 17b show the mean pressure and the unsteady dilatation,
respectively, at the center plane. Although very little three-dimensional variation
seems to occur in the mean pressure in this plane, local variations still occur,
especially near the dump plane. The mean dilatation field (not shown) also shows
a similar behavior, with three-dimensional variations near the dump plane that
quickly smooth out as the exit plane is approached.
On the other hand, the unsteady dilatation shows significant three-dimensional
structure and a periodic wavelike structure propagating from the inlet. The unsteady
a)
b)
Fig. 11.17 Mean pressure and unsteady dilatation fields in the center x–y plane:
a) mean pressure and b) unsteady dilatation.
310 S. MENON
pressure troughs (not shown) are closely related to the unsteady dilatation crests.
Time-series analysis indicates that these fields exhibit axially moving, hemispher-
ical wavelike structures that grow from the two inlets. However, halfway into the
combustor, the disturbances from the two cups start to interfere and loose their
coherence. The radial crossflow caused by the geometrical convergence introduces
transverse disturbances that interact with the axially moving waves, so that further
downstream, both unsteady pressure and dilatation fields exhibit disturbances that
are more axially oriented.
Analysis of these fields shows that for the test conditions employed here, the
unsteady dilatation and the unsteady pressure fields are only weakly correlated with
the vortex–flame structures in the entire combustor. As a result, no enhancement
of the acoustic fluctuations occurs and the combustion process is stable. Since the
test conditions were chosen for a stable condition, this observation is encouraging.
V. Conclusion
This chapter summarizes some observations of AVF interaction in gas-turbine
engines. Since swirl is a key feature in all operational gas-turbine combustors, all
observed AVF phenomena in these combustors are somehow affected by the type
and intensity of swirl in the inlet. In addition, the fuel-injection method, fuel–air
mixing (especially in liquid-fueled systems), flame stabilization, and combustor
geometry (multi-injector) are also important parameters in gas-turbine systems.
Understanding the sensitivity of the combustor’s performance to these parameters
requires a comprehensive and integrated experimental and simulation strategy that
is only now becoming a reality.
Acknowledgments
The results reported in this chapter have been obtained with support from Army
Research Office, General Electric Power Systems, and General Electric Aircraft
Engine Company. The simulations reported here were carried out by C. Stone, V.
Sankaran, and N. Patel.
References
1
Mongia, H. C., Held, T. J., Hsiao, G. C., and Pandalai, R. P., “Challenges and Progress
in Controlling Dynamics in Gas Turbine Combustors,” Proceedings of the Combustion
2
Correa, S. M., “Power Generation and Aeropropulsion Gas Turbines: From Combustion
Science to Combustion Technology,” Proceedings of the Combustion Institute, Vol. 27, The
Combustion Inst., Pittsburgh, PA, 1998, pp. 1793–1807.
3
Chu, B. T., and Kovasznay, L. S. G., “Non-Linear Interactions in a Viscous Heat-
Conducting Compressible Gas,” Journal of Fluid Mechanics, Vol. 3, 1958, pp. 494–
514.
4
Propellant Rocket Engines,” Liquid Rocket Engine Combustion Instability, edited by V. Yang
and W. E. Anderson, Progress in Astronautics and Aeronautics, AIAA, Washington, DC
1995, pp. 3–37.
5
Vuillot, F., “Vortex-Shedding Phenomena in Solid Rocket Motors,” Journal of Propul-
sion and Power, Vol. 11, No. 4, 1995, pp. 626–639.
6
Apte, S., and Yang, V., “Unsteady Flow Evolution and Combustion Dynamics of
Homogeneous Solid Propellant in a Rocket Motor,” Combustion and Flame, Vol. 131,
No. 1–2, 2002, pp. 110–146.
7
Schadow, K. C., and Gutmark, E., “Combustion Instability Related to Vortex Shedding in
Dump Combustors and their Passive Control,” Progress in Energy and Combustion Science,
Vol. 18, No. 2, 1992, pp. 117–132.
8
Menon, S., and Jou, W.-H., “Large-Eddy Simulations of Combustion Instability in an
Axisymetric Ramjet Combustor,” Combustion Science and Technology, Vol. 75, No. 1,
1991, pp. 53–72.
9
Yu, K. H., Trouvé, A., and Daily, J. W., “Low-Frequency Pressure Oscillations in a
Model Ramjet Combustor,” Journal of Fluid Mechanics, Vol. 232, No. 11, 1991, pp. 47–72.
10
Poinsot, T., Trouve, A. C., Veynante, D., Candel, S. M., and Esposito, E. J., “Vortex-
Driven Acoustically Coupled Combustion Instabilities,” Journal of Fluid Mechanics,
Vol. 177, No. 4, 1987, pp. 265–292.
11
Jou, W.-H., and Menon, S., “Modes of Oscillation in a Nonreacting Ramjet Combustor
Flow,” Journal of Propulsion and Power, Vol. 6, No. 5, 1990, pp. 535–543.
12
Takahashi, F., Schmoll, W. J., Trump, D. D., and Goss, L. P., “Vortex-Flame Interactions
and Extinction in Turbulent Jet Diffusion Flames,” Proceedings of the Combustion Institute,
13
Santoro, V. S., Kyritsis, D. C., and Gomez, A., “An Experimental Study of Vortex-
Flame Interactions in Counterflow Spray Diffusion Flames,” Proceedings of the Combustion
14
Renard, P.-H., Thevenin, D., Rolon, J. C., and Candel, S., “Dynamics of Flame/Vortex
Interactions,” Progress in Energy and Combustion Science, Vol. 26, No. 3, 2000, pp. 225–
282.
15
Dowling, A. P., “Vortex, Sound and Flames—A Damaging Combination,” Aeronautical
Journal, Vol. 104, No. 1033, 2000, pp. 105–116.
16
Paschereit, O. C., Gutmark, E., and Weisenstein, W., “Excitation of Thermoacoustic
Instabilities by Interaction of Acoustics and Unstable Swirling Flow,” AIAA Journal, Vol. 38,
No. 6, 2000, pp. 1025–1034.
17
ing Combustion Instabilities in Low NO Gas Turbines,” Proceedings of the Combustion
18
Lee, T. W., and Santavicca, D. A., “Experimental Diagnostics for the Study of Combus-
tion Instabilities in Lean Premixed Combustors,” Journal of Propulsion and Power, Vol. 19,
No. 5, 2003, pp. 735–750.
19
Cohen, J. M., Wake, B. E., and Choi, D., “Investigation of Instabilities in a Lean
Premixed Step Combustor,” Journal of Propulsion and Power, Vol. 19, No. 11, 2003,
pp. 81–88.
20
Kim, W.-W., Menon, S., and Mongia, H. C., “Large-Eddy Simulation of a Gas Turbine
Combustor Flow,” Combustion Science and Technology, Vol. 143, No. 1, 1999, pp. 25–62.
21
Kim, W.-W., and Menon, S., “Numerical Simulations of Turbulent Premixed Flames in
the Thin-Reaction-Zones Regime,” Combustion Science and Technology, Vol. 160, No. 1,
2000, pp. 119–150.
22
Sankaran, V., and Menon, S., “LES of Spray Combustion in Swirling Flows,” Journal
of Turbulence, Vol. 3, No. 11, 2002.
312 S. MENON
23
Stone, C., and Menon, S., “Open Loop Control of Combustion Instabilities in a Model
Gas Turbine Combustor,” Journal of Turbulence, Vol. 4, No. 1, 2003.
24
Huang, Y., Sung, H.-G., Hsieh, S.-Y., and Yang, V., “Large Eddy Simulation of Com-
bustion Dynamics of Lean Premixed Swirl-Stabilized Combustors,” Journal of Propulsion
25
Sommerer, Y., Galley, D., Poinsot, T., Ducruix, S., Lacas, F., and Veynante, D., “Large-
Eddy Simulation and Experimental Study on Flashback and Blow-Off in a Lean Partially
Premixed Swirled Burner,” Journal of Turbulence, Vol. 5, No. 1, 2004.
26
Rockwell, D., and Naudascher, E., “Self-Sustained Oscillations of Impinging Shear
Layers,” Annual Review of Fluid Mechanics, Vol. 11, 1979, pp. 67–94.
27
Stow, S. R., Dowling, A. P., and Hynes, T. P., “Reflection of Circumferential Modes in
a Choked Nozzle,” Journal of Fluid Mechanics, Vol. 467, 2002, pp. 215–239.
28
Peters, N., Turbulent Combustion, Cambridge Monographs on Mechanics, Cambridge
Univ. Press, U.K., 2000.
29
Sankaran, V., and Menon, S., “Structure of Premixed Turbulent Flames in the Thin-
Reaction-Zones Regime,” Proceedings of the Combustion Institute, Vol. 28, The Combus-
tion Inst., Pittsburgh, PA, 2000, pp. 203–209.
30
Sankaran, V., and Menon, S., “Subgrid Combustion Modeling of 3-D Premixed Flames
in the Thin-Reaction-Zone Regime,” Proceedings of the Combustion Institute, Vol. 30, The
31
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, R. T. Edwards,
Philadelphia, 2001.
32
Lieuwen, T., “Modeling Premixed Combustion-Acoustic Wave Interactions: A Re-
33
Clanet, C., Searby, G., and Clavin, P., “Primary Acoustic Instability of Flames Propa-
gation in Tubes: Cases of Spray and Premixed Combustion,” Journal of Fluid Mechanics,
Vol. 385, No. 4, 1999, pp. 157–197.
34
Ducruix, S., Durox, D., and Candel, S., “Theoretical and Experimental Determination
Institute, Vol. 28,The Combustion Inst., Pittsburgh, PA, 2000, pp. 765–772.
35
Borghi, R., “Turbulent Combustion Modeling,” Progress in Energy and Combustion
Science, Vol. 14, No. 4, 1988, pp. 245–292.
36
Poinsot, T., Veynante, D., and Candel, S., “Quenching Processes and Premixed Turbu-
lent Combustion Diagrams,” Journal of Fluid Mechanics, Vol. 228, No. 8, 1991, pp. 561–
606.
37
Dunlap, R., and Brown, R. S., “Exploratory Experiments on Acoustic Oscillations
Driven by Periodic Vortex Shedding,” AIAA Journal, Vol. 19, No. 3, 1981, pp. 408–409.
38
McMurtry, P. A., Riley, J. J., and Metcalfe, R. W., “Effects of Heat Release on the
Large-Scale Structure in Turbulent Mixing Layers,” Journal of Fluid Mechanics, Vol. 199,
No. 2, 1989, pp. 297–332.
39
Williams, F. A., Combustion Theory, 2nd ed., Benjamin/Cummings Publishing Com-
pany, Metro Park, CA 1985.
40
Candel, S. M., “Combustion Dynamics and Control: Progress and Challenges,” Pro-
ceedings of the Combustion Institute, Vol. 29, The Combustion Inst., Pittsburgh, PA, 2002,
pp. 1–28.
41
Ducruix, S., Schuller, T., Durox, D., and Candel, S., “Combustion Dynamics and Insta-
bilities: Elementary Coupling and Driving Mechanisms,” Journal of Propulsion and Power,
Vol. 19, No. 5, 2003, pp. 722–734.
42
Sinibaldi, J. O., Mueller, C. J., and Driscoll, J. F., “Local Flame Propagation Speeds
Along Wrinkled Unsteady Stretched Premixed Flames,” Proceedings of the Combustion
43
Hussain, A. K. M. F., “Coherent Structures—Reality and Myth,” Physics of Fluids,
Vol. 26, No. 10, 1983, pp. 2816–2850.
44
Metcalfe, R. W., Hussain, A. K. M. F., Menon, S., and Hayakawa, M., “Coherent Struc-
tures in a Turbulent Mixing Layer: A Comparison Between Direct Numerical Simulations
and Experiments,” Turbulent Shear Flows, Vol. 5, 1987, pp. 110–123.
45
Rivero, A., Ferre, J. A., and Giralt, F., “Organized Motions in a Jet in Crossflow,”
Journal of Fluid Mechanics, Vol. 444, No. 10, 2001, pp. 117–149.
46
Lilley, D. G., “Swirl Flows in Combustion: A Review,” AIAA Journal, Vol. 15, No. 8,
1977, pp. 1063–1078.
47
Michalke, A., “On the Inviscid Instability of the Hyperbolic Tangent Profile,” Journal
of Fluid Mechanics, Vol. 19, No. 4, 1964, pp. 543–556.
48
Mechanics, Vol. 16, 1984, pp. 365–424.
49
Gutmark, E., and Ho, C.-M., Preferred Modes and the Spreading Rate of Jets,” Physics
of Fluids, Vol. 26, No. 10, 1983, pp. 2932–2938.
50
Stone, C., and Menon, S., “Adaptive Swirl Control of Combustion Instability in Gas
Turbine Combustors,” Proceedings of the Combustion Institute, Vol. 29, The Combustion
51
Jeong, J., and Hussian, F., “On the Identification of a Vortex,” Journal of Fluid
Mechanics, Vol. 285, 1995, pp. 69–94.
52
Mongia, H. C., “TAPS - A 4th Generation Propulsion Combustor Technology for Low
Emissions,” AIAA Paper 03-2657, Jan. 2003.
53
Menon, S., “CO Emission and Combustion Dynamics Near Lean Blow-Out in Gas
Turbine Engines,” ASME GT2004-53290, June 2004.
54
Panda, J., and McLaughlin, D. K., “Experiments on the Instabilities of a Swirling Jet,”
Physics of Fluids, Vol. 6, No. 1, 1994, pp. 263–276.
55
Lucca-Negro, O. and O’doherty, T., “Vortex Breakdown: A Review,” Progress in Energy
and Combustion Science, Vol. 27, No. 4, 2001, pp. 431–481.
56
Chen, R.-H., and Driscoll, J. F., “The Role of the Recirculation Vortex in Improving
Fuel-Air Mixing within Swirling Flames,” Proceedings of the Combustion Institute, Vol. 22,
57
Billant, P., Chomaz, J.-M., and Huerre, P., “Experimental Study of Vortex Break-
down in Swirling Jets,” Journal of Fluid Mechanics, Vol. 376, No. 12, 1998, pp. 183–
219.
58
Ruith, M. R., Chen, P., Meiburg, E., and Maxworthy, T., “Three-Dimensional Vortex
Breakdown in Swirling Jets and Wakes: Direct Numerical Simulation,” Journal of Fluid
Mechanics, Vol. 486, No. 8, 2003, pp. 331–378.
59
Eggenspieler, G. and Menon, S., “LES of Premixed Combustion and Pollutant Emis-
sion in a DOE-HAT Combustor,” Journal of Propulsion and Power, Vol. 20, No. 6, 2004,
pp. 1076–1086.
60
Syred, N., and Beer, J. M., “Combustion in Swirling Flows: A Review,” Combustion
and Flame, Vol. 23, No. 2, 1974, pp. 143–201.
61
Ling, W., Troutt, J. N., and Crowe, C. T., “Direct Numerical Simulation of a Three-
Dimensional Temporal Mixing Layer with Particle Dispersion,” Journal of Fluid Mechanics,
Vol. 358, No. 3, 1998, pp. 61–85.
314 S. MENON
62
Lazaro, B. J., and Lasheras, J. C., “Particle Dispersion in the Developing Free Shear
Layer, Part 1: Unforced Flow Turbulent Channel Flow,” Journal of Fluid Mechanics,
Vol. 235, 1992, pp. 143–178.
63
Eaton, J. K., and Fessler, J. R., “Preferential Concentration of Particles by Turbulence,”
International Journal of Multiphase Flow, Vol. 20, No. 1, 1994, pp. 169–209.
64
Sankaran, V., and Menon, S., “Vorticity-Scalar Alignments and Small-Scale Structures
in Swirling Spray Combustion,” Proceedings of the Combustion Institute, Vol. 29, The
Combustion Inst., Pittsburg, PA, 2002, pp. 577–584.
65
Smith, T. M., and Menon, S., “The Structure of Premixed Flames in a Spatially Evolving
Turbulent Flow,” Combustion Science and Technology, Vol. 119, No. 1–6, 1996, pp. 77–106.
66
Menon, S., Stone, C., and Patel, N., “Multi-Scale Modeling for LES of Engineering
Designs of Large-Scale Combustors,” AIAA Paper 2004-0157, Jan. 2004.
67
Menon, S., McMurtry, P., and Kerstein, A. R., “A Linear Eddy Mixing Model for Large
Eddy Simulation of Turbulent Combustion,” LES of Complex Engineering and Geophysical
Flows, edited by B. Galperin and S. Orszag, Cambridge Univ. Press, Cambridge, U.K., 1993.
68
Menon, S., and Calhoon, W., “Subgrid Mixing and Molecular Transport Modeling
for Large-Eddy Simulations of Turbulent Reacting Flows,” Proceedings of the Combustion
69
Menon, S., and Pannala, S., “Subgrid Combustion Simulations of Reacting Two-Phase
Shear Layers,” AIAA Paper No. 98-3318, July 1998.
70
Chakravarthy, V., and Menon, S., “Large-Eddy Simulations of Turbulent Premixed
Flames in the Flamelet Regime,” Combustion Science and Technology, Vol. 162, No. 1,
2001, pp. 1–48.
Chapter 12
Physics of Premixed Combustion-Acoustic

Wave Interactions
Timothy C. Lieuwen∗
Nomenclature
A FL = flame-surface area
A LC = limit-cycle amplitude
c = speed of sound
d = flame thickness
E = energy
E a = overall activation energy
f = frequency
F = flame-transfer function
G = flame-area-transfer function
h = enthalpy
h R = heat of reaction per unit mass of reactant
Ia = net acoustic energy flux out of the flame
k = wave number
K a = Karlovitz number
Le = Lewis number
L F = flame length
ṁ = mass flow rate
M = Mach number
M = total mass
Ma = Markstein number
N = dimensionless length scale defined in Eq. (12.34)
p = pressure
Q = heat-release rate
r = radial coordinate
Copyright c 2005 by the author. Published by the American Institute of Aeronautics and Astro-
∗ Associate Professor, School of Aerospace Engineering.
315
316 T. C. LIEUWEN
R = jet or flame radius

R = reflection coefficient
s = flame coordinate along the nominal flame surface
Sc = stretched flame speed
S1 , SL = laminar flame speed
St = Strouhal number (= fL f /u o )
St2 = modified Strouhal number defined in Eq. (12.23)
Stc = convective Strouhal number (= ωL f /u c )
t = time
T = temperature
T = transmission coefficient
Tb = burned-gas temperature
u = velocity
uc = phase speed
uo = mean flow velocity
V = volume
Ẇk = consumption rate of the kth species
x = axial coordinate
Y = mass fraction
Greek
β = ratio of the flame length to radius, β = L f /R
ε = normalized amplitude of velocity disturbance, ε = u/u o
εf = disturbance amplitude for flashback defined in Eq. (12.33)
φ = equivalence ratio oscillation
γ = specific heats ratio
κ = stretch rate
η = nondimensional disturbance convection velocity defined in Eq. (12.24)
λ = wavelength
= mean temperature ratio across the flame

ν = normalized mass burning rate response to acoustic pressure
perturbations
θ = momentum thickness
θE = dimensionless overall activation energy
= incident angle of acoustic waves
ρ = density
σ = flame-brush thickness
τ = retarded time
τM = dimensionless timescale defined in Eq. (12.34)
ω = angular frequency
ω̇ = volumetric reactant consumption rate
ξ = axial flame position
Superscripts
( ) = fluctuating quantities
(¯ ) = mean quantities
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 317
Subscripts
1 = upstream side of the flame
2 = downstream side of the flame
a = acoustic disturbances
b = flame base, burned gas
c = conical flames
F = flame, fuel
in = inlet value
v = vortical disturbances
s = entropy disturbances
o = mean quantities
ox = oxidizer
w = wedge flames
I. Introduction
T HIS chapter provides an overview of the physics of acoustic interactions

with a premixed combustion process. Such interactions play important roles
in the characteristic unsteadiness of turbulent combustion systems found in most
processing, power-generating, and propulsion applications. The basic problem
of interest is depicted in Fig. 12.1. A premixed combustion process, stabilized
by, for example, a pilot or bluff body, with a characteristic dimension L is per-
turbed by an acoustic or fluid mechanic disturbance with frequency f and phase
speed u c .
Several key questions this chapter addresses are as follows: 1) How does a
flame respond to an acoustic or vortical perturbation; in particular, what is the
subsequent heat-release fluctuation? 2) How does this response scale with the
flame dimension L F , frequency f , disturbance-phase speed u c , method of flame
stabilization, or amplitude of perturbation ε? 3) What are the differences between
acoustic interactions with laminar and turbulent flames? 4) What are the effects of
acoustic disturbances on inherent flame instabilities? 5) What are the relative roles
of chemical kinetic and large-scale adjustments of flame location on its overall
heat release?
This chapter focuses on these interactions without consideration of the larger
system in which they occur. As such, many other important issues are not ad-
dressed here. These issues include 1) acoustic characteristics of the overall com-
bustion system, 2) mechanisms through which the flame–acoustic interactions
couple with the overall system to become self-exciting, 3) interactions of acoustic
waves with solid fuels,1 liquid sprays,2 and nonpremixed gaseous and liquid-fueled
flames.3
This chapter is organized in the following manner. The background section
briefly describes the different regimes of premixed combustion (Sec. II.A), the
characteristics of the flow and thermodynamic perturbations that disturb the flame
(Sec. II.B), and the mechanism by which these disturbances are generated at the
flame (Sec. II.C). Sections III.A and III.B then focus on the effects of flow distur-
bances on premixed flames within the flamelet and well-stirred reactor regimes,
respectively.
318 T. C. LIEUWEN
Acoustic disturbances
LF
Fluid mechanic
disturbances
Fig. 12.1 Interaction of flow disturbances with a turbulent, premixed flame.
II. Background
A. Combustion Regimes
Acoustic wave–flame interactions involve unsteady kinetic, fluid mechanic, and
acoustic processes over a large range of timescales. Fundamentally different phys-
ical processes may dominate in different regions of the relevant parameter space,
depending on the relative magnitudes of various temporal–spatial scales. The
different regimes of interaction between acoustic waves, the combustion process,
and broadband turbulent fluctuations can be readily visualized with the combustion
diagram in Fig. 12.2.4 The regions denoted by wrinkled and corrugated flamelets
correspond to situations in which the reactions occur in thin sheets that retain their
laminar structure. These sheets become increasingly wrinkled and multiconnected
with increasing values of u /SL , where u is the fluctuating velocity and SL is the
laminar flame speed. Acoustic or vortical flow disturbances push these reaction
sheets around, causing additional wrinkling of the flame over well-defined spatial
and temporal scales. In addition, the local propagation velocity of these sheets into
the reactants is modulated by local strain rate, pressure, and temperature fluctua-
tions in the wave.
In the distributed reaction zone, the smallest turbulent length scales are of the
same order as the flame thickness and thus alter the laminar flame structure. The
well-stirred reactor regime corresponds to the limit in which mixing occurs much
more rapidly than chemical kinetics, and reaction occurs homogeneously over a
distributed volume. Some debate exists about the characteristics of the combustion
process in the regions noted by well-stirred reactor and distributed reaction zone.5
100 Re =10
4
RMS velocity/ flame speed (u'/SL)
Da=1
Well Stirred Reactor
10 Distributed
Re =102 Ka=1
Reaction Zone
Corrugated
Flamelets
1
Wrinkled
Re =1 Flamelets
0.1
1 10 100 1000
Integral length scale/flame thickness ( /d)
Fig. 12.2 Turbulent-combustion diagram.
Acoustic disturbances modulate the local thermodynamic quantities in the well-

stirred reactor regime and the reactants’ residence time.
Consider the ratios of the spatial and temporal scales involved in acoustic–
flame interactions. Note first the following length scales: the thickness of a laminar
methane–air flame at standard conditions varies between ∼0.1 and 1 cm.6 On the
other hand, the acoustic wavelength of a 100-, 1000-, and 10,000-Hz sound wave
at standard conditions is 3.3 m, 33 cm, and 3.3 cm. At higher temperatures, these
wavelengths are even larger. Given this disparity between flame and acoustic length
scales, the flame front essentially appears as a discontinuity to the acoustic wave.
As such, the fluid dynamics of the flows up- and downstream of the flame can
often be treated separately from the dynamics of the flame structure. The situation
is quite different with respect to the relevant timescales. Forming a flame-response
timescale τ M from the ratio of the laminar flame thickness and flame speed leads to
values of between τ M ∼0.002 and 0.07 s for methane–air flames. These values are
of magnitudes similar to acoustic perturbations with frequencies between 20 and
500 Hz. Thus, the interior flame structure and, consequently, quantities such as the
flame speed do not respond in a quasi-steady manner to acoustic perturbations.
This issue is addressed further in Sec. III.A.4.
B. Disturbance Field Features

This section describes the characteristics of the flow and thermodynamic os-
cillations (e.g., p , T , ρ , u , etc.). This is important because, as will be shown
subsequently, the response of a flame to an acoustic or vorticity velocity distur-
bance of a given magnitude is quite different.
It is useful to decompose an arbitrary disturbance field into three canonical
types of disturbances7−9 : vortical, entropy, and acoustic. In other words, each fluc-
tuating quantity can be decomposed as: p = pa + pv + ps , ρ = ρa + ρv + ρs ,
320 T. C. LIEUWEN
and u = u a + u v + u s , where the subscripts a, v, and s denote acoustic, vortical,

and entropy disturbances, respectively. Several characteristics of these disturbance
modes should be noted.
First, acoustic disturbances propagate with a characteristic velocity equal to the
speed of sound. In a uniform flow, vorticity and entropy disturbances are convected
at the bulk flow velocity u o . Consequently, in low-Mach-number flows, these dis-
turbances have substantially different length scales. Acoustic properties vary over
an acoustic length scale, given by λa = c/ f , whereas entropy and vorticity modes
vary over a convective length scale, given by λc = u o / f . Thus, the entropy and
vortical mode wavelength is shorter than the acoustic wavelength by a factor equal
to the mean-flow-Mach number λc /λa = u o /c = M. This can have important im-
plications on acoustic–flame interactions. For example, a flame whose length L f
is short relative to an acoustic wavelength, that is, L f λa , may be of the same
order of, or longer than, a convective wavelength. Thus, a convected disturbance,
such as an equivalence-ratio oscillation, may have substantial spatial variation
along the flame front that results in heat-release disturbances generated at differ-
ent points of the flame that are out of phase with each other. Studies often find
that a Strouhal number, defined as St = ωL f /u o , is a key parameter that affects
the flame response to perturbations. Note that St is proportional to the ratio of the
flame length and convective wavelength, St = 2π L f /λc . A flame whose length is
much less than an acoustic or convective wavelength is referred to as acoustically
or convectively compact.
Second, entropy and vorticity disturbances propagate with the mean flow and
diffuse from regions of high to low concentration. In contrast, acoustic distur-
bances, being true waves, reflect off boundaries, are refracted at property changes,
and diffract around obstacles. In general, the reflection of acoustic waves from
multidimensional flame fronts results in a complex, multidimensional acoustic
field in the vicinity of the flame. We make this point because analytical studies
often assume that the acoustic field is one-dimensional. This is not the case in
reality, although it may be a reasonable approximation under certain conditions
or in specific regions of the flame. A planar-incident wave impinging on a flame
front not only generates planar (i.e., one-dimensional) reflected and transmitted
waves, but also multidimensional disturbances that are often evanescent, or spa-
tially decaying. To illustrate, Fig. 12.3 reproduces a figure from Ref. 10 showing
computed velocity vectors in the vicinity of an axisymmetric conical flame per-
turbed by a plane sound wave at a frequency below the duct-cutoff mode. Note
that the acoustic field has strong two-dimensional characteristics near the base of
the flame but it reverts to a one-dimensional structure up- and downstream of the
flame. Similar observations have been made from experiments.11
Third, in a homogeneous, uniform flow, these three disturbance modes propagate
independently in the linear approximation. Finite amplitude disturbances do inter-
act, however; for example, the interaction of two vortical disturbances generates
an acoustic disturbance.7 Coupling between small-amplitude perturbations occurs
at boundaries (e.g., through the no-slip condition) or in regions of inhomogeneity.
Experiments have highlighted the significance of vortical mode interactions with
flame fronts.12,13 Figure 12.4 shows a simulated result of such an interaction,14
where the flame is disturbed by vortex structures that are periodically shed off
the rapid expansion. In general, these vorticity oscillations are manifested as
large-scale, coherent structures that arise from the growth of intrinsic flow
Fig. 12.3 Instantaneous pressure contours (solid lines) and velocity vectors in a dump-
stabilized combustor geometry for which the flame was excited from upstream. Aver-
age flame location given by heavy inclined line. Adapted from Lee and Lieuwen.10
instabilities. The phase velocity and growth rate of the flow instabilities is strongly
affected by the amplitude of forcing and the relationship between the acoustic-
forcing frequency and the intrinsic flow instability. Acoustic excitation often causes
their shedding rate to lock in to the forcing frequency or one of its harmonics.
When the forcing frequency is much lower than the natural shedding frequency,
a collective-interaction phenomenon occurs in which the flow instabilities form
at their natural rate but subsequently coalesce to form a vortex whose forma-
tion frequency coincides with the forcing frequency.15 For example, if the forcing
Flame
Vorticity
iso-countours
Fig. 12.4 Computation of flame disturbed by vortical structure.14 Image courtesy of

S. Menon and C. Stone.
322 T. C. LIEUWEN
Fig. 12.5 Dependence of shear-wave convection velocity and growth rate in a jet flow
on Strouhal number and ratio of boundary-layer thickness to jet radius.
frequency is 10 times lower than that of the intrinsic instability, 10 vortices would
discretely form but subsequently merge into a single larger vortex.
The characteristics of the instability waves that grow and merge to form these
large-scale structures are a function of the specific characteristics of the burner-
exit shear layer, such as coflow velocity, and specifically of the receptivity of this
shear layer to external disturbances. For example, the phase speed of the convected
vortical instability waves is not necessarily equal to the flow velocity but varies
with frequency and shear-layer characteristics. The instability wave-growth rate
similarly varies with frequency and the shear-layer characteristics.
To illustrate, Fig. 12.5 plots Michalke’s16 theoretical curves of the dependence of
the phase speed u c of shear-layer instability waves in a jet flow on Strouhal number,
Sθ = f θ/u o , for several values of the momentum thickness θ, jet radius R, and
ratio R/θ . The figure shows that, for all R/θ values, the ratio u c /u o equals unity
and 0.5 for low and high Strouhal numbers. For thin boundary layers, for example,
R/θ = 100, the phase velocity actually exceeds the maximum axial-flow velocity
in a certain Sθ range. The dispersive character of the instability-wave convection
velocity has been confirmed by a variety of measurements. For example, Baillot
et al.17 measured u c /u o values of 0.88 and 0.98 at 35 and 70 Hz, respectively, on
a conical Bunsen flame. Similarly, Durox et al.18 measured u c /u o = 0.5 values at
150 Hz in an axisymmetric wedge flame.
In general, the disturbance field may have both acoustic and vortical components,
whose relative magnitudes depend strongly on the vortex-shedding dynamics at
the burner-shear layer. For example, Richards and coworkers11 found that the
disturbance field transitioned from a convected character to one with an acoustic
character at low and high frequencies ( f > 100 Hz), respectively.
Even in the absence of convected vorticity waves, the impact of the fluctuating
flame position on the acoustic field causes the acoustic disturbance field to have
a convected character. The convected character is because the flame response to

the acoustic field and the acoustic field disturbing the flame are coupled. For
large-amplitude disturbances, the flame develops large corrugations (as seen in
Fig. 12.7) that convect with a phase speed proportional to the axial-flow velocity.
These convecting flame wrinkles have an impact on the character of the interior
acoustic field.
C. Sound, Vorticity, and Entropy Generation by Unsteady Heat Release

Oscillations in heat release generate acoustic, vorticity, and entropy perturba-
tions. Sound generation is manifested as the broadband combustion roar of turbu-
lent flames19,20 and, in the context of combustion instabilities, by discrete tones. In
terms of sound generation, a flame can be thought of as a distribution of monopoles
whose local source strength is proportional to the unsteady rate of heat release. The
fundamental mechanism for this sound generation is the unsteady gas expansion
as the mixture reacts.
Unsteady heat release also generates entropy and vorticity disturbances.
Whether these disturbances significantly affect the dynamics of the combustor
depends on the downstream configuration. If the combustor area remains rela-
tively constant so that the flow passes out of the system unrestricted, such as an
open-ended pipe, the disturbance convects out of the system and is dissipated in the
atmosphere. This behavior is in contrast to the acoustic disturbance that is usually
strongly reflected by such a boundary. However, if the flow is accelerated (e.g., by
passing through a nozzle), the entropy disturbance generates sound.21
Even in the absence of heat-release perturbations, the presence of steady heat-
release introduces important coupling between the acoustic, vortical, and entropy
modes. First, an acoustic oscillation incident on a flame generates entropy and
vorticity disturbances.22 The vorticity disturbance is generated through several
mechanisms. The baroclinic mechanism, which occurs if the wave is obliquely
incident on the flame, is probably the most significant. It is caused by the mis-
alignment of the mean density gradient and the fluctuating pressure gradient (i.e.,
∇ ρ̄ × ∇ p = 0 ). Also, the unsteady wrinkling and subsequent curvature of the
flame front induced by the acoustic perturbations causes additional generation of
unsteady vorticity.23
Entropy and vorticity disturbances impinging on a flame excite acoustic waves
if their phase speed along the flame front (not the flow speed) is supersonic.24 In
low-Mach-number flows, this can occur if the flame is nearly orthogonal to the
flow.
III. Heat-Release Response to Flow and Mixture Perturbations

For a given disturbance, what is the response of a flame and, in particular, what is
the resultant fluctuation in local and global heat-release rate? The objective of this
section is to address this question. To illustrate, consider the data shown in Fig. 12.6,
which plots the amplitude of the C H ∗ chemiluminescence oscillations (presum-
ably related to heat release) from a swirling, premixed flame as a function of the
disturbance velocity u . The figure shows that the relationship between C H ∗ and
velocity-oscillation amplitudes is linear at disturbance amplitudes u /u o < 0.20.
324 T. C. LIEUWEN
0.30
0.25
CH*'/CH*o 0.20
0.15
0.10
0.05
0.00
0.00 0.05 0.10 0.15 0.20 0.25 0.30
u'/uo
Fig. 12.6 Measured dependence of C H ∗ –velocity amplitude relationship on distur-

bance amplitude at φ = 0.90 ( ) and φ = 0.87( ).
The objective of a linear analysis is to determine the proportionality constant or

transfer function G = C H ∗ /u , relating these perturbations and its dependence
on geometry, frequency, and mixture properties.
At u /u o values > 0.20, the C H ∗ /u relationship is no longer independent of
amplitude. Modeling and understanding these large-amplitude dynamics requires
an understanding of the flame’s nonlinear dynamics.
The rest of this section describes current understanding of and analytical ap-
proaches for modeling these linear and nonlinear flame dynamics. Secs. III.A and
III.B focus on these dynamics in the flame sheet and well-stirred reactor regime,
respectively.
A. Flame Sheets
1. Basic Concepts and Analytical Framework
We begin this section with a derivation of the fundamental equations describing
the dynamics of flame sheets. This approach for treating unsteady flame problems
was apparently first introduced by Markstein24 and Marble and Candel,25 and
subsequently developed by many other authors.17,26−33 More detailed treatments
can be found in Refs. 24 and 34.
Consider a flame front of arbitrary shape whose instantaneous surface is de-
scribed by the parametric equation f (x , t) = 0. It is assumed that the surface is
continuous with a uniquely defined normal at each point. Markstein derives the
following kinematic equations, which relate the flame-surface position to the local
flow and flame-burning velocities.24
∂f
+ u 1 · ∇ f − S1 |∇ f | = 0 (12.1)
∂t
∂f
+ u 2 · ∇ f − S2 |∇ f | = 0 (12.2)
∂t
where S and u denote the flame speed relative to the gases and flow velocity,
respectively. Subscripts 1 and 2 denote the value of each quantity on the up- and
downstream side of the flame, respectively. Either of the preceding two expres-
sions are often referred to as the G-Equation in the flame-dynamics literature. The
flowfields up- and downstream of the flame are coupled across the front by the
relations24 :
Mass: ρ1 S1 = ρ2 S2 (12.3)
Normal momentum: p1 + ρ1 S12 = p2 + ρ2 S22 (12.4)
Tangential momentum: u 1 − u 2 ) × ∇ f = 0
( (12.5)

u 1 · u 1 u 2 · u 2
Energy: ρ1 S1 h 1 + = ρ2 S2 h 2 + (12.6)
2 2
where ρ and h denote density and enthalpy, respectively. The dynamics of the
thermodynamic and flow variables up- and downstream of the flame are described
by the mass, momentum, and energy-conservation equations.
In many cases, the quantity of primary interest is the overall heat release. (For
acoustically compact flames, only the spatially integrated heat release is impor-
tant. However, for high-frequency oscillations, where the flame is not acoustically
compact, its spatial distribution is also important.) The global heat-release rate of
the flame is given by

Q(t) = ρ1 S1 h R dAFL (12.7)
S
where the integral is performed over the flame surface AFL and h R is the heat
release per unit mass of reactant. Equation (12.7) shows the four fundamentally
different ways of generating heat-release disturbances in a premixed flame: fluctu-
ations in density, flame speed, heat of reaction, or flame area. As noted by Clanet
et al.,35 they can be classified based on either their modification of the local internal
structure of the flame (such as the local burning rate) or its global geometry (such
as its area).
Fluctuations in the mass flow rate of reactive mixtures into the flame, cor-
responding to ρ1 S1 in Eq. (12.7), is the most basic mechanism of heat-release
oscillation. These density fluctuations could be caused by both acoustic and en-
tropy fluctuations. The flame’s burning rate S1 is sensitive to the perturbations
in pressure, temperature, strain rate, or mixture composition that accompany the
acoustic wave. These pressure and temperature fluctuations are usually generated
by acoustic perturbations, whereas the strain-rate fluctuations are associated with
acoustic or vortical velocity fluctuations.
Flame-area fluctuations are associated with disturbances in the flame’s position
and orientation that, in turn, are generated by fluctuations in either the local burning
rate or flow velocity. To illustrate the disturbance of a flame by an acoustic velocity
disturbance, Fig. 12.7 shows a photograph from Ducruix et al.36 of a simple Bunsen
flame disturbed by acoustic-flow oscillations generated by a loudspeaker placed
upstream of the flame. The figure clearly shows the large distortion of the flame
front that is evidenced by the pronounced cusp in the center of the flame. This
flame disturbance is convected downstream by the mean flow, so that it varies
spatially over a convective wavelength.37
326 T. C. LIEUWEN
Fig. 12.7 Photograph of flame disturbances generated by acoustic velocity

oscillations.36 Photograph courtesy of S. Ducruix, D. Durox, and S. Candel, Centre
National de La Recherche Scientifique and Ecole Centrale de Paris.
Finally, fluctuations in heating value h R are driven by variations in reactive

mixture composition.
2. Effects of Flame on the Disturbance Field

Acoustic–flame interactions are highly coupled because the flame has a strong
impact on the values of the acoustic oscillations that are disturbing it. Although
this point was briefly noted in Sec. II.B, this section considers these effects in more
detail.
Consider a nominally flat, vertically oriented flame front in a low-Mach-number
flow whose instantaneous position is described by the equation x = ζ (y, t) (see
Fig. 12.8). It is disturbed by an acoustic-plane wave, incident at an angle ,
whose wavelength is much larger than the flame thickness. This problem was first
Instantaneous
flame position
Cold Reactants x=ζ(y,t)
Transmitted
Reflected Wave Wave
Θ
Hot Products
Incident Acoustic
Wave Convected vortical
and entropy disturbances
Fig. 12.8 Illustration of planar flame disturbed by incident acoustic wave.

reported by Manson,38 who calculated the reflection and transmission coefficients

of a planar flame modeled as a simple temperature discontinuity and generalized
by Chu,38 who also considered sound amplification by perturbations in S1 , h r ,
γ , and entropy of the incoming mixture. This problem is treated by decomposing
each term in Eqs. (12.1–12.6) into the sum of a mean and fluctuating term and
retaining only linear terms in fluctuations; for example,
ρ1 S1 = (ρ̄ 1 + ρ1 )( S̄ 1 + S1 ) ≈ ρ̄ 1 S̄ 1 + ρ̄ 1 S1 + ρ1 S̄ 1 (12.8)
As shown in Ref. 22, this leads to the following approximate expressions coupling
the axial velocity u and pressure p across the flame:

u 2 u 1 S1 γ − 1 p1
− = (
− 1)Ms − (12.9)
c̄1 c̄1 S̄ 1 γ p̄1

p2 = p1 (12.10)
where Ms , γ , and
refer to the laminar-flame-speed Mach number, Ms = S̄ 1 /c̄1 ,
specific heat ratio, and mean temperature ratio across the flame. This equation
neglects variations in γ and terms of O(Ms2 ). To this order, the unsteady pressure
is continuous across the flame. However, there is a jump in unsteady velocity
across the flame; that is, the flame looks like an acoustic volume source or a
monopole. The terms on the right side of Eq. (12.9) quantifying this jump are
related to the flame’s unsteady rate of heat release and result in the amplification
of acoustic waves. This jump is directly proportional to the temperature jump
(or, more fundamentally, the gas-expansion ratio) across the flame and the
flame-speed Mach number Ms which typically has quite low values (∼0.001 for
a stoichiometric methane–air flame). Assuming a typical acoustic scaling, that
is, p ∼ ρcu , it can be seen that the second fluctuating term on the right side
results in a velocity increment across the flame that is on the order of Ms and,
thus, quite small. The relative magnitudes of the S1 term on the right side of
Eq. (12.9) and the fluctuating velocity-perturbation quantities on the left depend
on the specific processes causing the flame-speed perturbation. More detailed
analyses in Sec. III.A.4 suggest that flame-speed perturbations caused by pressure
and/or temperature fluctuations are of similar magnitude S1 / S̄ 1 ∼ O( p / p̄). Thus,
acoustic-wave amplification induced by the pressure or temperature sensitivity of
the flame speed is nonzero, but it is of O(M S ) and therefore quite weak.
Assuming that the source terms on the right side of Eq. (12.9) are small, the
acoustic field can be accurately calculated by ignoring them. As such, the leading-
order calculation of the acoustic field is equivalent to replacing the flame front
with a passive-temperature discontinuity, as assumed by Mason.38 Ignoring these
terms does not allow one to calculate the slight amplification or damping of sound
waves at the flame, but it does allow for an accurate calculation of the acoustic
field that is disturbing the flame. We will proceed in a sequential fashion, first
considering the leading-order problem, then considering the higher-order effects
needed to calculate the acoustic amplification and damping by the flame.
Consider first the problem of an acoustic wave of pressure amplitude PI im-
pinging normally (i.e., = 0) on the flame. The acoustic pressure and velocity
328 T. C. LIEUWEN
are given by the following plane-wave equations:
p1 (x, t) = PI (eik1 x + Reik1 x ) p2 (x, t) = PI Teik1 x (12.11)

PI ik1 x PI T ik1 x
u 1 (x, t) = (e − Reik1 x ) u 2 (x, t) = e (12.12)
ρ̄1 c̄1 ρ̄1 c̄1
where R and T are the reflection and transmission coefficients, denoting the am-
plitude of the acoustic waves reflected from and transmitted through the flame.
Matching the pressure and axial velocity at the flame by using the leading-order
approximation of Eqs. (12.9) and (12.10), p2 = p1 and u 2 = u 1 , leads to the fol-
lowing solution for R and T:
(ρ̄2 c̄2 /ρ̄1 c̄1 ) − 1 2(ρ̄2 c̄2 /ρ̄1 c̄1 )

R= T= (12.13)
(ρ̄2 c̄2 /ρ̄1 c̄1 ) + 1 (ρ̄2 c̄2 /ρ̄1 c̄1 ) + 1
Neglectingvariations √ in molecular weight and γ across the flame, note that

ρ̄2 c̄2 /ρ̄1 c̄1 ≈ T̄1/T̄2 = 1/
. This result shows that the flame’s impact on the
acoustic field (as manifested in R and T) monotonically increases with the jump
in gas impedance across the flame, which is closely related to the square root of
the temperature increase. To illustrate, consider a flame where
= T̄2 /T̄1 = 4, so
that R = −1/3 (the negative sign means that the reflected wave is out of phase
with the incident wave). This implies that the acoustic pressure and velocity at
the flame are lower and higher, respectively, than that of the incident wave, that

is, pflame /PI = 2/3 and u flame /(PI /ρ̄1 c̄1 ) = 4/3. If the acoustic wave is incident
on the same flame from downstream, the pressure and velocity are higher and

lower, respectively, than those of the incident wave, that is, pflame /PI = 4/3 and

u flame /(PI /ρ̄2 c̄2 ) = 2/3. This implies that the presence of the flame alters the value
of the quantities that are disturbing it and causing its heat-release rate to oscillate.
Although not calculated here, the flame can exhibit an even more substantial impact
on the local acoustic field if the acoustic wave is obliquely incident; that is, = 0.22
Consider next the impact of including the acoustic source terms on the right side
of Eq. (12.9). This inclusion results in the addition of energy to the acoustic field
by unsteady heat-release processes. In cases where the acoustic wave is obliquely
incident on the flame, vorticity is also produced via the baroclinic mechanism
because of the misaligned fluctuating pressure and mean density gradients. The
energy in these vorticity fluctuations is derived from the incident acoustic field itself
and acts as a source of acoustic damping. This situation is analogous to acoustic-
wave damping at a rigid surface, where vorticity oscillations are excited through
the no-slip boundary condition. The net acoustic energy flux out of the flame Ia
is controlled by competing acoustic-energy production and dissipation processes.
Equation (12.14) is reproduced from Ref. 22 and illustrates these production (first
two terms) and dissipation (third term) terms:
2
Ia p p S1 √ p u 2u
= (
− 1)Ms (2 − γ ) + −
p̄1 c̄1 p̄ p̄ S̄ 1 p̄ c2
(12.14)
R
r
R
Lf
ξ (r , t ) ξ (r , t )
Fig. 12.9 Illustration of conical (left) and wedge-shaped (right) flame geometries.
Depending on the temperature ratio across the flame, the magnitude and phase of
the flame-burning velocity response, and the angle of incidence between the wave
and the flame, the acoustic disturbance can be damped or amplified. Plots showing
typical results can be found in Ref. 22.
Equation (12.14) only describes the acoustic-field energy balance. Energy
is also added to the vortical and entropy fields. Although not shown here, Ref.
22 includes expressions for these convected waves. Although the vortical wave
couples with the acoustic field at O(Ms ) (resulting in acoustic damping and
vorticity amplification), at this order entropy waves are forced disturbances and
do not have an impact on the acoustic or vortical fields. Full coupling between all
three disturbances occurs at O(Ms2 ).
Note that all energy amplification and damping processes in this equation are
relatively small, being of O(Ms ). Flame-area fluctuations, which are discussed in
the next section, are usually a much stronger source of acoustic energy.
3. Flame-Area Response to Flow Perturbations

This section describes the dynamics of the flame-surface area, which, as noted
in Eq. (12.7), constitutes a mechanism for heat-release fluctuations. Flame-area
disturbances are generated by variations in flame-front orientation that, in turn,
are generated by disturbances either in the approach flow velocity or in the flame
speed. In this section, we restrict attention to area disturbances arising from flow-
velocity perturbations. Flame-speed perturbations, which introduce heat-release
disturbances through perturbations in both flame area and consumption rate, are
considered in Sec. III.A.4. Consider the geometries shown in Fig. 12.9. On the left
is a conical flame stabilized on a tube, such as a Bunsen flame. On the right is an
axisymmetric wedge flame, stabilized on a bluff body. The flames have axial and
radial dimensions given by the flame length L f and radius R.
The instantaneous flame-sheet location at the radial location r is given by ξ (r, t),
assumed to be a single-valued function of r ; thus, the flame-position surface f (x , t)
[see Eqs. (12.1)–(12.3)] is defined as f (x , t) = y − ξ (r, t). This assumption nec-
essarily limits the range of amplitudes that can be treated with this formulation.
With Eq. (12.1), the flame dynamics are described by

2
∂ξ ∂ξ ∂ξ
=u−v − S1 +1 (12.15)
∂t ∂r ∂r
330 T. C. LIEUWEN
where u and v are axial and radial velocity, respectively. Note that a differential
element of flame surface is related to the flame position through the relation

2
∂ξ
dAFL = 1+ dr (12.16)
∂r
We assume that the mean velocity is uniform and purely axial (i.e., v̄ = 0), that
v = 0, and that the mean flame speed is constant. Although these assumptions are
not necessary to proceed with the analysis, they do yield more transparent results
that retain many of the basic phenomena of interest.
We next focus on the flame area’s linear dynamics. Nonlinear dynamics are
considered later. As shown in the equation below, the linear solution to the equation
for ∂ξ /∂r , or the flame-surface area, can be decomposed into two canonical
components: the homogeneous solution (second term on the right side) containing
the influence of boundary conditions and the particular solution caused by spatial
nonuniformities in flow forcing (or flame speed). To simplify the equation, a flame
coordinate along the nominal flame surface position s is introduced:
s
∂ξ (s, t) 1 ∂ s − x 1 s
= u x ,t− dx + u s = 0, t − − u base
∂s uo ∂s uo uo uo
0
(12.17)
where u base denotes the velocity of the end of the flame sheet at the attachment point.
A spatially uniform velocity disturbance ∂u /∂s = 0 excites only the homogeneous
solution. This disturbance can be understood by first assuming that the flame edge
moves exactly in step with the particle velocity u (s = 0) = u base . In this case,
the entire flame moves up and down in a bulk motion without change in flame
orientation or area. However, if a flame-anchoring boundary condition is imposed,
for example, u base = 0, such that the flame remains fixed at a point, the flow
disturbance excites a flame-front disturbance that originates at the boundary and
propagates along the flame front.
If the disturbance flowfield is spatially nonuniform, ∂u /∂s = 0, the particular
solution is excited. This results in waves originating at the spatial location(s) of
flow nonuniformity that also propagate along the flame at roughly the mean flow
velocity.
As will be shown subsequently, the flame area acts as a low-pass filter to flow
disturbances, so that the amplitudes of the two canonical solutions individually
decay with frequency as roughly 1/f, but, in general, do not become identically
zero. As such, the transfer function relating the response of the flame area to a
spatially uniform velocity disturbance (where only the homogeneous solution is
excited), (A /Ao )/(u /u o ) has a value of unity at zero frequency and decays with
frequency. In contrast, when the flame is perturbed by a spatially nonuniform dis-
turbance (so that both the homogeneous and particular solutions are excited), the
flame area consists of a superposition of the two solutions. As such, although
each solution decreases with frequency, their sum has oscillatory behavior in
cases where they constructively interfere and even causes the transfer function
(A /Ao )/(u /u o ) to exceed unity. This result was first predicted and then experi-
mentally confirmed by Schuller et al.40 and Durox et al.,18 respectively. In addition,
the two solutions can destructively interfere, and in certain cases, exactly cancel
each other so that the resulting transfer function (A /Ao )/(u /u o ) identically equals
zero.
We now consider linear solutions to the nondimensionalized form of Eq. (12.15),
that is,

1 + β 2 ∂ξ 2
∂ξ ∂r
+ = u(ξ, t) (12.18)
∂t 1 + β2
for time harmonic velocity perturbations, given as
u(ξ, t) = u o + u cos [kc ξ − ωo t] (12.19)
where kc = ωo /u c and u c is the phase velocity of the disturbance The variables t,

r , u, and ξ are nondimensionalized by u o /L f , R, u o , and L F (note that the value
of L F and R refer to their nominal values without imposed oscillations), where
u o is the mean axial velocity. Three dimensionless parameters naturally arise. The
Strouhal number, St = (ωo L F )/u o , velocity perturbation, ε = u /u o , and ratio of
the flame length to radius, β = L F /R.
Assume that the flame remains anchored at the base, u base = 0:
ξ (r = 1, t) = 0 (12.20)
The effects of a nonstationary flame-anchoring point is addressed in Sec. III.A.7.

Consider the instantaneous flame-surface area, which, for a conical flame, is
given by
2
1
r 1 + β 2 ∂ξ
∂r
dr
Ac (t) 0
=2 (12.21)
Ac,o 1 + β2
where subscripts c and w are used to denote axisymmetric conical and wedge
flames, respectively. It is shown in Ref. 41 that the solution for the conical flame
area transfer function, G c = (Ac /Ac,o )/(u /u o ) is
G c (St2 , η) = G c,BC + G c,Flow

exp (i St2 ) − 1 − i St2 1 − exp (iηSt2 ) + iηSt2
=2 +2
(η − 1) St22 η (η − 1) St22

2 exp (iηSt2 ) − 1
= 2 1− exp (i St2 ) +
St2 (1 − η) η
(12.22)
332 T. C. LIEUWEN
where
St (1 + β 2 )
St2 = (12.23)
β2
uo β 2
η= (12.24)
uc 1 + β 2
Note that the contribution of the flow nonuniformity and boundary conditions are
explicitly separated, following the discussion of Eq. (12.17).
The solution for the wedge-flame-transfer function, G w = (Aw /Aw,o )/(u /u o ),
is
G w (St2 , η) = G w,BC + G w,Flow

1 − (1 − i St2 ) exp (i St2 ) (1 − iηSt2 ) exp (iηSt2 ) − 1
=2 +2
(η − 1) St22 η (η − 1) St22

2
= × [η − 1 + i(i + St2 )η exp (iω)
η(η − 1)St22
+ (1 − iηω) exp (iηω)]
(12.25)
Thus, the linear flame-transfer functions for both the conical and wedge flames,
Eqs. (12.22) and (12.25), only depend on two parameters, St2 and η. The term η
couples the effect of flame angle and phase speed of the disturbances. Alternatively,
the η effects can be captured by defining another Strouhal number based on the
convective velocity u c of the flow disturbances Stc . Stc naturally arises in the two
transfer functions, Eqs. (12.22) and (12.25), and equals ηSt2 . These two Strouhal
numbers are related to the amount of time taken for a flow (Stc ) and flame-front
(St2 ) disturbance (which is ultimately created by a flow disturbance) to propagate
the flame length, normalized by the acoustic period.
Before looking at the total flame-transfer functions, it is useful to understand the
characteristics of its two contributing flow-forcing and boundary condition terms.
Their ratio is given by
G c,Flow 1 − exp (iηSt2 ) + iηSt2

= (12.26)
G c,BC η exp (i St2 ) − 1 − i St2
G w,Flow (1 − iηSt2 ) exp (iηSt2 ) − 1

= (12.27)
G w,BC η 1 − (1 − i St2 ) exp (i St2 )
The magnitude of this ratio is identical for both wedge and conical flames (see
Fig. 12.10). The phase of this ratio is different for conical and wedge flames and
plotted in Fig. 12.11.
It is instructive to analyze the characteristics of this ratio for limiting values of the
parameters η and St2 . First, note that in the η → 0 limit (i.e., a spatially uniform dis-
turbance), the flame dynamics for both the wedge and conical flames is controlled
Fig. 12.10 Strouhal number dependence of the magnitude of the ratio of the transfer
functions caused by the flow-forcing and boundary-condition terms for different values
of η.
Fig. 12.11 Strouhal number dependence of the phase of the ratio of the transfer
functions caused by the flow-forcing and boundary-condition terms for wedge flames
for different values of η. Shaded regions indicate points where boundary-condition
and flow-forcing terms are in phase.
334 T. C. LIEUWEN
exclusively by the boundary-condition term, irrespective of Strouhal numbers:

G c,Flow G w,Flow
Lim = Lim =0 (12.28)
η→0 G c,BC η→0 G w,BC
This result can be anticipated from the preceding discussion and reflects the fact that
only the homogeneous solution is excited when the flow disturbance is uniform.
In the St2 → 0 limit, the relative contribution of the two terms is determined by
the value of η:

G c,Flow G w,Flow
Lim = Lim = −η (12.29)
St2 → 0 G c,BC St2 → 0 G w,BC
The boundary-condition and flow-forcing terms dominate when η <1 and η >1,
respectively. For long flames (β 1), this physically corresponds to situations
in which the disturbance-phase velocity is greater than and less than the mean
flow velocity, respectively. The two terms tend toward equal magnitudes when
η = 1. These points can be clearly observed in Fig. 12.10. Note also that the flow-
disturbance and boundary-condition terms are 180 deg out of phase for low St2
values (see Fig. 12.11).
In the St2 1 limit, the contribution from both the boundary conditions and
flow-forcing term are equal, as shown in Fig. 12.10 and in Eq. (12.30):

G c,Flow
Lim = −1
St2 → ∞ G c,BC

G w,Flow
Lim = − exp [i(η − 1)St2 ] (12.30)
St2 →∞ G w,BC
Equation (12.30) also shows that, in this limit, the relative magnitude contribu-
tion of these two terms is independent of η (assuming that the ηSt2 product does
not simultaneously go to zero). Moreover, the two terms are always out of phase for
conical flames irrespective of the Strouhal number and η; typical relative phases
range between 140 and 220 deg. In contrast, for wedge flames the phase differ-
ence between the two contributions monotonically increases with St2 , as shown in
Fig. 12.11 (the shaded bands in the figure indicate regions of constructive inter-
ference).
The dependence of the magnitude and phase of the conical flame-transfer func-
tion G c (St2 , η) on St2 is plotted in Figs. 12.12 and 12.13, respectively. Consider
the magnitude results first. Note that the transfer-function gain is identical in the
cases in which η = 0 or 1. Physically, this corresponds to cases in which the dis-
turbance velocity is uniform (η = 0) or its phase speed matches the flame-front
disturbance velocity (η = 1). The gain-transfer function differs for all other
disturbance-phase velocity cases. Note also that the gain value is always less than
one and generally decreases monotonically with St2 , although some ripple occurs
at higher St2 values because of constructive and destructive interference between
G c,Flow and G c,BC . The transfer-function phase starts at zero degrees at low St2
and initially increases monotonically with St2 .
Fig. 12.12 Axisymmetric conical linear transfer function G c (St2 , η) amplitude de-
pendence on the reduced Strouhal number (St2 ) for different values of η.
For wedge flames, the transfer-function gain G w (St2 , η) is plotted in Fig. 12.14.
Note that all gain values tend toward values of unity at low St2 . However, only
in the uniform velocity case, η = 0, does the gain monotonically decrease with
increases in St2 . In all other cases, the gain increases to values of greater than unity
because of constructive interference between G w,Flow and G w,BC .
Fig. 12.13 Axisymmetric conical linear transfer function G c (St2 , η) phase depen-
dence on the reduced Strouhal number (St2 ) for different values of η.
336 T. C. LIEUWEN
Fig. 12.14 Axisymmetric wedge linear transfer function Gw(St2 , η) amplitude depen-
dence on the reduced Strouhal number (St2 ) for different values of η.
Another striking feature is the resonance-like behavior at η = 1, where the

wedge-flame response does not decrease with St2 but tends toward a constant
value of two. This case corresponds to exact coincidence of flame-front and flow-
disturbance velocity. In reality, curvature effects on flame speed that are neglected
in this analysis, which increase with St2 , cause the transfer function to decrease at
higher values of St2 .
In general, the relationship between an unsteady heat-release rate and velocity
has a complex dynamic. However, for St 1 (convectively compact flame), the
A (t) relationship can be put in terms of a simple n–τ model:
A (t)/Ao = n u u (t − τ ) (12.31)
where n u = 1/ S̄ 1 , τ conical = [(η + 1)L F ]/3u o , and τ wedge = [2(η + 1)L F ]/3u o .
Eq. (12.31) indicates that the time response of the flame area to perturbations in
acoustic velocity is delayed by a retarded time τ . This retarded time equals the
time taken for the mean flow to convect some fractional distance of the flame
length, which is equivalent to replacing the distributed flame by a concentrated
source at this location; for example, for a conical flame this effective position of
concentrated heat release is L eff ≈ (η + 1)L F /3.
We next turn to the response of the flame area in the general, nonlinear case. Note
that in the linear case the transfer function is described by only two parameters;
that is, G Lin = G(St2 , η). For the general nonlinear case, however, the gain G also
depends on ε and β; that is, G = G(St2 , η, β, ε). Before considering specific
results, several general conclusions that can be obtained from analysis of the
equations should be considered. The key mechanism of nonlinearity is illustrated
in Fig. 12.15. In this illustration, a flame is perturbed into a corrugated front but
then allowed to relax back to its steady-state, planar position. Flame propagation
Fig. 12.15 Sketch of a flame that is initially wrinkled (top), showing the destruction
of flame area by kinematic restoration processes (bottom).
normal to itself smoothes out the wrinkle, so that its area eventually returns to being
constant in time. As shown by the dashed lines in the bottom sketch, opposed flame
branches merge to form a sharp cusp and propagate forward to destroy flame area.
The cusp-formation time τ cusp of a front with a corrugation of length scale λ f is
proportional to λ f /S1 .
The rate of flame-area destruction depends nonlinearly on the amplitude of the
flame-front disturbance. Large-amplitude corrugations are smoothed out at a rela-
tively faster rate than small-amplitude perturbations. In the same way, short-length-
scale corrugations are smoothed out faster than long-length scales. As discussed
further below, corrugation smoothing is the reason that nonlinearity is enhanced
at higher disturbance frequencies, which generate shorter-length-scale flame cor-
rugations.
Consider the effects of these nonlinearities on the flame disturbances generated
at the boundaries and regions of flow nonuniformity, as discussed after Eq. (12.17).
If only the homogeneous solution of Eq. (12.17) is excited, as in a spatially uniform
velocity-perturbation field, nonlinear effects always cause the nonlinear transfer
function relating flame area (at the disturbance frequency) and velocity perturba-
tions, (A /Ao )/(u /u o ), to monotonically decrease with disturbance amplitude. In
other words, the linear transfer function is always larger than the nonlinear transfer
function.
If the velocity field is nonuniform, the effects of nonlinearity on both the par-
ticular and homogeneous solutions causes the overall solution characteristics to
depend on whether the two solutions lie in a region of constructive or destructive
interference. If they lie in a region of constructive interference, a conclusion simi-
lar to the previous one holds: the transfer function (A /Ao )/(u /u o ) decreases with
disturbance amplitude. Opposite behavior may occur if the two solutions destruc-
tively interfere, because they are affected unequally by nonlinearity. Nonlinearities
have a longer time to destroy flame area for the boundary-condition term that prop-
agates the entire flame length, as opposed to the flow nonuniformity terms that are
excited at each point along the flame. As will be shown subsequently, the result is
that the nonlinear transfer function can actually exceed its linear value.
338 T. C. LIEUWEN
Next, consider the dominant factors that affect flame-area nonlinearity. Note
that these nonlinearities arise fromthree sources. The first factor is the nonlinear
flame dynamics through the term 1 + β 2 (∂ξ/∂r )2 in Eq. (12.18). The second
factor is the static nonlinearity introduced through the dependence of the flame area
on flame-position gradient through a term with the same form, 1 + β 2 (∂ξ/∂r )2
[see Eq. (12.21)]. In both of these cases, the nonlinearity is purely geometric in
origin and is introduced by the relationship between the instantaneous flame-front
normal and flame-position gradient. The third nonlinearity is caused by the flow
forcing itself and the dependence of the disturbance velocity at the flame front on
the flame position u(ξ, t).
The fact that the first two sources of nonlinearity are identical can be used to write
the final expressions for the flame area, Eq. (12.21), in a revealing form. By substi-
tuting Eq. (12.18) into Eq. (12.21), note that the term (1 + β 2 (∂ξ/∂r )2 )/(1 + β 2 )
which appears in both the area integrals can be written as

1 + β 2 (∂ξ /∂r )2 ∂ξ
= u(ξ, t) − (12.32)
1 + β2 ∂t
Thus, the explicit form of the nonlinearity disappears. Nonlinearities in flame-front

dynamics are included in the ∂ξ/∂t term, whereas those caused by the flow-forcing
nonlinearity noted previously are included in the u(ξ, t) and ∂ξ/∂t term. Based
on Eq. (12.32), the following observations can be made regarding the effects of
various parameters on nonlinearity in the flame’s response to flow perturbations.
a. Strouhal Number. At low Strouhal numbers, the unsteady term in
Eq. (12.32) is negligible. Moreover, the ξ dependence of the velocity field u(ξ, t),
is weak in the limit of low Stc , at least for the velocity fields considered here. Thus,
the flame area’s velocity response remains linear for low Strouhal numbers. This
point shows that the flame’s nonlinear area response is an intrinsically dynamic
phenomenon; its quasi-steady response is linear. An alternative way to state this
argument follows from noting that the Strouhal number is related to the ratio of
the time for a disturbance created at the flame base to convect the length of the
flame L F /u o to the flame-front cusp formation time τ cusp ∼ λ F /S1 . If this ratio is
small, the flame wrinkle will not have enough time to form a cusp, which is closely
associated with nonlinearity.
b. Flow Uniformity. Nonlinearities in the u(ξ, t) term are directly caused by
nonuniformity in flow disturbances. Thus, the contribution of this term to flame-
area nonlinearities is suppressed in the η → 0 limit.
c. Boundary Conditions. If the flame remains anchored at the attachment
point, as it is in this section, then ∂ξ/∂t is identically zero at this point for all
time. As such, the flame-area perturbations in the vicinity of the attachment point
[where ξ ≈ 0 ⇒ u ≈ u(0, t)] exhibit a linear dependence on velocity amplitude.
Nonlinearities only arise at points of the flame that are spatially removed from the
attachment point. As such, the axisymmetric conical flame exhibits a more linear
velocity response than the axisymmetric wedge flame for comparable values of ε,
because most and very little, respectively, of the flame area is concentrated near
the attachment point.
d. Flame Aspect Ratio. When β 1, that is, when the flame is very long,
the flame dynamics are approximately described by the equation
∂ξ ∂ξ
± = u(ξ, t)
∂t ∂r
In this case, the flame dynamics are linear, although the flow-forcing term does not
need to be linear. Thus, β is an important nonlinearity parameter for this problem;
that is, the flame’s area response can be anticipated to exhibit a linear dependence
on the perturbation velocity for much larger ε values at large β values.
The rest of this section presents typical results comparing the linear and nonlin-
ear flame-transfer function. These results were obtained by numerically integrating
the governing equation (12.18).41 The nonlinear flame-transfer function was deter-
mined by computing the flame area only at the forcing frequency (because higher
harmonics are also excited) via the Fourier transform.
The boundary condition, Eq. (12.20), cannot be used for disturbance-velocity
magnitudes in which the instantaneous flow velocity is lower than the flame speed.
In this case, the flame flashes back and Eq. (12.20) must be replaced by a different
condition (see Sec. III.A.7). Results are shown in the following text for disturbance
amplitudes up to the point of flashback, given by the disturbance amplitude ε = εf ,
where
1
εf = 1 − (12.33)
1 + β2
Figure 12.16 plots the St2 dependence of the nonlinear transfer-function gain for
a wedge flame. The gain-transfer functions are normalized by their linear values,
Fig. 12.16 Strouhal number dependence of the ratio of the magnitude of the flame
area–velocity transfer function to its linear value for the axisymmetric wedge flame,
β = 1, η = 0.
340 T. C. LIEUWEN
Fig. 12.17 Strouhal number dependence of the ratio of the magnitude of the flame-
area–velocity-transfer function to its linear value for an axisymmetric wedge and
conical flame, β = 1, ε/ε f = 0.99.
G/G Lin . Results are shown for a uniform-velocity field η = 0 and a value of β = 1.
As predicted previously, the response tends to its linear value in all cases at low
St2 . Note the substantial reduction in the flame area relative to its linear value; that
is, there is a substantial degree of gain saturation. In agreement with the Strouhal
number argument (Sec. III.A.3.a), the degree of nonlinearity increases with St2 .
For the present case, the gain for the wedge flames decreases by about 75% at
ε = εf . Although the phase of the area response is not shown here it exhibits little
amplitude dependence, varying by a total of about 8 deg at ε = εf .
Figure 12.17 shows the St2 dependence of the gain and phase of the nonlinear
transfer function for a wedge and conical flame at a given velocity amplitude for a
range of η values. The results are shown for a velocity amplitude of ε/εf = 0.99.
Note that, consistent with the boundary conditions argument (Sec. III.A.3.c), the
wedge exhibits a far more nonlinear response than a conical flame.
In the η = 0 case, nonlinearity causes a monotonic decrease in transfer function
with disturbance amplitude. This result is not true, in general, because of the inter-
actions between the boundary-condition and flow-forcing nonuniformity solutions
noted previously. To illustrate, Fig. 12.18 shows the flame response for a wedge
flame when η = 2 (i.e., disturbances are traveling at approximately half the mean
flow speed). Note that the gain results are not normalized by their linear value here.
The gain result indicates that, in the 6 < St2 < 8 range, the nonlinear transfer func-
tion actually exceeds its linear value. This result can be understood by noting that
this behavior occurs in the vicinity of the regions where the linear-transfer func-
tion achieves a minimum. At these St2 values, the contributions attributable to the
boundary conditions and the flow-forcing terms exactly cancel each other, leading
to zero gain. As the velocity amplitude is increased, nonlinearities cause the gain
attributable to both the boundary conditions and the flow-forcing terms to decrease.
Because the individual gain decreases by different amounts, the total gain does not
Fig. 12.18 Strouhal number dependence of the magnitude of the flame-area–velocity-

transfer function for the axisymmetric wedge flame, β = 2, η = 2.
go to zero at the St2 value at which the linear gain is zero but actually shifts to a
higher St2 value in the ε = 0.2ε f case. At higher disturbance levels, the two terms
never exactly cancel and the gain does not go to zero. Rather, a monotonic decrease
occurs in the gain of the transfer function with the increase in velocity amplitude.
These results are consistent with the related measurements of Durox et al.18 Analo-
gous behavior also occurs in conical flames, although less dramatically. In addition,
unlike the η = 0 case, the phase exhibits a stronger amplitude dependence.
Although these results have focused on theoretical predictions, they are generally
found to be in good agreement with experiments, assuming that the correct velocity
characteristics are used in the model.
These results have implications on the type of bifurcations that may be ob-
served in unstable combustors in cases in which heat-release nonlinearities are
the dominant source of nonlinearity (see the discussion in Chap. 1). In situations
in which the gain curves resemble those qualitatively shown in Fig. 12.16, only
supercritical bifurcations will occur and only a single stable-limit-cycle amplitude
ALC is possible. In situations in which the gain exceeds, then is less than, the
linear gain, multiple stable solutions for the instability amplitude may exist, and
subcritical bifurcations are possible. Depending on the operating conditions and
frequency, both types of gain curves can be obtained. This can be seen in Fig. 12.19,
which plots the dependence of A /Ao vs ε at the two conditions, St2 = 2.5 and
St2 = 6.25. Note the similarity in shape of these curves with those plotted in
Chapter 1 for the sub- and supercritical bifurcations. This similarity implies that
unstable combustors driven by these flame-area-fluctuation mechanisms may or
may not exhibit hysteresis and triggering, depending on the operating condition
and frequency.
342 T. C. LIEUWEN
A’/Ao
Fig. 12.19 Dependence of flame-area fluctuation, A /A0 on velocity amplitude ε for

wedge flames when η = 2.
Another point of interest is that the linear gain and nonlinear saturation ampli-
tude are not related; for example, it is not possible to draw definitive conclusions
about the flame’s saturation amplitude based on how strongly it responds to low-
amplitude fluctuations. It is possible to find regions in which change in a particular
parameter causes the saturation amplitude to either increase, not change, or de-
crease with variations in the linear gain. For example, Fig. 12.20 illustrates an
example in which the linear gain and nonlinear saturation amplitude have opposite
trends. This case corresponds to a situation in which the flame length is doubled
at a constant frequency. For example, such a trend could be associated with a
decrease in flame speed. This example has clear implications on the applications
of linear-stability analyses to inferring instability amplitude trends; increases in
instability-growth rate do not necessarily imply increases in combustion-instability
amplitude. The amplitude could also stay the same or, as just discussed, could ac-
tually decrease.
4. Flame-Speed Response to Perturbations

This section describes the various mechanisms of flame-speed perturbations,
which, as noted in Eq. (12.7), constitute a mechanism for heat-release fluctuations.
It considers the effects of fluctuating pressure and temperature, then strain rate,
and finally mixture composition.
Consider first the response of the flame speed to the unsteady pressure and
temperature variations in an acoustic wave. Several analyses,42−46 have studied
the internal structure of a flat flame perturbed by an acoustic wave with high-
activation energy asymptotics and single-step kinetics. Many of these results are
summarized by McIntosh,47 who emphasizes the different characteristics of the
interaction, depending on the relative magnitudes of the length and timescales of
Fig. 12.20 Dependence of flame-area fluctuation A /A0 on velocity amplitude ε for

wedge flames when η = 0 for two different conditions.
the acoustic wave and flame preheat and reaction zone. Following McIntosh,47
define the following ratios of these length and timescales:
diffusion time acoustic wavelength

τM = N≡ (12.34)
acoustic period diffusion length
These ratios are related by the Mach number of the flame-burning velocity:
SL 1
Ms = = (12.35)
cu τM N
Also, define the dimensionless overall activation energy:
Ea
θE = (12.36)
Rg Tb
where E a is the overall activation energy, Rg is the gas constant, and Tb is the
burned-gas temperature. Four different regimes exist whose characteristics depend
on the relative magnitudes of these parameters:
1) N 1/Ms (i.e., τ M 1). Acoustic wavelength is much longer than the

flame thickness and the flame responds in a quasi-steady manner to acoustic
disturbances.
2) N ∼ O(1/Ms ) (i.e., τ M ∼ O(1)). Acoustic wavelength is much larger than the
flame thickness, but acoustic- and flame-response times are commensurate.
344 T. C. LIEUWEN
10 0
-10
-20
phase (ν)
1 -30
|ν|
-40
-50
0.1 -60
0.01 0.1 1 10
Flame response time/Acoustic Period, τ
Fig. 12.21 Normalized mass burning-rate response to acoustic pressure perturba-

tions ν adapted from McIntosh.42
3) N ∼ O(1/θ E2 Ms ) (i.e., τ M ∼ O(θ E2 )). Fast timescale acoustic oscillations af-

fect inner-reaction zone. Spatial pressure gradients are not important in the com-
bustion zone.
4) N ∼ O(1) [i.e., τ M ∼ O(1/Ms )]. Pressure gradients occur over the same
length scale as flame thickness.
The regime of most interest to unstable combustors is likely parameters 1 and

2. For example, a frequency of 400 Hz roughly corresponds to a τ ∼ 1 value in a
stoichiometric methane–air flame. For these cases, McIntosh derives the following
expression relating the mass burning rate and acoustic pressure perturbation42 :

m p 2θ E (γ − 1) (−iτ M )(s − 1 + 1/
)
≡υ= θ E (
− 1)
m̄ p γ
[Le(s − 1) + (1 − r )] − 2s(1 − r )
(12.37)
where we assume an exp (−iωt) time dependence, Le is the Lewis number, and

s = 1 − 4iτ M /Le and r = 1 − 4iτ M (12.38)
Figure 12.21 plots the dependence of the flame-speed response ν. It increases

roughly with θ E , dimensionless frequency τ M , and flame temperature jump
.
The Lewis number dependence is quite weak for Le values near unity. This result
illustrates that the mass burning rate response is substantially larger than its quasi-
steady value in the physically interesting τ ∼ O(1) case. Although these analyses
are most relevant to the flamelet-combustion regimes, McIntosh suggests that they
could also be applied to the distributed reaction regime, where the laminar flame
thickness is replaced by the thickness of the thickened reaction zone.
Apparently, no complementary experimental investigations to critically assess
these predictions have been undertaken.
Consider next the effect of strain-rate fluctuations, also introduced by acoustic
or vortical velocity perturbations, on the flame speed.48 Flame strain can increase
or decrease the flame speed; this sensitivity is quantified by the Markstein number
Ma. For weakly stretched flames, a linearized expression for this dependence is:
Sc /SL = 1 − Ma K a (12.39)
where Sc and SL denote the stretched and unstretched flame speeds, K a = κ/SL d
is the Karlovitz number, d is the flame thickness, and κ is the stretch rate, given by:
κ = ∇t · u − Sc / (12.40)
where ∇t and denote the tangential divergence of the velocity field at the flame
surface and its radius of curvature, respectively.
Equation (12.40) shows that, in the steady case, the burning-velocity depen-
dence on curvature (second term) and hydrodynamic stretch (first term) combines
into a single term.49 Quasi–steady-strain fluctuations cause flame-speed oscilla-
tions about its nominal value with a Markstein number-dependent magnitude.
However, in the general unsteady case, Joulin’s50 analysis predicts that the flame-
speed sensitivity to the two terms in Eq. (12.40) have different frequency-response
characteristics. The unsteady strain effect diminishes with frequency, whereas the
unsteady-curvature term is independent of frequency. The latter prediction has
apparently not been assessed experimentally or computationally. The former pre-
diction is consistent with Im and Chen’s51 calculations, which predicted that the
flame-speed response to strain-rate fluctuations attenuates as the frequency in-
creases (see Fig. 12.22). This figure plots the instantaneous flame-consumption
speed as a function of the instantaneous stretch rate. The unclosed line that spans
the entire range of y-axis values corresponds to the steady-state result and shows the
consumption-speed augmentation by stretch. The filled circles correspond to the in-
stantaneous correspondence between consumption speed and stretch rate when the
stretch rate is oscillated sinusoidally. The 10-Hz case closely follows the steady
line. With increasing frequency, however, the amplitude of consumption-speed
oscillations monotonically decreases and becomes quite small at 1000 Hz. This
result emphasizes the importance of dynamic effects and reinforces a point from
Sec. II.A that the flame response is not quasi-steady, even at relatively low fre-
quencies.
Although unsteady stretch effects have not been systematically evaluated in
acoustically forced flames, they may be responsible for the filtering phenomenon
experimentally observed by Baillot and coworkers.52−54 When the Bunsen flame is
forced with high-frequency, low-amplitude disturbances, they observed that flame
wrinkles are only evident at the flame base and quickly decay with at axial locations
farther downstream. This behavior may be caused by the increased importance of
the flame’s curvature-dependent burning velocity and the very short convective
wavelengths of the imposed disturbances at these higher frequencies.
Finally, consider the effect of mixture-composition (i.e., equivalence ratio) os-
cillations on the flame speed. Insight into this sensitivity can be gained from the
steady-state dependence of flame speed on equivalence ratio. First, the flame speed
has a maximum value (i.e., ∂S1 /∂φ = 0) under near-stoichiometric conditions,
implying that the flame speed under such conditions is insensitive to equivalence
ratio oscillations. Second, because the flame-speed sensitivity to equivalence ratio
346 T. C. LIEUWEN
Sc/SL
Instantaneous Karlovitz Number

Fig. 12.22 Dependence of instantaneous flame-consumption speed Sc on instanta-
neous Karlovitz number at several frequencies of oscillation. Calculation performed
at φ = 0.4 for a hydrogen–air flame. Image courtesy of H. Im.51
∂ S1 /∂φ generally grows as φ decreases from unity, the amplitude of the flame-
speed oscillations generated by a fixed φ disturbance grows with decreases in the
mean equivalence ratio.
Similar to the fluctuating strain case, the flame’s response to dynamic fluctua-
tions in mixture composition decreases with increases in frequency. Sankaran and
Im studied the dynamic response of lean methane–air-premixed flames to such
fluctuations and also found substantial dynamic effects in phase.55 For example,
at 400 Hz, their analysis predicts a large-phase shift between the instantaneous
equivalence ratio and flame speed, with the effect that the flame speed actually
increases or decreases with decreases or increases in the equivalence ratio.
The overall response of the flame’s heat release Q(t) to flame-speed perturba-
tions is complex, because the flame’s position, and therefore surface area, is also
affected. As will be shown subsequently, the contributions of the flame-speed and
flame-area perturbations constructively and destructively interfere, respectively,
depending on the frequency of oscillations. We describe next an analysis56 of the
response of a conical flame to a convected φ disturbance that illustrates these cou-
pled dynamics. The analysis is similar to that described in Sec. III.A.3, except
now the flame-speed fluctuation terms are retained and the velocity disturbances
are neglected. To simplify the expressions from Ref. 56, it will be assumed that
β 1, so that (1 + β 2 )/β 2 ≈ 1 and St2 ≈ St.
Assume that the equivalence-ratio perturbation is convected with the mean flow
velocity and, thus, has an axial distribution given by
φ (x, t) = φb exp [−iω(t − x/ū)] = φb exp (−iωt) exp [i St(1 − r/R)]
(12.41)
φb in Eq. (12.41) denotes the perturbation in equivalence ratio at the flame base. The
flame-speed perturbation is related to the perturbation in mixture stoichiometry by

dS1
S1 = · φ (12.42)
dφ φ
From Eq. (12.7), the total heat-release perturbation is given as

Q ρ d AFL S1 d AFL h R d AFL A
= 1 + + + FL (12.43)
Q̄ ρ̄1 d AFL S̄1 d AFL h̄ R d AFL AFL
Assume that the equivalence-ratio disturbance occurs at constant density; that is,
ρ1 = 0. Define the following flame-transfer functions to perturbations in equiva-
lence ratio, Fφ
Q φ / Q̄
Fφ = = FH + FS = FH + (FS,dir + FA ) (12.44)
φb /φ̄
where

d(h R /h̄ R ) 2
FH = {1 + i St − exp (i St)}
d(φ/φ̄) φ̄ St 2

d(S1 / S̄ 1 ) 2
FS,dir = {1 + i St − exp (i St)} (12.45)

d(S1 / S̄ 1 ) 2
FA = {1 − (1 − i St) exp (i St)}
Q φ denotes the heat-release response to equivalence-ratio perturbations. Note that

these transfer functions are solely a function of the Strouhal number and the sen-
sitivities of the heat of reaction and flame speed to the equivalence ratio. The
amplitude and phase dependence of these transfer functions, Eq. (12.45), are plot-
ted in Fig. 23. These characteristics are described in more detail later.
The equivalence-ratio transfer function Fφ has three contributing terms [see
Eq. (12.44)]. The first term FH is caused by perturbations in heat of reaction. The
second term FS is caused by perturbations in flame speed. Note that perturbations
in flame speed are again divided into two factors; one is directly generated by the
flame-speed sensitivity to the equivalence ratio FS,dir and the other is caused by
the subsequent fluctuation in flame-surface area FA . These calculations assume a
quasi-steady relationship between the equivalence ratio and flame speed; that is,
348 T. C. LIEUWEN
Fig. 12.23 Dependence of flame-transfer function gain on Strouhal number (φ = 1).
that d(S1 / S̄1 )/d(φ/φ̄) is independent of frequency. The additional dynamics of

this relationship can be incorporated in a straightforward manner, however.
As in the velocity-disturbance case, it is useful to examine the time-domain
relationship between equivalence-ratio and heat-release disturbances. In contrast
to the velocity-perturbation case, the dynamics of Q φ cannot be described, in
general, by an n − τ model, even in the St 1 limit. This relationship is due
to the possible negative-phase dependence of Fφ on St when St 1; that is, the
flame can not respond before the equivalence-ratio perturbation reaches it. The
low St dynamics of Q φ is given by
dφb (t)
Q φ (t)/ Q̄ = n H φb (t − τ H ) + n S (12.46)
dt
where

d(h R /h̄ R ) LF 1 L F d(S1 / S̄1 )
nH = , τH = , ns =
dφ φ̄ 3ū 3 ū dφ φ̄
Q φ (t) is delayed or advanced depending on the combined effect of τ H and a

temporal rate of change of flame-speed perturbations as shown in Eq. (12.46).
To quantify the dependence of the heat of reaction and the flame speed
on the equivalence ratio we used the following correlation from Abu-Off
and Cant57 for methane: S1 (φ) = Aφ B exp[−C(φ − D)2 ], h R (φ) = [2.9125 ×
106 min (1, φ)]/(1 + φ 0.05825), using the coefficient values A = 0.6079, B =
−2.554, C = 7.31, D = 1.230.
These correlations were used to generate the results in Figs. 12.23, 12.24, and
12.25. Note that FH decreases monotonically from its maximum response at St =
0. In contrast, the heat-release response to flame-speed perturbation FS vanishes at
St = 0. This disappearance is caused by the exact cancellation of the flame-speed
and area-perturbation terms that have equal magnitudes but opposite phases. This
Fig. 12.24 Dependence of flame-transfer function phase on Strouhal number (φ = 1).
zero response at St = 0 can be understood from quasi-steady arguments; that is,

the flame area fluctuates with the same magnitude and opposite phase as the flame-
speed oscillations. This zero response can also be understood because the flame-
speed and area-perturbation terms account for the flame’s response to a mixture
with a constant heat of reaction. For example, two substoichiometric flames with
the same flow of fuel but differing amounts of air release the same amount of heat
for quasi-steady states, although the flames have different areas. As such, slow-
timescale perturbations may affect the flame’s local consumption rate, but the
resultant heat-release perturbation is exactly balanced by the resultant variations
in flame area.
The latter transfer function FS increases with Strouhal number from zero because
of changes in the relative phase of the terms FS,dir and FA . It reaches a global max-
imum at St ∼ 4.5 where, as shown in Fig. 12.24, the two perturbations reinforce
each other. As the Strouhal number increases further, FS decreases in an oscillatory
Fig. 12.25 Dependence of flame-transfer function gain on mean equivalence ratio.

350 T. C. LIEUWEN
pattern caused by the alternating phase relationship between FS,dir and FA . The
total heat-release response Fφ increases until St ∼ 4 and decreases in an oscillatory
manner.
Figure 12.25 plots the effect of the mean equivalence ratio on the flame-transfer
function. It shows that mixture stoichiometry has little effect on the transfer-
function magnitude for St 1 and at subsequent minima. In most cases, however,
the flame response increases with a decreasing equivalence ratio because of the
increased sensitivity of the flame speed to the equivalence ratio for lean mixtures
referred to previously. Although not shown, the heat-release response can either
lead or lag the φ perturbation, depending on the mean φ value.
An important conclusion to be drawn from these results is the importance of
both the local and global effect of a perturbation on the overall flame response. For
example, a flame-speed perturbation causes not only a local change in heat-release
rate per unit area, but also the overall flame area. The transfer-function results
illustrated previously show that inclusion of both effects is crucial in modeling the
overall flame response.
The nonlinear heat-release response to high-amplitude flame-speed disturbances
has not been analyzed to date. A partial accounting for the nonlinear dependence
of the φ disturbance on the acoustic field was modeled in a quasi-steady sense
by Peracchio and Proscia.58 They assumed the following relationship for the re-
sponse of the instantaneous mixture composition leaving the nozzle exit to velocity
perturbations:
φ̄
φ(t) = (12.47)
1 + ku (t)/u
where k is a constant with a value near unity. They also utilized a nonlinear
relationship relating the heat release per unit mass of mixture to the instantaneous
equivalence ratio, similar to the correlations used to derive the preceding results.
5. Wrinkled Flame Effects

An important question that must be addressed is how much the preceding results,
derived for smooth, laminar flame fronts, can be generalized to highly corrugated,
turbulent flames. Although many turbulent flame effects have not been treated to
date, several key features have been worked out in a series of experimental and
theoretical papers.
We first consider the effect of the flame on the acoustic field. Analytical treat-
ments of this topic modeled the flame as a dynamically evolving discontinuity in
temperature with a pressure-sensitive flame speed.59,60 These studies prescribed
the flame position and, thus, did not consider the fully coupled flame–wave dynam-
ics. They show that the key difference between sound wave scattering and laminar
and turbulent fronts is the following. Within the linear approximation, a coherent,
monochromatic wave incident on a laminar flame is scattered as a monochromatic
disturbance of the same frequency. In the turbulent flame, the same incident wave
generates scattered coherent and incoherent disturbances. The incoherent distur-
bances have a distributed spectrum that is roughly symmetric about the incident
-2
10
-3
10 f i=15 kHz
PSD (arb. units)
-4
10
-5
10 f i=7.5 kHz
-6
10
-20 -10 0 10 20
f-f i
Fig. 12.26 Measured spectra of scattered acoustic field excited by 7.5- and 15-kHz
incident waves. Adapted from Lieuwen.61
wave frequency f i . The broadened spectrum of the scattered field is caused by the
randomly moving flame front, resulting in Doppler-shifted scattered waves. These
characteristics are clearly illustrated by data plotted in Fig. 12.26, which shows
the spectra of 7.5- and 15-kHz sound waves scattered from a turbulent flame. The
narrow band, coherent peak at the incident wave frequency, and distributed side
bands are clearly evident in the figure.61 Note the broader bandwidth of the in-
coherent sidebands for the 15-kHz sound waves. This result can be understood
by noting that a harmonically oscillating acoustic wave incident on a reflecting
surface moving with a Mach number, M̄, generates reflected waves oscillating at
the Doppler-shifted frequency
· n)
(1 + M
f refl = f drive (12.48)

(1 − M · n)

where n denotes the unit normal direction of the incident wave. Noting that the
Mach number of flame-front motion is very small, this expression can be written
as
· n )2 1/2
( f − f drive )2 1/2 = ( f )2 1/2 ≈ 2 f drive ( M (12.49)
This equation shows that the bandwidth of the scattered waves f grows with rms
flame-front velocity and incident-wave frequency.
If the flame does not add energy to the acoustic field, the energy in the incoherent
sidebands is derived from the coherent wave. Thus, the wrinkled characteristics
of the flame act as a potential source of damping of coherent acoustic energy. In
cases in which the flame amplifies sound waves, the overall energy balance of the
352 T. C. LIEUWEN
Acoustic
Disturbance
Flame
Fig. 12.27 Image of instantaneous pressure field and flame front. Reproduced with
permission from A. Laverdant and D. Thevenin.63
coherent field is determined by a competition between these driving and damp-

ing processes. Although not reproduced here, example calculations are given in
Ref. 62.
This damping mechanism is primarily kinematic in nature, as the phase of
the scattered waves differs from point to point along the flame front because
of differences in distance the wave travels before impinging on the flame and
reflecting. Phase mismatch between disturbances originating from different points
of the flame results in destructive interference between these different waves. This
distortion of the acoustic field by a wrinkled flame front can be clearly seen in the
computational result in Fig. 12.27 which plots the spatial pressure-field distribution
of an initially planar acoustic wave after impinging on a wrinkled flame.63
In general, the characteristics of the scattered field depend on the statistical
distribution of the flame front about its average position. In the limit where the
scales of flame wrinkling are much smaller than a wavelength, only the turbulent
flame-brush thickness is important and the coherent scattered field is damped by the
amount 1–2(kσ cos i )2 , where k = ω/c, σ , and i are the acoustic wavenumber,
flame-brush thickness, and relative angle between the incident wave and average
flame position. Besides the reduction in amplitude, the coherent field has a phase
offset relative to its smooth-surface value if the flame position is asymmetrically
distributed about its mean position. These expressions predict, then, that turbulent
flame effects grow with increases in frequency or turbulent flame-brush thickness.
The increase of energy in the incoherent field on frequency can be seen from
the data in Fig. 12.28. Because the total energy in the scattered field is limited by
the energy in the incident wave plus any small amplification from the flame, the
energy in the incoherent field saturates at high frequencies. The saturation at high
frequencies corresponds to situations in which no energy remains in the scattered
coherent field, and all resides in the incoherent sidebands.
Return to the opening question of this section regarding the generality of results
obtained from laminar studies to turbulent flames. Apparently, the answer to this
0
10
Incoherent Scattered Power
-1
10
-2
10
0.1 0.2 0.3 0.4 0.5
σ/λ
Fig. 12.28 Dependence of scattered incoherent power on ratio of flame-brush thick-

ness σ and acoustic wavelength λ. Adapted from Lieuwen.61
question lies in the value of the ratio of the disturbance wavelength, whether acous-
tic or convective, to the turbulent flame-brush thickness. For typical longitudinal
mode instabilities, the ratio of σ/λ is often very small, implying that wrinkled
flame effects provide only a small correction from laminar-flame analyses. At
these same frequencies, however, the convective wavelength could potentially be
of the same magnitude as σ . This potential similarity implies that the conclusions
of Sec. III.A.3 or III.A.4, where convective disturbances or flame wrinkles vary
over a convective scale, could be modified in the turbulent case. In the same way,
the response of flames to high-frequency acoustic waves, such as during screeching
instabilities, could also be substantially different than the response anticipated from
laminar-flame analyses.
6. Acoustic Field Interactions with Inherent Flame Instabilities

Even in the absence of acoustic oscillations, premixed flames are often un-
steady because of intrinsic instabilities. These instabilities are significant because
their interaction with externally imposed acoustic oscillations results in qualita-
tive changes in the flame’s dynamics. We briefly introduce these instabilities in
the following text; detailed discussions and analysis can be found in Williams77
or Clavin.49
We focus on three basic categories of intrinsic premixed-flame instabilities:
body force, hydrodynamic, and diffusive thermal. Not covered here are multistep
chemistry effects, which can also introduce additional instabilities of a purely
kinetic nature, as noted in Sec. III.B.1, and Saffman-type instabilities, which appear
in flames propagating through thin channels.
The body-force instability is analogous to the classical buoyant mechanism in
which a heavy fluid resting above a lighter one is destabilized by the action of
gravity. In the same way, flames propagating upward divide a higher- and lower-
density region and, thus, are unstable. Similar instabilities can be induced by
acceleration of the flame sheet, either through a variation of the burning velocity
354 T. C. LIEUWEN
or an externally imposed flow perturbation. As will be discussed further in the

next section, the latter mechanism plays an important role in certain acoustic–
flame interaction phenomena, in which the acceleration is provided by the acoustic
velocity field.
The hydrodynamic, or Darrius–Landau instability, has an underlying mechanism
that is purely fluid mechanic in nature. Any front dividing two gases of different
densities that propagates at a constant velocity normal to itself with respect to
the more dense gas is unstable for all wavelengths of perturbation.77 This mecha-
nism is caused by gas expansion across the flame which causes the incident-flow
streamlines to diverge and/or converge in front of a flame disturbance that is convex
and/or concave to the unburned gas. The resulting flow divergence or convergence
causes the flow to locally decelerate or accelerate, respectively, causing the distur-
bance to grow further. The dependence of the local burning velocity on the radius
of flame curvature stabilizes short-wavelength perturbations. Longer-wavelength
perturbations are stabilized by gravity for downward-propagating flames.
The diffusive-thermal instability is caused by the effect of flame-front curvature
on the diffusion rates of heat and reactive species. For example, a disturbance
that causes the front to bulge toward the unburned gas results in defocusing of
the conductive heat flux that heats the incoming mixture. In the same way, it
results in focusing the diffusive flux of the deficient reactant into the flame. If the
heat conductivity and limiting reactant diffusivity are equal (i.e., a unity Lewis
number, Le = D T /D M ), then these effects balance so that the burning velocity is
unaltered. For mixtures with Lewis numbers less than about unity, this mechanism
is destabilizing. In addition, in multiple reactant systems, variations in the relative
diffusion rates of reactants can introduce variations in mixture composition at the
flame, also causing instability.
Two key interactions of acoustic waves with flame instabilities have been
noted.35,64−69 These interactions are the stabilization of the Darrius–Landau in-
stability by acoustic perturbations and a new parametric instability that occurs at
large-velocity oscillations.
Both instabilities can be observed in flames propagating down a pipe filled
with reactive mixture. Photographs obtained by R. C. Aldredge of the resulting
sequence of flame characteristics are illustrated in Fig. 12.29. As the flame prop-
agates down the tube, it develops a cellular shape caused by the Darrieus–Landau
instability (see top image in Fig. 12.29). Self-excited acoustic oscillations at the
natural frequency of the tube, caused by interaction of the acoustic wave with
the flame front, may also appear. These oscillations grow and can result in a re-
markable restabilization of the flame front, where the cellular structure disappears
and the flame reverts to a nearly planar front (see middle image in Fig. 12.29).
Measurements indicate that the flame’s propagation speed slows down substan-
tially because of the reduction in surface area and has a value that is close to the
laminar-burning velocity.64 In addition, the growth rate of the oscillations declines
markedly. Analysis indicates that this behavior is caused by stabilization of the
Darrieus–Landau flame instability by the oscillatory acceleration imposed by the
acoustic field.66,69
If the acoustic amplitude grows further, a violent secondary instability can occur
(see bottom image in Fig. 12.29). The nearly planar flame develops small, pulsating
cellular structures whose amplitude increases rapidly. These cellular structures
Fig. 12.29 Sequence of flame-front characteristics as it propagates downward in a

tube.68 Successive images show flame wrinkled by Darrieus–Landau instability mech-
anism (top), “planarization” of flame by low-amplitude velocity oscillations (middle),
and parametric instability induced by large-amplitude acoustic oscillations (bottom).
Images courtesy of R. C. Aldredge.
oscillate at half the period of the acoustic oscillations. This parametric acoustic
instability is caused by the periodic acceleration of the flame front by the unsteady
velocity field, which separates two regions of differing densities. With increased
amplitudes, these organized cellular structures break down into a highly disordered,
turbulent front. In the case in which the ambient flowfield is highly turbulent, Vaezi
and Aldredge70 found that the parametric instability still appeared. In addition, they
noted that for sufficiently high-turbulence levels, the appearance of the parametric
instability did not result in additional acceleration of the flame front. This finding
is in contrast with the case in which the ambient flowfield is quiescent, where the
parametric instability results in substantial flame acceleration.
Markstein first recognized that the period-doubling behavior occurring during
the parametric instability was indicative of a parametrically pumped oscillator, in
which the parametric excitation is caused by the oscillatory acceleration field. The
flame-front dynamics can be described by a parametric oscillator equation of the
356 T. C. LIEUWEN
form66
d2 y(k, t) dy(k, t)
A +B + [Co − C1 cos(wt)]y(k, t) = 0 (12.50)
dt 2 dt
where A, B, and C are coefficients defined in Ref. 66, k is the wave number
of the perturbation, and ω is the frequency of imposed oscillation. The damping
coefficient, B, is always positive, whereas the coefficient Co is negative if the
planar flame front is nominally unstable. In the case in which Co is negative,
this equation has the properties that the solution is unstable in the absence of
imposed oscillations (i.e., C1 = 0), is stabilized in the presence of small but finite
amplitude perturbations, and is destabilized in the presence of large-amplitude
parametric oscillations.
7. Flame Anchoring, Flashback, and Extinction

The dynamics of the flame-attachment point have a significant impact on the
overall flame kinematics. This point can be appreciated by the discussion in
Sec. III.A.3, which noted that the solution for the flame area was controlled by the
superposition of the boundary condition (i.e., the flame-attachment condition) and
flow–flame speed nonuniformity. The amplitude of flame disturbances generated
at its attachment point are directly affected by the extent to which this point does
or does not move in phase with the gas-particle velocity.
The dynamics of the flame-attachment point in an oscillatory flowfield is not
well understood. Most studies of laminar-flame dynamics have assumed that the
flame base remains motionless (i.e., ζ attachment point (t) = 0). The good agreement
between these models and experiments provides an indirect indication that such
an assumption is reasonable. It is not clear, however, what the appropriate boundary
condition is for a flame that is not attached at a fixed geometric point, such as in
a swirling flame that stands off from the burner and attaches at a flow-stagnation
point.
The flame-anchoring boundary condition is known to be amplitude and fre-
quency dependent at high forcing amplitudes. Baillot52 found that laminar, conical
Bunsen flames subjected to high-amplitude, low-frequency velocity perturbations
exhibited a variety of transient flame-holding behavior, such as flashback, asym-
metric blowoff, unsteady lifting, and reanchoring of the flame. In addition, they
noted that its response is asymmetric and extremely disordered. However, at high
frequencies and forcing amplitudes, the flame remains firmly attached, but its over-
all shape dramatically changes. They found that the flame becomes collapsed with
a rounded-off tip region, and for sufficiently high-forcing intensities (u /ū > 1),
the flame’s mean shape becomes hemispherical.54
Unsteady flashback phenomena, which occur when the instantaneous-flow ve-
locity falls below the flame speed, can be captured within the flame-sheet approach
considered here. For example, Dowling71 modeled this with a nonlinear boundary
condition which assumed that the flame remained anchored (i.e., ζ attachment point (t) =
0) and propagated upstream when the total gas velocity exceeded and fell below,
respectively, the flame speed.
Flow oscillations and the resulting oscillatory strain rates also cause local or
global flame extinction. Such extinction events in unsteady flows have been demon-
strated in several studies in fundamental geometries, such as counterflow flames.72
In addition, the flame may extinguish at locations of sharp curvature, such as in
cusps. Such extinction phenomena are routinely observed in turbulent flames.
Unsteady extinction and reignition of local or global regions of the flame intro-
duces nonlinearity in the acoustic–heat-release relationship. This point was em-
phasized in the measurements of the growth of the instability amplitude by Poinsot
et al.,73 who observed that the point of saturation of its amplitude corresponded
to the point where the instantaneous chemiluminescence emissions reached zero
during part of a cycle. Modeling extinction requires treating the internal flame
structure, and is considerably more involved than the simple front-tracking ap-
proach detailed in this chapter. For this reason, existing models have handled
extinction in a phenomenological manner. For example, Dowling74 observed that
the instantaneous heat release cannot go negative, thus limiting its magnitude to
100% of its mean value. She incorporated these observations into a phenomeno-
logical model for the finite amplitude response of a flame to velocity perturbations
in which heat-release saturation occurred at amplitudes at which the instantaneous
heat-release value went to zero.
B. Well-Stirred Reactors
1. Basic Analytical Framework
Returning to the combustion regime diagram in Fig. 12.2, consider next the
opposite extreme to the flamelet regime, the “well-stirred reactor” (WSR) regime.
It has been suggested that flame–acoustic interactions in this regime can be modeled
by generalizing the steady WSR equations to include nonsteady effects. Unsteady
reactor models are also routinely used to study kinetically driven instabilities
in multistep chemical mechanisms and extinction and ignition phenomenon.75,76
These unsteady reactor equations can be derived from a straightforward spatial
integration of the conservation equations over the WSR region by assuming that
all spatial quantities are uniform77 :
dM
= ṁ in − ṁ (12.51)
dt
dE
= ṁ in h in − ṁh (12.52)
dt
dMk
= ṁ in Yk,in − ṁYk − Ẇk (12.53)
dt
where ṁ and Ẇk are the mass flow rate and consumption rate of the kth species,
respectively. M, E, and Mk denote the total mass, total energy, and total mass
of the kth species in the reactor, respectively. The subscript in denotes the inlet
value. The steady-state characteristics of the well-stirred reactor are controlled by
the ratio of the chemical kinetic time to the reactor residence time, given by the
ratio of the mass flow rate and reactor volume, τres = ṁ/V . The reactor volume or
residence time cannot, in general, be specified by simplified analysis, because it
358 T. C. LIEUWEN
is determined by reaction and mixing rates; prior studies have used experimental
and computational analysis to determine these quantities, which are then used as
inputs to simplified models.78
Note that the preceding equations assume that the perturbations are spatially
uniform. If the flame zone is acoustically compact, such an approximation may
be adequate to describe acoustic perturbations. Entropy and vorticity fluctuations
could potentially be of much shorter length scales than the reactor size, how-
ever, indicating that the perturbation variables and flow-strain field are spatially
distributed in the reactor.
The total heat release from the reactor is given by the volume integral

Q(t) = ω̇h r dV (12.54)
v
where ω̇ and V denote the volumetric reactant consumption rate and reactor vol-
ume. This expression is analogous to that of Eq. (12.7), except it is on a volumetric,
rather than surface-area basis. The terms ρ 1 S1 and d A are replaced by ω̇ and d V ,
respectively. Note that heat-release fluctuations are generated by reaction rate,
reactor volume, and heat of reaction fluctuations.
2. WSR Response to Flow Perturbations

Consider first the factors affecting the reaction rate ω̇. Some understanding of
the sensitivity of ω̇ to flow and mixture perturbations can be gained from the
following generic global reaction-rate expression:
b c −E u /RT
ω̇ = AY af Yox p e (12.55)
where A is a pre-exponential factor, Y F and Yox denote fuel and oxidizer mass
fractions, and a, b, and c denote sensitivity coefficients. Whereas sensitivity of ω̇
to disturbances in any of the quantities in Eq. (12.55) can be determined from the
value of the exponential coefficients, it must be emphasized that any disturbance
does not occur in isolation. For example, fluctuations in temperature have an impact
on the residence time or reactor fuel and oxidizer concentration. The sensitivities
of ω̇ to perturbations in pressure, mass flow rate, inlet temperature, and equivalence
ratio have been discussed in Refs. 79, 80, and 81, whose results are summarized
subsequently.
Because the sensitivity of the Arrhenius term e−Eu /RT to temperature varia-
tions grows with temperature, the effect of temperature perturbations in the inlet-
reactant stream on ω̇ grow with increases in mean reactor temperature, such as
with increases in mean equivalence ratio. Opposite sensitivity is obtained with
equivalence-ratio oscillations, whose effect on ω̇ oscillations grows with decreas-
ing equivalence ratio. This point is analogous to the dependence of the flame speed
on the equivalence ratio discussed in Sec. III.A.4 and can be understood by noting
that ω̇ reaches a maximum near φ = 1 (i.e., it has no sensitivity to φ perturba-
tions) and decreases as φ becomes leaner. Both of these sensitivities were deter-
mined assuming that the reactor residence time was fixed; that is, oscillations in
reactive-mixture composition or temperature did not affect the residence time.
Blowout
Volumetric Heat Release Rate

(minimum residence
time)
Maximum
Reaction Rate
Decreasing
Residence Time
Reactor Temperature
Fig. 12.30 Dependence of rate of reaction (solid line) and convection (dashed line) on
reactor temperature. Steady-state reactor solution occurs at high-temperature inter-
section of two points. After Park et al.82
The effects of neglecting these variations could be substantial, however, as will be

discussed subsequently.
Also, although not explicitly shown in Eq (12.55), the reactor residence time has
a strong impact on reaction rate. Decreases in residence time lower the percentage
of fuel and oxidizer reacted, reducing the temperature. This sensitivity was ana-
lyzed by Park et al.82,83 They show that the phase and gain relation between reactor
residence time and heat-release oscillations qualitatively changes above and below
the point of maximum reactor heat release. They present an intuitive method of
explaining this result (see Fig. 12.30). The steady-state reactor conditions are de-
termined by the point where the rate of heat release by chemical reaction equals the
net rate of convection of energy out of the reactor. The dependence of these two
rates on reactor temperature are indicated by the solid and dashed lines in the
figure, respectively. Energy convection rate curves are drawn at several residence
times. Note that maximum reaction rate occurs at a certain value of residence time.
Consider the effect of small residence-time perturbations at mean residence
times above and below this maximum value. As indicated in the figure, these
perturbations result in reaction-rate oscillations that are out of and in phase with
the residence time and perturbation, respectively. Thus, as the combustion process
approaches the blowout point, it will pass through this point of phase reversal.
In addition, starting at the maximum reactor temperature, note that the sensitivity
of reaction-rate oscillations to residence-time perturbations decreases with reactor
temperature up to this maximum reaction-rate point where it becomes zero. Further
reductions in reactor temperature are accompanied by a corresponding rise in
sensitivity all the way to the blowout point.
Returning to Eq. (12.55), consider the effect of variations in reactor volume
and heating value. The heating value is a function of reactant composition and,
thus, is affected by equivalence ratio oscillations. The direct sensitivity of Q(t) to
the reactor volume V is straightforward, the two being linearly related. However,
remember that perturbations in either quantity affect the reaction rate and, in the
case of hr , the reactor volume as well (i.e., higher heat content reactant may
result in faster kinetics, reducing the reactor volume).
360 T. C. LIEUWEN
To extend the reactor approach to situations where the flame was convectively
noncompact and, thus, flow disturbances varied substantially over the flame region,
Lieuwen et al.84 treated the combustion zone as a distribution of infinitesimal, inde-
pendent reactors whose input conditions were given by that of the local flow at the
associated location. Although this heuristic treatment allowed for a consideration
of important noncompactness effects, its basic assumption of reactor independence
is questionable (e.g., as noted in Sec. III.A.3, disturbances generated at one point of
a flamelet convect downstream and affect its dynamics at other points). However,
as is the more general problem with these reactor-based approaches, it is not clear
how to incorporate these interaction effects in a rational manner.
Incorporating finite amplitude effects into unsteady, well-stirred reactor calcula-
tions is straightforward, although it may require numerics for time stepping through
Eqs (12.51–12.53).79,81 The WSR equations are often used for model problems in
nonlinear dynamics studies because of the complex, even chaotic dynamics85 they
exhibit.
3. Conceptual Problems with Current WSR Models

Two major conceptual issues associated with reactor models should be empha-
sized. First, reactor models were proposed for unsteady combustion systems based
on their utility in correlating certain steady-state combustion characteristics, such
as blowout conditions or pollutant emissions. This does not necessarily imply that
they are useful for predicting its dynamic characteristics for the following reason.
It is likely that the recirculation regions that stabilize many high-intensity flames
have distributed reactorlike properties; hence, the success in reactor models in cor-
relating blowout behavior. However, in many cases, it is also possible that other
parts of the flame have “flameletlike” properties. In these situations, a model that
is only valid in a small region of the combustion process may very accurately
describe its blowoff characteristics but not its other unsteady dynamics.
Second, it is difficult to rationally model the interactions between separate re-
actorlike regions in space that see different disturbance values (such as mixture
composition), and the interdependence between reaction rate and reactor residence
time. This second point seems particularly severe, as can be illustrated by the fol-
lowing points. Consider two identical reactors fed by the same fuel flow rates,
but at different pressures, p1 and p2 . Assuming that all the fuel is reacted to form
products, it is clear that the total heat release of both reactors is also the same.
Now, assume that the pressure in either reactor oscillates in time between the two
values; that is, p(t) = p1 + ( p2 − p1 )sin ωt. Assuming that the frequency is low
enough, the reactor will respond in a quasi-steady manner, implying that the to-
tal rate of heat release remains constant. A similar argument can be made for
a reactor disturbed by other fluctuations, such as temperature. Only fluctuations
in the heat content of the inlet fuel stream will cause a quasi-steady fluctuation
in reactor heat release. So what is happening? Clearly, the changes in pressure
or temperature influence the reaction rate ω̇. However, in the quasi-steady case,
increases in reaction rate must be accompanied by reductions in overall reaction
volume; that is, the same amount of heat is released but over a smaller volume.
This discussion shows that quasi-steady perturbations that do not affect the heat-
ing content of the inlet stream do not introduce fluctuations in overall heat release.
Thus, any fluctuation in heat release that does occur is a dynamic effect; that is, the
pressure perturbations referred to previously could potentially cause heat-release
oscillations at sufficiently large frequency ω. In this case, it is then necessary to
model the dynamics of the global reaction-zone response to the perturbation. This
situation is analogous to that encountered in flamelet studies (see Sec. III.A.4), in
which flame speed and area fluctuations are coupled; for example, quasi-steady
fluctuations in flame speed do not cause the global heat release to oscillate be-
cause of the accompanying oscillations in flame area. The difference is that these
coupled dynamics can be reasonably modeled in the flamelet case from first prin-
ciples. It is not clear how to couple these dynamics in WSR models, given their
phenomenological nature.
IV. Conclusion
The ultimate goal of this work is to develop models that can predict the qualita-
tive, and preferably quantitative, dependence of flame response in realistic com-
bustors on geometric and fuel composition parameters. Reasonable, quantitative
predictions of flame–acoustic interaction phenomena have been demonstrated
for a few simple configurations, such as the laminar Bunsen burner of Ducruix
et al.36 or Baillot et al.,53 or the nominally flat flame of Searby and Rochwerger.66
These successes illustrate the rapid progress being made in modeling kinematic
processes in flame–acoustic interactions. In addition, progress is being made in
developing hybrid models that use computational simulations to determine various
components of the combustor system–flame interactions.86 The development of
accurate, predictive combustion-response models for realistic (i.e., turbulent)
configurations has not been achieved, however, and remains a key challenge.
The subsequent discussion suggests some requirements needed to achieve this
capability.
First and most generally, it seems critical that better coordination between mod-
els and experiments be achieved. At present, a significant part of the relevant liter-
ature consists of essentially decoupled theoretical models or experimental studies,
even in rather fundamental configurations. For example, a substantial number of
fundamental studies have theoretically investigated the response of flat, laminar
flames to pressure perturbations.42−47 No serious effort appears to have been ini-
tiated to subject these predictions to experimental scrutiny. Similarly, although
equivalence-ratio oscillations are known to play an important role in exciting heat-
release oscillations, no experimental work has been performed to examine the
accuracy of models that relate them to the resultant heat-release oscillations. Even
though these highly fundamental studies may be far removed from practical flames,
they are prerequisite building blocks toward modeling realistic systems.
Second, work is needed to develop simplified models of vortex–flame interac-
tions. The existing theoretical work on this subject is largely numeric. Analytical
methodologies for modeling unstable, reacting shear flows have been developed87
and need to be extended to incorporate the unsteady flow effects on the flame.
Third, predicting the response of flames to finite amplitude waves is immature.
Substantial progress could be made by a set of experiments that isolate the key non-
linear processes that modelers need to focused on. Interpretive guidance of these
results can be achieved by parallel systematic studies of potential nonlinearities.
362 T. C. LIEUWEN
In addition, effects such as the stabilization or parametric destabilization of flames

discussed in Sec. III.A.6 may cause finite amplitude acoustic oscillations to sub-
stantially change the “mean” characteristics of the turbulent flame with which it
is interacting. These effects merit further investigation.
Fourth, flame–acoustic wave interactions in realistic environments occur in a
very noisy atmosphere where the flame is a highly perturbed front, even in the
absence of coherent acoustic oscillations, and executes large oscillations about
its mean position. Any model of the response of laminar-flame fronts to velocity,
equivalence ratio, or vortical disturbances needs to be generalized to include the
fact that the average flame is highly unsteady. For example, the successful work
performed to date on laminar, Bunsen flames should be extended to turbulent-flow
situations. Fundamental issues, such as how the transfer functions measured by
Ducruix et al.36 change with increasing turbulence levels, need to be addressed.
Fifth, the interactions of flames with thickened flamelets, distributed reaction
zones, or well-stirred reactions needs to be considered. As emphasized previously,
current well-stirred reactor models are largely phenomenological and have many
significant conceptual problems. Progress in this area will require clarification of
the nature of the combustion process in this regime by the turbulent-combustion
community.
Acknowledgments
This work was supported by the National Science Foundation, General Electric,
the U.S. Department of Energy, and Georgia Institute of Technology.
References
1
De Luca, L., Price, E., and Summerfield, M., Nonsteady Burning and Combustion Sta-
bility of Solid Propellants, Progress in Aeronautics and Astronautics, Vol. 143, AIAA,
Washington, DC, 1992.
2
Sirignano, W., Fluid Dynamics and Transport of Droplets and Sprays, Cambridge Univ.
Press, Cambridge, England, U.K., 1999.
3
Harrje, D., Reardon, F. (eds.) Liquid Propellant Rocket Instability, SP-194, NASA, 1972.
4
Peters, N., “Laminar Flamelet Concepts in Turbulent Combustion,” Proceedings of the
5
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, Edwards,
Philadelphia, PA, 2001.
6
Turns, S., An Introduction to Combustion, McGraw–Hill, New York, 2000.
7
Chu, B. T., and Kovasnay, L. S. G., “Nonlinear Interactions in a Viscous, Heat Conduct-
ing; Compressible Gas,” Journal of Fluid Mechanics, Vol. 3, 1958, pp. 494–514.
8
Kovaszay, L. S. G., “Turbulence in Supersonic Flow,” Journal of the Aeronautical Sci-
ences, Vol. 20, No. 10, 1953, pp. 657–674.
9
Jou, W. H., and Menon, S., “Modes of Oscillation in a Nonreacting Ramjet Combustor
Flow,” Journal of Propulsion and Power, Vol. 6, No. 5, 1990, pp. 535–543.
10
Lee, D. H., and Lieuwen, T., “Acoustic Nearfield Characteristics of a Conical, Premixed
Flame,” Journal of the Acoustical Society of America, Vol. 113, No. 1, 2003, pp. 167–177.
11
Ferguson, D., Richards, G., Woodruff, S., Bernal, S., and Gautam, M., Proceedings of
the 2nd Joint Meeting U.S. Sections of the Combustion Institute, Pittsburgh, PA, 2001.
12
Schadow, K., and Gutmark, E., “Combustion Instability Related to Vortex Shedding
Science, Vol. 18, 1992, pp. 117–132.
13
Coats, C., “Coherent Structures in Combustion,” Progress in Energy and Combustion
Science, Vol. 22, 1996, pp. 427–509.
14
Stone, C., and Menon, C., “Swirl Control of Combustion Instabilities in a Gas Turbine
Combustor,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 29, 2002.
15
Ho, C. H., Nosseir, N. S., “Dynamics of an Impinging Jet. Part 1. The Feedback
Phenomenon,” Journal of Fluid Mechanics, Vol. 105, 1981, pp. 119–142.
16
Michalke, A., Zeitschrift für Flugwissenschaften (in German), Vol. 19, 1971.
17
Baillot, F., Durox, D., and Prud’homme, R., “Experimental and Theoretical Study of a
Premixed Vibrating Flame,” Combustion and Flame, Vol. 88, 1992, pp. 149–168.
18
Durox, D., Schuller, T., and Candel, S., “Combustion Dynamics of Inverted Conical
Flames,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 30, 2004.
19
Strahle, W., “On Combustion Generated Noise,” Journal of Fluid Mechanics, Vol. 49,
No. 2, 1971.
20
Putnam, A. A, “Combustion Roar of Seven Industrial Burners,” Journal of the Institute
of Fuel, Vol. 49, 1976, pp. 135–138.
21
Marble, F., and Candel, S., “Acoustic Disturbance from Gas Non-uniformity Convected
Through a Nozzle,” Journal of Sound and Vibration, Vol. 55, 1977, pp. 225–243.
22
Lieuwen, T., “Theoretical Investigation of Unsteady Flow Interactions With a Premixed
Planar Flame,” Journal of Fluid Mechanics, Vol. 435, 2001, pp. 289–303.
23
Emmons, H. W., “Flow Discontinuities Associated with Combustion,” Fundamentals
of Gas Dynamics, Vol. III. High Speed Aerodynamics and Jet Propulsion, edited by H. W.,
Emmons, Princeton Univ. Press, Princeton, NJ, 1958, p. 584.
24
Markstein, G. H., Nonsteady Flame Propagation, Pergamon Press, New York, 1964.
25
Marble, F. E., and Candel, S. M., “An Analytical Study of the Non-Steady Behavior
of Large Combustors,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 17,
1978, pp. 761–769.
26
Subbaiah, M. V., “Nonsteady Flame Spreading in Two Dimensional Ducts,” AIAA
Journal, Vol., 21, No. 11, 1983, pp. 1557–1564.
27
Poinsot, T., and Candel, S. M., “A Nonlinear Model for Ducted Flame Combustion
Instabilities,” Combustion Science and Technology, Vol. 61, 1988.
28
Yang, V., and Culick F. E. C., “Analysis of Low Frequency Combustion Instabilities in
a Laboratory Ramjet Combustor,” Combustion Science and Technology, Vol. 45, pp. 1–25.
29
Boyer, L., and Quinard, J., “On the Dynamics of Anchored Flames,” Combustion and
Flame, Vol. 82, 1990, pp. 51–65.
30
Fleifel, M., Annaswamy, A. M., Ghoniem, Z. A., and Ghoniem, A. F., “Response of a
Laminar Premixed Flame to Flow Oscillations: A Kinematic Model and Thermoacoustic
Instability Results,” Combustion and Flame, Vol. 106, 1996, pp. 487–510.
31
Ashurst, W., and Sivashinsky, G., “On Flame Propagation through Periodic Flow
Fields,” Combustion Science and Technology, Vol. 80, 1991, pp. 159–164.
32
Joulin, G., Sivashinsky, G., Pockets in Premixed Flames and Combustion Rate, Com-
bustion Science and Technology, Vol. 77, 1991, pp. 329–335.
33
Aldredge, R., “The Propagation of Wrinkled, Premixed Flames in Spatially Periodic
Shear Flow,” Combustion and Flame, Vol. 90, No. 2, 1992, pp. 121–133.
34
Matalon, M., and Matkowsky, B., “Flames as Gas Dynamic Discontinuities,” Journal
of Fluid Mechanics, Vol. 124, 1982, pp. 239–260.
364 T. C. LIEUWEN
35
Clanet, C., Searby, G., and Clavin, P., “Primary Acoustic Instability of Flames Propa-
gating in Tubes: Cases of Spray and Premixed Combustion,” Journal of Fluid Mechanics,
Vol. 385, 1999, pp. 157–197.
36
Ducruix, S., Durox, D., and Candel, S., “Theoretical and Experimental Determinations
of the Transfer Function of a Laminar Premixed Flame,” Proceeding of the Combustion
Institute, Vol. 28, 2000.
37
Blackshear, P., “Driving Standing Waves by Heat Addition,” Proceedings of Combus-
tion Symposium, Vol. 4, 1953, pp. 553–566.
38
Mason, N., “Contribution to the Hydrodynamical Theory of Flame Vibration,” Pro-
ceedings of the Seventh International Congress on Applied Mechanics, Vol. 2, 1948,
pp. 187–199.
39
Chu, B. T., “On the Generation of Pressure Waves at a Plane Flame Front,” Proceedings
40
Schuller, T., Durox, D., and Candel, S., “A Unified Model for the Prediction of Lam-
inar Flame Transfer Functions: Comparisons between Conical and V-Flame Dynamics,”
Combustion and Flame, Vol. 134, 2003, pp. 21–34.
41
Preetham, and Lieuwen, T., “Nonlinear Flame-Flow Transfer Function Calculations:
Flow Disturbance Celerity Effects,” AIAA Paper 2004–4035.
42
McIntosh, A. C., “Pressure Disturbances of Different Length Scales Interacting with
Conventional Flames,” Combustion Science and Technology, Vol. 75, 1991, pp. 287–309.
43
Peters, N., and Ludford, G. S. S., “The Effect of Pressure Variations on Premixed
Flames,” Combustion Science and Technology, Vol. 34, 1983, pp. 331–344.
44
Van Harten, A., Kapila, A., and Matkowsky, B. J., “Acoustic Coupling of Flames,”
SIAM Journal of Applied Mathematics, Vol. 44, No. 5, 1984, pp. 982–995.
45
Keller, D., and Peters, N., “Transient Pressure Effects in the Evolution Equation for
Premixed Flame Fronts,” Theoretical and Computational Fluid Dynamics, Vol. 6, 1994,
pp. 141–159.
46
Ledder, G., and Kapila, A. K., “The Response of Premixed Flames to Pressure Pertur-
bations,” Combustion Science and Technology, Vol. 76, 1991, pp. 21–44.
47
McIntosh, A. C., “Deflagration Fronts and Compressibility,” Philosophical Trans-
actions of the Royal Society of London, Series A: Mathematical and Physical Sciences,
Vol. 357, 1999, pp. 3523–3538.
48
Huang, Z., Bechtold, J., and Matalon, M., “Weakly Stretched Premixed Flames in
Oscillating Flows,” Combustion Theory Modeling, Vol. 2, 1998, pp. 115–133.
49
Clavin, P., “Dynamic Behavior of Premixed Flame Fronts in Laminar and Turbulent
Flows,” Progress in Energy and Combustion Science, Vol. 11, 1985.
50
Joulin, G., “On the Response of Premixed Flames to Time Dependent Stretch and
Curvature,” Combustion Science and Technology, Vol. 97, 1994, pp. 219–229.
51
Im, H. G., and Chen, J. H., “Effects of Flow Transients on the Burning Velocity
of Laminar Hydrogen/Air Premixed Flames,” Proceedings of the Combustion Institute,
Pittsburgh, PA, Vol. 28, 2000, pp. 1833–1840.
52
Bourehla, A., and Baillot, F., “Appearance and Stability of a Laminar Conical Premixed
Flame Subjected to an Acoustic Perturbation,” Combustion and Flame, Vol. 114, 1998,
pp. 303–318.
53
Baillot, F., Bourehla, A., and Durox, D., “The Characteristic Method and Cusped Flame
Fronts,” Combustion Science and Technology, Vol. 112, 1996, pp. 327–350.
54
Durox, D., Baillot, F., Searby, G., and Boyer, L., “On the Shape of Flames Under
Strong Acoustic Forcing: A Mean Flow Controlled by an Oscillating Flow,” Journal of
Fluid Mechanics, Vol. 350, 1997, pp. 295–310.
55
Sankaran, R., and Im, H., “Dynamic Flammability Limits of Methane/air Premixed
Flames with Mixture Composition Fluctuations,” Proceedings of the Combustion Institute,
Pittsburgh, PA, Vol. 29, 2002, pp. 77–84.
56
Cho, J. H., and Lieuwen, T., “Laminar Premixed Flame Response to Equivalence Ratio
Oscillations,” Combustion and Flame, Vol. 140, No. 1–2, pp. 116–129.
57
Abu-Off, G. M., and Cant, R. S., “Reaction Rate Modeling for Premixed Turbulent
Methane-air Flames,” Proceedings of the Joint Meeting of Spanish, Portuguese, Swedish
and British Sections of the Combustion Institute, Madeira, 1996.
58
Peracchio, A. A., and Proscia, W. M., “Nonlinear Heat Release/Acoustic Model
for Thermo-Acoustic Instability in Lean Premixed Combustors,” American Society of
Mechanical Engineers, Paper 98-GT-269.
59
Lieuwen, T., “Theory of High Frequency Acoustic Wave Scattering by Turbulent
Flames,” Combustion and Flame, Vol. 126, No. 1–2, 2001, pp. 1489–1505.
60
Lieuwen, T., “Analysis of Acoustic Wave Interactions with Turbulent Premixed
Flames,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 29, 2002.
61
Lieuwen, T., Neumeier, Y., and Rajaram, R., “Measurements of Incoherent Acoustic
Wave Scattering from Turbulent Premixed Flames,” Proceedings of the Combustion Insti-
tute, Pittsburgh, PA, Vol. 29, 2002.
62
Lieuwen T., and Cho., J. H., “Coherent Acoustic Wave Amplification/Damping by
Wrinkled Flames,” Journal of Sound and Vibration (to be published).
63
Laverdant, A., and Thevenin, D., “Interaction of a Gaussian Acoustic Wave with a
Turbulent Premixed Flame,” Combustion and Flame, Vol. 134, 2003, pp. 11–19.
64
Searby, G., “Acoustic Instability in Premixed Flames,” Combustion Science and Tech-
nology, Vol. 81, 1992, pp. 221–231.
65
Markstein, G., “Flames as Amplifiers of Fluid Mechanical Disturbances,” Proceedings
of the Sixth National Congress on Applied Mechanics, 1970, pp. 11–33.
66
Searby, G., and Rochwerger, D., “A Parametric Acoustic Instability in Premixed
Flames,” Journal of Fluid Mechanics, Vol. 231, 1991, pp. 529–543.
67
Pelce, P., and Rochwerger, D., “Vibratory Instability of Cellular Flames Propagating
in Tubes,” Journal of Fluid Mechanics, Vol. 239, 1992, pp. 293–307.
68
Vaezi, V., and Aldredge, R., “Laminar Flame Instabilities in a Taylor-Couette Com-
bustor,” Combustion and Flame, Vol. 121, 2000, pp. 356–366.
69
Bychkov, V., “Analytical Scalings for Flame Interaction with Sound Waves,” Physics
of Fluids, Vol. 11, No. 10, 1999, pp. 3168–3173.
70
Vaezi, V., and Aldredge, R. C., “Influences of Acoustic Instabilities on Turbulent-Flame
Propagation,” Experimental Thermal and Fluid Science, Vol. 20, 2000, pp. 162–169.
71
Vol. 394, 1999, pp. 51–72.
72
Sung, C. J., and Law, C. K., Combustion and Flame, Vol. 123, 2000, pp. 375–388.
73
Poinsot, T., Veynante, D., Bourienne, F., Candel, S., Esposito, E., and Surget, J., Pro-
ceedings of the Combustion Institute, Pittsburgh, PA, Vol. 22, 1988.
74
Dowling, A. P., “Nonlinear Self-Excited Oscillations of a Ducted Flame,” Journal of
75
Park, Y., and Vlachos, D., “Isothermal Chain-Branching, Reaction Exothermicity, and
Transport Interactions in the Stability of Methane/Air Mixtures,” Combustion and Flame,
Vol. 114, 1998, pp. 214–230.
76
Kalamatianos, S., Park, Y., and Vlachos, D., “Two-Parameter Continuation Algorithms
for Sensitivity Analysis, Parametric Dependence, Reduced Mechanisms, and Stability Cri-
teria of Ignition and Extinction,” Combustion and Flame, Vol. 112, 1998, pp. 45–61.
366 T. C. LIEUWEN
77
Williams, F., Combustion Theory, Addison Wesley, Redwood City, CA, 1985.
78
Sturgess, G., Hedman, P., Sloan, D., and Shouse, D., “Aspects of Flame Stability in a
Research Dump Combustor,” American Society of Mechanical Engineers, Paper 94-GT-49.
79
Richards, G. A., Morris, G. J., Shaw, D. W., Keely, S. A., and Welter, M. J., “Thermal
Pulse Combustion,” Combustion Science and Technology, Vol. 94, 1993, pp. 75–85.
80
Janus, M. C., and Richards, G., “Results of a Model for Premixed Combustion Oscil-
lations,” Proceedings of the 1996 AFRC International Symposium, 1996.
81
Lieuwen, T., Neumeier, Y., and Zinn, B. T., “The Role of Unmixedness and Chemical
Kinetics in Driving Combustion Instabilities in Lean Premixed Combustors,” Combustion
Science and Technology, Vol. 135, 1998, pp. 193–211.
82
Park, S., Annaswamy, A., and Ghoniem, A., “Heat Release Dynamics Modeling of
Kinetically Controlled Burning,” Combustion and Flame, Vol. 128, 2002, pp. 217–231.
83
Park, S., Annaswamy, A., and Ghoniem, A., “Dynamic Characteristics of Kinetically
Controlled Combustion and their Impact on Thermoacoustic Instability,” Proceedings of
ICDERS 2001.
84
85
Rhode, M. A., Rollins, R. W., Markworth, A. J., Edwards, K. D., Nguyen, K., Daw, C. S.,
Thomas, J. F., “Controlling Chaos in a Model of Thermal Pulse Combustion,” Applied
Physics, Vol. 78, No. 4, 1995, p. 2224.
86
Polifke, W., Poncet, A., Paschereit, C. O., Dobbeling, K., “Reconstruction of Acous-
tic Transfer Matrices by Instationary Computational Fluid Dynamics,” Journal of Sound
Vibration, Vol. 245, No. 3, 2001, pp. 483–510.
87
Wee, D., Park, S., Annaswamy, A., and Ghoniem, A., “Reduced Order Modeling of
Reacting Shear Flow,” AIAA Paper 2002-0478, 2002.
88
Bellows, B., and Lieuwen T., “Nonlinear Response of a Premixed Combustor to Forced
Acoustic Oscillations,” AIAA Paper 2004-0455.
IV. Modeling and Diagnostics
Chapter 13
Acoustic Analysis of Gas-Turbine Combustors
Ann P. Dowling∗ and Simon R. Stow†

University of Cambridge, Cambridge, England, United Kingdom
I. Introduction
C OMBUSTION instability has become a major issue for gas turbine manufac-
turers. Stricter emission regulations, in particular, on nitrogen oxides, have
led to the development of new combustion methods such as lean premixed, prevap-
orized (LPP) combustion to replace the traditional diffusion flame. However, LPP
combustion is much more liable to generate strong oscillations that can damage
equipment and limit operating conditions. In this chapter, methods to investigate
combustion instabilities are reviewed (see also Dowling and Stow1 ). The emphasis
is on gas-turbine applications and LPP combustion. The flow is modeled as a one-
dimensional mean with linear perturbations. Calculations are typically done in the
frequency domain. The techniques described lead to predictions for the frequencies
of oscillations and the susceptibility to instabilities in which linear disturbances
grow exponentially in time. Appropriate boundary conditions are discussed, as is
the change in the linearized flow across zones of heat addition and/or area change.
Many of the key concepts are first introduced by considering one-dimensional
perturbations. Later, higher-order modes, in particular, circumferential waves, are
introduced and modal coupling is discussed. The modeling of a simplified combus-
tion system, from compressor outlet to turbine inlet, is described, as is the potential
for acoustic absorbers to control the instability. The approaches are simple and fast
enough to be used at the design stage. The effect of nonlinearity is discussed along
with techniques for predicting the amplitude of the resulting limit cycles.
LPP gas-turbine combustors have the great advantage of very low nitrous oxide
(NOx) emissions, but they are susceptible to instability. These instabilities involve
coupling between the rate of combustion and acoustic waves in the combustor. Un-
steady combustion generates acoustic waves that alter the inlet flow rates of fuel
and air. At lean premixed conditions, this changed fuel–air ratio leads to significant
∗ Professor of Mechanical Engineering, Department of Engineering. Senior Member AIAA.
† Research Associate, Department of Engineering.
369
370 A. P. DOWLING AND S. R. STOW
unsteady combustion. If the phase relationship is suitable,2 self-excited oscillations

grow. Because acoustic waves play such a central role in this phenomenon, the
frequencies of the combustion oscillations tend to be close to the acoustic reso-
nance frequencies of the combustion system. Although the coupling between the
combustion and the acoustics modifies the frequencies of oscillation, under many
circumstances the shift in frequency is small. A complete analysis of this phe-
nomenon requires the capability to model and understand the acoustic modes of
the combustion system and to couple these to a flame model that describes the
unsteady combustion response to these acoustic disturbances. Although the drive
for low emissions has made gas-turbine combustors particularly susceptible to in-
stability, such oscillations have long been an issue for other combustion systems,
for example, rocket motors.3, 4
We start with the equations of motion and investigate the forms of linear dis-
turbances. In a region of uniform mean flow, these forms are found to consist
of acoustic, vortical, and entropic disturbances. We begin by investigating one-
dimensional disturbances, in which these linearized waves are functions of a single
spatial variable and time, propagating in a duct of uniform cross-sectional area.
After application of appropriate boundary conditions, the mode shape and reso-
nant frequencies are determined. The analysis is gradually developed by adding
incrementally various effects that characterize gas-turbine combustors. These ef-
fects include unsteady heat addition, mean temperature gradients, and a mean flow
velocity. We investigate how these effects alter the frequencies of oscillation and
the mode shapes. In this chapter we concentrate on an acoustic analysis of gas-
turbine combustors. The discussion of flame models is in Chapter 12, and, as an
illustrative example, we consider the dependence of unsteady heat release on the
fuel–air ratio, which is widely recognized as the major cause of instability in LPP
combustors. However, the techniques we describe could be used with any flame
model for other configurations.
The one-dimensional examples introduce many of the key concepts but need
extension to be applicable to annular combustors, in which the longest combus-
tor dimension can be its circumference. If the longest combustor dimension is
its circumference the lowest-resonance frequency is associated with modes that
propagate in the azimuthal direction. We, therefore, extend the modal analysis
to annular and cylindrical geometries. Then, the axial-phase speed of acoustic
waves is usually a function of frequency and some modes are cut off, decaying
exponentially with axial distance.
In a LPP combustor, the acoustics from the compressor exit to turbine entry
can influence the combustion instabilities. We note how this combustion system
can be represented by a series of annular and cylindrical ducts and describe how
these ducts can be joined to determine the resonance frequencies of the complex
system.5–10 When the geometry is no longer axisymmetric, modal coupling may
occur, and we describe the influence of modal coupling on the frequencies of
instability and the mode shape. Self-excited combustion oscillations occur when
the energy gained by an acoustic wave through its interaction with the unsteady
combustion exceeds the energy lost at the boundaries of the combustor. This insta-
bility can be eliminated if the dissipation of acoustic waves within the combustor
is increased sufficiently through the introduction of acoustic absorbers, such as
perforated plates and Helmholtz resonators. The effects of such passive dampers
are discussed, along with their design requirements.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 371
Finally, we note that, according to linear theory, oscillations grow or decay

exponentially with time. Nonlinearity soon becomes important for the growing
disturbances and leads to stable, finite amplitude limit cycles. We show how simple
theories, based on a describing-function approach, can predict the frequency and
amplitude of the limit cycle and highlight the important physics.
II. Linearized Equations of Motion

We will start from the full equations of motion and derive their linearized form.
For a compressible viscous fluid in the absence of external forces, conservation of
mass and momentum lead to the Navier–Stokes equations,
Dρ
∇ ·u=0
+ ρ∇ (13.1a)
Dt
Du ∂σi, j
ρ ∇p+
= −∇ ei (13.1b)
Dt ∂x j
where p is the pressure, ρ is the density, u is the velocity, and σi, j is the viscous
stress tensor. Here D/Dt is the material derivative ∂/∂t + u · ∇ and ei represents
the unit vector in the direction of coordinate i. For a perfect gas, we have the gas
law p = Rgas ρT , where T is the temperature, Rgas = c p − cv is the gas constant,
and c p and cv are the specific heats at constant pressure and volume, respectively.
The internal energy per unit mass e is equal to cv T , and the enthalpy h is c p T =
e + p/ρ. Conservation of energy gives the energy equation,

D 1 ∂
ρ e + u2 = −∇ ∇ · ( pu) + q + ∇ · (k∇∇T ) + (σi, j u i ) (13.2)
Dt 2 ∂x j
where k is the conductivity and q is the rate of heat added to the fluid per unit
volume. By using Eq. (13.1b), this can be written as
Dh Dp ∂u i
ρ = ∇ T ) + σi, j
+ q + ∇ · (k∇ (13.3)
Dt Dt ∂x j
We define entropy S by the thermodynamic relation Dh = T DS + (1/ρ) D p.

Hence, Eq. (13.3) gives that
DS ∂u i
ρT ∇ T ) + σi, j
= q + ∇ · (k∇ (13.4)
Dt ∂x j
showing that it is heat release, heat transfer, and viscous effects that lead to an
entropy increase for a material particle. Taking the curl of Eq. (13.1b) and using
Eq. (13.1a) gives an equation for the development of the vorticity, ξ = ∇ × u,

D ξ ξ 1 1 1 ∂σi, j
= · ∇ u + 3∇ρ × ∇ p + ∇ × ei (13.5)
Dt ρ ρ ρ ρ ρ ∂x j
The first term on the right-hand side describes how the stretching of vortex lines
intensifies the local vorticity, and the last term clearly represents generation of
vorticity by viscous effects. The second term shows that vorticity can be created
when the pressure gradient and density gradient are not aligned. An example of
this would be an acoustic pressure oscillation with a component normal to a flame
front (density gradient), so that, for instance, circumferential waves will generate
vorticity at combustion zones.
We will now assume inviscid flow (σi, j ≡ 0). We will also assume the fluid is an
ideal gas (i.e., in addition to being a perfect gas, it does not conduct heat), and we
take c p and cv to be constant. From the preceding definition of entropy, we find that
S = cv log( p/ρ γ ) (plus an arbitrary constant that we set to zero), where γ = c p /cv
is the ratio of specific heats. We take the flow to be composed of a steady uniform
mean flow (denoted by overbars) and a small perturbation (denoted by primes),
p(x, t) = p̄ + p (x, t) (13.6)
and similarly for the other flow variables. From Eqs. (13.1), (13.4), and (13.5), the
linearized equations for these perturbations are
D̄ρ
∇ · u = 0
+ ρ̄∇ (13.7a)
Dt
D̄u 1
+ ∇ p = 0 (13.7b)
Dt ρ̄
D̄S
ρ̄ T̄ = q (13.7c)
Dt
Dξξ
=0 (13.7d)
Dt
where D̄/Dt = ∂/∂t + ū · ∇ and we have used ξ̄ξ = 0. Combining Eqs. (13.7a–
13.7c) and using S = cv p / p̄ − c p ρ /ρ̄ = 0 leads to the inhomogeneous wave
equation,
1 D̄2 p 2 γ − 1 D̄q
− ∇ p = (13.8)
c̄2 Dt 2 c̄2 Dt
where c is the speed of sound. We see that the vorticity equation (13.7d) is not
coupled to either the pressure or the entropy. For no unsteady heat release, the
pressure equation (13.8) and entropy equation (13.7c) are also uncoupled. Any
perturbation can then be thought of as the sum of three types of disturbances11 : 1)
an acoustic disturbance that is isentropic and irrotational; 2) an entropy distur-
bance that is incompressible and irrotational; and 3) a vorticity disturbance that is
incompressible and isentropic. These three types of disturbances are independent
and can be considered separately. For the pressure (acoustic) disturbance, we have
S = 0 and ξ = 0; hence, ρ = p /c̄2 . Since q = 0, Eq. (13.8) becomes the wave
equation with convection for p ,

1 D̄2
− ∇2 p = 0 (13.9)
c̄2 Dt 2
and the corresponding u is given by Eq. (13.7b). Disturbances of this type are
acoustic waves; relative to the fluid, they propagate at the speed of sound. For
the entropic disturbance, p = 0 and u = 0. From Eq. (13.7c) we see that the
disturbance is stationary relative to the fluid, that is, it is convected with the mean
flow. This disturbance can be thought of as an entropy wave, and is sometimes
referred to as a convected hot spot. For the vortical disturbance, p = ρ = 0 and
∇ · u = 0, and Eq. (13.7d) shows that this type of disturbance (a vorticity wave)
is also convected with the mean flow. If the mean flow is zero, only acoustic
disturbances propagate.
A. Conditions Across a Flame Zone

We now consider the effect of a thin flame zone in the plane x = 0, where we
take the rate of heat release per unit area to be Q A . A discontinuity will occur in
the flow parameters across the flame; we denote conditions at x = 0− and x = 0+
by subscripts 1 and 2, respectively. From Eqs. (13.1) and (13.3), we find that
ρ2 u 2 = ρ1 u 1 (13.10a)
p2 + ρ2 u 22 = p1 + ρ1 u 21 (13.10b)
ρ2 u 2 H2 = ρ1 u 1 H1 + Q A (13.10c)
where H = h + 12 u 2 is the stagnation enthalpy. To calculate the mean flow, we

assume that Q̄ A is known (from knowledge of the fuel type, equivalence ratio, etc.).
A flame model is used to describe the dependence of Q A on the flow perturbations
(see Chap. 12).
B. Boundary Conditions
At the inlet and outlet of the combustion system, there are boundary conditions
that the perturbations must satisfy. If the outlet discharges into the atmosphere or a
large plenum chamber (as is often the case for combustor test rigs), we may model
this as an open end, taking p (r, θ, t) = 0. If the inlet is supplied by a plenum
chamber we may treat this also as an open end [ p (r, θ, t) = 0] and, in addition,
assume that no entropy or vorticity disturbances are present. The compressor exit
and turbine inlet of a gas turbine can be modeled as a choked inlet and choked
outlet, respectively, to the combustion system. The nozzle guide vanes at the entry
to the turbine are choked, that is, the mean flow velocity accelerates to the local
speed of sound. At the compressor exit of a gas-turbine combustor, the flow is
nearly choked (meaning that the mass and energy flow rates are nearly constant
irrespective of downstream pressure perturbations) and so a choked inlet boundary
condition provides an approximation. At a choked outlet, the nondimensional mass
flow rate (defined as the mass flow rate multiplied by the square root of stagnation
temperature and divided by the stagnation pressure) is constant and, for one-
dimensional perturbations, Marble and Candel12 showed that at a compact choked
outlet this condition reduces to
u ρ p
2 + − =0 (13.11)
ū ρ̄ p̄
Stow et al.13 have shown that this condition still applies for circumferential-varying
disturbances in a narrow annular gap (disturbances in narrow annular gap geome-
tries are discussed in Sec. IV.C).
For a compact choked inlet, Stow et al.13 considered the interaction of the shock
position and the flow perturbations (see also Yang and Culick14 and Culick and
Rogers15 ), finding that, for one-dimensional disturbances, the perturbations in mass
flux and energy flux are zero just after the shock and that, for circumferential-
varying disturbances in a narrow annular gap, the angular-velocity perturbation
is also zero. From conservation of mass, energy, and angular momentum, these
quantities are also zero at the start of a straight duct with a low-Mach-number
mean flow M̄ 1 just downstream of the choking plane. This gives the inlet boundary
conditions:
ρ u p ρ u
+ = − + (γ − 1) M̄ 1 = w = 0 (13.12)
ρ̄ ū p̄ ρ̄ ū
For a weak shock, one would expect that there is negligible entropy production.
However, the equations imply that the (usually ignored) entropy perturbation down-
stream of the inlet is in fact comparable with the acoustic oscillations. In a frame
of reference moving with the shock, the acoustic perturbations are indeed, much
larger than the entropy disturbance, but viewed in a stationary frame close to the
shock, the discrepancy is not as great. After an area increase to a low-Mach-number
region, the acoustic perturbations are smaller still and are then of the same order as
the entropy perturbations. For circumferential-varying disturbances, a significant
vorticity perturbation is also produced.
Other analytical inlet and outlet boundary conditions, such as acoustically closed
ends (u = 0) or semi-infinite (nonreflecting) pipes, can also be used. Alternatively,
the acoustic impedance of the inlet or outlet can be measuring experimentally by
using microphones and an acoustic source driven over a range of frequencies.
This approach is similar to the measurement of the transfer matrix for a premixer
discussed in Sec. V.B.
III. One-Dimensional Disturbances

A. Plane Wave Solutions
As an introductory example, let us first consider a duct with a uniform cross-
sectional area, a mean temperature, and a density with no mean flow, in which the
unsteady flow parameters are just functions of the axial space coordinate x and
time t. Then the general solution of the wave equation (13.9) can be written in the
form
p (x, t) = f (t − x/c̄) + g(t + x/c̄) (13.13)
where the functions f (t) and g(t) are arbitrary. From the one-dimensional form of
the linearized momentum equation (13.7b), the particle velocity in the x direction
is given by
∂u 1 ∂ p 1 ∂ 1 ∂
=− = f (t − x/c̄) − g(t + x/c̄) (13.14a)
∂t ρ̄ ∂ x ρ̄ c̄ ∂t ρ̄ c̄ ∂t
that is,
1
u (x, t) = ( f (t − x/c̄) − g(t + x/c̄)) (13.14b)
ρ̄ c̄
For perturbations of frequency ω, it is convenient to write f (t) = Re( f̂ eiωt ),

where the circumflex denotes a complex amplitude. With this notation
p̂(x) = f̂ e−iωx/c̄ + ĝeiωx/c̄ (13.15a)

1 −iωx/c̄
û(x) = f̂ e − ĝeiωx/c̄ (13.15b)
ρ̄ c̄
The resonant frequencies follow from application of appropriate boundary condi-

tions at the ends of the duct. For example, with a large plenum attached to the duct
end at x = 0 and a restriction at x = l, as illustrated in Fig. 13.1, the appropriate
boundary conditions are
p̂(0) = û(l) = 0 (13.16)
Equation (13.15a) then leads to ĝ = − f̂ , and it follows directly from Eq. (13.15b)
that

ωl
cos =0 (13.17a)
c̄
1 2
x=b
Fig. 13.1 Boundary conditions in the model problem.

with solutions

1 π c̄
ω = ωn = n − (13.17b)
2 l
for integer n 1. These are the resonant frequencies ωn of the duct, describing
the oscillations in which the pressure oscillates without decay. The corresponding
mode shapes are

(2n − 1)π x
p̂(x) = An sin (13.18a)
2l

iAn (2n − 1)π x
û(x) = cos (13.18b)
ρ̄ c̄ 2l
for an arbitrary constant An .
B. Unsteady Heat Addition

With heat addition at a rate q(x, t) per unit volume, the pressure perturbations
satisfy an inhomogeneous one-dimensional wave equation that follows from setting
ū = 0 in Eq. (13.8):
1 ∂ 2 p ∂ 2 p γ − 1 ∂q
− = (13.19)
c̄2 ∂t 2 ∂x2 c̄2 ∂t
The term on the right-hand side describes how the unsteady addition of heat gen-
erates pressure disturbances. For a specified rate of heat release q (x, t), this inho-
mogeneous wave equation could be solved to determine the resultant sound field.
However, combustion instabilities are caused by feedback when the rate of heat
release is affected by the flow perturbations it generates. We can illustrate these
effects through simple model problems.
We again consider a flow that satisfies the boundary conditions of Eq. (13.16), but
we now suppose that the rate of heat release responds to the flow in specified ways.
Example 1
Suppose that the rate of heat-release perturbation q (x, t) is influenced by the
local pressure but lags it by a time delay τ . It is convenient to write the constant
of proportionality as 2α/(γ − 1), that is,
2α
q (x, t) = p (x, t − τ ) (13.20)
γ −1
The form of the pressure perturbation can be determined by substituting for q (x, t)
in Eq. (13.19) and seeking a separable solution, p (x, t) = Re( p̂(x)eiωt ). This
substitution leads, after application of the boundary conditions, to p̂(x) of the
form given in Eq. (13.18), and the equation for the resonant frequency ω is
ω2 + 2iωαe−iωτ − ωn2 = 0 (13.21)
where ωn is defined in Eq. (13.17).

When α = 0, the roots of Eq. (13.21) are the undamped resonant organ-pipe
frequencies ωn .
When α = 0, τ = 0, the quadratic equation (13.21) for ω can be readily solved
to give
1/2
ω = −iα ∓ ωn2 − α 2 (13.22)
ω is now complex. Because the time dependence is eiωt , − Im(ω) is the growth rate
of the disturbances. Here eiωt = exp[αt ∓ i(ωn2 − α 2 )1/2 t], showing that the oscil-
lations grow exponentially in time if α is positive. We have recovered Rayleigh’s
criterion2 from this particular example. Unsteady heat release in phase with the
pressure perturbation has a destabilizing effect and tends to increase the amplitude
of the perturbations. In contrast, for negative α, that is, heat release in antiphase
with the pressure, the oscillations are damped.
When α = 0, τ = 0, Eq. (13.21) would, in general, need a numerical solution,
and some results are shown in Fig. 13.2. In Fig. 13.2 and subsequently, a normalized
frequency f N = Re(ω)/ω1 and a normalized growth rate g N = − Im(ω)/ω1 are
used. However, the general characteristics of the solution can be investigated by
considering small α, and determining the roots iteratively. We have already noted
1.03
normalized frequency, fN
1.02
1.01
0.99
0.98
0.97
0 0.2 0 .4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω1 τ / (2π)
0.02
normalized growth rate, gN
0.01
0.01
0.02
0.03
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω1 τ / (2π)
Fig. 13.2 Variation with τ of the root of Eq. (13.21) near ω1 : ———, α/ω1 = 0.01;
– – –, α/ω1 = 0.02; and –··–, α/ω1 =−0.01. a) Frequency. b) Growth rate.
that, for α = 0, a root of Eq. (13.21) is at ω = ωn . For small α, this root moves to
ω = ωn + ε, where ε is small; substitution into Eq. (13.21) shows that
ε = −iαeiωn τ = −iα cos(ωn τ ) − α sin(ωn τ ) (13.23)

From this we see that any α cos(ωn τ ) > 0 leads to a positive growth rate, that is,
any unsteady heat release with −π/2 < phase(q̂/ p̂) < π/2 is destabilizing. It is
also clear that the resonant frequency is shifted whenever α sin(ωn τ ) = 0. Rate of
heat release in quadrature (±90 deg) with the pressure alters the frequency, and
unsteady rate of heat release leading the pressure (+90 deg) tends to increase the
frequency, and reduces the frequency when it lags the pressure. This effect was
noted by Rayleigh.2 These analytical predictions for small α are confirmed by
the numerical results shown in Fig. 13.2. For α > 0, the growth rate is increased
for cos(ω1 τ ) > 0, that is, (2n − 12 )π < ω1 τ < (2n + 12 )π and decreased when
cos(ω1 τ ) is negative. Also the frequency is decreased for sin(ω1 τ ) > 0 and
increased for sin(ω1 τ ) < 0, and the behaviors are reversed for negative α. For
nonzero τ , Eq. (13.21) becomes transcendental and has additional solutions that
are primarily related to τ −1 rather than the downstream geometry. For example,
for small |α| these are at Im(ω) → ∞ and Re(ω) ∼ 2mπ/τ for negative α and
(2m + 1)π/τ for positive α, where m is an integer. These are the even and odd
harmonics for the convection time τ . The choice of even and odd comes from
a balance of the right-hand side and the first term on the left-hand side in Eq.
(13.19); the first term is much larger than the second term on the left-hand side,
which represents the axial variation and hence the effect of the geometry.
This simple example illustrates that combustion instability is a genuinely cou-
pled problem. Both the acoustics and the unsteady combustion must be considered.
The coupling between them affects both the frequency and the susceptibility to
self-excited oscillations. At certain conditions, linear perturbances are predicted
to grow exponentially with time. In practice, nonlinear effects, the most signifi-
cant of which is usually a saturation in the heat-release response,16 lead to a finite
amplitude limit-cycle oscillation.
However, this first example is an oversimplification of what occurs in practice.
In LPP gas turbines, it is not the unsteady pressure that has the greatest influence on
the rate of heat release. Rather, the rate of heat release is related to the instantaneous
fuel–air ratio, which is most affected by the velocity of the airstream near the fuel
bars. See Chapter 12 for a discussion of the main causes of unsteady combustion.
Moreover, the heat release tends to be localized rather than distributed throughout
the duct as in example 1. We can illustrate again the influence of these effects
through an example.
Example 2
We now consider that the unsteady heat release is concentrated at a single axial
plane x = b and is related to the oncoming air velocity there with a time delay τ ,
q (x, t) = Q (t)δ(x − b) (13.24a)

β ρ̄ c̄2
Q (t) = − u (t − τ ) (13.24b)
γ −1 1
where Q (t) is the rate of heat release per unit area and subscript 1 denotes condi-
tions just upstream of this region of heat release, that is, u 1 (t) = u(b− , t). Chap-
ter 12 discusses forms of the unsteady heat release. In this chapter, we note that the
nondimensional number β can be expected to range from 0 to about 10 and that
in a LPP system τ is typically the convection time from fuel injection to its com-
bustion. For simplicity, u 1 has been taken to be the velocity just upstream of the
flame. However, for consistency, with τ being the fuel-convection time, the flame
model should really be referenced to the perturbations at the fuel-injection point,
as is done in example 5. However, the distance between these points is typically
short compared with the wavelengths, so the phase difference between them will
be small, although they may differ in magnitude by the area ratios.
With the rate of heat release q (x, t) as given in Eq. (13.24a), Eq. (13.19) reduces
to the homogeneous wave equation in the regions x < b and x > b. Integration of
Eqs. (13.7b) and (13.19) across x = b gives
x=b+
p x=b− = 0 (13.25a)
x=b+
∂p γ − 1 dQ
=− 2 (13.25b)
∂ x x=b− c̄ dt
Equation (13.25b) is equivalent to
x=b+ γ −1
u x=b− = Q (t) (13.25c)
ρ̄ c̄2
relating the volumetric expansion to the instantaneous rate of heat release. After
substitution for the particular Q (t) in Eq. (13.24b), we obtain
u (b+ , t) = u (b− , t) − βu (b− , t − τ ) (13.26)
We will consider solutions with time dependence eiωt and want to find the resonant
frequencies ω and the mode shapes.
In x < b, the solution of the homogeneous wave equation that satisfies the inlet
boundary condition p̂(0) = 0 is
p̂(x) = A sin(kx) (13.27a)

i
û(x) = A cos(kx) (13.27b)
ρ̄ c̄
where k is the wave number ω/c̄ and the complex constant A has yet to be deter-
mined. Similarly, in x > b, the boundary condition û(l) = 0 leads to
p̂(x) = B cos(k(l − x)) (13.28a)

i
û(x) = B sin(k(l − x)) (13.28b)
ρ̄ c̄
1
08
06
04
02
0
0 01 02 03 04 05 06 07 08 09 1
β
Fig. 13.3 Variation of frequency with β for the root of Eq. (13.30) near ω1 , taking τ =
0, b=l/10: ——, exact solution; and – – –, one-term Galerkin approximation (13.39).
The pressure jump condition (13.25a) then gives
A sin(kb) = B cos(k(l − b)) (13.29)
whereas the velocity jump condition (13.26), on division by Eq. (13.29), gives
tan(kb) tan(k(l − b)) = 1 − βe−iωτ (13.30)
The resonant frequencies follow from a numerical solution of Eq. (13.30). Their
dependence on β and τ is shown in Figs. 13.3 and 13.4.
For β = 0, the roots are at ω = ωn . As β varies, for τ = 0, the rate of heat release
is in quadrature with the pressure perturbation [note the 90-deg phase difference
between p and u in Eq. (13.27)] and so shifts only the frequency of oscillation.
A time lag is required for the unsteady heat release to destabilize the system. For
0.1
0.05
0.05
0.1
0.15
0.2
0 0.2 0 .4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω1 τ / (2π)
Fig. 13.4 Variation of growth rate with τ for the root of Eq. (13.30) near ω1 , taking
b=l/10: ——, β= 0.2; – – –, β= 0.4; –··–, β= 0.6; and · · · ·, β= 0.8.
τ = 0, the unsteady heat release affects both the growth rate and the frequency
of oscillation. Perturbations grow in time if, in this undamped system, the rate of
heat release has a component in phase with pressure perturbation. It is clear from
the form of the heat release in Eq. (13.24b) and the mode shape in Eq. (13.28) that
this requires
−π < Re(ωτ ) − phase(β cot(kb)) < 0 (13.31)
These bands of instability are clearly seen in Fig. 13.4.

The mode shapes follow from the substitution for B from Eq. (13.29) into
Eq. (13.28) and have the form

C sin(kx)/ sin(kb) for 0 x b
p̂(x) = (13.32)
C cos(k(l − x))/ cos(k(l − b)) for b x l
where the constant C is arbitrary.
C. Galerkin Series
Another way of solving the inhomogeneous wave equation Eq. (13.19) is through
a Galerkin expansion. The Galerkin expansion involves, the expansion of the pres-
sure perturbation as a Galerkin series:
∞
p (x, t) = ηm (t)ψm (x) (13.33)
m=1
where the functions ψm (x) are the eigensolutions or normal modes of the homoge-
neous wave equation that satisfy the same boundary conditions as p . In general,
these functions are orthogonal, and we will denote their eigenfrequencies by ωm .
Substitution for the pressure perturbation from Eq. (13.33) into Eq. (13.19) then
leads to

∞ 2
d ηm ∂q
+ ω 2
η m ψ m (x) = (γ − 1) (13.34)
m=1
dt 2 m
∂t
After multiplication by ψn (x) and integration with respect to x, the orthogonality

of ψn (x) shows that Eq. (13.34) becomes

d 2 ηn γ −1 l
∂q
+ ωn2 ηn = ψn (x) dx n = 1, . . . , (13.35)
dt 2 En 0 ∂t
l
where E n = 0 ψn2 dx. Equation (13.35) is a complicated system of equations,

because q (x, t) is related to the local flow and so involves all the unknown coef-
ficients ηm (t).
To make the analysis tractable, it is usually assumed that ∂q /∂t is small in mag-
nitude and needs only to be evaluated approximately. The method is described by
Culick and Yang.17 When ∂q /∂t = 0, the nth mode is p̂(x) = ηn (t)ψn (x) with fre-
quency ωn . This acoustic approximation is used when evaluating ∂q /∂t, replacing
the pressure and velocity perturbations by ηn (t)ψn (x) and (η̇n (t)/ρ̄ωn2 ) dψn /dx,
respectively, where the dot denotes a time derivative. If the second derivatives
of the amplitudes arise, they are replaced by the zeroth-order approximation,
η̈n (t) ≈ −ωn2 ηn (t). The errors introduced by these approximations can be checked
by applying the method to find the lowest frequency of oscillation in example 2.
Example 2 by Galerkin Series

After applying Culick’s rules, the rate of heat release in Eq. (13.24) leads to
∂q β c̄2 dψ1
(x, t) = η1 (t − τ ) (b)δ(x − b) (13.36)
∂t γ −1 dx
and substitution into Eq. (13.35) gives
d2 η1 β c̄2 dψ1
2
+ ω 2
1 η 1 = η1 (t − τ ) (b)ψ1 (b) (13.37)
dt E1 dx
The solutions ψn of the homogeneous wave equation are given in Eq. (13.18) and
ψ1 (x) = sin(π x/2l), leading to E 1 = 12 l. Equation (13.37), therefore, simplifies
to

d2 η1 β c̄2 π πb πb
2
+ ω1 η1 = 2 η1 (t − τ ) cos
2
sin (13.38)
dt l 2l 2l
The frequency of oscillation ω can be found by substituting η1 (t) = Ceiωt into

Eq. (13.38) to give

β c̄2 π −iωτ πb
ω2 = ω12 − e sin (13.39)
2l 2 l
The root of this equation is shown as a dashed line in Fig. 13.3 for the particular
case τ = 0. Comparison with the exact solution given in Eq. (13.30) shows that
the one-term Galerkin expansion gives the correct frequency and gradient at β = 0
but that it rapidly diverges from the exact solution as β increases. The divergence
is not really surprising; this method treats the shift in frequency as small, but it can
be substantial for the type of combustion response typical of LPP systems. The
inadequacy of the one-term Galerkin for a more complicated model problem was
discussed by Dowling.18 Annaswamy et al.19 noted that three terms in the Galerkin
series were needed to model the system dynamics for feedback control.
D. Temperature Gradients
So far, our examples have been somewhat artificial; they have had an unsteady
heat release q (x, t) and yet the mean temperature has been uniform. In practice,
of course, heat release is associated with temperature gradients and the mean
temperature and density are functions of position. We will introduce these effects
through discussion of the zero-mean-flow case. Then the momentum equation
(13.1b) ensures that the mean pressure is uniform and for linearized perturbations,
∂u
ρ̄ ∇ p
= −∇ (13.40)
∂t
in an inviscid flow. We show in the Appendix that the mass conservation equation
(13.1a) and the entropy equation (13.4) can be combined to give
1 ∂ p γ −1
= ∇ · u + q (13.41)
ρ̄ c̄2 ∂t ρ̄ c̄2
when heat conduction and viscous effects are neglected. Eliminating u from Eqs.
(13.40) and (13.41), we obtain

1 ∂ 2 p 1 γ − 1 ∂q
∇·
− ρ̄∇ ∇p = (13.42)
c̄2 ∂t 2 ρ̄ c̄2 ∂t
In this equation, ρ̄ and c̄ vary with position, but ρ̄ c̄2 = γ p̄ is uniform if the small
dependence of γ on temperature is neglected. We can illustrate the influence of
temperature variation by extending example 2 to the case in which the mean
temperature rises from T̄1 to T̄2 across the zone of heat release at x = b, with
corresponding changes in sound speed and mean density.
Example 3
Consider one-dimensional linear disturbances of frequency ω in the system
illustrated in Fig. 13.5. Just as in example 2, we again apply the boundary conditions
(13.16) and the flame model (13.24).
T1 T2
ρ1 ρ2
c1 c2
x=b
Fig. 13.5 System for example 3.

Outside the flame zone x = b, the solutions of the homogeneous wave equa-
tion (13.42), satisfying the appropriate boundary conditions, have the same form
as in example 2 provided the local mean flow variables are used. Hence, using
Eq. (13.27) and Eq. (13.28), we can write in x < b
p̂(x) = A sin(k1 x) (13.43a)

i
û(x) = A cos(k1 x) (13.43b)
ρ̄ 1 c̄1
and in x > b
p̂(x) = B cos(k2 (l − x)) (13.43c)

i
û(x) = B sin(k2 (l − x)) (13.43d)
ρ̄ 2 c̄2
where k1 = ω/c̄1 and k2 = ω/c̄2 .

Integration of Eqs. (13.40) and (13.42) across the region x = b with q (x, t) =
Q (t)δ(x − b) leads to
x=b+
p x=b− = 0 (13.44a)
+
1 ∂ p x=b γ − 1 dQ
=− (13.44b)
ρ ∂ x x=b− ρ̄ 1 c̄21 dt
Equation (13.44b) is equivalent to
x=b+ γ −1
u x=b− = Q (t) (13.44c)
ρ̄ 1 c̄21
After substituting for the particular Q (t) in Eq. (13.24b) and using Eq. (13.43),
we obtain
ρ̄ 2 c¯2
tan(k1 b) tan(k2 (l − b)) = (1 − βe−iωτ ) (13.45)
ρ̄ 1 c¯1
A comparison with Eq. (13.30) shows that the varying temperature effects appear
not only in the wave numbers k1 and k2 , which account for propagation effects,
but also in the factor ρ̄ 2 c¯2 /(ρ̄ 1 c¯1 ), which describes the impedance change across
the flame zone. The solid line in Fig. 13.6 shows how the temperature variations
affect the resonant frequency. A typical LPP gas-turbine combustor operates with
a temperature ratio of about 3 (T̄1 ∼ 700 K, T̄2 ∼ 2000 K).
So far, we have assumed that the flame is compact, that is, axially short com-
pared with the wavelengths of the perturbations. If the flame is not compact, we
may approximate the axial heat-release distribution by discretizing into a series of
compact flames, each having the form described earlier. Between these series, there
2
1.8
1.6
1.4
1.2
1
1 1.5 2 2.5 3 3.5 4 4.5 5
T2 / T1
Fig. 13.6 Variation of frequency with mean temperature ratio for solution near ω1
taking β = τ = 0, b = l/10: ———, root of Eq. (13.45); – – –, uniformly distributed
heat release between x = 0 and x = 2 b approximated by 10 temperature jumps;
and –··–, the same but using 5 jumps.
is assumed to be no heat release and we use the usual wave propagation (13.15).
The dashed and dashed–dotted lines in Fig. 13.6 show results for applying this
approximation when, instead of a compact flame at x = b, we have a uniformly
distributed heat release between x = 0 and x = 2b. It leads to a 6% shift in the
frequency at a temperature ratio 3. An alternative approach is to seek a continuous
analytical solution. Exact solutions can be found for particular temperature distri-
butions (such as linear variations and power laws)20–25 and also for certain area
variations.26, 27
E. Mean Flow
Most combustion systems involve a mean flow that brings fresh reactants into
the combustion zone. The Mach number of the oncoming flow is so small (typically
less than 0.1) that it is tempting to neglect this mean velocity. The errors introduced
by such an approximation are investigated in this section.
A mean flow has two main consequences. Trivially, it affects the speed of prop-
agation of the acoustic waves, with one-dimensional disturbances then traveling
downstream with speed c̄ + ū and upstream at c̄ − ū. In addition, the mean flow ad-
mits the possibility of convected entropy and vorticity disturbances. These modes
are coupled by the requirement of conservation of mass, momentum, and energy
across zones of heat release.
Example 4
These effects may be illustrated by extending example 3 to include a mean flow.
For definiteness we again apply an open-end inlet boundary condition p (0) = 0.
At the downstream end, we assume an area restriction in which the flow becomes
choked, and so Eq. (13.11) is the appropriate boundary condition. Note that the
hard-end boundary condition u = 0 is recovered from Eq. (13.11) as ū tends to
zero. The heat release will again be considered as concentrated at the fixed plane
x = b, the rate of heat release per unit of cross-sectional area being denoted by
Q (t) with Q (t) given by the particular flame model in Eq. (13.24).
Upstream of the zone of heat release, acoustic waves are propagating in both
directions, and the flow is isentropic. The pressure perturbation is the general
solution of the wave equation with convection (13.9); this gives

p (x, t) = Aeiωt e−iωx/(c̄1 (1+ M̄ 1 )) − eiωx/(c̄1 (1− M̄ 1 )) (13.46)
for disturbances of frequency ω and M̄ 1 = ū 1 /c̄1 . For this isentropic flow ρ =

p /c̄21 and for a perfect gas c p T = p /ρ̄. The velocity fluctuation follows directly
from the momentum equation (13.7b):

ρ̄ 1 c̄1 u (x, t) = Aeiωt e−iωx/(c̄1 (1+ M̄ 1 )) + eiωx/(c̄1 (1− M̄ 1 )) (13.47)
The fluxes of mass, momentum, and stagnation enthalpy into the combustion zones
[defined in Eq. (13.10)] can be expressed in terms of the unknown complex A
through Eqs. (13.46) and (13.47).
Downstream of the region of heat release, there might be a convected hot spot
in addition to plane sound waves, and so

p (x, t) = eiωt Ce−iωx/(c̄2 (1+ M̄ 2 )) + Deiωx/(c̄2 (1− M̄ 2 )) (13.48a)

ρ̄ 2 c̄2 u (x, t) = eiωt Ce−iωx/(c̄2 (1+ M̄ 2 )) − Deiωx/(c̄2 (1− M̄ 2 )) (13.48b)

p (x, t) S ρ̄ 2 iω(t−x/ū 2 )
ρ (x, t) = 2
− e (13.48c)
c̄2 cp
p (x, t) S c̄22
c p T (x, t) = + eiω(t−x/ū 2 ) (13.48d)
ρ̄ 2 (γ − 1)c p
for b x l and M̄ 2 = ū 2 /c̄2 . C and D are the amplitudes of the acoustic waves,
S is the amplitude of the entropy wave or convected hot spot, and no vorticity
waves occur in this one-dimensional example. The wave amplitudes C, D, and S
can be found in terms of A through Eqs. (13.10a–13.10c).
Care needs to be taken to recover the jump conditions for zero mean flow from
Eq. (13.10). In the limit ū 1 , ū 2 → 0, Eq. (13.10b) clearly simplifies to p2 = p1 ,
the zero-mean-flow jump condition [Eqs. (13.25a) and (13.44a)]. At first sight
one might assume that Eq. (13.10a) gives ρ̄ 2 u 2 = ρ̄ 1 u 1 as ū 1 , ū 2 tend to zero.
That is wrong. Note it is incompatible with Eq. (13.44c). The resolution of this
apparent inconsistency is that the strength of the entropy wave S enters the jump
conditions (13.10) only in the product ū 2 S. In the limit ū 2 → 0, S tends to infinity,
in such a way as to keep the product ū 2 S and, hence, ū 2 ρ2 and ū 2 T2 , finite. For
low-Mach-number mean flows, very large entropy fluctuations occur downstream
of the flame zone. To see these fluctuations mathematically it is convenient to first
use Eq. (13.10a) to recast Eq. (13.10c) into the form
ρ̄ 2 ū 2 (c p T2 + ū 2 u 2 ) = Q + ρ̄ 1 ū 1 (c p T1 + ū 1 u 1 )
(13.49)
− ( H̄2 − H̄1 )(ρ̄ 1 u 1 + ρ1 ū 1 )
for linear perturbations. After using Eq. (13.48d) to expand c p T2 and taking the
limit ū → 0, this equation simplifies to
ρ̄ 2 ū 2 c̄22
Seiω(t−b/ū 2 ) = Q − c p (T̄2 − T̄1 )ρ̄ 1 u 1 (13.50)
c p (γ − 1)
Physically, Eq. (13.50) shows that entropy is generated unsteadily at the combus-
tion zone whenever Q = c p (T̄2 − T̄1 )ρ̄ 1 u 1 , that is, whenever there is unsteadiness
in the rate of heat addition per unit mass. In particular, the preceding assertion that
ū S remains finite for small ū is confirmed. Equation (13.48c) clearly shows that,
in this limit, the left-hand side of Eq. (13.50) is equal to −ū 2 c̄22 ρ2 /(γ − 1), and
hence, the equation can be rearranged to give
γ −1
ū 2 ρ2 = − Q + (ρ̄ 1 − ρ̄ 2 )u 1 (13.51)
c̄22
where we have used the perfect gas relationships to rewrite c p (γ − 1) × (T̄2 − T̄1 )
ρ̄ 1 /c̄22 as ρ̄ 1 − ρ̄ 2 . Finally, substitution for ū 2 ρ2 in the equation of mass conserva-
tion leads to
γ −1
ρ̄ 2 u 1 = ρ̄ 2 u 2 − Q (13.52)
c̄22
thereby recovering the zero-mean-flow jump condition Eq. (13.44c).

Once Q has been related to the unsteady flow by a flame model and linear flow
perturbations expressed in terms of waves, the three equations describing conser-
vation of mass, momentum, and energy across the flame zone can be rearranged to
determine the downstream wave amplitudes C, D, and S in terms of the upstream
wave amplitude A.
With the wave amplitudes known, the flow perturbation at any position in the duct
can be written down by using Eq. (13.48). For a general value of ω, the flow will
not satisfy the exit boundary condition (13.11). It is therefore necessary to iterate
in ω to find the complex values of ω for which the exit boundary condition is met.
These are the frequencies of the thermoacoustic oscillations. Only disturbances
with these particular frequencies can exist as free modes of the duct/flame. The
mode shapes are determined in this linear theory but not the level of the oscillation.
In other words, a single-wave amplitude, A say, is arbitrary, but then all other wave
amplitudes are given in terms of A.
Combustion usually occurs in a low-Mach-number flow, and 2π ū 2 /ω, the wave-
length of the entropy wave, can be very short indeed for high-frequency distur-
bances. Then, turbulent mixing and diffusion tend to smooth out the entropy fluc-
tuations as they convect downstream. As a consequence, although a strong entropy
1
0.8
0.6
0.4
0.2
0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
M1
Fig. 13.7 Variation of frequency with Mach number for lowest-frequency mode,
taking Q̄= 0: ———, acoustic mode when diffusion attenuates the entropy waves by
the combustor exit; hence, mode is near ω1 and – – –, including entropy waves; hence,
mode is a low-frequency convection mode.
fluctuation may be generated in the flame zone, the amplitude of a high-frequency

entropic disturbance may be negligible by the time the wave reaches the exit of the
combustor. Judgment is needed, based on the ratio of mixing to convection time,
to decide whether the entropy waves persist as far as the downstream contraction.
If they do not persist that far, ρ should be replaced by its acoustic contribution
p /c̄2 in the downstream boundary condition (13.11). We would expect the entropy
fluctuations to be of importance only for the lowest-frequency acoustic mode, if
at all. Figure 13.7 shows the effects of a mean flow on the lowest acoustic mode
of oscillation when the entropy wave has diffused before the exit contraction. The
frequency varies only very slightly with Mach number, that is, the variation is
order Mach number squared and is 5% at a Mach number of 0.2.
An additional consequence of a mean flow is that it admits a different mode of
oscillation, one with a much lower resonance frequency (typically, 40–150 Hz for
aeroengines), where the period of oscillation is set by the convection time of the
entropy fluctuations from the flame zone to the exit nozzle and the propagation of
an acoustic wave back upstream.28 This acoustic wave causes unsteady combustion
through its effect on the inlet velocity. The unsteady combustion leads to entropy
waves or local hot spots. At these low frequencies, the entropy wavelengths are
long and the waves undergo little attenuation, generating sound as they are con-
vected through the downstream contraction. The acoustic wave propagates back
upstream, thus, completing the cycle. Only the first few harmonics of this type
of mode will be present because, as already discussed, at higher frequencies the
entropy waves will diffuse. An example of such a convection mode is shown as a
dashed line in Fig. 13.7. The frequency is approximately proportional to the Mach
number.
In this section, we have introduced some of the parameters that affect one-
dimensional acoustic waves in gas-turbine combustors. In many industrial gas
turbines, where the combustors are long, the most unstable modes are indeed plane,
but even these combustors support more complex modal solutions. Aeroengine
combustors are often annular with a short axial length. Then the lowest frequency
(and often the most unstable) modes are associated with circumferential waves.
We discuss these more general modes in the next section.
IV. Modal Solutions

We now consider perturbations that are three-dimensional. We consider two
geometries relevant to gas turbines: first, a cylindrical duct and, second, an annular
duct. Particular attention is given to the special case of the latter geometry when
the annular gap is small. This limit often occurs in practical applications and the
acoustic waves then have a particularly simple form.
A. Cylindrical Duct
Using cylindrical polar coordinates x, r , and θ, we are interested in a straight
cylindrical duct 0 r b. Because we are assuming that the mean flow is uniform,
we must have v̄ = w̄ = 0. We look for separable solutions for the three types of
disturbance mentioned earlier. The general solution is a sum of such separable
solutions.
We first consider a pressure disturbance. We seek a separable solution by sub-
stituting p = F(t)X (x)B(r )(θ) into Eq. (13.9) to give

F F X 2X

2 X (r B ) −2
+ 2ū + ū − c̄ + +r =0 (13.53)
F FX X X rB
where the prime denotes a derivative with respect to the argument. We see that
solutions take the form F(t) = eiωt , X (x) = eikx , and (θ ) = einθ , with
(r B ) + (λ2 − n 2r −2 )r B = 0 (13.54)
where λ2 = (ω + ūk)2 /c̄2 − k 2 . For continuity in θ the circumferential wave num-

ber n must be an integer. The axial wave number k and complex frequency ω may
take any complex value, but they are dependent. The general solution of Eq. (13.54)
is B(r ) = c1 Jn (λr ) + c2 Yn (λr ), where Jn and Yn are the Bessel functions of the
first and second kind, respectively. Since Yn is singular at r = 0, we must have
c2 = 0, and the rigid wall boundary condition v(b) = 0 implies
dJn
(λb) = 0 (13.55)
dr
For a given n, this gives an infinite number of discrete solutions for λ. The solutions
are all real,29 and without loss of generality we may take λ 0. We define λn,m to
be the (m + 1)th solution. The full solution can be expressed as an acoustic wave
of the form30
p = A± eiωt+inθ +ik± x Bn,m (r ) (13.56a)

1
ρ = A± eiωt+inθ+ik± x Bn,m (r ) (13.56b)
c̄2
k±
u = − A± eiωt+inθ+ik± x Bn,m (r ) (13.56c)
ρ̄α±
i dBn,m
v = A± eiωt+inθ+ik± x (r ) (13.56d)
ρ̄α± dr
n
w = − A± eiωt+inθ+ik± x Bn,m (r ) (13.56e)
r ρ̄α±
with Bn,m (r ) = Jn (λn,m r ). [Note that the perturbations as given in Eq. (13.56) will
be complex, but it is assumed that we take the real part.] Here α± = ω + ūk± ,
M̄ω ∓ (ω2 − ωc2 )1/2

k± = (13.57)
c̄(1 − M̄ 2 )
and M̄ is the mean Mach number (which is assumed to be less than unity). Also,
ωc = c̄λn,m (1 − M̄ 2 )1/2 is the complex cutoff frequency of the duct, and A± , which
may be complex, are the wave amplitudes. For real ω > ωc , A+ represents a
downstream-propagating wave and A− represents an upstream-propagating wave.
For real ω < ωc the waves are cut off. Defining the square root in Eq. (13.57)
to be a negative imaginary number, A+ now represents a downstream-decaying
disturbance and A− represents an upstream-decaying disturbance. For complex ω,
a combination of these behaviors is seen.
The separable solutions for an entropy disturbance are entropy waves of the
form
1
ρ = − A E eiωt+inθ+ik0 x E(r ) (13.58)
c̄2
with p = u = v = w = 0, where k0 = −ω/ū and E(r ) can be any function of

r . For a vorticity disturbance, the solution can be thought of as a sum of two types
of vorticity wave, one in which the radial velocity is zero and one in which the
circumferential velocity is zero.13 The first type has the form
n
u = A V eiωt+inθ+ik0 x V (r ) (13.59a)
ρ̄ c̄
k0 r
w = − A V eiωt+inθ+ik0 x V (r ) (13.59b)
ρ̄ c̄
with p = ρ = v = 0, whereas perturbations in the second type can be expressed

as
1 dW
u = A W eiωt+inθ+ik0 x (r ) (13.60a)
ρ̄ c̄r dr
ik0
v = − A W eiωt+inθ +ik0 x W (r ) (13.60b)
ρ̄ c̄r
with p = ρ = w = 0. The only restrictions on V (r ) and W (r ) are that V (0) =
W (0) = W (b) = 0.
In this section we have assumed that the duct wall is rigid. The case of a compliant
duct wall is discussed by Eversman,30 as is the case of a nonuniform mean flow.
B. Annular Duct
Many gas turbines, particularly aeroengines, have an annular geometry. Hence,
we will now consider the form of perturbations that can occur in the gap between
two rigid-walled concentric cylinders a r b. The acoustic waves are the same
as for a cylindrical duct except that now31
dYn dJn
Bn,m (r ) = (λn,m b)Jn (λn,m r ) − (λn,m b)Yn (λn,m r ) (13.61)
dr dr
and λn,m 0 is now the (m + 1)th solution of
dJn dYn dJn dYn
(λn,m a) (λn,m b) = (λn,m b) (λn,m a) (13.62)
dr dr dr dr
from the rigid-wall boundary conditions on r = a and r = b. [By using an approach
similar to that given by Watson29 to prove that Jn only has real zeros, it can be
shown that the solutions of Eq. (13.61) are again all real.] The entropy waves are
unchanged. For the vorticity waves, there is now no restriction on the function
V (r ), whereas for W (r ) we have W (a) = W (b) = 0.
C. Narrow Annular Gap

In annular gas turbines, the radial gap of the combustor is typically shorter
than the axial length and much shorter than the circumference. In such situations
we may approximate the flow by considering the case when the annular gap is
narrow, that is, a ≈ b. For m = 0, Bn,0 (r ) can be approximated as constant; hence,
in particular, v = 0, and it can be shown that λn,0 ≈ n/R, where R = 12 (a + b).
The higher-order radial modes, m > 0, are highly cut off (meaning that they have
rapid axial decay) and can be ignored. Comparison with full solutions confirms
the expected radial independence. For the entropy and vorticity waves, E(r ) and
V (r ) should be taken to be constant, whereas W (r ) should be discarded. For more
details on this approximation and its applicability see Stow et al.13
We now illustrate modal solutions, specifically, circumferential modes in a nar-
row annular gap, with an example. As before we consider a uniform straight duct,
length l. However, we now impose a mean flow and change the inlet and outlet
boundary conditions to be choked. Also, we assume that the duct has a narrow
annular cross section. Entropy waves (and vorticity waves if n = 0) will be

generated at the inlet and convected with the mean flow to the outlet where they can
interact with the acoustic waves. However, because the entropy and vorticity waves
have a short wavelength [see Eqs. (13.58)–(13.60)], if the duct is long they are likely
to be diffused away by mixing processes before they reach the combustor outlet.
Hence, initially we ignore the influence of these waves at the downstream bound-
ary. When the Mach number in the duct is taken to be small, the choked inlet and
outlet boundary conditions give u ≈ 0. Hence for plane waves, n = 0, the resonant
modes of the duct for integer m are approximately the organ-pipe resonances,
mπ c̄
ω ≈ ω̃m = (13.63a)
l
mπ x
p̂(x) ≈ Am cos (13.63b)
l
iAm mπ x
û(x) ≈ − sin (13.63c)
ρ̄ c̄ l
a) 0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6
b) 0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6
Fig. 13.8 Frequencies and growth rates of resonant modes of a duct with choked
outlet: ×, choked inlet with entropy and vorticity waves included; ◦, choked inlet with
convected waves dissipated; +, open inlet; and – – –, cutoff frequency of the duct for
n = 1. a) n = 0. b) n = 1.
where we have taken p (t, x, θ ) = p̂(x)eiωt+inθ . For circumferential waves, n = 0,

for integer m, ω is given by
1/2
ω ≈ ω̃m2 + ωc2 (13.63d)
where ωc = n c̄/R is the cutoff frequency of the duct, with the mode shapes also
approximated by Eqs. (13.63b) and (13.63c). In particular, for a given n the lowest
frequency mode is close to the cutoff frequency and has a pressure perturbation
that is roughly uniform axially. The frequencies [= Re(ω)/(2π )] and growth rates
[= − Im(ω)] of the modes for n = 0 and 1 are shown as circles in Fig. 13.8. The
pressure distribution for the second n = 1 mode (m = 1) at a sequence of times in
its oscillation period (T = 1/frequency) is shown in Fig. 13.9. Axially the mode
is a standing half-wave, whereas circumferentially it is a spinning whole wave.
All the modes have a negative growth rate, because the choked inlet and choked
outlet boundary conditions do not give a perfect reflection of acoustic waves and
are, therefore, sources of damping. If entropy and vorticity wave propagation is
included, many more modes are found, as denoted by crosses in Fig. 13.8. The
modes are roughly ū/(2l) Hz apart, that is, Re(ω) ≈ π ū/l. The least stable
modes, that is, those with the largest growth rates, are found to be close to the
modes when entropy and vorticity waves are ignored. For comparison with the
preceding examples, results for an open inlet/choked outlet are shown as pluses
in the figure. As we would expect the frequencies lie midway between the choked
inlet/choked outlet frequencies. In this case, neither entropy nor vorticity waves
are generated by the open inlet, so neither are present in the duct. Also, the growth
rates are less negative here because the open inlet gives no damping.
a) spin direction
spin direction 1
p′ (arbitrary scale)
1
1
1
z /R
0
0
z /R
0
0
−1
−1 1
1 −1
1 −1 0
1 0.5
0 0.5 y /R −1 0 x/l
y / R −1 0 x/l
b) spin direction spin direction

1 1
1 1
z /R
z /R
0 0
0 0
−1 −1
1 −1 1 −1
1 1
0 0.5 0 0.5
y/R −1 0 x/ l y/ R −1 0 x/ l
Fig. 13.9 Time sequence of pressure distribution in thin annular duct for second
mode in Fig. 13.8b (choked inlet with convected waves dissipated). a) t = 0. b) t = T/4.
c) t = T/2. d) t = 3T/4.
fuel injection premix duct and swirler unit
plenum combustor
compressor outlet turbine inlet
Fig. 13.10 Typical gas-turbine geometry.
V. Application to Gas-Turbine Combustors

So far, we have described the modal analysis of simple cylindrical and annular
ducts and have shown how, with appropriate boundary conditions, it leads to their
resonant frequencies. However, the geometry of gas-turbine combustors is far from
simple. The acoustics of the gas turbine from compressor exit to turbine entry may
play a role in combustion instability. A typical geometry is illustrated in Fig. 13.10.
The high-speed flow at the compressor exit is slowed down in a diffuser and made
more uniform in preparation for fuel addition and combustion. At the downstream
end of the diffuser, the air is accelerated through premixing ducts where fuel is
added, and the premixed fuel and air then enter a combustion chamber where it is
burned. Although this geometry is complex, it is made up of a series of annular and
cylindrical ducts. The flow passage is annular at compressor exit, the premix ducts
have small cross-sectional areas in which only one-dimensional waves propagate,
and the combustion chamber may be either annular or cylindrical. Our previous
analysis is therefore relevant provided we know how to join ducts of different
cross-sectional areas. We can illustrate the approach by discussing the simple
quasi-one-dimensional geometry in Fig. 13.11.
A. Plenum
We investigate the form of linear disturbances in the geometry of a plenum
section, premixing ducts, and combustor. In this example, we will assume that the
frequency of oscillation is sufficiently low that only plane waves carry acoustic
energy, with all higher-order modes decaying exponentially with axial distance.
At the inlet, representing compressor exit, the flow is nearly choked, which leads
to inlet boundary conditions for the linear waves of frequency ω (as discussed in
Sec. II.B). The relative wave strengths at A–A (Fig. 13.11) are then completely
determined.
Equations (13.7c), (13.7d), and (13.9) describe how those waves develop along
the plenum, hence, determining the unsteady flow at entrance to the premix duct.
choked end premix duct and swirler unit

A S
plenum combustor
A’ x1 x2
fuel injection
Fig. 13.11 Simple quasi-one-dimensional combustor.
B. Premix Duct and Flame

There are two main approaches to relating the perturbations in the plenum and
combustor: One approach is purely acoustic and often relies on empirical inputs,
whereas the second approach is model based through appropriate application of
the equations of conservation of mass, momentum, and energy.
The acoustic approach involves determination from experiment32 or simple
models33, 34 of the transfer matrix N(ω), which relates pressure and velocity per-
turbations at the entrance to the premixer (denoted by subscript 1) to perturbations
downstream of the combustion zone (denoted by subscript 2):

p̂ 1 p̂ 2
= N(ω) (13.64)
û 1 û 2
A schematic diagram is shown in Fig. 13.12. The 2×2 matrix N depends on the
details of the geometry and the flow between x1 and x2 . For example, for the duct
with a uniform cross-sectional area, and a flow with negligible mean flow and the
premix duct
1 3 2
combustion zone
Fig. 13.12 Schematic diagram of premix duct and combustion zone (for definition of
transfer matrix).
flame model in Eq. (13.24b), we found [see Eqs. (13.25a) and (13.26)]

1 0
N= (13.65)
0 (1 − βe−iωτ )−1
In the case in which the premix duct is short and has a small cross-sectional area and
no combustion occurs, the flow in the premix duct is effectively incompressible and
the pressure difference across it can be related to the rate of change of momentum
in the premix duct. For negligible mean flow, the relationship is A3 ( p̂ 1 − p̂ 2 ) =
(∂/∂t)(ALρu)3 = iωρ̄ A3 L 3 û 3 , where A represents the cross-sectional area, L is
the effective axial length, and subscript 3 denotes flow within the premix duct.
From conservation of mass, A1 û 1 = A2 û 2 = A3 û 3 . Hence, we have

1 iωρ̄ A2 L 3 /A3
N= (13.66)
0 A2 /A1
For more realistic conditions, N can be investigated through carefully chosen

experiments. Such experiments typically involve introducing an acoustic source
at an upstream location S in Fig. 13.11. The source could be an in-line siren or
wall-mounted loudspeakers. By driving the source at a range of frequencies, p̂ 1 (ω),
û 1 (ω), p̂ 2 (ω), and û 2 (ω) can be measured. However, the impedance Z 2 (ω) = p̂ 2 /û 2
is specified by the downstream geometry and so, for a particular downstream
geometry, only the product of N [Z 2 , 1]T can be investigated. Measurements are
needed with two different downstream impedances if all four coefficients are to
be found. In practice, this can be done by making measurements with two differ-
ent downstream lengths, or alternatively, a single length with two different exit
conditions, for example, open and constricted. The advantage of this approach is
that it does not rely on any modeling, assuming only that the perturbations are
linear. It therefore gives an accurate representation of the jump or joining condi-
tions across any geometry of premix ducts and combustion zone. Its disadvantages
are: it provides little physical insight, and measurements must be made with the
flow between x1 and x2 representative of full-scale conditions, not only in terms
of geometry, but also with the correct inlet temperature, pressure, mass flow rate,
and rate of combustion. This method has been used successfully by Paschereit
et al.32 to characterize the pressure–velocity relationship across a premix duct and
combustion zone in a geometry similar to that shown in Fig. 13.11.
An alternative approach is based on conservation equations.35–38 The pre-
mixer geometry may be modeled by several compact area changes connected
by straight ducts. At an area increase, the mass and energy flows are unchanged,
and momentum flow is increased by the axial force on the walls; hence, we may
take
A2 ρ2 u 2 = A1 ρ1 u 1 (13.67a)
H2 = H1 (13.67b)
A2 p2 + A2 ρ2 u 22 = A2 p1 + A1 ρ1 u 21 (13.67c)
where subscripts 1 and 2 denote the flow parameters and areas before and after
the area change, respectively. Here the pressure on the abrupt expansion has been
taken to be p1 ; however, some pressure recovery could be included through the use
of a loss coefficient. To find the perturbations after the area increase, Eq. (13.67) is
linearized in the usual way to give a transfer matrix relating the downstream and
upstream flow.
An area decrease can be assumed to be isentropic, hence,
γ γ
p2 /ρ2 = p1 /ρ1 (13.68)
and conservation of mass and energy give Eqs. (13.67a) and (13.67b) as before.
For no mean flow, the jump conditions at any area change simplify to
[ p]21 = [Au]21 = 0 (13.69)
The flame is also treated as compact, and so Eq. (13.10) applies across it.
However, this approach needs a flame model relating the instantaneous rate of heat
release to the oncoming flow. Flame models are discussed in Chapter 12, but here
we note that they can be determined either by analytical descriptions of the flame
dynamics9, 39 or through numerical36 or experimental investigations40–42 of the un-
steady combustion response to inlet flow disturbances. Measurements carried out
at low and high pressure have remarkably similar forms42 but different amplitudes,
supporting the idea that the flame-transfer function can be investigated by suitably
scaled experiments or through local computational fluid dynamics (CFD) solutions.
C. Combustor
Once the fluxes of mass, momentum, and energy are known in the combustor
just downstream of the zone of combustion, the strengths of the linear waves can
be calculated. Equations (13.7c), (13.7d), and (13.9) describe how those waves
develop along the combustor, thus determining the flow at exit. For a general
value of frequency ω, this will not satisfy the downstream boundary condition.
The resonant frequencies are the values of ω at which the downstream boundary
condition is satisfied.
Example 5
We now consider an example of a complete system consisting of a plenum,
premix system, and combustor, similar to that shown in Fig. 13.11 except that the
combustor has an open end. Details of the geometry are given in Table 13.1. A
simple flame model,
Q̂ m̂ i
= −k e−iωτ (13.70)
Q̄ m̄ i
is used at the start of the combustor, where m i is the air-mass flow at the fuel-
injection point (taken to be at the start of premixer). The circles in Fig. 13.13
denote the resonant modes of the geometry for k = 0. Several modes are seen, all
of which are stable as we would expect because there is no unsteady heat release.
The premix duct provides sufficient blockage that it acts approximately like a hard
Table 13.1 Geometry and flow conditions for simple

combustor (based on an atmospheric test rig)
Description Value
Choked inlet, mass flow rate 0.05 kg s−1

Choked inlet, temperature 300 K
Plenum, cross-sectional area 0.0129 m2
Plenum, length 1.7 m
Premixer, cross-sectional area 0.00142 m2
Fuel-injection point, fuel-convection time 0.006 s
Premixer, length 0.0345 m
Combustor, cross-sectional area 0.00385 m2
Flame zone, temperature after combustion 2000 K
Combustor, length 1.0 m
Open outlet, pressure 101,000 Pa
150
100
50
0
Im ω (s 1 )
50
100
150
growth rate,
200
250
300
350
400
450
0 50 100 150 200 250 300 350 400 450 500 550 600
frequency (Hz)
Fig. 13.13 Resonant modes of simple combustor: ×, modes for k= 1; ◦, k= 0, that

is, no unsteady heat release; and ———, variation between these two values.
end (u = 0, maximum pressure amplitude) to disturbances in the plenum, which

means that there is a family of resonant frequencies consisting of resonances of the
plenum. In the figure, these resonances are seen at 110, 203, 289, 416, and 511 Hz,
the first being the fundamental half-wave mode and the others being its harmonics.
The mode at 337 Hz is the first of a family of combustor modes. Taking the front
face of the combustor to be a closed end gives only a very crude approximation
because the discrepancy in area between the combustor and the premixer is not
as large as for the plenum. The mode is somewhere between a quarter-wave and
a half-wave resonance of the combustor (its mode shape is very similar to that in
Fig. 12.14f). The low-frequency mode at 30 Hz is a resonance of the geometry as
a whole, specifically, a quarter-wave.
We now introduce unsteady heat release by setting k to be unity. The resulting
modes are denoted by crosses in Fig. 13.13 (the lines show the variation for k
between 0 and 1). The unsteady heat release has little effect on some modes, but,
in general, the growth rates are increased, pushing the modes into instability. In
addition to the original modes, a new set of modes is associated with the flame
model. These modes are closely related to the additional modes for non-zero τ
found in example 1. Their frequencies are approximately 1/τ , 2/τ , and 3/τ , and
their growth rates become large and negative as k tends to zero. The mode shapes
for k = 1 are shown in Fig. 13.14.
D. Annular Combustors
We now consider an annular gas turbine for which the plenum and combustor
have a narrow annular gap cross section, as discussed earlier. Hence, we take the
perturbations to have the form of a circumferential mode. Wave propagation in
the plenum and combustor is given by Eqs. (13.7c), (13.7d), and (13.9) as before
(see also Sec. IV). When joining annular ducts of different inner and outer radii,
considering conservation laws in a thin sector of the transition leads to the same
flux relationships as for plane waves, with the addition that the angular-momentum
flux is unchanged.10 Hence, if the premix region also had an annular geometry,
the perturbations for a circumferential mode could be found in much the same
way as described earlier for plane waves. Typically, however, the premix region
consists of a large number of identical premix ducts that are evenly distributed
around the circumference. Hence, there is a loss of axisymmetry, and we might
expect that this would interact with the circumferential wave in the plenum to
produce circumferential waves of other orders, that is, modal coupling would
occur. In fact, any additional modes will be high order and decay rapidly with
axial distance (see Sec. VI). Thus, it is valid to consider a single circumferential
wave of a selected order in the plenum. The premix ducts will usually also be
annular; however, they will have a much smaller cross section than the plenum
and combustor and so, for frequencies of interest, the perturbations in them will
be one dimensional. The circumferential wave in the plenum produces identical
perturbations in the ducts, except that each perturbation is phase shifted. The
equations relating the perturbations in the plenum to those in the premix ducts
are similar to those for a simple area decrease, with adjustments due to the change
from a circumferential disturbance to a set of one-dimensional perturbations. The
propagation of these one-dimensional disturbances along the premix ducts can
a) b)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
c) d)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
e) f)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
g) h)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
j)
i)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
Fig. 13.14 Mode shapes for simple combustor, k= 1. a) 30-Hz mode. b) 104-Hz mode.
c) 168-Hz mode. d) 203-Hz mode. e) 300-Hz mode. f) 312-Hz mode. g) 396-Hz mode.
h) 415-Hz mode. i) 495-Hz mode. j) 514-Hz mode.
be found as before. At the inlet to the combustor, the ring of phase-shifted one-
dimensional disturbances creates a circumferential wave of an order identical with
that in the plenum. The resonant modes for circumferential waves of this selected
order can then be found by investigating the propagation of this circumferential
mode through the combustor and determining the resonant frequencies at which
the downstream boundary condition is satisfied.
VI. Modal Coupling
In uniform cylindrical and annular ducts, the solutions in Eq. (13.56) for different
values of n and m are independent and can be considered separately. However,
nonuniformities can lead to a coupling of these modes. For instance, if the duct
has an area change, but remains axisymmetric, the circumferential modes, that is,
different values of n, are still independent but the radial modes, that is, different
values of m, become coupled. Consider, for example, a circular duct that has an
abrupt area increase at x = 0. We denote conditions in x < 0 by superscript (1)
and in x > 0 by superscript (2). The duct is then r b(1) for x < 0 and r b(2)
for x > 0, with b(2) > b(1) . For no mean flow, only acoustic waves are present, and
so from Eq. (13.56) for a given n, we may write for x < 0
∞
+ −
p = eiωt+inθ A+(1)
n,m e
ikn,m x
+ A−(1)
n,m e
ikn,m x
Bn,m (r ) (13.71a)
m=1
and for x > 0
∞
+ −
p = eiωt+inθ A+(2)
n,m e
ikn,m x
+ A−(2)
n,m e
ikn,m x
Bn,m (r ) (13.71b)
m=1
with similar expressions for the other flow variables. Miles43 and Alfredson44 con-
sidered this problem for plane waves; however, the extension to n = 0 is straight-
forward (as is the extension to annular ducts). At x = 0, we must have continuity of
p and u for 0 r b(1) (continuity of ρ , v , and w follow from continuity of p ),
and on the rigid wall b(1) r b(2) we require u = 0. This continuity leads to a
linear system of equations relating A±(1) ±(2)
n,m and An,m . The amplitudes for one value
m are found to depend on those for all other values of m, meaning that the radial
modes are coupled. In Eq. (13.71) we included all the radial modes; however, in
practice, for m that is sufficiently large the waves will be highly cut off and so
can be ignored. Hence, we can approximate using a finite number of radial modes,
for example, 0 < m < M. Some examples of results for n = 0 modes in a duct in
which the area doubles are shown in Fig. 13.15.
The radial variations of the magnitudes of the pressure and axial velocity on
either side of the area change are shown; the solid and dashed lines denote the
values in the larger and smaller area regions, respectively. These results are for
M = 5; as more radial modes are included, the matching becomes better and the
solutions more accurate.
A similar approach was used by Akamatsu and Dowling45 to consider three-
dimensional combustion instabilities in a cylindrical combustor with a ring of
a) 1.25 b) 1.25
1 1
u′ (arbitrary scale)
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r / b(2 ) r / b(2 )
Fig. 13.15 Radial variation for n = 0 and n = 5: ———, x = 0+ ; – – –, x = 0− ; and

–··–, r = b(1) . a) Pressure magnitude and b) axial velocity magnitude.
premix ducts. Oscillations in the premix ducts were assumed to be one dimen-
sional, and these were treated as point sources when joining to the combustion
chamber. The loss of radial symmetry here led to a coupling of the radial modes
in the combustor. Perhaps surprisingly, because the premixers were identical and
evenly distributed circumferentially, the circumferential modes remained uncou-
pled. Similarly, Evesque and Polifke46 found that circumferential modes became
coupled only when their premix ducts were nonidentical. In fact, it can be shown
that a ring of identical premix ducts does not introduce coupling of circumferential
modes provided that N is less than half the number of ducts. In other words, any
coupling occurs in high-order modes that decay rapidly with axial distance and
are not of practical interest.
Coupling of circumferential modes in a narrow annular gap has been considered
by Stow and Dowling.47 The presence of Helmholtz resonators in the geometry
destroys the axisymmetry causing modal coupling. We now describe their method
of solution because the approach, in general, should be applicable to finding linear
resonances in problems
with modal coupling. We write p (t, x, θ) = p̂(x, θ )eiωt
with p̂(x, θ) = ∞ n=−∞ p̂ n (x)e , and similarly for the other variables. For |n|
inθ
large (for example, |n| > N ), the mode will be highly cut off; hence, in a way sim-
ilar to the radial
N modes earlier, we approximate circumferential modes by taking
p̂(x, θ) = n=−N p̂ n (x)einθ . At the inlet of the geometry, there are boundary con-
ditions that apply to each mode independently. These define the perturbations for
each circumferential mode n except for an unknown parameter λn . For instance, if
it is an open end, p̂ = 0 for all θ , implying that p̂ n = 0 for all n, and so we may set
A+ −
n = −An = λn (with no entropy or vorticity waves). Here, λ = [λ−N , . . . , λ N ]
T
describes the relative amplitude and phase of the modes at the inlet and must be
found as part of the solution. Similarly, at the outlet, there is a boundary condi-
tion that applies to each mode independently. We define µn to be the error in this
boundary condition for circumferential mode n; for example, for an open end we
may take µn = p̂ n . Given ω and λ , all the circumferential components at the inlet
are known. We can step through the geometry calculating all the circumferential
modes at each sec. before continuing to the next. For the solution thus found, each
mode will have an error at the outlet µn . We must find ω and λ to satisfy µn = 0,
thus, giving a resonance of the geometry. For a given ω, we define the matrix M
to be such that Mn,m is the value of µn for the solution with λi = δi,m . For a gen-
eral λ , µn = Mn,m λm because the perturbations are linear. Hence, for the correct
values of ω and λ , Mλλ = 0. For a solution to exist, λ = 0, and so this implies that
det M = 0. Thus the procedure to find a complex resonant frequency ω is to first
guess the value of ω and calculate the matrix M and then iterate the value of ω to
achieve det M = 0. For this value of ω, a λ will exist giving Mλ λ = 0. Finally, this
correct λ is calculated by using an inverse iteration method (Mλ λnew = λ old ). The
mode shape for the resonance can then be calculated using this λ . As before, the
resonant frequency and growth rate are given by ω.
VII. Acoustic Absorbers

Passive control of combustion instability can significantly reduce amplitudes,
even causing modes to become stable. The use of passive control is discussed in
Chapter 17. Here we concentrate on two particular passive-control devices that can
damp oscillations by absorbing acoustic energy, namely, Helmholtz resonators and
perforated liners. We will describe how these devices can be used and how they
may be included in the linear models introduced earlier.
Helmholtz resonators are damping devices that can be used to tackle combustion
instability in gas turbines (see, for example, Refs. 48 and 49) and many other
applications in which one might wish to reduce acoustic oscillations. Figure 13.16
shows an example of the reduction in the amplitude of combustion instability when
a Helmholtz resonator is used. A Helmholtz resonator consists of a large volume
connected via a short neck to a duct, such as a combustion chamber, in which
the oscillations occur. The mass of air in the neck and the stiffness of air in the
resonator volume act as a mass–spring system, which has a resonant frequency
dependent on the volume of the resonator, the length and cross-sectional area of
the neck, and the speed of sound (see, for example, Ref. 50). If the perturbations
in the duct are close to this frequency, the fluctuating pressure at the neck entrance
170
sound pressure level (dB)
160
150
140
130
120
110
0 100 200 300 400 500 600 700 800 900 1000
frequency (Hz)
Fig. 13.16 Power spectra of experimental results for an adjustable-volume Helmholtz

resonator: ———, resonator volume is minimum, hence, damping is negligible; and
———, resonator volume to suppress oscillations.
will cause large velocity oscillations into and out of the resonator. These velocity
oscillations dissipate energy leading to a damping of the acoustic perturbations
in the duct. This source of damping is a nonlinear effect (see Ref. 51), relying
on the velocity oscillations in the neck to have a large enough amplitude so that
significant kinetic energy is dissipated in the unsteady jets that form. Specifically,
the acoustic energy of the velocity oscillations at the neck is converted to vortical
energy and ultimately dissipated as heat. If the pressure perturbations in the duct
are low amplitude, this mechanism gives negligible damping. Hence, it can reduce
the amplitude of an existing instability but cannot stabilize the mode.
In gas-turbine applications, there will be a mean flow through the combustor
and hence across the neck of the Helmholtz resonator, and the requirement to cool
the resonator may lead to an additional flow through the neck; these flows lead to
additional sources of damping. In this situation, the acoustic waves modulate the
vortex shedding at the neck and lead to a linear source of damping in the sense that
the proportion of acoustic energy absorbed is independent of the sound pressure.
This mechanism therefore has the potential to stabilize a mode. However, there
is a danger with this configuration that, in some frequency ranges, generation of
sound instead of absorption can occur because of vortices being shed from the
upstream lip of the neck and impinging on the downstream lip. This problem
can be overcome by using a downstream lip that is rounded, not sharp-edged.
Alternatively, introducing a sufficiently strong cooling flow through the neck into
the combustor can remove the problem, because the vortices are then driven away
from the downstream lip.
To include a Helmholtz resonator in an acoustic calculation of the type described
earlier, one can consider conservation of unsteady mass, momentum, and energy
between the point in the duct just upstream of the resonator and the point just down-
stream. However, one needs to account for the mass flow perturbation m̂ entering
the duct through the neck of the resonator. In no mean flow, as stated previously,
nonlinear effects are important and so m̂ is not simply linearly proportional to
the amplitude of the oscillations. Hence, a nonlinear calculation is now required,
such as the describing-function approach considered in the next section. With a
mean crossflow and/or neckflow, the system remains linear and the calculation
techniques described previously can be applied directly. In this case, writing p̂ 1
and p̂ 2 for the pressure perturbations in the combustor and inside the resonator,
respectively, we may define the Rayleigh conductivity κ of the neck by
iωm̂
κ= (13.72)
p̂ 2 − p̂ 1
The rate of decrease of mass inside the resonator must equal m̂ and so, assuming
isentropic conditions there, we have m̂ = −iωV p̂ 2 /c̄2 , where V and c̄ are the
volume of the resonator and the speed of sound inside, respectively. If the mean
flow through the neck is much larger than the crossflow, the Rayleigh conductivity
for a circle aperture found by Howe et al.52 can be used, with a correction to account
for the length of the neck. Conversely, if the only mean flow is across the neck,
the result by Howe53 is applicable. The case in which the cross- and throughflows
are comparable has been considered by Dupère and Dowling.54 Details on the
modeling of Helmholtz resonators with and without mean flows and a discussion of
practical issues for their use in real combustors are given by Dupère and Dowling.55
The placement of Helmholtz resonators can be an important consideration. For
maximum effect, they should be placed where the amplitude of pressure oscillation
is high, whereas at a pressure node they would have no effect. In relation to this
effect, circumferential modes in annular combustors using only a single resonator
give no damping. A circumferential standing wave is produced (the clockwise and
anticlockwise spinning circumferential modes becoming coupled) which aligns
itself to have a pressure node at the resonator neck. Hence, at least two resonators
are required to damp such an oscillation. The azimuthal placement of resonators
to achieve the best damping of circumferential waves has been investigated by
Stow and Dowling.47 A drawback of using Helmholtz resonators is that they give
good damping only over a relatively small-frequency band. If there are several
modes of instability, several resonators may be required. However, in a situation
in which at different operating conditions a single frequency dominates, but in
which the frequency varies (either continuously or suddenly) as the operating con-
ditions are changed, an interesting alternative is the use of a Helmholtz resonator
that can retune itself to damp the current instability. This form of actively tuned
passive damping or semiactive control has been investigated by Wang.56 Figure
13.16 shows experimental results for an atmospheric rig similar to that shown in
Fig. 13.11, with an adjustable-volume Helmholtz resonator attached to the com-
bustion chamber. A feedback algorithm was used to tune the resonator, leading
to a more than 15-dB reduction in peak amplitude. [We thank Dr. Chuan-Han
Wang (Cambridge University Engineering Department) for permission to show
this figure.]
Liners with bias flow have the potential to damp oscillations over a much
broader range of frequencies than Helmholtz resonators have. This type of acous-
tic absorber was investigated for plane waves by Eldredge and Dowling,57 who
found that more than 80% absorption can be acheived. Eldredge58 extended their
analysis to higher-order modes. The configuration is as follows. A section of the
duct in which we wish to damp oscillations (for example, the combustion chamber)
is replaced by a liner consisting of an array of holes, through which a mean flow
passes into the duct. This flow can form part of the cooling of the duct. On the
other side of the liner we could simply have a duct, a large chamber, or one or more
additional liners supplying the flow. This setup changes the performance of the
liner but not the underlying principles. The mechanism of absorption is very simi-
lar to the case of a Helmholtz resonator with a mean flow through the neck, namely
the conversion of acoustic energy to vortical energy in the shed vortices, which is
then dispersed. However, the liner does not rely on matching a resonant frequency
and can absorb over a large frequency range. As with Helmholtz resonators, the
liner is most effective if located at a region of large pressure oscillations. To include
such a liner in the linear models described earlier, the liner must be discretized
axially. For instance, one may represent the liner as a series of compact regions
containing the holes separated by straight ducts. The perturbations at these hole
regions can be calculated by using the Rayleigh conductivity of the holes in much
the same way as for a Helmholtz resonator with neck flow; the main difference is
the treatment of the perturbations on the other side of the liner. If there is a large
chamber on the other side, we may assume that the pressure oscillation is zero;
150
Im ω (s 1 )
100
50
0
growth rate,
50
100
150
0 50 100 150 200 250 300 350 400 450 500 550 600
frequency (Hz)
Fig. 13.17 Resonant modes of simple combustor: ◦, with Helmholtz resonator; and
×, without Helmholtz resonator.
if a duct or secondary liner is present, this must be modeled as part of the linear
calculations.
As an example of the use of acoustic dampers in linear acoustic models, we now
consider adding a Helmholtz resonator to example 5 (with k = 1). The resonator is
placed halfway along the combustor and is assumed to have a mean neck outflow
of 10 ms−1 . (The crossflow in the combustor is negligible compared with this.) The
neck is taken to have a radius of 7 mm and length of 30 mm, and the temperature
in the volume is set to be 1000 K. We seek to damp the most unstable mode, which
is at 168 Hz; hence, we set this to be the resonant frequency of the Helmholtz
resonant by taking its volume to be 1.24 × 10−3 m3 . The resulting resonant modes
are plotted as circles in Fig. 13.17, with the modes without the resonator shown
as crosses (see also Fig. 13.13). We find that the growth rate of the targeted mode
is reduced, which indicates damping, and there is a small shift in the frequency to
166 Hz. Also an additional mode has appeared at 159 Hz; this mode is associated
with the resonator and is highly damped. The resonator has only a minor effect
on the frequency of the other modes, however, many have a significant increase
or decrease in their growth rate. This change is perhaps surprising, in particular,
where the growth rate is increased because a Helmholtz resonator with a mean
flow through the neck never generates acoustic energy. The effect is caused by the
fact that the resonator, independent of any damping effects, alters the acoustics of
the combustor because of the inertia of the mass of air in the neck. Although this
inertia has only a small effect on the frequency, it is enough to change the difference
between acoustic energy gained from and lost at the combustor boundaries, which
is much more sensitive. This has a direct effect on the growth rate.
VIII. Limit-Cycle Prediction

The models presented so far apply to small linear oscillations, not to the large-
amplitude limit cycles that cause problems in gas turbines. Such linear models can
provide useful information. First, the models give predictions of linear instability
boundaries. An oscillation will always be small to begin with, and if it is linearly
stable it will not grow to form a limit cycle. Second, the frequency of a linear
mode usually provides a good approximation to that of the resulting limit cycle.
Damage is often the result of a oscillation frequency being close to the structural
resonant frequency of a component of the gas turbine, so knowledge of potential
frequencies can be very useful.
However, to obtain predictions of the amplitudes of oscillation, the nonlinear
effects that limit the size of the perturbations must be modeled. The limit cycle usu-
ally involves oscillations in which the pressure perturbation is small in comparison
with the mean pressure, and the fluctuating velocity is much less than the speed of
sound. These conditions ensure that the acoustic waves are still linear. The main
nonlinearity is usually in the combustion response to velocity and equivalence ratio
fluctuations which can be of the order of their mean. The time-domain Galerkin
method can be extended to include nonlinearity and hence give limit-cycle solu-
tions.59 The transfer matrix and conservation approaches can be converted to the
time domain (at least for plane waves), and so, similarly, once nonlinear effects
are included, these can be used to predict limit cycles. However, a faster and sim-
pler method is to remain in the frequency domain and use a describing-function
approach as follows (see also Dowling16 ). The main effect determining the limit-
cycle amplitude is likely to be a saturation of the heat-release oscillation from the
flame. Consider a flame being forced by a time-varying input (such as the air-mass
flow at the fuel-injection point m i ) at a single frequency. The heat release from
the flame will be periodic with the same frequency, but at a high forcing am-
plitude the response may contain multiple frequencies, because nonlinearity can
generate the harmonics of the forcing frequency. However, we expect the flame
will respond less to high-frequency disturbances, suggesting that these harmonics
are not important in the feedback loop between the heat release and the acoustics
of the geometry. Hence, the flame can be characterized by a nonlinear flame-
transfer function relating the heat-release component at the forcing frequency to
the flow perturbation as a function of both frequency and amplitude of forcing. For
example,
Q̂ m̂ i
= T (ω, A) (13.73)
Q̄ m̄ i
where T is the nonlinear transfer function and A = |m̂ i |/m̄ i is the forcing ampli-
tude. Typically, increasing A will decrease the magnitude of T because of satura-
tion effects, whereas the effect on the phase of T is often less significant. We have
already seen in example 5 that decreasing the gain of the flame-transfer function
has a stabilizing effect. For a mode that is linearly unstable, the amplitude will
initially increase with time, thus reducing the gain and hence reducing the growth
rate. Eventually we reach a point where the growth rate is zero. This mode is a
stable limit cycle; at lower amplitudes the oscillations are still growing, whereas at
higher amplitudes they will decay. One may assume that elsewhere in the geometry
nonlinear effects are less important and that the linear models are still applicable
there. Instead of solving for complex ω in the linear problem, the solution now
amounts to finding real ω and A such that the exit boundary condition is satisfied.
This solution gives the limit-cycle frequency, amplitude, and dimensional mode
shape.
We now give a simple illustrative example of a nonlinear flame-transfer function.

Written in the time domain, the linear flame model in Eq. (13.70) becomes
m i (t − τ )
Q L (t) = −k Q̄ (13.74)
m̄ i
The subscript L denotes that this a linearized result. One could consider nonlinear
effects on the equivalence ratio, convection time, and flame response (see Stow
and Dowling60 ); however, here we will assume a simple saturation in Q(t):

Q L (t) for |Q (t)| α Q̄
Q (t) = (13.75)
α Q̄ sign Q L (t) for |Q (t)| > α Q̄
This nonlinear flame model is very similar to the one used in a time-domain
approach by Bellucci et al.61 To obtain the nonlinear transfer function we need
to convert to the frequency domain. Setting m i (t) = A cos(ωt)m̄ i , the transfer
function is found by calculating the component of Q (t) at frequency ω,

ω 2π/ω
T (ω, A) = Q (t)e−iωt dt (13.76)
π A Q̄ 0
In this model, for A α/|k| no saturation occurs and so the transfer function
is the same as the linearized version, that is, T (ω, A) = TL (ω) = −ke−iωτ . For
A > α/|k|, it can be shown that
φ/ω (π −φ)/ω
2ωe−iωτ
T (ω, A) = − αe−iωt dt + |k|A cos(ωt)e−iωt dt
πA −φ/ω φ/ω
(13.77)
with φ = cos−1 (1/β), where β = |k|A/α is a scaled amplitude parameter. Evalu-

ating these integrals and combining with the low-amplitude result gives

1 for β 1
T (ω, A)/TL (ω) = 2 cos−1 (1/β) 2(1 − 1/β 2 )1/2 (13.78)
1 − + for β > 1
π πβ
Figure 13.18 shows the variation of T /TL with β. We see that for A > α/|k|
the flame response decays monotonically, tending to zero for large amplitudes.
Note that in this model the phase of the heat release is unaffected by the forcing
amplitude.
As an illustration, we now consider a limit-cycle calculation for the geometry
in example 5. In the linear calculations, we found that there were several unstable
modes for k = 1 in the flame model. The describing-function approach is not
strictly applicable unless there is a single dominant frequency in the limit cycle,
so it is questionable to use the technique for this case. However, if instead we take
12
08
T TL
06
04
02
0
0 1 2 3 4 5 6 7 8 9 10
β
Fig. 13.18 Variation of nonlinear flame-transfer function (normalized by linear

value) with amplitude parameter β=|k|A/α.
k = 0.4 there is only one linearly unstable mode in the frequency range considered.
This is at 290 Hz and has a growth rate of 30 s−1 . Using the nonlinear flame model
in Eq. (13.78) with α set to be 0.1, we find that the corresponding limit cycle has
a frequency of 288 Hz and amplitude A = 0.65. The dimensional mode shape is
shown in Fig. 13.19.
IX. Conclusion
A series of model problems with very simple geometries has been considered
to demonstrate an acoustic analysis of the various components that make up a
gas-turbine combustion system.
The form of the coupling between the heat release and the unsteady flow has
been demonstrated to have a crucial effect on the frequency of oscillation. A one-
term Galerkin series expansion is not adequate to determine this frequency shift
for the sorts of unsteady combustion response typical of gas-turbine combustors.
The effect of the mean temperature ratio across the combustion zone can be signif-
icant. Mean flow effects are not significant for Mach numbers less than about 0.2;
4000
3000
p′ (Pa)
2000
1000
0
0 0 25 05 0 75 1 1 25 1 5 1 75 2 2 25 25 2 75
x (m)
Fig. 13.19 Limit-cycle mode shape for simple combustor.

however, a mean flow does introduce the possibility of a new mode of oscillation
at a much lower frequency where the period of oscillation is set primarily by the
time taken for the convection of entropy waves, or hot spots. Higher-order modes
in the annular and cylindrical ducts bring in the possibility that the modes are cut
off. We have described how a typical LPP combustion system can be built up and
analyzed through connection of a series of cylindrical and/or annular ducts. In
many geometries the premix ducts provide sufficient blockage that these modes
of oscillation are close to separate modes of the plenum and combustor with a
hard or approximately constant velocity boundary condition at the premixer. We
have also noted that modal coupling may occur when the geometry is no longer
axisymmetric, and we have seen how to include acoustic absorbers in the models.
The linear models discussed in this chapter relate to small oscillations only, and
hence they give predictions of the stability of modes but not the amplitude of the
resulting limit cycles. However, these models can still provide important infor-
mation to gas-turbine designers and operators. Furthermore, we have seen that by
using describing-function analysis these models can be extended to give amplitude
predictions. The great advantage of the approaches presented here is their speed.
Many geometry configurations and operating conditions can be investigated in a
relatively short time.
Appendix: Derivation of Eq. (13.41)

When heat conduction and viscous effects are neglected, the entropy Eq. (13.4)
simplifies to
DS
ρT =q (13.A1)
Dt
Replacing S by the perfect gas form S = cv log p − c p log ρ, we obtain
ρT cv D p Dρ
= −c p T +q (13.A2)
p Dt Dt
After substitution for Dρ/Dt from the equation of mass conservation, we obtain
cv D p
∇ ·u+q
= c p Tρ∇ (13.A3)
Rgas Dt
which is equivalent to
Dp
∇ · u + (γ − 1)q
= c2 ρ∇ (13.A4)
Dt
since c2 = γ Rgas T and Rgas /cv = γ − 1. Equation (13.41) is the linearized form
of Eq. (13.A4).
References
1
Dowling, A. P., and Stow, S. R., “Acoustic Analysis of Gas Turbine Combustors,” Journal
2
Rayleigh, L., The Theory of Sound, 2nd ed., Vol. 2, Macmillan, London, 1896, pp. 226–
227.
3
Yang, V., and Anderson, W. E. (eds.), Liquid Rocket Engine Combustion Instability,
Vol. 169, Progress in Astronautics and Aeronautics, AIAA, New York, 1995.
4
De Luca, D., Price, E. W., and Summerfield, M. (eds.), Nonsteady Burning and Com-
bustion Stability of Solid Propellants, Vol. 143, Progress in Astronautics and Aeronautics,
AIAA, New York, 1992.
5
Keller, J. J., “Thermoacoustic Oscillations in Combustion Chambers of Gas Turbines,”
6
Hsiao, G. C., Pandalai, R. P., Hura, H. S., and Mongia, H. C., “Combustion Dynamic
Modeling for Gas Turbine Engines,” AIAA Paper 98-3380, July 1998.
7
Hsiao, G. C., Pandalai, R. P., Hura, H. S., and Mongia, H. C., “Investigation of Combus-
tion Dynamics in Dry-Low-Emission (DLE) Gas Turbine Engines,” AIAA Paper 98-3381,
July 1998.
8
Lovett, J. A., Chu, W.-W., and Shah, S. N., “Modeling of Combustion Chamber Acoustics
and Control of Combustion Instabilities in Gas Turbines,” 6th International Congress on
Sound and Vibration, July 1999.
9
Dowling, A. P., and Hubbard, S., “Instability in Lean Premixed Combustors,” Journal
of Power and Energy, Vol. 214, No. 4, 2000, pp. 317–332.
10
Stow, S. R., and Dowling, A. P., “Thermoacoustic Oscillations in an Annular Combus-
tor,” American Society of Mechanical Engineers, Paper 2001-GT-0037, June 2001.
11
Chu, B.-T., and Kovasznay, L. S. G., “Non-linear Interactions in a Viscous Heat-
Conducting Compressible Gas,” Journal of Fluid Mechanics, Vol. 3, Feb. 1958, pp. 494–
514.
12
Marble, F. E., and Candel, S. M., “Acoustic Disturbance from Gas Non-uniformities
Convected Through a Nozzle,” Journal of Sound and Vibration, Vol. 55, No. 2, 1977,
pp. 225–243.
13
a Choked Nozzle,” Journal of Fluid Mechanics, Vol. 467, Sept. 2002, pp. 215–239.
14
Yang, V., and Culick, F. E. C., “Analysis of Unsteady Inviscid Diffuser Flow with a
Shock Wave,” Journal of Propulsion and Power, Vol. 1, No. 3, 1985, pp. 222–228.
15
Culick, F. E. C., and Rogers, T., “The Response of Normal Shocks in Diffusers,” AIAA
Journal, Vol. 21, No. 10, 1983, pp. 1382–1390.
16
Vol. 394, Sept. 1999, pp. 51–72.
17
Propellant Rocket Engines,” Liquid Rocket Engine Combustion Instability, edited by
V. Yang, and W. E. Anderson, Vol. 169, Progress in Astronautics and Aeronautics, AIAA,
New York, 1995, pp. 3–37.
18
Dowling, A. P., “The Calculation of Thermoacoustic Oscillations,” Journal of Sound
19
Annaswamy, A. M., Fleifil, M., Hathout, J. P., and Ghoniem, A. F., “Impact of Linear
Coupling on Design of Active Controllers for Thermoacoustic Instability,” Combustion
20
Cummings, A., “Ducts with Axial Temperature Gradients: An Approximate Solution
for Sound Transmission and Generation,” Journal of Sound and Vibration, Vol. 51, 1977,
pp. 55–67.
21
Jones, H., “The Mechanics of Vibrating Flames in Tubes,” Proceedings of the Royal
Society of London Series A, Vol. 353, No. 1675, 1977, pp. 459–473.
22
Sujith, R. I., Waldherr, G. A., and Zinn, B., “An Exact Solution for One-dimensional
Acoustic Fields in Ducts with an Axial Temperature Gradient,” Journal of Sound and
Vibration, Vol. 184, No. 3, 1995, pp. 389–402.
23
Kumar, B, M., and Sujith, R. I., “Exact Solution for One-dimensional Acoustic Fields
in Ducts with Polynomial Mean Temperature Profiles,” Journal of Vibration and Acoustics,
Vol. 120, No. 4, 1998, pp. 965–969.
24
Karthik, B., Kumar, B. M., and Sujith, R. I., “Exact Solutions to One-dimensional
Acoustic Fields with Temperature Gradient and Mean Flow,” Journal of the Acoustical
Society of America, Vol. 108, No. 1, 2000, pp. 38–43.
25
Sujith, R. I., “Exact Solutions for Modeling Sound Propagation Through a Combustion
Zone,” Journal of the Acoustical Society of America, Vol. 110, No. 4, 2001, pp. 1839–
1844.
26
Eisenberg, N. A., and Kao, T. W., “Propagation of Sound Through a Variable-Area Duct
with Steady Compressible Flow,” Journal of the Acoustical Society of America, Vol. 49,
No. 1, 1971, pp. 169–175.
27
Subrahmanyam, P. B., Sujith, R. I., and Lieuwen, T., “A Family of Exact Transient
Solutions for Acoustic Wave Propagation in Inhomogeneous, Non-uniform Area Ducts,”
Journal of Sound and Vibration, Vol. 240, No. 4, 2001, pp. 705–715.
28
Zhu, M., Dowling, A. P., and Bray, K. N. C., “Self-excited Oscillations in Combustors
with Spray Atomisers,” Journal of Engineering for Gas Turbines and Power, Vol. 123,
No. 4, 2001, pp. 779–786.
29
Watson, G. N., A Treatise on the Theory of Bessel Functions, 2nd ed., Cambridge
University Press, Cambridge, England, UK, 1944, p. 482.
30
Eversman, W., “Theoretical Models for Duct Acoustic Propagation and Radiation,”
Aeroacoustics of Flight Vehicles: Theory and Practice, edited by H. H. Hubbard, Vol. 2,
Acoustical Society of America, New York, 1994, pp. 101–163.
31
Tyler, J. M., and Sofrin, T. G., “Axial Compressor Noise Studies,” SAE Transactions,
Vol. 70, No. 31, 1962, pp. 309–332.
32
Paschereit, C. O., Schuermans, B., Polifke, W., and Mattson, O., “Measurement of
Transfer Matrices and Source Terms of Premixed Flames,” Journal of Engineering for Gas
Turbines and Power, Vol. 124, No. 2, 2002, pp. 239–247.
33
Ohtsuka, M., Yoshida, S., Inage, S., and Kobayashi, N., “Combustion Oscillation Anal-
ysis of Premixed Flames at Elevated Pressures,” American Society of Mechanical Engineers,
34
Hobson, D. E., Fackrell, J. E., and Hewitt, G., “Combustion Instabilities in Industrial
Gas Turbines—Measurements on Operating Plant and Thermoacoustic Modeling,” Journal
of Engineering for Gas Turbines and Power, Vol. 122, No. 3, 2000, pp. 420–428.
35
Dowling, A. P., “Thermoacoustic Instability,” 6th International Congress on Sound and
Vibration, July 1999, pp. 3277–3292, http://icsv6.dat.dtu.dk.
36
Krüger, U., Hürens, J., Hoffmann, S., Krebs, W., and Bohn, D., “Prediction of Ther-
moacoustic Instabilities with Focus on the Dynamic Flame Behavior for the 3A-Series Gas
Turbine of Siemens KWU,” American Society of Mechanical Engineers, Paper 99-GT-111,
June 1999.
37
Krüger, U., Hürens, J., Hoffmann, S., Krebs, W., Flohr, P., and Bohn, D., “Prediction and
Measurement of Thermoacoustic Improvements in Gas Turbines with Annular Combustion
Systems,” American Society of Mechanical Engineers, Paper 2000-GT-0095, May 2000.
38
Lovett, J. A., and Uznanski, K. T., “Prediction of Combustion Dynamics in a Staged
Premixed Combustor,” American Society of Mechanical Engineers, Paper GT-2002-30646,
June 2002.
39
Ni, A., Polifke, W., and Joos, F., “Ignition Delay Time Modulation as a Contribution to
Thermo-acoustic Instability in Sequential Combustion,” American Society of Mechanical
Engineers, Paper 2000-GT-0103, May 2000.
40
Lawn, C. J., “Interaction of the Acoustic Properties of a Combustion Chamber with
Those of Premixture Supply,” Journal of Sound and Vibration, Vol. 224, No. 5, 1999,
pp. 785–808.
41
Krebs, W., Hoffmann, S., Prade, B., Lohrman, M., and Büchner, H., “Thermoacoustic
Flame Response of Swirl Flames,” American Society of Mechanical Engineers, Paper GT-
2002-30065, June 2002.
42
Cheung, W. S., Sims, G. J. M., Copplestone, R. W., Tilston, J. R., Wilson, C. W.,
Stow, S. R., and Dowling, A. P., “Measurement and Analysis of Flame Transfer Function
in a Sector Combustor under High Pressure Conditions,” American Society of Mechanical
Engineers, Paper GT-2003-38219, June 2003.
43
Miles, J., “The Reflection of Sound Due to a Change in Cross Section of a Circular
Tube,” Journal of the Acoustical Society of America, Vol. 16, No. 1, 1944, pp. 14–19.
44
Alfredson, R. J., “The Propagation of Sound in a Circular Duct of Continuously Varying
Cross-sectional Area,” Journal of Sound and Vibration, Vol. 23, No. 4, 1972, pp. 433–442.
45
Akamatsu, S., and Dowling, A. P., “Three Dimensional Thermoacoustic Oscillation in
an Premix Combustor,” American Society of Mechanical Engineers, Paper GT-2001-0034,
June 2001.
46
Evesque, S., and Polifke, W., “Low-Order Acoustic Modelling for Annular Combustors:
Validation and Inclusion of Modal Coupling,” American Society of Mechanical Engineers,
Paper GT-2002-30064, June 2002.
47
Stow, S. R., and Dowling, A. P., “Modelling of Circumferential Modal Coupling Due
to Helmholtz Resonators,” American Society of Mechanical Engineers, Paper GT-2003-
38168, June 2003.
48
Gysling, D. L., Copeland, G. S., McCormick, D. C., and Proscia, W. M., “Combustion
System Damping Augmentation with Helmholtz Resonators,” Journal of Engineering for
Gas Turbines and Power, Vol. 122, No. 2, April 2000, pp. 269–274.
49
Bellucci, V., Flohr, P., Paschereit, C. O., and Magni, F., “On the Use of Helmholtz
Resonators for Damping Acoustic Pulsations in Industrial Gas Turbines,” Journal of Engi-
neering for Gas Turbines and Power, Vol. 126, No. 2, April 2004, pp. 271–275.
50
Dowling, A. P., and Ffowcs Williams, J. E., Sound and Sources of Sound, Ellis Horwood,
London, 1983.
51
Cummings, A., “Acoustic Nonlinearities and Power Losses at Orifices,” AIAA Journal,
Vol. 22, No. 6, 1984, pp. 786–792.
52
Howe, M. S., Scott, M. I., and Sipcic, S. R., “The Influence of Tangential Mean Flow
on the Rayleigh Conductivity of an Aperture,” Proceedings of the Royal Society of London
Series A, Vol. 452, 1996, pp. 2303–2317.
53
Howe, M. S., “On the Theory of Unsteady High Reynolds Number Flow Through a
Circular Aperture,” Proceedings of the Royal Society of London Series A, Vol. 366, 1979,
pp. 205–223.
54
Dupère, I. D. J., and Dowling, A. P., “The Absorption of Sound by Helmholtz Resonators
with and Without Flow,” AIAA Paper 2002-2590, June 2002.
55
Dupère, I. D. J., and Dowling, A. P., “The Use of Helmholtz Resonators in a Practical
Combustor,” Journal of Engineering for Gas Turbines and Power, Vol. 127, No. 2, April
2005, pp. 268–275.
56
Wang, C.-H., “Actively-Tuned Passive Control of Combustion Instabilities,” Ph.D.
Thesis, Cambridge Univ. Cambridge, England, U.K., 2004.
57
Eldredge, J. D., and Dowling, A. P., “The Absorption of Axial Acoustic Waves by
a Perforated Liner with Bias Flow,” Journal of Fluid Mechanics, Vol. 485, June 2003,
pp. 307–335.
58
Eldredge, J. D., “On the Interaction of Higher Duct Modes with a Perforated Liner
System with Bias Flow,” Journal of Fluid Mechanics, Vol. 510, July 2004, pp. 303–331.
59
Culick, F. E. C., “Some Recent Results for Nonlinear Acoustics in Combustion Cham-
bers,” AIAA Journal, Vol. 32, No. 11, Jan. 1994, pp. 146–169.
60
Stow, S. R., and Dowling, A. P., “Low-Order Modelling of Thermoacoustic Limit
Cycles,” American Society of Mechanical Engineers, Paper GT-2004-54245, June 2004.
61
Bellucci, V., Schuermans, B., Nowak, D., Flohr, P., and Paschereit, C. O., “Thermoa-
coustic Modeling of a Gas Turbine Combustor Equipped with Acoustic Dampers,” American
Society of Mechanical Engineers, Paper GT-2004-53977, June 2004.
Chapter 14
Three-Dimensional Linear Stability Analysis of Gas

Turbine Combustion Dynamics
Danning You∗ , Vigor Yang† , and Xiaofeng Sun‡

Nomenclature
a = speed of sound in mixture
C = constant-volume specific heat for liquid phase
Cf = coefficient of source term model, Equation (14.33)
Ch = coefficient of source term model, Equation (14.33)
CJ = coefficient of Bessel Function of the first kind, Equation (14.20) or
(14.25)
C p = constant-pressure specific heat for two-phase mixture
CY = coefficient of Bessel Function of the second kind, Equation (14.21)
f = Equation (14.12), or frequency
h = Source term in wave equation
I = acoustic intensity, Equations (14.62)
Jm = mth-order Bessel Function of the first kind
kmn = eigenvalue of mnth mode
L = chamber length
M = number of modes in azimuthal direction, or Mach number
N = number of modes in radial direction
N T = total number of cells
p = pressure
q = heat flux vector
Q̇ = rate of heat release
R = Gas constant for two-phase mixture
Rc = radius of circular chamber
∗ Postdoctoral Research Associate. Member AIAA.
† Distinguished Professor. Fellow AIAA.
‡ Professor. Member AIAA. Present address: Beijing University of Aeronautics and Astronautics,
Beijing, China.
415
416 D. YOU AND V. YANG
Rh = hub radius of annular chamber

R p Ru = combustion response coefficient, Eq. (14.47)
r = position vector
r = radial coordinate
S = cross-sectional area
s = entropy
T = temperature
t = time
u = velocity of the gas phase
u = axial velocity
v = radial velocity
w = circumferential velocity
x = axial coordinate
xo = x coordinate at cell interface
Ym = mth-order Bessel Function of the second kind
Greek Symbols
α = axial wave number
βa = reflection coefficient, Eq. (14.66)
βs = reflection coefficient, Eq. (14.68)
ρ = density of two-phase mixture
δ = Kronecker delta
ψ = normal mode function
θ = circumferential coordinate
η = series coefficient of Fourier-type expansion, Eq. (14.35)
γ = specific heat ratio for mixture
= frequency
σ̇ = rate of entropy generation
Overscripts and Superscripts

− = mean quantity

= perturbation quantity
∧ = fluctuation amplitude
+ = downstream running wave
− = upstream running wave
Subscripts
i = imaginary part
j = cell index
l = axial direction mode
m = circumferential direction mode
n = radial direction mode
r = real part
T = transverse plane
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 417
I. Introduction
C OMBUSTION instability, a phenomenon that manifests itself by the occur-

rence of well-organized flow oscillations in combustion chambers, has been
a serious concern in the development of gas-turbine engines, as discussed in Chap-
ters 1–9 of this volume. The instability is highly detrimental to combustor opera-
tion because it causes excessive vibration and heat transfer and, in extreme cases,
catastrophic failure. Extensive efforts have been made to understand, analyze, and
predict the characteristics of combustion instabilities in various operational
and laboratory systems. Two general theoretical approaches have been developed
and implemented: numerical simulations and analytical analyses. Each approach
has advantages that are complementary to the other. Numerical integration of the
complete conservation equations provides more accurate and thorough results for
well-posed problems and serves as the primary means of verifying the validity of
approximate methods. Substantial progress has been achieved for flow and flame
dynamics of single-element and multielement injector rigs. Brief reviews of recent
advances are given in the chapters by Huang et al. and Menon (Chaps. 10 and 11).
Most existing analytical models for treating combustion instabilities are based on
a wave equation of some kind that characterizes the oscillatory flowfield in a cham-
ber. The equation, along with its boundary conditions, is then solved by using either
Green’s function1,2 or Galerkin3−5 methods. The latter is now considered to be
the standard method for investigating combustion instabilities in solid-propellant
rocket motors,6 in which the mean flow property variations and Mach number are
assumed to be small. Under such conditions, the frequencies and spatial variations
of unsteady motions deviate only slightly from the classical acoustic field obtained
for the same geometry but lack any source terms. The acoustic field can conve-
niently be expressed as a synthesis of the normal modes with time-varying ampli-
tudes. However, for many practical combustion devices, such as gas-turbine main
combustors and augmenters, the large Mach number and mean flow property vari-
ations in the chamber prohibit the use of the standard Galerkin method. Although
several analyses of combustion instabilities have recently been conducted for gas-
turbine engines, most of them focused on one-dimensional thermoacoustic insta-
bilities in relatively simple geometries. What is known about three-dimensional
oscillations commonly observed in operational systems is limited.
The purpose of this chapter is to establish a three-dimensional linear stability
analysis that is capable of treating acoustic oscillations in complex geometries with
mean flow gradients. The work complements the acoustic analyses described in
the chapter by Dowling and Stow (Chap. 13), in which carefully selected examples
are given for longitudinal disturbances in straight ducts and model gas-turbine en-
gines, as well as for three-dimensional waves in cylindrical and annular chambers.
This chapter also discusses the effects of unsteady heat release, mean temperature
gradient, convective velocity, and acoustic resonator on acoustic wave motions to
introduce several key concepts of instability characteristics. The general approach
developed in this chapter allows us to treat a broader class of problems involving
complicated configurations and nonuniform flow distributions that are represen-
tative of operational gas-turbine combustors. The work proceeds in several steps.
First, a generalized wave equation that accommodates various distributed and
boundary source terms in gas-turbine combustors is derived. Second, to account
acoustic damper
diffuser shroud
turbine
flow vane
fuel unsteady oscillation
compressor nozzle combustion s
diffuser
swirler
primary air
cooling air
Fig. 14.1 Schematic of a gas turbine combustor.
for the effects of geometric and flow variations, the chamber is discretized into a
number of circular or annular cells along the axial direction. The cross-sectional
area and axial distributions of mean flow properties are assumed to be uniform
within each cell. A combined modal-expansion and spatial averaging technique
is then applied to solve for unsteady motions in one cell. The next step involves
matching the oscillatory flowfields in adjacent cells at the interface according to
the conservation laws. Finally, a set of equations is established by combining all
the interface and boundary conditions. The procedure eventually leads to deter-
mining the stability characteristics of the entire system of concern. The analysis
is validated against several well-defined problems for which either closed-form or
numerical solutions are available. A parametric study is also conducted to inves-
tigate the underlying mechanisms for driving instabilities in a model combustor.
II. Theoretical Formulation

For purposes of illustration, we consider a generic gas-turbine combustor, as
shown schematically in Fig. 14.1. Mixtures of fuel and air enter the primary com-
bustion zone after passing through injectors. The flow then accelerates in the
chamber, as a result of area reduction and heat release from chemical reactions,
and becomes nearly choked as it exits the combustor. Cooling air is added into the
secondary combustion zone from an outer passage through the combustor liners.
Hence, flow nonuniformities and temperature gradients exist throughout the com-
bustor. In certain designs, passive control devices, such as acoustic cavities, may
be employed along the combustor walls to suppress oscillations.
A. Wave Equation
A generalized wave equation that characterizes the acoustic motions of a two-
phase mixture in the combustion chamber is derived. Following the approach by
Culick and Yang,4,5 the conservation equations of mass, momentum, and energy
can be expressed in the following forms, written for the behavior of the gas phase:
∂ρ
+ u · ∇ρ = W (14.1)
∂t
∂u
ρ + ρu · ∇u = −∇ p + F (14.2)
∂t
∂p
+ u · ∇ p = −γ p∇ · u + P (14.3)
∂t
where u and p are the velocity and pressure of the gas phase, respectively but
where ρ and γ are the mass-averaged values of the condensed and gas phases. The
source terms W, F, and P include the effects of viscous dissipation, heat release,
and two-phase interactions. Their explicit expressions are given in Refs. 4 and 5.
To derive the wave equation, each dependent variable is decomposed into a
time-mean and a fluctuating quantity:
u(r , t) = ū(r ) + u (r , t)
ρ(r , t) = ρ̄(r ) + ρ (r , t) (14.4)

p(r , t) = p̄(r ) + p (r , t)
The density fluctuation contains two components: an isentropic part that propagates
in the form of an acoustic wave, and a nonisentropic part that results from entropy
oscillation and is convected with the local mean flow. Thus, from the equation of
state, we have

∂ρ ∂ρ
dρ = dp + ds (14.5)
∂p s ∂s p
After some straightforward manipulations to first-order approximation, the density

fluctuation can be written as
1 ρ̄
ρ = p − s (14.6)
ā 2 Cp
where C p is the constant-pressure specific heat of the mixture. Similarly, the tem-
perature fluctuation can be written as
1 ā 2
T = p + s (14.7)
ρ̄C p (γ − 1) C 2p
Substituting the decomposed variables into Eqs. (14.2) and (14.3) and linearizing
the result yields
∂u
∇ p = −ρ̄ − ρ̄(ū · ∇)u − ρ̄(u · ∇)ū − ρ (ū · ∇)ū + F (14.8)
∂t
∂ p
+ ū · ∇ p + u · ∇ p̄ = −γ̄ ( p̄ · ∇u + p · ∇ ū) + P (14.9)
∂t

We differentiate Eq. (14.9) with respect to time and substitute Eq. (14.8) for ∂u ∂t
to find the wave equation governing the oscillatory field in a two-phase mixture:
1 ∂ 2 p
∇ 2 p − =h (14.10)
ā 2 ∂t 2
where the inhomogeneous term h has the following form:
h = hI + hI I + hI I I (14.11)
And where
∂u
h I = − ∇ ρ̄ − ∇[ρ̄(ū · ∇)u ] − ∇[ρ̄(u · ∇)ū]
∂t
p 1 ∂ 1 ∂
− ∇ 2 (ū · ∇)ū + 2 (ū · ∇ p ) + 2 (u · ∇ p̄) (14.11a)
ā ā ∂t ā ∂t
γ̄ ∂ 1 ∂
+ 2 ( p̄∇ · u ) + 2 ( p ∇ · ū)
ā ∂t ā ∂t
ρ̄s
hI I = ∇ (ū · ∇)ū (14.11b)
C̄ p
1 ∂
h I I I = − 2 P + ∇F (14.11c)
ā ∂t
The first term, h I , represents the linear gasdynamic effect. The second term, h I I ,
arises from entropy fluctuations. The third term, h I I I , is related to processes,
such as two-phase interactions, combustion heat release, and viscous dissipation.
Equation (14.10) suggests that combustion instabilities can be viewed as classical
acoustic motions perturbed by various source terms in the chamber. The major
driving source inevitably arises from unsteady heat release because of its exceed-
ingly large contribution to the energy of the oscillatory field as compared with
other source terms.
B. Boundary Conditions
The boundary condition for Eq. (14.10) is set on the gradient of p , obtained
by taking the scalar product of the outward normal vector with the perturbed
momentum equation, Eq. (14.8).
n · ∇ p = − f
∂u
= n · − ρ̄ − ρ̄(ū · ∇)u − ρ̄(u · ∇)ū − ρ (ū · ∇)ū + F (14.12)
∂t
It can be treated conveniently by using an acoustic admittance function, Ad , which

relates the local velocity fluctuation to its pressure counterpart:
u /ā
Ad = n · (14.13)
p /γ p̄
If all perturbations are absent—with both functions h and f vanished—the wave

equation for classical acoustics in a closed volume having a rigid wall is recovered.
1 2 3 j j +1
Fig. 14.2 Discretization of combustion chamber into cells in axial direction.
C. Modal Expansion and Spatial Averaging

The wave equation has mixed terms that involve derivatives in all three spatial
directions. Direct treatment of this equation subject to inhomogeneous boundary
conditions presents serious challenges that arise from complex geometries, mean
flow gradients, and various distributed and boundary source terms of concern. To
circumvent this obstacle, the combustor is divided into a number of cells along
the axial direction, as shown in Fig. 14.2, such that the cross-sectional area and
mean axial-flow properties can be taken as uniform within each cell. Furthermore,
because the mean flow Mach number and variations of flow properties in the trans-
verse (both radial and azimuthal) directions are small for most practical systems,
the spatial structures of unsteady motions on the transverse plane deviate slightly
from the classical acoustic field obtained for the same cell geometry, but without
any source terms. The acoustic field in each cell can be synthesized as a Fourier-
type series in terms of the eigenfunctions for the transverse plane ψ, along with
temporal and axial variations. In the cylindrical coordinates, the expansion can be
expressed as

∞
∞
p (r , t) = [ψmn (θ, r )ηmn (x, t)] (14.14)
n=0 m=−∞
where subscripts m and n stand for the spatial variations in the circumferential and
radial directions, respectively. The problem then becomes solving for the series
coefficients, ηmn (x, t).
The eigenfunction, also called the normal mode, satisfies the Helmholtz equation
in the transverse plane,
∇T2 ψmn + kmn

2
ψmn = 0 (14.15)
subject to the following boundary condition for a rigid surface along the combustor
wall.
n · ∇T ψmn = 0 (14.16)
where kmn is the wave number. The transverse Laplacian operator ∇T2 in the cylin-
drical coordinates is defined as

1 ∂ ∂ 1 ∂2
∇T =
2
r + 2 2 (14.17)
r ∂r ∂r r ∂θ
The eigenfunction ψmn can be constructed to be orthonormal:

ψmn ψm n ds = δmm · δnn (14.18)
where the integral is performed throughout the entire cross section and where δ
denotes the Kronecker delta function.
For an annular duct with the inner and outer radii of Rt and Rh , respectively,
the eigenfunction can be expressed as
1
ψmn (θ, r ) = √ exp(imθ )[C J mn Jm (kmn r ) + CY mn Ym (kmn r )] (14.19)
2π
where

−1/2
Rt2 m2 Rh2 m2
C J mn = 1 − 2 2 Bm,t −
2
1− 2 2
Bm,h (14.20)
2 kmn Rt 2 kmn Rh
and where

dJm (kmn r ) dr
CY mn = − C J mn (14.21)
dYm (kmn r ) dr
r =Rt
The coefficients Bm,t and Bm,h are

dJm (kmn r ) dr
Bm,t = Jm (kmn Rt ) + Ym (kmn Rt )
dYm (kmn r ) dr
r =Rt

dJm (kmn r ) dr
Bm,h = Jm (kmn Rh ) + Ym (kmn Rh ) (14.22)
dYm (kmn r ) dr
r =Rt
The eigenvalues kmn can be determined by applying the boundary condition,

Eq. (14.16), at both the inner and outer walls:

dJm (kmn r ) dYm (kmn r )
·
dr r =Rt dr r =Rh

dJm (kmn r ) dYm (kmn r )
− · =0 (14.23)
dr r =Rh dr r =Rt
For a circular duct with radius Rc , the transverse eigenvalue takes the form
1
ψmn (θ, r ) = √ exp(imθ )C J mn Jm (kmn r ) (14.24)
2π
where

−1/2
Rc2 m2
CJ mn = 1 − 2 2 Jm (kmn Rc )
2
(14.25)
2 kmn Rc
The eigenvalue kmn can be obtained from the following wall condition:
d
Jm (kmn r ) = 0 at r = Rc (14.26)
dr
After obtaining the eigenfunction ψmn , a spatial-averaging technique equivalent

to the Galerkin method is applied to the transverse plane to solve for the series co-
efficient ηmn . By multiplying Eq. (14.15) by p and Eq. (14.10) by ψmn , subtracting
the results, and integrating over the cross section, we have

1 ∂ 2 p
p ∇T2 ψmn + p kmn
2
ψmn − ψmn ∇ 2 p + ψmn ds = − ψmn hds
ā 2 ∂t 2
(14.27)
Applying Green’s theorem and substituting boundary conditions (12) and (16) into
Eq. (14.27) yields

∂ 2 p 1 ∂ 2 p
p kmn
2
ψmn − ψmn + ψ mn 2 ds = − ψmn hds − ψmn f T dl
∂x2 ā ∂t 2
(14.28)

where f T = −n · ∇T p at the wall. The line integral dl is performed along the
surface of the cross section. Substitute Eq. (14.14) into Eq. (14.28) and rearrange
the result to yield

∂ 2 ηmn 1 ∂ 2 ηmn
2
kmn ηmn − + ψ 2
ds = − ψmn hds − ψmn f T dl
∂x2 ā 2 ∂t 2 mn
(14.29)
For linear stability analysis, each fluctuating quantity can be decomposed into a
spatial and a time-harmonic temporal part, i.e.,
ηmn (x, t) = η̂mn (x) · exp(i t), h (r , t) = ĥ (r ) · exp(i t), etc. (14.30)
The overhat ∧ denotes a complex function of spatial coordinates. The characteristic

modal frequency is also complex:
= r + i i (14.31)
The real part r represents the radian frequency of oscillation, and the imaginary
part i is called the damping coefficient, because its value determines the decay
rate of a particular acoustic mode. Consequently, the equation governing the axial
variation is derived as follows:

d2 η̂mn 2
+ − k 2
mn η̂mn = ψ mn ĥds + ψ fˆ
mn T dl (14.32)
dx 2 ā 2
The source terms ĥ and fˆT are functions of both the mean and oscillatory flow
properties. The latter consists of a series of transverse acoustic modes. For a reason-
able approximation, the acoustic mode coupling in evaluating the source terms in
Eq. (14.32) can be ignored because of the disparity of the length scales associated
with these modes. It can easily be shown that the cross-coupling terms are much
smaller and that only the specific mode of concern dominates. Thus, to facilitate
formulation, the surface and line integrals on the right-hand side of Eq. (14.32)
can be modeled as the products of the axial variation η̂mn and coefficients C h,mn
and C f,mn , which incorporate all the distributed and surface effects at a given cross
section through spatial averaging:

ψmn ĥds = C h,mn η̂mn (x)

ψmn fˆT dl = C f,mn η̂mn (x) (14.33)
Equation (14.32) then reduces to a second-order ordinary differential equation

with constant coefficients, whose solution η̂mn (x) takes the form
+
+ −
−
η̂mn (x) = pmn exp iαmn x + pmn exp iαmn x (14.34)
− +
where pmn and pmn are the complex amplitudes of the upstream and downstream
±
traveling waves, respectively. The axial wave number αmn is related to the fre-
quency , eigenvalue kmn , and source-term coefficients:
±
2 2 ± ±
αmn = 2
− kmn
2
− C h,mn αmn , · · · − C f,mn αmn , ··· (14.35)
ā
So far, by means of normal-mode expansion, the solution of the wave equation

has been given in the form of Eq. (14.14), with the eigenfunction provided by either
Eq. (14.19) or Eq. (14.24). The axial variation expressed in Eq. (14.34) is derived
by applying spatial averaging over each transverse plane. The acoustic pressure in
each cell can thus be explicitly expressed by combining these results:
∞
∞
+ −
p (r , t) = ei t + iαmn
ψmn (θ, r ) pmn − iαmn
e x + pmn e x (14.36)
n=0 m=−∞
The axial velocity fluctuation u can be obtained from the linearized x-

momentum equation:

1 ∞ ∞
α + + iαmn
p e
+
x
α − − iαmn
p e
−
x
u (r , t) = − ei t ψmn (θ, r ) mn mn
+ + mn mn −
ρ̄ n=0 m=−∞ + ūαmn + ūαmn
(14.37)
Similarly, the radial and circumferential velocity fluctuations are derived by sub-
stituting Eq. (14.36) into the linearized momentum equations in the radial and
circumferential directions, respectively:

ei t
∞ ∞ + −
+ iαmn − iαmn
∂ψmn (θ, r ) pmn e x pmn e x
v (r , t) = i · + + −
ρ̄ n=0 m=−∞ ∂r + ūαmn + ūαmn
(14.38)

ei t ∞ ∞ + −
+ iαmn x − iαmn x
pmn e pmn e
w (r , t) = − mψmn (θ, r ) · + + −
ρ̄r n=0 m=−∞ + ūαmn + ūαmn
(14.39)
In addition to the acoustic field, entropy fluctuation arising in the flame zone
must be treated with care. The following transport equation is used:
Ds ∇ ·q
=− + σ̇ (14.40)
Dt ρT
The first term on the right-hand side represents the rate of entropy change caused by
the heat flux, q. The second term stands for irreversibilities, which can be modeled
as follows if we only consider the entropy generated from heat release, Q̇:
Q̇
σ̇ = (14.41)
ρT
Following the procedure described in Ref. 7, we obtain the equation governing the
transport of entropy oscillation:

∞
∞

s = exp(i t) exp(−i x/ū) ŝmn + Cs,mn ψmn (r, θ ) (14.42)
m=−∞ n=0
where ŝmn is the amplitude of the entropy disturbance and where Cs,mn is a co-
efficient associated with the unsteady heat release represented by pressure and
velocity perturbations

+
Q̄ x Rp αmn Ru +
−
Cs,mn (x, r, θ ) = − + eiαmn x pmn
ρ̄ T̄ ū p̄ ρ̄ ā( + ūαmn )
−
Rp αmn Ru −
x −
+ − − e iαmn
p mn (14.43)
p̄ ρ̄ ā( + ūαmn )
The interaction between entropy fluctuation and mean-flow gradients, as shown

in Eq. (14.11b), represents an important source term in driving combustion insta-
bilities, especially in the low-frequency range. The phenomenon often occurs in
regions with rapid velocity variations, such as choked nozzles.8
D. Treatment of Inhomogeneous Terms

The stability analysis requires explicit modeling of the source terms ĥ and fˆT
in Eq. (14.32) and subsequently the coefficients C f,mn and C h,mn in Eq. (14.35)
to determine the wave characteristics in each cell. Because these terms depend
on the specific processes of concern, developing general expressions covering
all scenarios that may take place is impractical. In principle, these terms can be
expressed in terms of the mean and fluctuating quantities. The results are then
substituted into Eq. (14.33) to determine the coefficients, C f,mn and C h,mn , so as to
facilitate model closure. As examples, the source terms resulting from mean flow,
combustion heat release, and boundary conditions are formulated below.
1. Effect of Mean Flow

If a uniform mean flow in the axial direction is the only one considered, the
inhomogeneous term in the wave equation, Eq. (14.10), becomes
2ū ∂ 2 p ū 2 ∂ 2 p
h= + (14.44)
ā 2 ∂ x∂t ā 2 ∂ x 2
Substituting of Eq. (14.44) into Eq. (14.33) leads to

± 2 M̄ ± ± 2
C h,mn = ψmn
2
− αmn − M̄ αmn
2
ds (14.45)
ā
2. Effect of Combustion Heat Release

The second case is concerned with combustion heat addition, expressed as

h = − (γ − 1) i Q̇ /ā 2 (14.46)
Following common practice, the oscillatory heat-release rate can be conveniently

related to local pressure and velocity fluctuations, as follows:

Q̇ p u v w
= Rp + Ru + Rv + Rw (14.47)
Q̄ p̄ ā ā ā
where R p and Ru , Rv , and Rw are complex variables commonly referred to as the

pressure- and velocity-coupled response functions, respectively. Substituting the
oscillatory flow properties, Eqs. (14.36)–(14.39), into Eq. (14.47) and applying
Eq. (14.33) yields

±
C h,mn = ψmn G ±
h,mn ds (14.48)
where
i(γ − 1) Q̄
G±
hmn = −
ρ̄ ā 2
±
R p ρ̄ Ru αmn ψmn iRv ∂ψmn m Rw ψmn
× ψmn − ± + ± − ±
p̄ ā( + ūαmn ) ā( + ūαmn ) ∂r ār ( + ūαmn )
(14.49)
The formulation is thus closed after those combustion response functions are estab-
lished. Several empirical and analytical models, including the time-lag9 and flame
response10,11 models, were developed and employed to represent the combustion
responses in gas-turbine combustors.
3. Effect of Surface Condition

The third case treats the boundary effect, which may arise from the implemen-
tation of such passive control devices as Helmholtz resonators and quarter-wave
tubes, to suppress oscillations in gas-turbine engines. Detailed information about
the use of acoustic dampers is given in the chapter by Richards et al. (Chap. 17).
Since those devices are typically installed on the combustor wall, they can best be
modeled as boundary conditions for the wave equation. If the mean flow influence
is ignored, Eq. (14.12) reduces to
∂v
f T = ρ̄ = ρ̄(i )v̂ (14.50)
∂t
where v is the radial velocity fluctuation at the resonator entrance. It can be related
to the local pressure fluctuation by means of the acoustic admittance function
defined in Eq. (14.13). Thus, we have
f T = i Ad p̂/ā (14.51)
sˆ j −1
Rc,j
Rc,j-1 sˆ j
p +j −1 p+j
j-1 p−j −1 j
p−j
x0
0 0+
-
Fig. 14.3 Schematic of two adjacent cells with different cross-sectional areas.
Substituting of Eq. (14.51) into Eq. (14.33) gives rise to the formula for coefficient
C f,mn :

C f,mn = ψmn
2
i Ad /ā dl (14.52)
E. Matching Conditions
The oscillatory field in each cell must be matched with its counterpart in ad-
jacent cells by enforcing conservation laws at the interfaces. Figure 14.3 shows
schematically the fluctuating quantities on both sides of the interface at x0 between
two neighboring cells having different cross-sectional areas. The matching condi-
tions require continuities of mass, momentum, and energy fluxes over the region
0 ≤ r ≤ Rc, j−1 . It is assumed that Rc, j−1 ≤ Rc, j without loss of generality.
Mass flux:
(ρu)| j−1, x0− = (ρu)| j, x0+ (14.53)
Momentum flux:

p + ρu 2 j−1, x − = p + ρu 2 j, x + (14.54)
0 0
Energy flux:

1 1
C p T + u 2 = C p T + u 2 (14.55)
2 j−1, x0− 2 j, x0+
Because no mass flow is allowed to pass through the solid region Rc, j−1 ≤ r ≤ Rc, j
at the interface,
(ρu)| j, x0+ = 0, Rc, j−1 ≤ r ≤ Rc, j (14.56)

Equation (14.56) must be combined with the mass continuity equation (14.53) to
complete the mass-balance condition.
A general form of these matching conditions can be derived in terms of the
+ −
wave amplitudes pmn , pmn , and ŝmn and corresponding coefficients. The deriva-
tion is achieved by 1) decomposing the variables in Eqs. (14.53)–(14.56) into the
mean and fluctuating parts; 2) linearizing the results; 3) substituting the fluctuating
quantities given in Eqs. (14.6), (14.7), (14.36), (14.37), and (14.42) into the lin-
earized equations; and 4) combining terms with the wave amplitudes of the same
kind:

∞
∞

A+
m n pm+ n + A−
m n pm− n + Asm n ŝm n
m =−∞ n =0 j−1

∞
∞

+ A+pq p +pq + A−pq p −pq + Aspq ŝ pq =0 (14.57)
p=−∞ q=0 j
The explicit expressions of the coefficients in the preceding equation can be found
in Ref. 7.
Equation (14.57) indicates that the interfacial matching conditions contain in-
+ −
finite summations of unknown acoustic and entropy wave amplitudes, pmn , pmn ,
and ŝmn . In practice, a finite number of modes is sufficient because higher modes
will be either cut off or damped out. Thus, the total numbers of modes in the cir-
cumferential direction, M, and in the radial direction, N , can be selected to provide
a faithful solution. The number of unknowns in terms of wave amplitudes within
each cell is 3MN . For this reason, 3MN equations are required for each interface.
However, only three equations, Eqs. (14.53)–(14.55), have been formulated so far.
The additional equations can be constructed by employing orthonormal proper-
ties of eigenfunctions, according to the procedure given subsequently. A detailed
derivation is given in Ref. 7.
The general form for the momentum and energy interfacial conditions, Eqs.
(14.54) and (14.55), given in Eq. (14.57), is further manipulated by multiplying
the eigenfunction ψmn, j−1 and integrating the result over the cross-sectional area
S j−1 . This operation yields
M−1
N −1

A+
m n pm+ n + A−
m n pm− n + Asm n ŝm n ψmn, j−1 ds
m =−∞ n =0 j−1
S j−1
M−1
N −1

+ A+pq p +pq + A−pq p −pq + Aspq ŝ pq ψmn, j−1 ds = 0 (14.58)
p=−∞ q=0 j
S j−1
As a result of the orthonormal property of the transverse eigenfunctions,

Eq. (14.18), the double summation of m and n in the first integral of Eq. (14.58)
vanishes, thereby giving

+ + − −

Bmn pmn + Bmn pmn + Bsmn ŝmn j−1
M−1
N −1

+ + − −

+ A pq p pq + A pq p pq + Aspq ŝ pq ψmn, j−1 ds = 0
p=−∞ q=0 j
S j−1
m = 0, 1, . . . , M − 1 (14.59)
n = 0, 1, . . . , N − 1
Equation (14.59) can be rearranged in the following form:
+ + − −

Bmn pmn + Bmn pmn + Bsmn ŝmn j−1

M−1 N −1
+
+ B pq,mn p +pq + B − −
pq,mn p pq + Bs, pq,mn ŝ pq =0
p=−∞ q=0 j
m = 0, 1, . . . , M − 1 (14.60)
n = 0, 1, . . . , N − 1
Similarly, the general form of Eq. (14.57) for the mass interfacial condition,
Eqs. (14.53) and (14.56), can be rearranged by multiplying eigenfunction ψmn, j
and integrating the result over the cross-sectional area S j , yielding

M−1 N −1
+
+ − −
Bm n ,mn pm n + Bm n ,mn pm n + Bs,m n ,mn ŝm n
m =−∞ n=0 j−1
+ + − −

+ Bmn pmn + Bmn pmn + Bsmn ŝmn j = 0,
m = 0, 1, . . . , M − 1
n = 0, 1, . . . , N − 1 (14.61)
Consequently, 3MN matching conditions are obtained from the mass, momentum,
and energy balances and are given by Eqs. (14.60) and (14.61).
F. Boundary Conditions
The acoustic boundary conditions at the combustor inlet and outlet play an
important role in determining the stability behavior of the entire system and must
be carefully treated. The effect of the boundary can be effectively measured with
an acoustic admittance function Ad , which quantifies the magnitude and direction
of the energy flow across the boundary, as evidenced in the expression for the
acoustic intensity, I :
I = p u = Ad · p 2 /(ρ̄ ā) (14.62)
It is apparent that energy is delivered into the system if the pressure and velocity
fluctuations are in phase, thereby exerting a destabilizing influence. A more rig-
orous theory of linear combustion stability clearly shows that the growth rate and
frequency of oscillation depend on the real and imaginary parts of the admittance
function, respectively.4
By substituting the expressions of pressure and velocity fluctuations, Eqs.

(14.36) and (14.37), into Eq. (14.13) and manipulating the result, we obtain the
condition at the combustor entrance:
+ −
αmn + αmn −
Ad + ā + pmn + Ad + ā − pmn =0 (14.63)
+ ūαmn + ūαmn
The entropy fluctuation at the combustor inlet is assumed to be zero,
ŝmn = 0 (14.64)
At the combustor exit, the boundary conditions can be more conveniently spec-
ified by the reflection coefficients. The acoustic wave reflected from the boundary
consists of contributions from the incident acoustic and entropy disturbances. Thus,
the pressure of the reflected wave takes the form

p − = pa− + ps− = βa p + + βs ŝ· (γ p̄) C p (14.65)
where the acoustic reflection coefficient is defined as
βa = pa− / p + (14.66)
It can be related to the admittance function as follows:

+ −
āαmn āαmn
βa = − 1 + + 1+ − (14.67)
Ad ( + αmn ū) Ad ( + αmn ū)
The entropy reflection coefficient βs is defined as

βs = ps− γ p̄ ŝ/C p (14.68)
By applying Eq. (14.65) at the combustor exit, we can express the outlet boundary
equation as follows, in terms of the amplitudes of the incident and reflected acoustic
waves, as well as the entropy fluctuation:
+ + − − γ p̄
βa exp(iαmn d x) pmn − exp(iαmn d x) pmn + βs ŝmn exp (−i d x/ū) = 0
C̄ p
(14.69)
where d x is the axial length of the boundary cell.

The acoustic admittance function, or the reflection coefficient, is determined
by the characteristics of the boundary itself. For example, when the boundary is
connected with a plenum chamber such as a diffuser, the pressure fluctuation var-
nishes, and the admittance function becomes infinity. If the boundary is rigid, the
vanished velocity fluctuation results in a zero admittance function. For a situa-
tion between these two extremes, the admittance function can be obtained either
experimentally or analytically.
If the combustor exit is choked by a compact nozzle, the fluctuating quantities in
the axial direction at the nozzle entrance (or combustor exit) satisfy the following
relation.8
2u ρ p
+ − =0 (14.70)
ū ρ̄ p̄
The same condition is valid for circumferentially varying disturbances in a nar-

row annular gap.12 A simple manipulation of Eq. (14.70) leads to the reflection
coefficient at the nozzle entrance under a choked condition.
1 − (γ − 1) M̄/2 − M̄/2
βa = and βs = (14.71)
1 + (γ − 1) M̄/2 1 + (γ − 1) M̄/2
In the limit of zero Mach number, βa and βs approach unity and zero, respectively,
rendering an acoustically closed boundary.
G. System Equations
It has previously been shown that the number of unknowns in terms of the wave
+ −
amplitudes pmn , pmn , and ŝmn (n = 0, 1, · · · , N − 1 and m = 0, 1, · · · , M − 1)
for the oscillatory flowfield in each cell is 3MN , where M and N are the numbers
of the tangential and radial acoustic modes, respectively. If the combustor is divided
into NT cells along the length of the chamber, then in addition to the frequency ,
the total number of unknowns is 3 · N · M · NT unknowns. The number of equa-
tions is also 3 · N · M · NT , obtained by combining the (3 · N · M) · (NT − 1)
interfacial matching conditions, in the form of Eq. (14.60), and the 3MN inlet and
exit boundary conditions given in Eqs. (14.63), (14.64), and (14.69). As a result,
a set of equations governing the acoustic characteristics of the entire system is
established as follows:
  
C1 ··· ··· ··· ··· ··· D1
 . .. .. .. .. ..  . 
 .. . .   .. 
 . . .  
··· ··· C j−1
··· ··· ···   D j−1 
  
  =0 (14.72)
··· ··· ··· Cj ··· ···  Dj 
 .  . 
 . .. .. .. .. ..  . 
 . . . . . .  . 
··· ··· ··· ··· ··· C NT
DNT
where C j is a matrix consisting of the coefficients B of the matching conditions,

Eq. (14.60), at the interface between cells j − 1 and j. The column vector D j
+ −
contains the unknown variables pmn , pmn , and ŝmn . To find a nontrivial solution D,
the determinant of the matrix C must be zero. This condition establishes the char-
acteristic equation for the eigenvalue frequency . Once the frequency becomes
known, the spatial distributions of the acoustic and entropy waves can be obtained
straightforwardly from Eq. (14.72).
III. Solution Procedure

The previous section introduced a general three-dimensional linear acoustic
analysis of gas-turbine-combustion instability. The overall approach can be sum-
marized by the following steps.
1) Define the domain of concern.

2) Obtain the mean flow properties from numerical simulations or experimental
measurements.
3) Determine the acoustic-boundary conditions at the combustor inlet and outlet.
4) Discretize the combustor into a number of cells along the axial direction
according to the mean flow distribution and chamber geometry.
5) Calculate the bulk flow properties within each cell.
6) Determine the source terms arising from volumetric and boundary effects.
7) Construct the system equation and perform numerical calculations for the
eigenfrequencies that characterize the stability characteristics.
8) Calculate the acoustic and entropy fields on the basis of the predicted oscil-
lation frequency.
IV. Sample Studies

The analysis developed in the preceding sections was used to calculate acoustic
oscillations in a variety of environments. The results were compared with either
analytical or numerical solutions, where available, to assess the validity of the
overall approach. First, longitudinal acoustic waves in channels with geometric
and mean-temperature variations were obtained. Excellent agreement was ob-
served between frequencies and mode shapes predicted by the present analysis
and exact solutions. A detailed discussion can be found in Ref. 7. In the remaining
cases, three-dimensional acoustic fields in a step duct and a straight duct with a
temperature jump are treated to further validate the present analysis. The stability
characteristics of a model gas-turbine combustor are also investigated to examine
the underlying mechanisms for driving instabilities.
A. Acoustic Field in Step Duct

This case deals with the acoustic flowfield in a step duct with uniform mean
temperature, as shown in Fig. 14.4. The ratio of the cross-sectional area is four,
R2
R1
L1
L2
Fig. 14.4 Schematic of a chamber with sudden expansion.

Table 14.1 Acoustic oscillations frequencies

of step duct (R2 /R1 = 2, L1 = L2 = R1 )
Mode 1T 1T/1R
Frequency ANSYS 1.95 4.38

R2 /ā Present Analysis 1.96 4.40
thereby serving as a challenging test problem. The duct is discretized into two
cells, and the cross-sectional area of each cell is uniform. Three different acoustic
modes are studied: the first longitudinal (1L), the first tangential (1T), and the
mixed first tangential/first radial (1T/1R) mode. If all source terms are absent and
if the mean-flow Mach number is ignored, the frequency of oscillation depends
only on the sound speed and chamber configuration. In other words, the acoustic
characteristics in each cell can be determined by the following Helmholtz equation,
subject to appropriate boundary conditions:

d 2 η̂mn 2
+ − k mn η̂mn = 0
2
(14.73)
dx2 ā 2
If the duct is acoustically closed, the frequency of the 1L mode can be analytically
determined from the following equation:
R12 tan ( L 1 /ā) + R22 tan ( L 2 /ā) = 0 (14.74)
where L 1 , L 2 , R1 , and R2 denote the lengths and radii of the small and large
cells, respectively. The present analysis predicts the longitudinal-mode frequencies
identical to the analytical solutions of Eq. (14.74).
Results for transverse oscillations are validated against calculations by the finite
element software ANSYS.13 Table 14.1 compares the frequencies of the 1T and
1T/1R modes obtained from the present analysis and from ANSYS. The discrep-
ancy is less than 1%. Two azimuthal modes (i.e., M = 3) and two radial modes
(i.e., N = 3) were used to described the wave motion in each cell. Fig. 14.5 shows
the calculated acoustic pressure fields of the two modes. For pure transverse oscil-
lations in a straight duct with uniform temperature, the axial wave number in Eq.
(14.35) is zero. The oscillation frequencies normalized by ā/R2 for the 1T mode
of the simple small and large cells are 3.68 and 1.84, respectively. For the 1T/1R
mode, the normalized frequency of the large cell is 5.33. However, because of the
geometric change the axial wave number αmn for a duct with an area variation can-
not be zero. For the 1T mode, the acoustic field in the large cell dictates the wave
motion in the entire chamber. The frequency of 1.96 is close to that of a simple cell
with a diameter R2 (i.e., 1.84). This frequency yields a nontrivial complex axial
wave number in the small cell to satisfy the boundary condition, thereby leading to
an axially attenuated 1T wave. In other words, the 1T mode is cut off in the small
cell, in which the oscillation has a much lower amplitude, as shown in Fig. 14.5a.
Similarly, for the 1T/1R mode in the entire chamber, the prevalence of the wave
a) b)
1.00
0.60
0.20
c) d)
−0.20
−0.60
−1.00
Fig. 14.5 Distributions of acoustic pressures in a step duct (R2 /R1 = 2, L1 = L2 =

R1 , axial coordinate extended for better resolution): a) and b) first tangential mode
from different views (ΩR2 /ā = 1.96); c) and d) first tangential/first radial mode from
different views (ΩR2 /ā = 4.40).
motion in the large cell results in a cutoff phenomenon in the small cell, as shown
in Fig. 14.5b. The normalized frequency of 4.40 for the whole duct falls between
3.68 (i.e., the 1T mode for a simple small cell) and 5.33 (i.e., the 1T/1R mode for
a simple large cell).
The coexistence of different modes with structures corresponding to the 1T
mode in the small cell and the 1T/1R mode in the large cell at a single frequency
(4.40) represents a phenomenon called modal coupling. In the present case, the
coupling arises from the abrupt change in the cross-sectional area, which leads to
mode transition in the axial direction, as evidenced in the acoustic pressure field
on a longitudinal plane along the centerline shown in Fig. 14.6. The transition
is influenced by the cell length. For example, when the lengths of both cells are
reduced by one-half, the frequencies of the entire duct become 2.13 for the 1T
mode and 4.75 for the 1T/1R mode. This increase in frequency is attributed to the
stronger transition caused by the shorter length. However, when the duct length
is increased, the frequencies of transverse oscillations decrease as predicted. In
any case, the frequency of the entire duct always lies between those of the two
individual cells.
B. Acoustic Field in Straight Duct with Temperature Jump

The second case studies the acoustic wave in a constant-area duct with a
step-change in temperature at x = L 1 , i.e., T = T1 for x ≤ L 1 and T = T2 for
x > L 1 , as shown in Fig. 14.7. Other conditions remain identical to those in
the first case. The calculation only involves two cells corresponding to the low-
and high-temperature regions. The frequency of the 1L mode can be analytically
a) b) 1.00
0.60
0.20
−0.20
−0.60
−1.00
Fig. 14.6 Distributions of acoustic pressures in step duct (R2 /R1 = 2, L1 = L2 =

R1 ) showing mode transition: a) first tangential mode (ΩR2 /ā = 1.96) and b) first
tangential/first radial mode (ΩR2 /ā = 4.40).
determined from the following equation:
ā1 / tan ( L 1 /ā1 ) + ā2 / tan ( L 2 /ā2 ) = 0 (14.75)
The prediction from the present analysis exactly matches the analytical solution
of Eq. (14.75). The normalized 1T frequencies for different cell lengths L 1 are
given in Fig. 14.8. Two azimuthal and two radial modes are used to represent the
wave motion in each cell. If L 1 = 0, the situation corresponds to an acoustic wave
propagating at the speed ā2 in a straight duct. Thus, the normalized frequency
/ (ā2 kmn ) becomes unity. As L 1 increases, the effect of the lower sound speed
ā1 becomes stronger. The frequency decreases and finally reaches its minimum
of ā1 /ā2 as L 2 approaches zero. Similar to the step-duct case, mode transition
phenomena are observed in the acoustic pressure field shown in Fig. 14.9. The 1T
mode is attenuated in the high-temperature section.
C. Combustion Instability in Swirl-Stabilized Combustor

This case is concerned with combustion instabilities in a lean-premixed swirl-
stabilized combustor typical of gas-turbine applications, as shown schematically in
Fig. 14.10. The model includes an axisymmetric chamber connected upstream with
R
T1 T2
L1 L2
Fig. 14.7 Schematic of a straight chamber with temperature jump.

1
a1 / a2 = 0.9
0.9
Ω/(a2k mn )
0.8
a1 / a2 = 0.7
0.7
0.6
a1 / a2 = 0.5
0.5
0 0.25 0.5 0.75 1
L 1/(L 1 + L2)
Fig. 14.8 Normalized frequency of first tangential mode in a straight duct with tem-
perature jump.
Z
0.2
Y X
1 −0.2
0
−1
Fig. 14.9 Acoustic pressure field of first tangential mode in a straight duct with tem-
perature jump (ā1 /ā2 = 0.5, L1 = L2 = R); cross-sections at x = 0(z > 0 is blanked),
x = L, and slice z = 0.
swirl combustion choked

injector chamber exit
45 mm
235 mm
38 mm
Fig. 14.10 Schematic of a swirl-stabilized combustor.
an inlet annulus and downstream with a choked nozzle. Broda et al.14 performed
extensive experiments on this combustor to obtain a stability map for the range
of operating conditions that is conducive to the occurrence of instabilities. When
the inlet air temperature exceeded a threshold value and the equivalence ratio fell
into a certain range, substantial pressure oscillations occurred, with their limiting
amplitudes being about 20% of the mean quantity. The underlying mechanisms for
driving instabilities are discussed in detail in the chapter by Huang et al. (Chap. 10)
in this volume.
Two cases are investigated herein. Case 1 deals with a stable operating con-
dition with an inlet temperature of 600 K. Case 2 corresponds to an unstable
situation with an inlet temperature of 660 K. The equivalence ratio remains at
0.573 for both cases. Figure 14.11 shows the physical domain of concern. The
chamber length is selected for convenient specification of the boundary condition.
At the inlet, the admittance function can be obtained from an impedance-tube
experiment for the swirler.14 At the outlet, the boundary condition for a choked
compact nozzle is applied, as given in Eq. (14.71). The mean flow properties
can be acquired from the numerical simulation of the conservation equations by
using either the Reynolds-averaged Navier–Stokes (RANS) or large-eddy simu-
lation(LES) approach.15 Figure 14.12 shows the mean-temperature contours and
streamlines on a longitudinal plane for two different inlet temperatures. A cen-
tral toroidal zone and a corner recirculation zone exist in both cases because of
the swirling effect and the geometrical configuration. In case 1, the flame spreads
from the corner of the centerbody to the chamber wall. In case 2, the flame is
anchored by both the corner-recirculating flow and the center-recirculating flow
and forms a compact enveloped shape, which is in sharp contrast with the shape of
case 1.
The combustion responses of these two flames to acoustic perturbations were
comprehensively analyzed by You et al.11 All known factors affecting the un-
steady heat release were examined, including the heat of the reaction, density,
flame speed, and flame-surface area. Briefly, the fluctuation of the heat of the re-
action is attributed to changes in the mixture-equivalence ratio resulting from flow
disturbances. The density fluctuation, mainly arising from pressure perturbation,
has a negligible effect on unsteady heat release, as compared with the other
20 mm
20.3 mm
45 mm
9.53 mm
224 mm
Fig. 14.11 Physical domain of a model combustor.

a)
b)
Fig. 14.12 Mean temperature contours and streamlines in swirl-stabilized com-

bustor15,16 : a) simple flame (Tin = 600K,S = 0.76, φ = 0.57, p = 0.463 MPa); b) en-
veloped flame (Tin = 660K,S = 0.76, φ = 0.57, p = 0.463 MPa).
2.5
♦ pressure (experiment)
pressure
2.0
Normalized amplitude
velocity
1.5
♦
1.0 ♦
0.5
♦
0.0
0 0.1 0.2
Axial coordinate (m)
Fig. 14.13 Distributions of acoustic pressure and velocity oscillations in swirl-
stabilized combustor; first longitudinal mode.
Table 14.2 Calculated oscillation frequencies and damping

coefficients for swirl-stabilized combustor
Case 1 (Tin = 600 K) Case 2 (Tin = 660 K)

Frequency Damping Frequency Damping
Mode (Hz) coefficient (s−1 ) (Hz) coefficient (s−1 )
1L 1,645 2.1 1,753 −21

1T 10,610 0.9 11,310 −3.3
1R 22,297 3.5 24,236 −4.5
a) b) c)
d) e) f)
g)
Fig. 14.14 Distributions of acoustic pressure in swirl-stabilized combustor; first tan-

gential mode: a–f) contours on cross sections at x = 0, 0.02, 0.04, 0.08, 0.16, and
0.24 m; g) contours on x–r plane.
a) b) c)
d) e) f)
g)
Fig. 14.15 Distributions of acoustic pressure in swirl-stabilized combustor; first ra-

dial mode: a–f) contours on cross sections at x = 0, 0.02, 0.04, 0.08, 0.16, and 0.24 m;
g) contours on x–r plane.
three factors. The oscillation of flame speed, similar to that of the heat of reaction,
is also caused by the equivalence-ratio fluctuation for a given chamber and flow
condition. The mechanisms of flame surface-area fluctuation are relatively com-
plicated and are primarily dictated by local velocity perturbations. The analytical
forms of the response functions derived in Ref. 11 accommodate the effects of
spatial variations in chamber geometry and mean flowfield and can be effectively
incorporated in the present stability analysis.
The computational domain was discretized into one and six axial cells in the
inlet and combustor, respectively. Since oscillations with frequencies greater than
the 1R mode can be ignored, only two azimuthal (M = 3) and two radial (N = 3)
modes are considered in each cell. Another calculation involving a total of 14 cells
was also performed for comparison. The slight discrepancy of 0.6% in predicted
frequencies between the two cases confirms the validity of the use of seven cells in
determining the system stability behavior. Table 14.2 summarizes the calculated
oscillation frequencies and damping coefficients of the first longitudinal, first tan-
gential, and first radial modes for two different inlet temperatures. The acoustic
motion in case 1 is stable, whereas that in case 2 is unstable. The phenomena
are consistent with experimental observations14 and numerical simulations.16 The
calculated frequency of 1753 Hz in Case 2 matches closely the experimental value
of 1750 Hz. The spatial distribution of the 1L mode shown in Fig. 14.13 further
demonstrates the validity of the present analysis. The spatial distributions of the
1T and 1R modes are shown in Figs. 14.14 and 14.15. A detailed discussion of
the underlying mechanisms of driving combustion instabilities in the chamber is
given in the chapter by Huang et al. (Chap. 10) in this volume.
V. Conclusion
In this chapter, a three-dimensional linear acoustic analysis of gas-turbine com-
bustion instability was established. The work provides an effective means of un-
derstanding, analyzing, and predicting the stability characteristics of gas-turbine
combustors with geometric and mean flow variations. The purpose is to construct
a general framework within which the effects of all known quantities, including
both distributed and boundary source terms, can be assessed quantitatively. The
approach was extensively validated against problems for which either analytical
or numerical solutions, augmented by experimental measurements, are available.
Sample calculations were also conducted to investigate the effects of chamber
geometry, mean-flow distributions, and unsteady heat release on the behavior of
oscillatory flowfields.
VI. Acknowledgments
The work reported in this chapter was sponsored in part by Rolls-Royce plc,
in part by the Air Force Office of Scientific Research under Grant No. F49620-
99-0290, in part by the NASA Marshall Space Flight Center under the Grant
NAG8-187, and in part by Pennsylvania State University. The encouragement and
support from M. S. Anand and B. Bullard are gratefully acknowledged.
References
1
Culick, F. E. C., “Stability of High-Frequency Pressure Oscillations in Rocket Combus-
tion Chambers,” AIAA Journal, Vol. 1, No. 5, 1963, pp. 1097–1104.
2
Mitchell, C. E., “Analytical Models for Combustion Instability,”Liquid Rocket Engine
Combustion Instability, edited by V. Yang and W. E. Anderson, Progress in Astronautics
and Aeronautics, Vol. 169, AIAA, Washington, DC, 1995, pp. 403–430.
3
Zinn, B. T., and Powell, E. A., “Nonlinear Combustion Instability in Liquid Propel-
lant Rocket Engines,” Proceedings of the 13th Symposium (International) on Combustion,
Combustion Institute, Pittsburgh, PA, 1971, pp. 491–503.
4
Culick, F. E. C., and Yang, V., “Prediction of the Stability of Unsteady Motions in Solid
Propellant Rocket Motors,” Chapter 18, Nonsteady Burning and Combustion Stability of
Solid Propellants, edited by L. De Luca and M. Summerfield, Progress in Astronautics and

Aeronautics, Vol. 143, Washington, DC, 1992, pp. 719–779.
5
Propellant Rocket Engines,” Chapter 1, Liquid Rocket Engine Combustion Instability, edited
by V. Yang and W. E. Anderson, Progress in Astronautics and Aeronautics, Vol. 169,
Washington, DC, 1995, pp. 3–37.
6
Nickerson, G. R., Culick, F. E. C., and Dang, L. G., “Standard Stability Predic-
tion Method for Solid Rocket Motors,” Air Force Rocket Propulsion Lab., AFRPL
TR-83-017, 1983.
7
You, D, “A Three-Dimensional Linear Acoustic Analysis of Gas-Turbine Combustion
Instability,” Ph.D. Thesis, Dept. of Mechanical Engineering, Pennsylvania State Univ.,
University Park, PA.
8
Marble, F. E., and Candel, S. M., “Acoustic Disturbance from Gas Non-Uniformities
Convected Through a Nozzle,” Journal of Sound and Vibration, Vol. 55, No. 2, 1977,
pp. 225–243.
9
Crocco, L., and Cheng, S. I., Theory of Combustion Instability in Liquid Propellant
Rocket Motors, AGARD Monograph No. 8, Butterworths Scientific Publications, London,
1956.
10
Vol. 394, 1999, pp. 51–72.
11
You, D., Huang, Y., and Yang, V., “A Generalized Model of Acoustic Response of Turbu-
lent Premixed Flame and Its Application to Gas-Turbine Combustion Instability Analysis,”
Combustion Science and Technology, Vol. 177, 2005, pp. 1109–1150.
12
a Choked Nozzle,” Journal of Fluid Mechanics, Vol. 467, 2002, pp. 215–239.
13
ANSYS, Inc. Corporate (2003), “The ANSYS 7.1 Users Documents,” http://www.
ansys.com/services/ documentation/index.htm.
14
Study of Combustion Dynamics of a Premixed Swirl Injector,” Proceedings of the Com-
bustion Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1849–1856.
15
Huang, Y., Sung, H., Hsieh, S., and Yang, V., “Large-Eddy Simulation of Combus-
tion Dynamics of Lean-Premixed Swirl-Stabilized Combustor,” Journal of Propulsion and
Power, Vol. 19, No. 5, 2003, pp. 782–794.
16
Huang, Y., and Yang, V., “Bifurcation of Flame Structure in a Lean-Premixed Swirl-
Stabilized Combustor: Transition from Stable to Unstable Flame,” Combustion and Flame,
Vol. 136, No. 3, 2004, pp. 383–389.
Chapter 15
Implementation of Instability Prediction in Design:

ALSTOM Approaches
Christian Oliver Paschereit∗

Hermann-Föttinger-Institute, Berlin University of Technology,
Berlin, Germany
and
Bruno Schuermans , Valter Bellucci‡ , and Peter Flohr§
†
ALSTOM Power, Ltd., Baden, Switzerland
I. Introduction
F OR stationary gas turbines, the drive for lower emissions of oxides of nitrogen
during the past decade has lead to the widespread use of lean premix burners
and convectively cooled combustion chambers. These technological changes have
resulted in a reduced stability of flame anchoring and lower acoustic damping.
Consequently, modern gas turbines are more susceptible to combustion-driven
oscillations and the importance of thermoacoustic phenomena in gas turbine com-
bustors has increased sharply.
To prevent acoustic instabilities, accurate models are needed to describe the dy-
namic properties of the combustion process and the propagation of acoustic waves.
This information can then even be used in the design process of a combustor to op-
timize combustor design. Knowing the stability borders of a combustor allows us to
define the operational concept for the gas turbine without exceeding instability am-
plitudes that would harm the combustor. An additional advantage is the reduction
of commissioning time and expensive and time-consuming engine testing.
The thermoacoustic system can then be modeled as a network of acoustic el-
ements, in a similar way to that described by Munjal1 and Polifke et al.2 To de-
termine the stability borders of an acoustic network containing measured transfer
functions, no transient properties of the transfer functions are required; thus, the
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights
reserved.
∗ Chaired Professor.
† Group Leader.
‡ Senior Scientist.
§ Department ad interim Leader.
445
446 C. O. PASCHEREIT ET AL.
transfer matrices only have to be measured as a function of the real part of the fre-
quency. A detailed description is given in Chapter 12 by Ann Dowling. The whole
system can then be assessed either in the frequency domain or in the time domain.
For gas-turbine manufacturers, stability borders are of crucial interest. Because
operating conditions such as power, flame temperature, and pilot to premix fuel
ratio affect the pulsation behavior of the combustor, knowledge of the combustor
stability map is one of the most important points to allow smooth operation of the
engine. The modeling of pressure pulsation amplitudes as well as the mode distri-
bution is the load input for the mechanical integrity assessment of the combustor
and thus is the basis for lifetime calculations. Modern gas turbines are required to
work with increasing operation hours between between major overhauls. The use of
advanced simulation tools in the design process allows the delivery of combustors
that exceed customer’s requirements.
A general method to predict stability of complex thermoacoustic systems in the
frequency domain was presented by Schuermans et al.3 The method used to solve
for the stability borders was based on a method described by Lang et al.4 and
relies on solving the complex eigenvalues of the resulting system; however, the
equations are solved by using a numerical and graphical approach that enables one
to find the stability borders of networks of any complexity. The stability borders
of an atmospheric combustion test facility with variable length and variable exit
conditions were predicted and compared with experimental results. The influence
of the thermal power of the combustion process on the transfer function is measured
and corresponds with basic physical understanding of the thermoacoustic process.
Although the network modeling of the combustor acoustics and the methods
that assess combustor stability require fast and efficient models, the main task
in combustor stability modeling remains properly describing the burner and the
flame. Without this knowledge, confidence in these tools remains low. A proper
description of the flame as a function of its operational parameters is needed to
increase confidence about the simulation.
Thermoacoustic flame models have been derived that describe the interaction
between the acoustic waves and the combustion process; they are discussed in de-
tail in Chapter 12 by Lieuwen. However, because of the highly three-dimensional
flowfield of swirl-stabilized combustion and the interaction between heat release
and the flow-field, these flame models often are unrealistic. An experimental de-
termination of the transfer matrix is still preferred. The measured transfer matrix
can be used directly in the network model, or it can be used to validate analytically
derived transfer functions.
Paschereit et al.5,6 and Schuermans et al.7 developed a method to experimentally
determine transfer matrices of swirl-stabilized flames. The method is based on a
technique described by Cremer,8 Bodén and Åbom,9 and Lavrentjev and Åbom,10
who applied these techniques to describe the acoustic properties of fans in flow
ducts. The element “burner and flame” is considered to be a black box that takes
into account the complex interaction between turbulent flow, flow instabilities,
and unsteady heat release. The property of the burner to produce sound by the
interaction of burner flow instabilities and unsteady heat release was taken into
account by measuring its source term.
This chapter describes how these approaches were extended by using computa-
tional fluid dynamics to determine the main parameters of the analytical transfer
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 447
+ + +
hood burner and flame area change nozzle choked exit
Fig. 15.1 Network representation of a combustion system.
function. The integration into a fast and efficient simulation of combustor pulsation
properties is described, and application to real engines is demonstrated.
II. Network Representation of Thermoacoustic Systems

A thermoacoustic system can be represented as a network of acoustic elements.
Each element gives a simple linear relation between the acoustic quantities on
both sides of the element. For example, a gas-turbine combustion system—with
air supply, burner, fuel supply, flame, combustion chamber, cooling air channels,
and so on—can be modeled as a network of these elements. In this work, burner
is defined as the element in which fuel injection and mixing of fuel and air takes
place and “flame” is the heat release zone. These elements are shown in Fig. 15.1
for a very simple combustion system. Such a network representation may also
contain an active-control feedback loop that could consist of a sensor, a controller
and an actuator.
A network describing a gas-turbine system would be more complex, containing
elements for annular ducts and having side branches representing, for example,
cooling channels and fuel supply lines. Such a network representation may also
contain an active-control feedback loop that could consist of a sensor, a controller,
and an actuator. Solving the wave equations for one dimension with mean flow
yields a relation for the Riemann invariants. These Riemann invariants can be
regarded as sound waves traveling in both the upstream and downstream directions
and are related to acoustic pressure and velocity by p̂(ω, t) = f + g and û(ω, t) =
f − g, respectively. According to this definition, p is the acoustic pressure (in
Pascal), normalized by the characteristic impedance ρc.
Mathematically, each element of the network may be described by its transfer
matrix: a 2 × 2 matrix giving a linear time-invariant relation between the acoustic
pressure and velocity fluctuations on both sides of the element. This relation is
shown in Eq. 15.1. The incoming and outgoing Riemann invariants of an acoustic
element are related by a matrix that is referred to as the scattering matrix, as
expressed in Eq. (15.2):

p̂d T11 T12 p̂u ps
= + (15.1)
û d T21 T22 û u us

fd SC11 SC12 fu fs
= + (15.2)
gu SC21 SC22 gd gs
Here, the subscripts u and d indicate upstream and downstream properties. The
elements of the transfer matrix [Eq. (15.1)] and the scattering matrix [Eq. (15.2)]
are complex-valued and a function of the angular frequency ω. Both descriptions
are equivalent and can be transformed from one description to the other. It has been
assumed that the acoustic waves are longitudinal and propagate one-dimensionally.
This assumption is valid because wave lengths corresponding to the frequency
range of interest are much larger than the non-axial dimensions of the test rig. The
hat in the equations denotes the complex-valued amplitude of acoustic pressure
and velocity.
The transfer matrix of the burner with flame is of crucial importance because
an interaction between acoustic fluctuation and heat release by the flame (which
may be a driver for thermoacoustic instabilities) takes place in this element. This
interaction could possibly be the result of coherent vortices building up and break-
ing down, resulting in fluctuating flame surface area and causing fluctuating heat
release. This coupling can also be caused by pressure and velocity fluctuations at
the burner location, resulting in fluctuations in the fuel–air ratio, and therefore in
oscillating heat release.
The transfer matrix and the scattering matrix in Eqs. (15.1) and (15.2) describe
a passive element, that is, an element that does not generate sound itself but that
amplifies and reflects or transmits incoming signals. If an acoustic element contains
an independent source of acoustic energy, these relations are not valid any more,
since the sound generated by the element has to be added to the outgoing waves
of the element.
In the case of gas-turbine burners, the source term contains noise generated by
the turbulent flow in the burner and flame, possibly involving large-scale hydrody-
namic structures. The source term consists of “colored” noise caused by turbulent
flow; it is expected to have a preferred frequency that depends on the Strouhal
number and a magnitude that depends on the mean flow velocity. If relations such
as Eqs. (15.1) and (15.2) can be found for all elements in an acoustic network,
either analytically or experimentally, then these relations can be combined to-
gether with the appropriate boundary conditions in one linear system of equations
[Eq. (15.3)]:
Sr = q (15.3)
Here, S represents the system matrix: a square matrix containing the coefficients
of all the transfer and scattering matrices of the individual network elements.
The vector of unknowns, r , contains the unknown quantities: Riemann invariants
( f and g) or acoustic pressure and velocity ( p and u) at every node of the network.
The forcing vector q contains the excitation signals or inputs of the system. They
can be the source terms but can also be the excitation signal of, for example, a
loudspeaker or fuel flow actuator. The matrix S and vectors r and q are functions of
the angular frequency, ω. If the transfer matrices and source terms of all network
elements are known, Eq. (15.3) can be solved for r to find the response of the
system to acoustic excitation or to perform a stability analysis.
The stability of the system can be determined from the homogeneous system of
equations derived from network analysis or, equivalently, by evaluating the Nyquist
diagram of the open-loop transfer function and using the method developed by
Polifke et al.2 If the system is unstable or close to the instability border, extremely
high amplitudes can be expected. However, even if the criterion of stability is ful-
filled, the magnitude of the oscillations may be unacceptably high if the forcing by
inhomogeneities goes into resonance with one of the otherwise stable eigenmodes
of the system. To determine the pressure amplitudes in this case, the response
of the system to the source term has to be calculated, which is equivalent to solv-
ing the nonhomogeneous system of equations in which the source terms appear
on the right-hand side. If one of the preferred frequencies of the source term is in
proximity to one of the resonance frequencies of the system, high amplitudes will
result.
III. Experimental Determination of Transfer Matrices and Source Terms

Because of the highly three-dimensional flowfield and the interaction among
flow (instabilities), heat release, and acoustics, the derivation of analytical rela-
tions for the transfer matrix and the source term of the burner and flame is diffi-
cult. A method has been developed to determine the transfer matrices and source
terms of these elements experimentally. A detailed explanation of the measurement
method and test facility are given by Paschereit.5 The combustion facility (shown in
Fig. 15.2) is equipped with loudspeakers upstream and downstream of the burner
to apply an acoustic excitation to the flow. Water-cooled microphones are used to
measure the pressure fluctuations on both sides of the burner.
The combustion chamber has a variable geometry at the downstream end to
adjust the reflection coefficient at that end. A lower reflection coefficient results
in more loss of acoustic energy at this boundary, and thus results in more-stable
combustion and lower pressure pulsations. To obtain accurate measurements of
the scattering matrix, an exit with a low reflection coefficient is chosen to ensure
stable combustion. When measuring transfer matrices, the response of the system
to acoustic excitation by a loudspeaker is measured with microphones. However,
the microphones not only measure the sound generated by the loudspeakers but also
measure combustion noise and pressure fluctuations caused by local turbulence at
combustion air
cooling air
4 loudspeakers (downstream)
water-cooled
water-cooled
sensor holders
sensor holders
(downstream)
(upstream)
air-cooled
swirl-stabilized combustion
burner glass tubes chamber
4 loudspeakers (upstream)
adjustable end
Fig. 15.2 Experimental arrangement of the combustion test facility.

the microphone positions. The pressure signal can thus be considered as the sum
of three different contributions, as shown in Eq. 15.4:
P = Pe + Ps + Pr (15.4)
The contribution p e represents sound that is the response of the system to the
excitation by the loudspeaker. This part of the sound is used when determining
transfer matrices. The system is forced with a sequence of pure tones at distinct
frequencies, so p e also consists of pure tones. The amplitude, p e , can be obtained
from the averaged cross power spectrum between the excitation signal and the
microphone signal.
The contribution p s represents sound that is the response of the system to the
source term. The source term consists of noise generated by the turbulent flow
through the burner and of noise generated in the swirl-stabilized flame. This part
of the noise is independent of and thus uncorrelated with the speaker signal. How-
ever, this part is not random; it is coherent in the sense that there is a high correlation
of this part of the signal at different axially spaced microphone positions. This part
of the sound is needed to determine the source term. The term p s can be obtained
from the averaged cross power spectrum between the microphone signals and a
reference microphone signal but only in the case in which no excitation by the
loudspeakers occurs.
The contribution pr represents random noise caused, for example by local tur-
bulence generating pressure fluctuations at the microphone location. This part is
uncorrelated with the speaker signal and the source term. Because of the local char-
acter of the turbulence, these signals have a high correlation only for very small
axial spacing of the microphones. By taking the averaged cross spectra between
microphone signals and a reference signal, this part of the noise will be averaged
out if the axial spacing between the microphones is large enough.
In Fig. 15.3, the inputs of the system are the sources of sound, e, s and r .
The microphone signals p1 and p2 are the response of the system to all three
inputs. The one-dimensional transfer function of the burner is H , and G 1 and G 2
describe how acoustic waves are propagated and reflected by the geometry. The
transfer function H and the source-term s have to be determined now from the two
microphone signals. Since s and r are not correlated to the excitation signal e, these
s r
p p
e 1 2
G1 H
G2
Fig. 15.3 Simplified representation of the combustion system. Three different contri-
butions to the pressure signal are shown: p e , sound caused by external excitation; p s ,
sound generated by the burner itself, source term; pr , random noise caused by local
turbulence, for example.
contributions will vanish when averaging the cross spectra between the excitation
signal and the microphone signal, and thus p e can be obtained from this signal.
Using Eq. (15.5), the transfer function H can easily be found:
P̂ e2
H= (15.5)
P̂ e1
By calculating the cross spectra between the microphone signals and a signal of a
third reference microphone, only the random noise will be averaged out. In this way
pe,s can be obtained: it is the part of the microphone signal that is the response of the
source term and of the speaker signal. As shown in Eq. (15.6), the source term can
now be found by using the transfer matrix H , which was determined previously:
s = P̂ e,s e,s
2 − H p̂ 1 (15.6)
Once the transfer matrix H has been found, the source term can also be
determined from a second measurement, one in which the loudspeakers are shut
off. Equation (15.6) will still be valid for this case; the only difference is that e = 0.
The approach described above can be extended to systems with four pole-
transfer or scattering matrices. The multimicrophone method was used to obtain the
Riemann invariants from multiple axially distributed microphones; more details
about this technique are given in Paschereit.5 The scattering matrix can be calcu-
lated from the signals that are cross-correlated to the speaker signal by using Eq.
(15.2). Because the signals are cross-correlated to the speaker signal, the signal
does not contain any response of the source terms, and Eq. (15.2) is therefore
reduced by the source term.
Because four elements of the scattering matrix have to be found and since
Eq. (15.2) without the source term only provides two equations, at least two in-
dependent test states are needed to solve the system of equations. These two
independent test states are generated by forcing with speakers downstream and
upstream from the burner. The result is a system of equations [Eq. (15.7)], which
has to be solved for the four complex-valued elements of the scattering matrix:
e e
f d A f deB SC11 SC12 f u A f ueB
= (15.7)
gue A gue B SC21 SC22 gde A gde B
Here, subscripts A and B refer to test states A (upstream forcing) and B (down-
stream forcing). As previously indicated, the elements of the scattering matrix
are functions of the angular frequency, ω. After the scattering matrix has been
found, the source term can be found from the signals that were cross-correlated to
a reference microphone by using Eq. (15.8):
e,s
fs fd SC11 SC12 f ue,s
= − (15.8)
gs gde,s SC21 SC22 gde,s
This equation is also valid if there is no forcing from the speakers (e = 0).
A more straightforward way to determine the four elements of the scattering

matrix and the source terms is to solve Eq. (15.9) by using three independent test
states, A, B, and C:
   e,s 
f de,s f de,s e,s
f dC fu A f ue,s e,s
f uC
A B SC11 SC12 B fs
 =  e,s + (111 )
gue,sA gue,sB e,s
guC SC21 SC22 gd A gde,sB e,s
gdC gs
(15.9)
These three independent test states can be generated by forcing upstream, down-
stream, and on both sides of the burner at the same time.
A. Experimental Validation
The previously described method has been used to measure the source terms and
scattering matrix of a swirl-stabilized premix burner at certain operating conditions.
A first validation of the method can be done by modeling the combustion test rig as
a network of acoustic elements and comparing the results obtained in this way with
measured results. The acoustic network of the test facility consists of a measured
reflection coefficient, duct with flow, measured scattering matrix of the burner and
flame, measured source term, duct with flow, and a measured reflection coefficient.
The reflection coefficients were determined by calculating the ratios of the
Riemann invariants at the entrance and exit of the combustion system. The network
was simulated with a computer program. This program combines all the elements
(obtained analytically or experimentally) of an acoustic network into one system
of equations. The nonhomogeneous system of equations was then solved to obtain
the response of the system to excitation by the source term. Because of the linear
approach, the oscillation amplitudes are proportional to the magnitude of the source
term if the system is stable. If the system is not stable, the absolute values of the
oscillations cannot be predicted by using this linear approach.
To validate the measurement method and the network modeling, the system
of equations is solved to find the spectrum of the fluctuations at a certain po-
sition in the combustion chamber. The spectrum of one of the microphone sig-
nals has been plotted together with the result of this simulation in Fig. 15.4. The
frequency scale is normalized by dividing the actual frequency by the highest
frequency measured; the highest frequency corresponds to a Strouhal number
of St = 2.68. The pressure scale is normalized by dividing the pressure spectra
obtained from simulation and experiment by the highest value in the measured
spectrum. Two cases were considered: 1) nonreflecting boundary conditions, and
2) reflecting boundary conditions. The nonreflecting boundary conditions in the
experiment were achieved by an orifice at the exit of the test rig.11 The simulation
of the nonreflecting boundary conditions was almost identical to the measured data
(Fig. 15.4a). This outcome is not surprising, since almost all components of the
network model consist of measured elements. However, it shows that no errors
have been made when processing the raw experimental data to obtain transfer ma-
trices, source terms, and reflection coefficients; and it shows that modeling acoustic
networks with measured transfer matrices and source terms yields valid results.
A real test for checking the predictive capabilities of the method is to change
one of the boundary conditions of the test rig and then compare measured pressure
a) 1.0
0.8
0.6
0.4
0.2
0.2 0.4 0.6 0.8 1.0

Normalized frequency
b) 4
0.2 0.4 0.6 0.8 1.0

Fig. 15.4 The predicted spectrum of the pressure fluctuation (dashed lines) and the
measured fluctuations (dotted lines): a) in the combustion test rig with a nonreflecting
exit; b) in the combustion test rig with a reflecting exit.
spectra with computational results obtained with the adjusted reflection coeffi-
cient and the previously determined burner transfer matrix and source terms. The
boundary at the exit has been changed to an almost fully reflecting end, and the
pressure spectrum in the test rig has been measured at the same operating condi-
tions. The comparison between the predicted and measured spectrum is shown in
Fig. 15.4b.
Again, good agreement exists between the values predicted by the model and
the experimental data. By comparing the spectra measured in the test rig using
a reflecting end and a nonreflecting end, not only are higher overall amplitudes
observed but a shift in the resonance frequencies can also be seen. This outcome
occurs because the reflection coefficient is a complex quantity; by changing the
geometry of the end, the absolute values as well as the argument of the reflection
coefficient differ. This change in phase of the reflection coefficient can cause a
shift of the resonance frequency.
The pressure amplitudes in Figs. 15.4a and Fig. 15.4b are normalized, but the
scaling factor for all graphs is the same. The two dominant peaks in the spectra
roughly correspond to the quarter wave and three-quarter wave resonance mode
of the combustion chamber. Around the peak frequency, the predicted absolute
1.0 3
abs(T22)
arg(T22)
0.8 2
Normalized abs(T22)
1
0.6
arg(T22)
0
0.4
–1
0.2 –2
0.0 –3
0.5 1.0 1.5 2.0 2.5

Strouhal number
Fig. 15.5 Comparison of T22 element of the transfer function, measured at two dif-
ferent forcing amplitudes.
values do not match the experimental data well. An explanation of this mismatch
is that the assumption of linearity is no longer adequate, specifically for the source
term or the transfer matrix when the acoustic state of the system is at very high
amplitudes. In the case of a system driven by a source term, the response, that
is, the observed pressure spectrum, follows such nonlinear changes in the source
term. In the case of self-excited instability, modifications of the transfer matrix
caused by high amplitude will lead to nonlinear cycle limitation.
The linearity of the system has been assessed by forcing the system with two
different forcing levels. Figure 15.5 shows the T22 element of the transfer function
and proves its linearity.
IV. Modeling the Burner Transfer Matrix

A generic premix burner that resembles important features of a gas-turbine
burner is considered, as shown in Fig. 15.6. The preheated and compressed air
enters a mixing device from the burner plenum. In the mixer, where the flow
is significantly accelerated, fuel is injected and is homogeneously mixed with
the passing airstream. Additional swirl is often imparted on the burner flow to
increase the mixing efficiency. The fuel–air mixture then enters the combustor,
and a flame can stabilize at the recirculation zones that form at the flow expansion;
inner recirculation zones associated with strongly swirling flows may additionally
act as flame holders. Inhomogeneities in the fuel–air mixture are convectively
transported from the fuel injection points (i) into the flame front. Each of the fuel
injectors can be associated with a specific time delay that corresponds to local travel
times of the fuel particles. The additional smearing effect by turbulent diffusion is
schematically shown for the central fuel injector.
plenum fuel injector flame / combustor

and mixer
0 1 2
Fig. 15.6 Sketch of the premix burner. Reference points 0, 1 upstream- and down-
stream of the burner element are indicated, as well as reference points 1, 2, upstream-
and downstream of the flame front.
A. Lossy Flow Through Burner Nozzle

The flow in the burner element is described by the unsteady, incompressible
Bernoulli equation. In this model, the effect of unsteady fluctuations is associated
with inertia work; losses caused by the complex three-dimensional flow inside the
burner are taken into account by using an integral loss coefficient:
 2 
ω

p̂1  1 ρ0 c 0 M 0 1 − ζ − A0
A1
− i c0
L red  p̂ 0
=  (15.10)
û 1 A0 û 0
0 A1
where
1 1
A0 u(s)
L red = ds = ds (15.11)
0 A(s) 0 u0
is a virtual length of the oscillating air column inside the burner. The derivation of
Eq. (15.10) is based on the assumption of compactness, that is, no physical length
of the burner element. The transfer function of the cold burner was measured and
compared against the previously described model with good agreement.
B. Flame Model
A real flame does not have a steady position in the combustor, but varies in
position. This fluctuation is taken into account in the model. The flame is fed with
a premixed fuel–air stream, where the fuel injection takes place at the location i
inside the burner, before the flame front with upstream and downstream states
1 and 2; see Fig. 15.6. The common approach to model the acoustic behavior
of such a flame is based on the assumption that the acoustic and heat-release
fluctuations at the flame front are coupled with fluctuations in the fuel–air mixture
that are attributable to acoustic disturbances at the fuel injectors. This implies
the existence of a characteristic time lag τ , after which the fuel particles reach the
flame; thus, for the fuel–air mixture φ and its fluctuation in time φ (t),
φ1 (t) φ (t − τ )
= i (15.12)
φ1 φi
The heat release Q in the flame can be written as
Q = φ1 ρ1 S f h fuel (15.13)
The turbulent flame speed is not assumed to be constant but is assumed to be

affected linearly by the fuel supply, S f ∼ φ, which is reasonable for lean flames
(φ < 1). This result implies for the linearized perturbations from Eq. (15.13),
Q φ ρ
=2 1 + 1 (15.14)
Q φ1 ρ1
By using the Rankine–Hugoniot jump conditions across the flame and the pre-
ceding equations, we can formulate the flame model as
 
p̂ 2 1 ρ1 c1 TT21 − 1 M1 1 − 2e−iωτ p̂ 1
=  (15.15)
û 2 0 1 − TT21 − 1 2e−iωτ û 1
In particular, the amplitude of the velocity fluctuations across the flame has been
changed, which is important because no free parameter (such as the interaction
parameter n in the n–τ model) exists to adjust this condition.
The model is still based on a time-delay τ . It is an oversimplification for a
realistic flame, in which fuel injectors are spread over an axial distance inside the
burner, the flame is nonplanar, or both. As will be shown subsequently, the effect of
the time-delay spread can in fact significantly influence the stability characteristics
of a burner system. This effect can be incorporated in the flame model by dividing
the fuel inlet points of the burner into p submodels (an alternative that is not
pursued here is to divide the model at the flame front itself). The fuel from each of
the inlet points reaches the flame after a certain time delay τ j , and the following
relation for the velocity fluctuations across the flame can be derived:

p
2 T2 −iωτ j
û 2 = 1 − aj −1 e û 1 (15.16)
j=1
p T1
Here, only a burner configuration in which the fuel injectors are homogeneously
distributed is considered, and thus a j = 1 for the weight factors of each inlet point.
To illustrate the effect of the model, a situation with linearly distributed time
delays such that τ j [τmax − τ ; τmax ] is considered. In this case, the flame speed
model reads

T2 1 −iω(τmax −τ )
û 2 = 1 − 2 −1 e − e−iωτmax û 1 (15.17)
T1 iωτ
5 3
4 2
1
abs(T22)
arg(T22)
3
0
2
–1
1 –2
0 –3
0.2 0.4 0.6 0.8 1.0

4 3
2
3
1
abs(T22)
2 0 arg(T22)
–1
1
–2
0 –3
0.2 0.4 0.6 0.8 1.0

Fig. 15.7 T22 element of the flame transfer function, with measured (solid) and mod-
eled (dashed) values: a) best fit of the constant time-delay model; b) best fit for the
two-parameter model with linear distribution of time delays.
The improvement of this two-parameter model over the constant time-delay ap-
proach is shown in Fig. 15.7, in which both models have been used to fit experi-
mental results.
C. Computational Fluid Dynamics Analysis of Time Delays

Computational fluid dynamics (CFD) of the burner flow is now used to deter-
mine the model parameters τ j in the transfer function [Eq. (15.16)], or rather its
distribution f τ , to be used as input to the model [Eq. (15.17)]. It is very attractive
because it involves only the postprocessing of steady-state computations and di-
rectly provides the model parameters. The method employed here should perhaps
be contrasted with transient CFD, in which the full transfer function is obtained
directly from the unsteady response to forced perturbations (see, for example,
Fig. 15.8 The burner.
Polifke et al.).12 There, no assumptions on time-delay mechanisms are necessary;

and other potentially important effects, such as vortex shedding or flame front
kinematics, can be resolved. However, such methods are computationally very ex-
pensive and beyond the scope of this work. Only an overview is given here; details
can be found in Flohr et al.13
1. Numerical Setup
The CFD calculations of the experimental premix burner are based on the steady-
state three-dimensional Navier–Stokes equations with a second-order accurate fi-
nite volume solver. The geometry of the burner is schematically shown in Fig. 15.8.
The preheated and compressed air enters the swirler through the two inlet slots on
the cone shell in a circumferential direction. The resulting swirl flow breaks down
near the exit, leading to a recirculation region. Gaseous fuel is injected into the
passing airstream along the burner slots. Flame stabilization takes place in this inner
recirculation region formed by the vortex breakdown and in the outer recirculation
region formed by the dump plane. The combustor and the upstream plenum have
been modeled by an unstructured hybrid grid with approximately 600,000 cells.
The fuel injectors are not resolved explicitly in the CFD simulations. Instead, the
fuel injection is modeled by using numerical source terms that are placed shortly
downstream of the injection plane (see Fig. 15.9). It is an approximation of the real
configuration, in which the fuel emanates from an injection hole into the passing air
cross-stream. The fuel is injected into a grid cell through a numerical source term.
The expected fuel penetration at this location has been derived from the appropriate
injection pathline
injection cell
gas channel
Fig. 15.9 Cut through the cone shell at the inlet slot.
correlations and reference experiments. In this way, local variations in the mixture
fraction can be considered while keeping the computational cost reasonable.
The turbulent flow is modeled by the standard k − model and combustion is
incorporated by using a turbulent flame-speed closure model.14 In this model, the
combustion process is described by a reaction progress variable c, which takes
values between zero (fresh mixture) and one (burnt products); and the temper-
ature field is linked to the reaction progress variable and the local (mean) fuel
concentration.
The time delays τ j have been derived from the converged solution by injecting
particles into the flow at the location of the fuel injection and then tracking their
trajectories and measuring the travel times until they hit the flame front. The flame
front is defined somewhat arbitrarily where c = 0.5 for the reaction progress. We
assume complete burning at this location for the fuel tracer particles, and we do
not include effects of local variations in heat-release rates because of uncertainties
that are associated with the flame model itself.
To include the effect of turbulent diffusion on the displacement of the fuel par-
ticles for each of the fuel ports, a random velocity is added to the mean convective
velocity u, which is obtained from the CFD computation. The turbulent displace-
ment velocity is modeled by a Gaussian white-noise process dW(t) that is scaled
by the local turbulent kinetic energy k√in the flow, such that the total displacement
dx of the fuel particle is dx = udt + 2k/3dW(t). The time-step size dt has been
chosen such that the travel distance of the fuel particles, for each time step, was
small compared with the local grid resolution.
Each of the distributions presented subsequently has been obtained from tracking
q = 128 particle injection points distributed along the fuel-injection ports. Where
we added the effect of turbulent dispersion, n = 10 realizations have been used
for each injection location (n = 100 has also been used to check for convergence
in the statistics).
2. Analysis of CFD Results

In Fig. 15.10, four delay-time distributions obtained from the CFD of the burner
flow are shown. As previously stated, the fuel injectors are not resolved explicitly
in the CFD simulations but are modeled instead with source terms that are placed
shortly downstream of the injection plane (the fuel particles that are used for
determining the delay-time distribution have been injected into the flow at the same
0.5
A
B
0.4 C
0.3
f
τ
0.2
0.1
2 4 6 8 10 14
0.5 1 1.5 2 2.5 3 3.5
τ/ τ ref
0.5
B
0.4
0.3
f
τ
0.2 D
0.1
2 8 10 14
0.5 1 1.5 2 2.5 3 3.5
τ/ τ ref
Fig. 15.10 Top: normalized distribution of time delays, as obtained from particle
trajectories. Bottom: For configuration B, the effect of turbulent diffusion on redis-
tributing particle trajectories is shown (D).
location). The distributions A–C (top part of Fig. 15.10) explicitly study the effect
of variations in fuel penetration. The distribution B denotes the reference simulation
that is expected to match best with the experiment; A and C are simulations with
lower and higher fuel-penetration depths, respectively, and could be associated
with, for example, smaller and larger fuel ports. A change in fuel penetration will
change the convective times τ j because that different streamlines inside the burner
will be fed by the fuel if the penetration changes, and as a consequence, the flame
shape changes, as well.
All three variants show the same characteristic behavior, namely, a distributed
time delay with two distinct peaks, one at τ/τref ≈ 1, and one at τ/τref ≈ 2 to 2.5.
Only the peak at large times is affected by the modification in fuel penetration,
and this effect can in fact be correlated with changes in the shape of the flame,
which are not shown here. Unfortunately, when incorporating these distributions
into the flame model [Eq. (15.16)], the agreement with the experiment is found
to be unacceptable. The damping of higher-frequency modes is not captured ap-
propriately. Also, the main mode of instability, where the phase of the measured
transfer function cuts the frequency axis, at a normalized frequency of St = 0.26,
is not captured appropriately; As subsequently explained, CFD generally has a
tendency to overpredict time delays.
This situation changes, at least partially, if the effect of turbulent mixing is
included on the fuel particle trajectories (bottom part of Fig. 15.10). Again, the
peak on the right of the distribution density function f τ is primarily affected. It is
4 3
absolute 2
3 phase
1
arg(T22)
abs(T22)
2 0
–1
1
–2
–3
0
0.2 0.4 0.6 0.8 1.0
Fig. 15.11 The T22 flame transfer matrix element, derived from the time-delay dis-
tributions in Fig. 15.10; measured (solid) and modeled (dashed) values.
much flatter if turbulent diffusion is included. The peak at τ/τref = 1 appears to be

rather insensitive to any changes in fuel penetration or turbulent diffusivity. This
outcome is perhaps attributable to the fact that these particles arrive so quickly at
the flame that local deviations from the particle path have no significant effect.
The flame transfer function for the time-delay distribution with turbulent spread-
ing is in much better agreement with the measurements and is given in Fig. 15.11.
It captures the general trend of damping at the higher frequencies very well.
The remaining difference can be linked to errors in the CFD calculations. These
errors are likely to be caused by the choice of the turbulence model that has been
used to compute the underlying velocity fields. The k − turbulence model is
known to be too diffusive to fully capture peak values of the axial burner flow,
leading to overprediction of time delays. This error is known from comparative
studies in which different turbulence models have been compared against each
other for this burner flow; see Flohr et al.15 In that paper it was concluded that
peak velocities inside the burner nozzle ahead of the flame front were not captured
correctly by the k − turbulence model; and as a consequence, one would expect
here that time lags are overpredicted. An example of this behavior is shown in
Fig. 15.12. Different turbulence models in the CFD analysis are compared against
water-tunnel LDA measurements.
V. Reduced-Order Modeling of Complex Thermoacoustic Systems
Lumping the combustion system into several subsystems and combining the
subsystems in a network of acoustic elements allows for a combination of different
modeling techniques. The idea of such a lumped-element representation is not new
and such studies include Lang et al.,4 Dowling,16 Schuermans et al.7 Pankiewitz
and Sattelmayer,17 just to cite a few. However, in this new approach, a methodology
is developed that includes geometries of any complexity, and the resulting systems
can be analyzed in a time-efficient, straightforward manner.
First, we demonstrate how a state–space representation of geometries, without
combustion, can be obtained. As an example, the state–space representation of an
Fig. 15.12 Normalized axial velocity along the burner axis; 0 corresponds to the
burner exit position.
annular duct is derived. A comparison with results obtained from finite element
analysis is made. Interconnecting of several systems into a network of acoustic
systems is done by using linear fractional transforms. A stability analysis is then
made by evaluating the eigenfrequencies of the interconnected system. Then in
a second step, the validity of this approach is demonstrated on a very simple,
one-dimensional thermoacoustic system. The eigenfrequencies are solved for ana-
lytically and compared with the results obtained from modal expansion and linear
fractional transforms. A second validation is performed on a system consisting of
two annular ducts interconnected by one-dimensional tubes; the eigenfrequencies
are compared with results obtained from finite element analysis. A network model
of an annular, multiburner, gas-turbine combustion chamber is then derived.
A. Network Interconnections
To obtain a model of the acoustic behavior of a gas-turbine combustion system,
acoustic transfer functions need to be combined in a network of acoustic elements.
The resulting system can then be analyzed to assess its stability to calculate sta-
bility borders or to calculate frequency spectra. Two different methods for system
interconnection and subsequent analysis will be discussed here. The first one is
a typical frequency-domain approach. The second method yields a state–space
representation and can be analyzed either in the frequency domain or in the time
domain. As an example for both methods, the lumped-element representation of a
combustion system shown in Fig. 15.13, is discussed.
B. Frequency Domain Approach

We assume that all transfer matrices in Eq. 15.13 are known as a function of
frequency. For simplicity, only one-dimensional wave propagation is considered
p1 p2 p3 p4
P B F C E
u1 u2 u3 u4
Plenum Burner Flame us Combustion Exit
chamber
Fig. 15.13 Interconnection of subsystems of the combustion system; note that all
arrows represents vectors of input or output signals.
in this example. The elements in Fig. 15.13 are then all 2 × 2 transfer matrices,
except for P and E, which are 1 × 1 transfer functions. We also assume that all
transfer matrices or functions are stable. This assumption is generally a safe one;
the problem of thermoacoustic instabilities is not that one of the transfer functions
is unstable but that the interconnected system can become unstable under certain
conditions. A linear, stable system is completely characterized by its impulse
response or by its frequency response, which is the Fourier transform of the impulse
response. This property is exploited here to determine the stabiliy and stability
borders of the interconnected system.
All the transfer functions can be combined into one system of equations, as
shown in Eq. (15.18). The left-hand side of this equation consist of a large matrix
S(ω) that contains the transfer matrices and of a vector P̂(ω) that contains the
unknown pressures and velocities. The right-hand side of this equation contains
fˆ(ω) the source signals. In this example, the only nonzero entry is u s :
    
P −1 p1 0
 B11 B12 −1 0   u1   0 
B −1  p   0 
 21 B22 0  2  
 F11 F11 −1 0   u2   0 
    =   (15.18)
 F11 F11 0 −1   p3   0 
    
 C11 C11 −1 0   u3   us 
    
C11 C11 0 −1 p4 0
E −1 u4 0
Solving this system for a specific frequency seems to be straightforward, since

the solution is given by p̂(ω) = S−1 (ω) f̂ (ω). However, extreme care has to be taken
when interpreting the result. The result only has a direct physical interpretation if
the system of equations is stable. If the system is stable, then the solution p̂(ω)
represents the Fourier transform of the pressure and velocity signals. However,
if the system is unstable, this Fourier transform is undefined (not convergent). A
very annoying consequence is that the more unstable the system is, the smaller the
resulting p̂(ω) will be. Accessing the stability of the system before interpreting
the result of such an analysis is therefore extremely important. The stability of the
system can be assessed by analyzing the system’s complex-valued eigenvalues.
The eigenvalues are those ω C for which the determinant of the matrix S(ω)
vanishes.
If all transfer functions are known functions in the complex ω plane, then
det(S(ω) = 0) can be solved by using a numerical root-finding procedure. If the
imaginary parts of all roots ωn are larger than zero, then the system is stable. If one
or more roots have negative imaginary parts, then the system is unstable. Hence,
finding all roots of the equation is crucial. As previously mentioned, the measured
transfer functions are only known for real-valued frequencies. Two ways of cir-
cumventing this problem are possible. The first possibility is to fit a function to the
measured transfer function. Once the function is obtained, complex values can be
substituted for the frequency. Great care has to be taken with the choice of function
to be used for fitting the experimental data. To make physical and mathematical
sense, the function should be analytic for imaginary frequencies smaller than zero.
For practical applications, fulfilling these conditions can be very difficult; and for
that reason, this method is not pursued here. The second possibility is to look for
stability borders of the system as a function of the operation parameters, such as
flame temperature and power, instead of solving for parametres. The results is
a map indicating the value of the parameter (such as a temperature, combustor
length, or time delay) for which the system changes from stable to unstable. It will
be demonstrated here that to find stability borders, the transfer functions only need
to be known as a function of the real part of the frequency. A system is said to
be on a stability border if the imaginary part of the frequency equals zero. Thus,
solving Eq. (15.19) yields the stability border for an arbitrary parameter τ at R(ω);
clearly, no information on the dependence on the imaginary part of the frequency
is needed:
det[S{R(ω), (ω) = 0, τ }] = 0 (15.19)
Equation (15.19) is complex-valued and is a function of two parameters. Typi-

cally, the matrix S contains several complex exponentials, which makes it difficult
to solve the system. Numerical techniques like the Newton–Rhapson algorithm
cannot be applied directly, since the equation has, in general, many solutions.
If the system is not bracketed in a proper way, the algorithm may fail to find
all complex-valued roots in a certain domain. Therefore, a numerical–graphical
method for solving the system of equations is used. This method is discussed in
more detail in Schuermans and Paschereit.18 The resulting stability plot shows
the stability regions as a function of one parameter, τ . However, the method was
extended to calculate stability plots as a function of two parameters. The method
is essentially the same; but for the two-dimensional case, the stability borders of
one parameter are calculated for a range of values of the other parameter and the
results are plotted in one graph, resulting in a stability map.
Although this method provides the stability borders of the system, it does not
provide information on what side of the border the system is stable or unstable. If
some of the transfer functions in the system are only known as a function of the real
part of the frequency, then it is mathematically impossible to determine the side
of the border on which the system is stable or unstable. However, one might use
physical arguments to determine the side of the border on which stability occurs.
One could, for example, introduce additional damping into the system and reason
that areas that represent stability regions would increase and that areas representing
instability would decrease.
Another possibility is to determine explicitly the stability of the system in one
point at each side of the boundary. Nyquist plots are typically used for this purpose.
A Nyquist plot is a graphical method of determining the stability of dynamic

systems. To calculate a Nyquist plot, a closed-loop system, as represented in
Fig. 15.13, needs to be “cut” to obtain an open-loop system. The parametric plot
of the open-loop transfer function is then analyzed. If the open-loop system itself
is stable, then the system is stable if there are no clockwise encirclements of the
point −1 + i in the complex plane (simplified Nyquist criterion). This method is
suitable, since only the transfer function as a function of real-valued frequency
needs to be known. However, this criterion is only valid if the open-loop system
itself is stable. This causes a problem in the stability analysis: since it is not known
whether the open-loop system is stable, this method cannot be used without any
additional assumptions.
C. Frequency Domain Stability Analysis of a Gas-Turbine Combustor

The previously described methods for stability analysis have been applied to a
gas-turbine combustion system. A network model of the combustion system was
obtained by combining analytic transfer function models and fits to experimentally
obtained transfer functions. Stability maps have been made as a function of several
parameters in the model.
1. Effect of Operating Point

For stable burner operation over an entire load cycle of a gas turbine, investigat-
ing how the system changes as power and flame temperature are varied is of interest.
It was previously demonstrated that the linear time-delay model (τ , τmax ) captures
important features of the experimental transfer function. By fitting this model to
various operating points and using a quadratic interpolation between these points,
we display a stability map in Fig. 15.14. In this plot, unstable and stable regions
(for the most unstable frequency) can be identified. The black areas indicate the
2.5
Unstable
2.4
T2
2.3
T1
2.2
Stable
2.1
0.5 1.0 1.5 2.0 2.5
v /v
ref
Fig. 15.14 System stability for a typical test rig configuration; stability borders
for variations in burner velocity (i.e., power) vs flame temperature, are shown. The
simulations are based on the linear time-delay model.
v = v ref
Unstable
time
max in sec
v = 1.5 v ref
Stable
Convection
v = 2 v ref
Unstable
Mixing spread
n
Fig. 15.15 System stability for a typical test-rig configuration; stability borders for
variations in time delay spread vs convection time (i.e., 1/power) are shown.
stability border; the thickness of this border could be reduced for significantly
larger computing times. It is evident from this figure that both stable and unstable
regions can be expected for this system. This result also agrees with the exper-
imental observation. It also opens ways to stabilize and destabilize a system by
varying power levels or flame temperatures accordingly and can guide the operator
of a machine to avoid regions of unstable combustion.
2. Effect of Time-Lag Spread

The results of a stability analysis for a given typical operating point with a
hypothetical burner design in which both τmax and τ can be varied at will are
presented in Fig. 15.15. The simulations are based on the linear time-delay model.
Not surprisingly, the system’s stability strongly depends on the maximum time
delay. It is perhaps less obvious that for certain values of τmax , an increase in τ ,
which is plotted here in normalized form τn = τ/τmax , can change the burner
stability significantly, and this result opens possibilities to modify a given burner
configuration.
D. Time-Domain Approach
Although analysis of systems in the frequency domain seems very straight-
forward, assessing the stability of the system is, strictly speaking, impossible if
no information on the dependence of the functions on the imaginary part of the
frequency is available. Even if all transfer functions have been defined in the en-
tire complex-frequency plane, stability analysis is not straightforward, because it
requires a numerical search for all possible eigenvalues. Especially if thermoa-
coustic systems modeling is extended to multi-input multioutput systems, such as
multiburner systems, the numerical root-finding procedure may be forbiddingly
expensive. The problems with determining the stability of the frequency domain
thermoacoustic models have led to a time-domain formulation of the problem. The
idea is to represent all transfer functions as differential equations and to combine
them all in one system of equations. This interconnection can be done in a very
elegant and straightforward manner by making use of state-space representations
and Redheffer star products. A state-space representation of a dynamic system is
a system of first-order differential equations that is equivalent to one higher-order
differential equation. The advantage of using state-space systems is that they are
numerically very robust, and that extension to multi-input multioutput systems is
straightforward. A linear, time-invariant state-space system has the general struc-
ture give in Eq. (15.20):
ẋ(t) = Ax(t) + Bu(t) (15.20)

y(t) = Cx(t) + Du(t)
where u and y are vectors of input and output signals, x is an internal state vector
and the matrices A, B, C, and D represent the system. Any interconnection of such
systems again yields a system with the same general structure but with different
A, B, C, and D matrices. These matrices are real-valued and are independent of
time or frequency. The stability analysis of such a state-space system is thus very
straightforward: The system is stable if the matrix eigenvalues of A all have a
negative real part. The eigenvalues of a real-valued matrix are computed by using
standard methods available in linear algebra.
The entire system of Fig. 15.13 can now be modeled by interconnecting all the
outputs of the subsystems to the inputs of their “neighbors.” These subsystems
can be interconnected in a very convenient way by making use of the Redheffer
star product. The Redheffer star product is a matrix operation based on a linear
fractional transform.19 It is often used in control theory to model uncertainty in
systems but can be used to interconnect any network of state-space systems. The
interconnection of two ducts (or any other systems) H and G is then simply given
by H G, in which denotes the Redheffer star product, and is defined as

F1 (H, G 11 ) H12 (I − G 11 H22 )−1 G 12
H G =
G 21 (I − H22 G 11 )−1 H21 Fu (G, H22 )
in which Fu () and Fl () denote the upper and lower linear fractional transform,
defined as: Fl (M, g): = M11 + M12 g(I − M22 g)−1 M21 .
Thus, the system of Fig. 15.13 can easily be represented by the matrix
S = P B F C E. This system has no inputs or outputs; they could of course
be added, but they are not required for a stability analysis. The stability require-
ment of the system is then satisfied if the real parts of all eigenvalues of S are
negative.
Not only the stability analysis is very straightforward in this approach. Time-
domain simulations, even those including nonlinear elements, and frequency
responses can very easily be performed.
E. Modal Expansion
In this section, acoustic transfer functions are derived for geometries with mul-
tiple inputs and multiple outputs (MIMO). Starting from the wave equation with
sources on the surface f , but without sources in the volume,
∂ 2 p
c2 ∇ 2 p − =0 (15.21)
∂t 2
n̂ · ∇ p = − f (15.22)
A solution can be obtained by making use of Green’s functions. As shown in

Culick,20 the acoustic pressure at any point of the volume can be written as a
function of the sources, modal eigenvalues ωn , speed of sound c, and the eigen-
vectors ψ:

∞
c2 ψn (x)
p̂(x) = ψ(xs ) fˆ(xs )dS (15.23)
n=0
(ω2 − ωn2 ) s

where = ψ 2 dV . The eigenfunctions ψ and eigenfrequencies ωn can be ob-
tained analytically for simple geometries or numerically (e.g., finite element
method) for more-complex systems. If the source function is a source of acoustic
velocity on the boundary, it can be written as: fˆ (ω, xs ) = iωρ û s (ω, xs )
An acoustic transfer function can be defined as the ratio between the acoustic
pressure at a certain position x in the volume to the acoustic velocity acting as an
input on an area As centered on the boundary at xs . If the extent of the area As is
small compared with the wavelength, then Eq. (15.23) can be rewritten to obtain
the transfer function H (ω) between p̂(x) and û s (x0 ):
p̂(x) ∞
ψn (x)ψn (xs )
H (ω) = = iωρ As c2 (15.24)
û s (xs ) n=0
(ω2 − ωn2 )
Equation (15.24). relates the acoustic pressure at one location to the acoustic
velocity at one other location. This single input single output (SISO) representation,
which corresponds to the acoustic impedance, can easily be extended to the general
MIMO case. The K velocities at xin are then related to the J pressures at x out by
a J × K transfer matrix H : p(xout ) = H u(x in ), in which the elements of H are
given by
∞
ψn (xj )ψn (xk )
H jk = −iωρ Ak c2 (15.25)
(n=0)
(ω2 − ωn2 )
1. State-Space Representation
Because all elements of the transfer matrix H have the same eigenvalues (ωn ) the
transfer matrix can be expressed more conveniently by a state-space representation.
One mode of the SISO system of Eq. (15.24) can be represented as
ẋ(t) = An x(t) + Bn u(t) (15.26)

pn (t)
= Cn x(t) + Dn u(t)
ρc

−αn −ωn 0
An = , Bn j =
ωn −αn ψn (x j )

Cnk = 0 cAk ψn (xk ) , D = [0]
Note that α, the modal damping, has been introduced here. The value of α is
assumed to be small compared with ωn . Many different state-space representations
of a system are possible. The representation chosen here has the advantage that it
can easily be extended to the MIMO case. The state of the system is represented
by the 2 × 1 vector xn (this notation is chosen to be consistent with notation used
in control theory and should not be confused with geometrical position x). The
structure of the equations for the general case with N modes, J inputs, and K
outputs is the same as Eq. (15.26). However, the matrix A becomes a 2N × 2N
block-diagonal matrix and the matrices B and C become 2N × J and K × 2N
matrices, respectively:
       
ẋ1 A1 x1 B11 . . . B1J u1
 ..   .   .   . . ..   ... 
.  
 . = ..   ..  +  .. .. 
ẋ2N AN x2N BN 1 . . . BN J uj
       
p1 C11 . . . C1N x1 0 ... 0 u1
1  .   .
 ..  =  .. ..
. ..   ..  +  .. . . . ..   ... 
.   .   . .  
 (15.27)
ρc
pK CK 1 . . . CK N x2N 0 ... 0 uJ
For a realistic acoustic model of a combustion system, it is essential to take into

account the acoustic losses, or damping. The most important acoustic losses are
caused by dissipation on the boundaries of the system (e.g., air supply system, high
Mach-number combustor exit) and because of the process of converting acoustic
energy into vorticity. This latter mechanism is very important in the model of the
burner and is directly associated with the mean flow loss coefficient of the burner.
Acoustic losses purely associated with wave propagation through the combustion
chamber [taken into account by the parameter α in Eq. (15.26)] are generally
very small compared with the losses on the boundaries and in the burner element.
The acoustic losses are thus explicitly considered in the network model. This is
an advantage compared with the approach used e.g. by Annaswamy et al.,21 for
example where acoustic losses are not considered in the model.
2. State-Space Representation of an Annular Duct

To obtain an acoustic transfer function or matrix of some geometry, the eigen-
frequencies ωn and the values of the eigenvectors at the interface locations ψn (x)
need to be known. For practical (often very complicated) systems, the eigen-
frequencies and vectors can be obtained from a finite element analysis. In finite
element analysis, only a modal analysis, which is very computationally efficient, is
required. Moreover, only the modal values at one position on the interface locations
are required; thus, very little output is needed. For more simple geometries, the
eigenfrequencies and vectors can be obtained analytically. As an example, the
transfer function of an annular duct is derived. A combustion chamber of a gas
turbine can be represented as an annular duct with J input and output ports, J
being the number of burners.
The input–output relation is given by a transfer matrix relating J inputs to J
outputs. The required eigenvalues and vectors for a thin annular duct of length L,
mean diameter D, and height h are given by
!

2cm 2 πcn 2
ωn,m = + (15.28)
D L
πnx cos(mφ)
ψn,m = cos
L sin(mφ)
 L Dπ h


n,m = (2 − δkr on (n))(2 − δkr on (m))

 L Dπ h
2(1 + δkr on (n))δkr on (m)
in which n and m are the numbers of the longitudinal and azimuthal modes,
respectively. Thus, the mode is notated here as (n, m).
Because of the rotational symmetry of the annular duct, all eigenvalues are
two fold degenerate except for n = 0 and have two orthogonal eigenmodes. Sub-
stituting of Eq. (15.28) into Eq. (15.26) and Eq. (15.27) yields the state-space
representation of a thin annular duct with colocated inputs and outputs at one side
of the duct:
 
−α ωn,m
 ωn,m −α 
=
ωn,m 
An,m (15.29)
−α
ωn,m
−α
 
0 ... 0
 cos(mφ1 ) . . . cos(mφ j ) 
Bn,m =



0 ... 0
sin(mφ1 ) . . . sin(mφ j )
(2 − δkr on (n))(2 − δkr on (m)) T
Cn,m = B n,m (15.30)
JL
the D matrix being empty. This modal-expansion representation of the transfer

function, or impedance, of the annular duct has been compared with an analytic
0.4
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16 18
ω/c
Fig. 15.16 Frequency response of annular duct, calculated with Sysnoise (solid),
analytic solution (dotted), and modal expansion (dashed).
solution and with a solution obtained from the commercial finite element method
package Sysnoise (Fig. 15.16). The modal-expansion representation is mathemat-
ically equal to the analytic solution for N = M = ∞. However, for the result
presented in Fig. 15.16, values of N = 2 and M = 4 have been used. In Sysnoise,
the zero Hertz mode is not calculated because it causes numerical difficulties.22
As a consequence, the frequency response calculated by Sysnoise is incorrect for
the very low frequency regime.
This representation can easily be extended to the more general case with inputs
and outputs on both sides of the duct (at x = 0 and x = L). When doing so, it is
helpful to apply the following partioning of the B, C, and D matrices:
ẋ = Ax + BI u1 + Br ur (15.31)
pI
= C1 x + Dll u1 + D1r ur
ρc
pr
= Cr x + Dr1 u1 + Drr ur
ρc
in which l and r refer to the left- or right-hand side inputs and outputs. The matrix A
is the same as for the single-sided duct. The matrices B1 and C1 contain the values
of the eigenvector on the left side of the duct and are identical to the matrices B and
C in Eq. (15.29). The matrices Br and Cr contain the values of the eigenvectors
at x = L. Because cos (πnx/L) = (−1)n if x = L, the following expressions are
obtained for Br and Cr in annular ducts:
Brn,m = (−1)n B1n,m

Crn,m = (−1)n CIn,m
Although the D matrices are empty again, they are shown here to be consistent
with a more general notation of partitioned state-space systems:

H11 H11 A Bl Br
H= = Cl Dll Dlr
H21 H22
Cr Drl Drr
F. Modeling of Sources and and Nonlinearities

An additional issue is to take into account inputs to the system. Two types of
system inputs are considered:
1) Sources inherent to the combustion process: They consist of sound created
by turbulence that propagates to the acoustic far-field, but of which the generation
itself is not influenced by the acoustic field.
2) External excitation with fuel flow actuators, necessary for active control.
The frequency spectra of the combustion source terms have been determined
experimentally by Schuermans.3 A transfer function Hsource was then fitted to the
magnitude of the frequency spectra. A time-domain source signal can then be
obtained by filtering a white-noise signal w(t) with the transfer function Hsource .
In a multiburner configuration, the source terms of the individual burners are, by
definition, linearly independent. Thus, different white-noise sequences have to be
generated for each burner.
So far, the entire system is considered to be linear. It is very likely that the actual
system is not linear, especially when the linearized system is unstable. Therefore, a
nonlinear saturation of the heat release signal was included, similar to the approach
used in Pankiewitz and Sattelmayer.17
G. Examples
1. Can-Type Combustor
As a first example, the one-dimensional thermoacoustic system described in
Lang et al4 is analyzed. This system consists of a straight duct, closed on one side,
open on the other side, with a flame stabilized in the middle of the duct. The pressure
drop across the flame sheet is assumed to be negligible. The acoustic velocity jump
is modeled by the so-called n–τ model: u 2 (t) = u 1 (t) + nu 1 (t − τ ), in which τ is
a delay time and n is referred to as the interaction coefficient. The impedance of the
open end is simply Z 3 = 0. The n–τ model contains a delay and is thus of infinite
order. To avoid systems of infinite order, the time delay is approximated by a Padé
approximation, a technique commonly used in control theory. The upstream duct is
represented by a transfer function similar to Eq. (15.24) but becomes more simple
because m = 0 in the one-dimensional case. By using the Redheffer star product,
the system can be represented as: S = P ∗ F ∗ C ∗ Z. The eigenfrequencies or
poles of the system are the complex eigenvalues of the matrix S. The eigenvalues
of S corresponding to the first resonant mode have been calculated for several
values of τ . According to Lang et al.,4 the eigenvalues of the system are shown to
1.04 0.04
1.02 0.02
Real (ω)/ω0
Im (ω)/ω0
1.00 0.00
0.98 –0.02
0.96 –0.04
0 π 2π 3π 4π 0 π 2π 3π 4π
ωτ ωτ
Fig. 15.17 Eigenfrequencies of the one-dimensional combustion system as a funtion
of normalized time delay, τ , left: real part of frequency; right: imaginary part. Solid
line: analytic solution, dotted line: modal expansion.
be the roots of

ωL ωL
cos 2 − sin 2
ne−iωτ = 0 (15.32)
c c
which requires a numerical search. The results obtained through modal expansion
and the roots of Eq. (15.31) are both plotted in Fig. 15.17, in which ω0 is the
resonance frequency in the case n = 0. For large values of ω0 τ both curves deviate
because the order of the Padé approximation was relatively low (six).
2. Interconnection of Annular Ducts

Using the state-space representation and the linear fractional transforms, com-
plex MIMO systems can be interconnected in a straightforward manner. As an ex-
ample, two annular ducts are interconnected by 24 smaller one-dimensional ducts,
very similar to those Evesque and Polifke.23 It represents in essence a gas-turbine
combustion chamber geometry: The first annular duct corresponds to the plenum
chamber, the smaller ducts represent the burners, and the second annular duct rep-
resents the combustion chamber. The temperature in the second annular duct differs
from the temperature in the first duct. This geometry has been modeled in Sysnoise
and was also evaluated by calculating P ∗ B ∗ C, in which P represents the up-
stream cold, annular duct; B represents 24 parallel one-dimensional ducts, and C
is the downstream hot, annular duct. Thus, P and C have 24 inputs and 24 outputs,
whereas B has 2 × 24 inputs and 2 × 24 outputs. Another possibility is to calculate
the eigenmodes of the upstream geometry together with the one-dimensional ducts
and calculate the downstream geometry (consisting of the hot annular duct only).
These two geometries can be calculated separately in Sysnoise, and coupled after
calculating them. These modes can then be used directly to apply modal expansion
and represent the two subsystems in state-space. If P B is the upstream system
and C is the downstream system, the interconnected system can be represented as:
P B C. Note that once the eigenfrequencies ωn,ref are obtained for the geometry
at a specific temperature, the eigenfrequencies at different temperatures are
6
5
4
3
ω/c
2
1
0
0 5 10 15 20 25 30 35
Eigenvalue number
Fig. 15.18 Eigenfrequencies of the coupled duct with temperature jump obtained
directly by Sysnoise (o) compared with the modal expansion method based on
numerically obtained eigenvalues and vectors (x) and using analytically obtained
eigenvectors (+).
√
easily obtained from: ωn = (c/cref )ωn,ref . The scaled eigenvectors: ψ/ remain
unchanged. Although the system interconnected in this way approximates the
numerical solution, a perfect match is not obtained—even for a system of very high
order because the interconnection at each interface position is one-dimensional, and
some important three-dimensional effects close to the interface are neglected. This
phenomenon is well known from Helmholz resonator theory: a length-correction
factor (virtual length) has to be applied to compensate for the local deformation
of the potential field. We can easily do this by adding a one-dimensional duct
element at the interconnection. The virtual length is very small; thus, a zero- or
first-order expansion is sufficient. The value of the length correction has been set
to (As /n)1/2 , as described in Rienstra and Hirschberg.24 The eigenfrequencies
of the coupled annular ducts calculated in three different ways are plotted in
Fig. 15.18, the values on the x-axis correspond to the numbering of the modes.
3. Annular Combustion System

From the previous two examples, it is only a small step to a representation of an
annular gas-turbine combustion chamber. The lumped-element representation of
the annular combustion system is represented by the block diagram in Fig. 15.13.
The burner will be modeled as an L − ζ model, as described in Schuermans
et al.25 This model is derived from the unsteady Bernouilli equation. The param-
eter L is a measure of the amount of air fluctuating in the burner nozzle, the
parameter ζ represents the effect of dissipation of acoustic energy to the mean
flow. The block diagram of the L − ζ model is given in Fig. 15.19, in which,
Lζ (s) = [−(L/c)s + M(1 − ζ − (A1 /A2 )2 ]−1 . The values of L and ζ are ob-
tained from a fit to measured transfer functions as described in Schuermans et al.7
For one burner, this element (denoted by B) has two inputs ( p1 and p2 ) and two
outputs (u 1 and u 2 ). In a multiburner configuration with J burners, a block diago-
nal matrix has to be formed: B = diag[B1 , B2 , · · · B J ]. If geometrically different
burners are used, the elements B j will be different.
The flame module, in which the interaction of the combustion process with
the acoustic field is modeled, is again represented as an n–τ model. However,
p1 p2 p2 p3
- ρ 3 c3
ρ 2 c2
Lζ(s)
u1
A1
A2
u2 u2
( − 1) e
T3
T2
− iωτ
e
1 ω 2σ 2
2
-
u3
Fig. 15.19 Block diagrams of the burner model (left), and the flame model (right).
rather than having one delay, a distribution of time delays is assumed here. The
block diagram is shown in Fig. 15.19. The relation between ∞ the acoustic veloc-
ities accross the flame is then given by û 2 (s) = (1 − n 0 ξ (τ ) e−sτ dr ) û 1 (s), in
which τ is the convective time delay between fuel injection and consumption
and ξ (τ ) is the probability density distribution function of time delays. The in-
teraction coefficient is given by n = 1 − T2 /T1 . The distribution of time delays
can either be obtained numerically13,26 or from experimental fits.27 If a Gaussian
distribution of time delays is assumed with mean value τ and standard deviation
στ , then, after carrying out the integration, the flame model can be written as
û 2 (s) = (1 − nes2 σ τ e−sτ )û 1 (s). In this case, the values of τ and στ have been
1/2 2
generated from fits to experimentally obtained frequency responses of the flame

transfer function. The interconnection of the flame block is given in Fig. 15.19.
The flame subsystem will be denoted by F; the diagonal system containing the J
flame transfer function is then denoted by F.
With the plenum chamber represented by annular duct P and the combustion
chamber as C, all the submodules can be combined as S = P B F C.
VI. Application to a Gas-Turbine Combustor

A. Application of Dampers in a Silo Combustor
The method was applied to a silo gas-turbine combustor, the ALSTOM GT11N2
(Fig. 15.20). The GT11N2 has an electrical output of about 115 MW, operates at
a pressure ratio of 15.5:1 and has an exhaust mass flow of 399 kg/s. The net-
work model included burner asymmetries and asymmetric arrangement of damper
elements. Helmholtz resonators were applied for advanced damping and are mod-
eled by an analytical nonlinear model. The hood and combustor are represented
by means of three-dimensional finite element method (FEM) modal expansion.
For the L–ζ representation of burners, the end correction is obtained by FEM
applied to a combustor-burners-hood model, and the loss coefficient is obtained
from impedance tube acoustic measurements. The flame is modeled as previously
described by a gasdynamic discontinuity whose transfer function is measured in an
atmospheric combustion test rig. A time-lag model of the flame-transfer function
is fitted to the experimental data.
The FEM modal analysis of hood and combustor was applied to the real geome-
tries. As an example, two modes are shown in Fig. 15.21.
Hood air-supply channels and the combustor exit are assumed to be acoustically
closed. This assumption is justified by the large area jump between air-supply
Fig. 15.20 GT11N2 ALSTOM gas turbine.
-1
Fig. 15.21 Hood and combustor modes.
Hood
Burner Burner
pˆ j , uˆ j Flame Flame
Resonator
Combustor
Fig. 15.22 Thermoacoustic network of the GT11N2 gas turbine.
channels and hood and by the large-flow Mach number at the combustor exit. For
acoustic wavelengths that are much larger than burner and resonator dimensions,
the acoustic pressure p̂ j and normal acoustic velocity û j are assumed to be uniform
on the opening area A j centered on the boundary at x j (see Fig. 15.22).
Cooling air for the Helmholtz resonator is supplied by the hood and enters the
resonator through an opening located on the resonator volume. The neck mouth
communicates directly with the combustion chamber. Both the resonator neck and
volume are modeled as ducts in which plane acoustic wave propagation occurs.
The relation between acoustic impedances at duct extremities 1 and 2 is given by
Tijdeman28 :

i

ek − e−k + Z 2 ek + e−k
Z 1 = k (15.33)
e + e−k + Z 2 ek − e−k
where the factor is given by the Kirchhoff solution,

i +1 γ −1
=i+√ 1+ √ (15.34)
2Sh Pr
and where Sh and Pr are the shear number and Prandtl number, respectively. The
area jump between neck and resonator volume is modeled by forcing the continuity
of p̂ and û A. Furthermore, at the neck ends an additional transfer function must
be considered to account for end resistance and end reactance. The end resistance
is attributable to the area change pressure drop. The end reactance accounts for the
fluid mass inside the combustor that is involved in fluctuations by the air fluctuating
inside the neck. The Helmholtz dampers were also tested in the impedance lab in
atmospheric conditions and showed excellent agreement with theory.29
Resonators for two unstable modes Additional resonators to suppress

the third unstable mode
Normalized p̂ Normalized p̂
simulation simulation
engine data engine data
Normalized frequency Normalized frequency
Fig. 15.23 Application of the state-space modeling to a silo gas-turbine combustor.

Left: two unstable modes controlled by Helmholtz dampers. Right: suppression of the
third unstable mode by additional damper elements.
Two different setups were considered in the modeling and were compared with
engine data:
1) Helmholtz dampers designed to suppress two unstable modes. A strong
instability at St ≈ 0.7 was observed.
2) Additional Helmholtz dampers were designed to suppress the third unstable
mode.
Only a limited volume was available to mount the resonators. The additional
damping power was thus limited as well. The design task therefore had the extended
goal of suppressing the instability by the most efficient placement of the damper
elements. The results are displayed in Fig. 15.23. Good agreement was found
between modeling and measurement. The third unstable mode was effectively
suppressed by the additional Helmholtz damper.
VII. Conclusion
A thermoacoustic network analysis method was presented. The network uses
both measured and analytically derived transfer functions of the components in
the combustion system. Transfer functions of burners and flames were obtained
experimentally by forcing the combustion system with loudspeakers. The transfer
function is then obtained from microphone signals by using a cross-correlation
technique. The system is assumed to be linear and time-invariant. The linearity
assumption of the burner and flame acoustics is validated by determining ex-
perimentally the transfer functions at several acoustic pressure levels. The flame
properties were show to be linear in the amplitude range of interest.
The network modeling approach was tested by modeling an atmospheric com-
bustion test facility with measured burner and flame transfer functions. The in-
fluence of changing boundary conditions was then predicted by using this model.
The predicted and measured spectra corresponded very well.
A stability analysis was performed by determining the eigenfrequencies of the

system. This analysis showed that for certain combustor lengths and certain acous-
tic exit conditions, the thermoacoustic system would become unstable. An experi-
ment on a combustion facility with varying length showed that very high pressure
amplitudes occur at those combustor lengths at which the system was predicted
to be unstable. The linear approach presented in this paper can be used to pre-
dict instabilities but will fail to predict pressure amplitudes at these instability
frequencies, since amplitudes will be limited because of nonlinearities.
The influence of the thermal power of the combustion process on the transfer
function is investigated. According to the classical n–τ model, the transportation
time of a fuel particle between fuel injection in the burner and fuel consumption
in the flame will result in a phase shift between velocity fluctuations at the burner
and acoustic heat release in the flame. Since the velocity of the fuel–air mixture
in the burner is proportional to the power (at constant equivalence ratio), the
characteristic time delay of the combustion process is expected to decrease with
increasing power. This general behavior can clearly be seen when comparing
the transfer functions measured at several thermal powers. A flame model that
considers the three-dimensional properties of the flame is shown to be consistent
with measured transfer functions.
References
1
Munjal, M. L., Acoustics of Ducts and Mufflers, John Wiley & Sons, New York, 1986.
2
Polifke, W., Paschereit, C. O., and Sattelmayer, T., “A Universally Applicable Sta-
bility Criterion for Complex Thermo-Acoustic Systems,” VDI-Berichte, 1997, pp. 455–
460.
3
Schuermans, B., Polifke, W., and Paschereit, C. O., “Prediction of Acoustic Pressure
Spectra in Gas Turbines Based on Measured Transfer Matrices,” ASME Turbo Expo ’00,
Munich, Germany, May 2000.
4
Lang, W., Poinsot, T., and Candel, S., “Active Control of Combustion Instability,” Com-
bustion and Flame, Vol. 70, 1987, pp. 281–289.
5
Paschereit, C. O., Schuermans, B., Polifke, W., and Mattson, O., “Measurement of Trans-
fer Matrices and Source Terms of Premixed Flames,” ASME Turbo Expo ’99, Indianapolis,
IN, June 1999.
6
Paschereit, C. O., and Polifke, W., “Investigation of the Thermoacoustic Characteristics
of a Lean Premixed Gas Turbine Burner,” ASME Turbo Expo ’98, Paper 98-GT-582, June
1998.
7
Schuermans, B. B. H., Polifke, W., Paschereit, C. O., and van der Linden, J., “Prediction
of Acoustic Pressure Spectra in Combustion Systems Using Swirl Stabilized Gas Turbine
Burners,” ASME Turbo Expo ’00, Munich, Germany, May 2000.
8
Cremer, L., “The Second Annual Fairy Lecture: The Treatment of Fans as Black Boxes,”
Journal of Sound and Vibration, Vol. 16, 1971, pp. 1–15.
9
Bodén, H. and Åbom, M., “Modelling of Fluid Machines as Sources of Sound in Duct
and Pipe Systems,” Acta Acustica, 1995, pp. 549–560.
10
Lavrentjev, J., and Åbom, M., “Characterization of Fluid Machines as Acoustic Mul-
tiport Sources,” Journal of Sound and Vibration, Vol. 197, 1996, pp. 1–16.
11
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Excitation of Thermoacoustic
Instabilities by the Interaction of Acoustics and Unstable Swirling Flow,” AIAA Journal,
Vol. 38, 2000, pp. 1025–1034.
12
Polifke, W., Poncet, A., Paschereit, C. O., and Döbbeling, K., “Reconstruction of Acous-
tic Transfer Matrices by Instationary Computational Fluid Dynamics,” Journal of Sound
and Vibration, Vol. 245, 2001, pp. 483–510.
13
Flohr, P., Paschereit, C. O., and van Roon, B., “Using CFD for Time-Delay Modeling
of Premix Flames,” ASME Turbo Expo ’01, New Orleans, LA, June 2001.
14
Polifke, W., Flohr, P., and Brandt, M., “Modeling of Inhomogeneously Premixed Com-
bustion with an Extended TFC Model,” ASME Turbo Expo ’00, Munich, Germany, May
2000.
15
Flohr, P., and Paschereit, C. O., “Mixing Prediction in Premix Burners Using Industrial
LES Tools,” Symposium on Computational Modeling of Industrial Combustion Systems,
ASME International Mechanical Engineering Congress and Exposition, New Orleans, LA,
Nov. 2002.
16
and Vibration, Vol. 180, 1995, pp. 557–581.
17
Pankiewitz, C., and Sattelmayer, T., “Time Domain Simulation of Combustion Instabil-
ities in Annular Combustors,” ASME Turbo Expo ’02, GT-2002-30063, Amsterdam, The
Netherlands, June 2002.
18
Schuermans, B., and Paschereit, C. O., “Investigation of Thermoacoustic Oscillations
in Combustion Systems using an Acoustic Network Model,” Acoustics of Combustion;
EU-ROTHERM Seminar No. 67, Univ of Twente, Enschede, The Netherlands, July 2000.
19
Zhou, K., and Doyle, C., Essentials of Robust Control, Prentice-Hall, Upper Saddle
River, NJ, 1998.
20
Culick, F., “Combustor Dynamics: Fundamentals, Acoustics and Control,” Active Con-
trol of Engine Dynamics, Von Karman Institute for Fluid Dynamics, Rhode-Saint-Genese,
Belgium, RTO-EN-20, May 2001.
21
Annaswamy, A., Fleifil, M., Rumsey, J., Prasanth, R., Hathout, J., and Ghoniem, A.,
“Thermoacoustic Instability: Modelbased Optimal Control Design and Experimental Vali-
dation,” IEEE Transactions Control Systems Technology, Vol. 8, No. 6, 2000.
22
LMS International, “SYSNOISE Revision 5.4 documentation, Version 1.0,” LMS In-
ternational, Leuven, Belgium, May 1999.
23
Evesque, S., and Polifke, W., “Low-Order Acoustic Modelling for Annular Combustors:
Validation and Inclusion of Modal Coupling,” ASME Turbo Expo ’02, No. GT-2002-30064,
Amsterdam, The Netherlands, June 2002.
24
Rienstra, S., and Hirschberg, A., “An Introduction to Acoustics,” Report IWDE99–02,
TU Eindhoven, 1999.
25
Schuermans, B., Paschereit, C. O., and Polifke, W., “Modeling Transfer Matrices of
Premixed Flames,” ASME Turbo Expo ’99, Indianapolis, IN, June 1999.
26
Polifke, W., Kopitz, J., and Serbanovic, A., “Impact of the Fuel Time Lag Distribution
in Elliptical Premix Nozzles on Combustion Stability,” 7th AIAA/CEAS Aeroacoustics
Conference, Maastricht, The Netherlands, May 2001.
27
Bellucci, V., Paschereit, C. O., Flohr, P., and Schuermans, B., “Thermoacoustic Simu-
lation of Lean Premixed Flames Using an Enhanced Time-Lag Model,” 31th AIAA Fluid
Dynamics Conference (Aeroacoustics Section), Anaheim, CA, June 2001.
28
Tijdeman, H., “On the Propagation of Sound Waves in Cylindrical Ducts,” Journal of
Sound and Vibration, Vol. 39, 1975, pp. 1–33.
29
Bellucci, V., Schuermans, B., Nowak, D., Flohr, P., and Paschereit, C. O., “Thermoa-
coustic Modeling of a Gas Turbine Combustor Equipped with Acoustic Dampers,” ASME
Turbo Expo ’04, No. 2004-GT-53977, Vienna, Austria, June 2004.
Chapter 16
Experimental Diagnostics of Combustion Instabilities
Jong Guen Lee∗ and Domenic A. Santavicca†

I. Introduction
T HE problem of unstable combustion continues to be a critical issue that limits

the development of gas-turbine combustors for propulsion and land-based
power-generation applications.1,2 To a great extent, unstable combustion is a result
of the increased use of premixed combustors, which are inherently more susceptible
to unstable combustion than nonpremixed combustors. To develop combustors that
are capable of stable operation over their entire operating range, an understanding
of the mechanisms that initiate and sustain unstable combustion and their relative
importance at different operating conditions is essential.
Unstable combustion refers to self-sustained combustion oscillations at or near
the acoustic frequency of the combustion chamber, which are the result of the
closed-loop coupling between unsteady heat-release and pressure fluctuations.
That heat-release fluctuations produce pressure fluctuations is well known and well
understood,1–6 but the mechanisms whereby pressure fluctuations result in heat-
release fluctuations are not. In general, it is thought that flame–vortex interaction,7,8
feed-system coupling,9–13 and spray–flow interactions2 are the most important
instability-driving mechanisms in gas-turbine instabilities.
Flame–vortex interaction refers to the interaction between the flame front and
vortices that are periodically shed at the entrance to the combustor. As the vortex
passes through the flame front, the flame is stretched by the vortex. Depending
on the rate at which the flame is stretched and the local equivalence ratio, this
interaction can either increase the flame area and hence the rate of heat release, or
it can lead to local extinction and as a result decrease the rate of heat release.
Feed-system coupling refers to a modulation of the fuel flow rate caused by
pressure fluctuations in the combustor and fuel-delivery system. This modulation
results in a fluctuating fuel concentration that is convected to the flame front and
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights re-
served.
∗ Senior Research Associate.
† Professor of Mechanical Engineering.
481
482 J. G. LEE AND D. A. SANTAVICCA
produces a fluctuating rate of heat release. If the fuel fluctuation arrives at the flame
front in-phase with the pressure fluctuation, the resulting heat-release fluctuation
amplifies the oscillations, whereas, if the fuel fluctuation arrives at the flame front
out-of-phase with the pressure fluctuation, the resulting heat-release fluctuation
damps the oscillations.
Spray–flow interactions refer to several phenomena that act to drive unstable
combustion, including oscillations in droplet atomization and droplet vaporization
and spray–vortex interaction. Any of these phenomena can cause modulation of
the fuel concentration and/or fuel distribution and thereby produce fluctuations in
the rate of heat release.
To understand the role and relative importance of flame–vortex interaction, feed-
system coupling, and spray–flow interactions during unstable combustion, mea-
surements must be made that characterize the mechanisms, the resulting instability,
and the relationship between the two. Of particular importance are measurements
of the fluctuations in pressure, heat release, fuel concentration, and flame structure.
This paper discusses several diagnostic techniques that have been used to make
such measurements. Most of the techniques require optical access and therefore
are limited to use in laboratory-scale combustors or in full-scale single-nozzle
combustor test rigs where optical access is available. The purpose of this paper is
to discuss the application of these techniques to the study of combustion instabil-
ities and, in particular, to demonstrate how these techniques can be used to gain
an improved understanding of the mechanisms of unstable combustion in gas- and
liquid-fueled combustors. To date, these techniques have primarily been used to
study the mechanisms of unstable combustion under limit-cycle conditions, and
very few studies have been done of the mechanisms involved in the transition from
stable to unstable combustion.
II. Pressure Measurements

There are many ways to detect and characterize unstable combustion. The most
basic measurement is of the dynamic pressure in the combustor. This measurement
is typically made by using high-frequency response (up to 250 kHz), water-cooled,
piezoelectric pressure transducers. The magnitude, phase, and frequency of the
pressure fluctuations and various statistical properties14,15 can be determined from
a measurement of the combustor pressure vs time.
A typical pressure trace from a longitudinal-mode instability in a laboratory-
scale, lean premixed combustor is shown in Fig. 16.1a, along with the corre-
sponding frequency spectrum in Fig. 16.1b. This particular instability exhibits a
peak-to-peak pressure fluctuation of approximately 2 psi at a frequency of 360 Hz
with weaker pressure fluctuations at the second and third harmonics, that is, 720
and 1080 Hz. Proper interpretation of such measurements requires knowledge of
the mode of the instability that determines the location of the nodes and antinodes
of the pressure oscillation. The instability mode can be determined by measuring
the pressure at several locations in the combustor.16 For example, identification and
characterization of a longitudinal mode requires a minimum of three transducers
located along the length of the combustor that is, at the entrance, the exit, and
halfway between. To identify transverse or circumferential modes, it is necessary
to locate multiple transducers at specific circumferential positions.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 483
a)
2
1
p' (psi)
-1
-2
0.000 0.025 0.050 0.075 0.100
Time (sec)
b)
0
Power spectrum (dBVRMS)
-10
-20
-30
-40
-50
-60
-70
0 200 400 600 800 1000 1200 1400 1600
Frequency (Hz)
Fig. 16.1 Pressure fluctuation during unstable combustion: a) typical time trace
and b) power spectrum.
A simpler approach, which can sometimes be used to identify the mode, is to

estimate the acoustic frequencies of the different modes, which are given by the
speed of sound in the combustor divided by the corresponding dimension of the
combustor, and to compare the acoustic frequencies with the measured frequency.
If the acoustic frequencies of the different modes are well separated, one can often
match the measured frequency to that of a specific mode and, thereby, identify the
mode of the instability.
To accurately measure combustor pressure fluctuations, the pressure transducer
should be mounted flush with the inner wall of the combustion chamber. In some
combustors the design of the combustion chamber does not allow for this or there
might be concerns about exposing the transducer to the high temperatures of com-
bustion. In such cases it is necessary to isolate the transducer from the combustion
chamber by using a recess mount with a small-diameter passageway between the

transducer and the combustion chamber. When the pressure transducer is mounted
in this manner, it is important to account for the acoustic characteristics of the pas-
sageway, because they can alter the amplitude and phase of the measured pressure
signal.16 Another consideration when making pressure measurements is that the
interaction between the flame and the pressure wave results in a three-dimensional
acoustic field in the vicinity of the flame. Under some circumstances, the pressure
at the wall of the combustor, where it is often measured, can differ in amplitude
and phase by as much as 20% from the pressure at the flame.17,18
Combustor pressure measurements are often combined with other measurements
when characterizing unstable combustion. (Examples of such measurements are
presented in the following sections.) When combining combustor pressure mea-
surements with other measurements, it is important to phase-synchronize the mea-
surements with the pressure oscillation, in which case it is usually necessary to
electronically filter the pressure signal to eliminate higher harmonics and noise
from the pressure signal. When an electronic filter is used for this purpose, care
must be taken to account for the phase delay introduced by the filter to correctly
synchronize the two measurements.
In addition to measuring pressure fluctuations in the combustor, it is useful
to simultaneously measure pressure fluctuations in the nozzle and the fuel line.
These fluctuations result in fluctuations in the fuel flow rate, a phenomenon that
was discussed previously and is referred to as feed-system coupling.9−13 Such
measurements provide valuable information for assessing the role of feed-system
coupling as an instability-driving or -damping mechanism. They can also be used
as a guide when attempting to modify the nozzle or fuel-system geometry to alter
the relative phase of the equivalence ratio and heat-release fluctuations to suppress
-200
Phase angle of Pf with respect to Pc
-240
-280
-320
-360
0.55 0.60 0.65 0.70 0.75
Equivalence ratio
Fig. 16.2 Phase difference between fuel-line pressure (P f ) and combustor pressure
(Pc ) fluctuations vs equivalence ratio ( , original length fuel line; , extended length
fuel line).
0.6
0.5
0.4
Pc,rms (psi)
0.3
0.2
0.1
0.0
0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80
Equivalence ratio
Fig. 16.3 The rms fluctuation in combustor pressure vs equivalence ratio
( , original length fuel line; , extended length fuel line).
the instability. This has been successfully demonstrated in a single-nozzle research

combustor19 and a combustor equipped with single-nozzle industrial nozzle,20 re-
sulting in attenuation of the pressure oscillation and a shift of the instability range.
For example, Fig. 16.2 shows the phase difference between the pressure fluctuation
in the combustor Pc and the pressure fluctuation in the fuel line P f as a function
of the equivalence ratio for two different fuel-line geometries.20 The geometry
change in this case involved changing the length of the fuel line between the fuel
injector and an upstream choked orifice. The corresponding change in the stabil-
ity characteristics of this combustor are presented in Fig. 16.3, where it is shown
that the range of unstable combustion has shifted to higher equivalence ratios.
Comparing Figs. 16.2 and 16.3 reveals that, with both fuel-line geometries, the
strength of the instability increases as the phase difference between the fuel-line-
and the combustion-chamber-pressure fluctuations changes from approximately
−250 to −300 deg, indicating the importance of feed-system coupling and sug-
gesting strategies for suppressing this instability.
III. Chemiluminescence Measurements

A second measurement that has proven extremely useful in characterizing un-
stable combustion is of the naturally occurring flame chemiluminescence. Chemi-
luminescence is the radiative emission from electronically excited species formed
by chemical reactions.21,22 The intensity of the chemiluminescence emission is
directly related to the concentration of the electronically excited species, which is
determined by the competition between the chemical reactions that produce the
excited species and collisional quenching reactions. As discussed subsequently,
the intensity of the chemiluminescence emission from lean premixed hydrocar-
bon flames has been shown to be an indicator of the rate of heat release; hence,
10.0
CH*
Equiv. Ratio = 0.80

8.0
Vinlet = 45 m/s
Tinlet = 673K
Intensity (a.u.)
OH*
6.0
CO2*
4.0
2.0
0.0
300 350 400 450 500 550 600
Wavelength (nm)
Fig. 16.4 Chemiluminescence-emission spectrum from a lean premixed combustor

operating at 100 kPa on natural gas, at an equivalence ratio of 0.8, with an inlet
temperature of 673 K.
this technique has been widely used for measuring both local and overall rates of
heat release in lean premixed combustors under both stable and unstable operating
conditions.
The strongest chemiluminescence emission from lean hydrocarbon flames
comes from CH∗ , OH∗ , and CO∗2 (the asterisk indicates an excited species), whereas
in rich hydrocarbon flames strong chemiluminescence emission also comes from
C∗2 . Fig. 16.4 shows a chemiluminescence emission spectrum measured in the
laboratory-scale optically accessible lean premixed combustor that is illustrated
100mm dia x 500mm

stainless steel
combustor section
choked swirl vanes
inlet
19mm dia
exit
110mm dia x 375mm
fused silica
combustor section
Fig. 16.5 Schematic drawing of optically accessible lean premixed combustor.
schematically in Fig. 16.5.23 This combustor was operating on natural gas at an

equivalence ratio of 0.8, a pressure of 100 kPa, and an inlet temperature of 400◦ C.
As shown, the chemiluminescence emission from CH∗ (431 nm) and OH∗ (309 nm)
occur at distinctly different and relatively narrow-wavelength intervals, whereas
the CO∗2 chemiluminescence lies over a broad-wavelength interval (350–600 nm)
and overlaps the CH∗ and OH∗ chemiluminescence spectra. For diagnostic applica-
tions, there are several considerations to keep in mind when choosing among OH∗ ,
CH∗ , and CO∗2 chemiluminescence. First, the measured CO∗2 chemiluminescence
signal strength can be significantly increased over that of OH∗ and CH∗ chemilu-
minescence by using a very broad filter, for example, λ = 100–200 nm. Second,
to detect OH∗ chemiluminescence, which occurs below 350 nm, ultraviolet-grade
optics must be used. Last, elimination of the CO∗2 chemiluminescence from an
OH∗ or CH∗ chemiluminescence measurement requires the added complexity of
an independent measurement of the CO∗2 chemiluminescence background. The in-
dependent measurement of the CO∗2 chemiluminescence background has not been
done in most OH∗ and CH∗ chemiluminescence measurements that have been re-
ported; therefore, these measurements include a significant contribution from CO∗2
chemiluminescence.
Measurements of the chemiluminescence emission from lean premixed flames
have been used in numerous studies to indicate the location of the reaction zone and
to infer local and overall heat-release rates.24−40 The rationale for such measure-
ments is usually based on the experimental observation that, for a fixed equivalence
ratio, the intensity of chemiluminescence emission from the entire flame, hereafter
Overall CO2 Chemiluminescence Intensity
2.0 0.70
Tinlet = 650 K
1.5
0.65
(a.u.)
1.0
0.60
0.55
0.5
0.50
φ = 0.45
0.0
3 4 5 6 7 8 9
Fuel Flow Rate (scfm)

Fig. 16.6 Overall CO2 chemiluminescence emission vs fuel flow rate from a lean
premixed combustor operating on natural gas at 100 kPa with an inlet temperature
of 650 K.
8
Vinlet = 67 m/s Tinlet = 673 K
Intensity/ Fuel Flow Rate (a.u.)
0
0.4 0.6 0.8 1
Equivalence Ratio
Fig. 16.7 Overall CO2 chemiluminescence emission divided by fuel flow rate vs equiv-
alence ratio.
referred to as the overall chemiluminescence emission, increases linearly with the

fuel flow rate, where the slope increases with increasing equivalence ratio.24−28
This is illustrated in Fig. 16.6, which shows the overall CO∗2 chemiluminescence
emission, from a lean premixed combustor operating on natural gas, as a function
of the fuel flow rate for a range of fixed equivalence ratios from 0.45 to 0.70 at an
inlet temperature of 650 K and a pressure of 100 kPa. These results were obtained
in the laboratory-scale optically accessible lean premixed combustor shown previ-
ously in Fig. 16.5. The CO∗2 chemiluminescence was detected by imaging the entire
flame onto a photomultiplier tube through a glass filter (BG-40) that transmits over
the wavelength interval from 325 to 650 nm. The results shown in Fig. 16.6 indicate
that the overall chemiluminescence intensity is a function of both the fuel flow rate,
that is, the overall heat-release rate, and the equivalence ratio. (This observation has
important implications regarding the use of the overall chemiluminescence emis-
sion as a measure of the overall rate of heat release during unstable combustion.)
The effect of the equivalence ratio on the overall chemiluminescence emission is
shown more clearly in Fig. 16.7, which is a plot of the overall chemiluminescence
emission divided by the fuel flow rate vs the equivalence ratio for a constant in-
let temperature of 650 K and inlet velocity of 67 m/s. These measurements were
made in the same combustor described previously. This result indicates that at fuel
lean conditions the overall chemiluminescence emission increases exponentially
with the equivalence ratio, which can be attributed to the exponential temperature
dependence of the reaction rate for the formation of CO∗2 ·41
The fact that it is the flame temperature, and not the equivalence ratio per se, that
affects the intensity of the chemiluminescence emission is further evidenced by
the observation that the overall chemiluminescence emission can be increased by
increasing either the equivalence ratio or the inlet temperature.23,41 Measurements
in the same combustor of the overall OH∗ chemiluminescence and the overall CH∗
chemiluminescence also show a linear dependence on the fuel flow rate and an
50
Intensity / Fuel Mass Flow Rate 40 φ = 0.65
30 φ = 0.60
φ = 0.55
20
φ = 0.45
10
0
45 65 85 105 125
Combustor Inlet Velocity (m/s)
Fig. 16.8 Overall CO2 chemiluminescence intensity divided by fuel flow rate vs
combustor velocity.
exponential dependence on equivalence ratio.23 For these measurements, the OH∗
chemiluminescence was detected by using a bandpass filter centered at 307 nm with
a full width at half-maximum (FWHM) of 10 nm, and the CH∗ chemiluminescence
was detected with a bandpass filter centered at 430 nm and a FWHM of 10 nm.
In addition to the effects of fuel flow rate and equivalence ratio, some studies
have shown that turbulence reduces the intensity of the overall chemiluminescence
emission.24,27 This effect was not observed in tests conducted in the same lean
premixed combustor discussed previously (Fig. 16.5). These results are shown
in Fig. 16.8, which is a plot of the overall CO∗2 chemiluminescence intensity
divided by the fuel flow rate vs the combustor inlet velocity for constant values of
equivalence ratio at an inlet temperature of 650 K and a pressure of 100 kPa. As
shown, the inlet velocity was increased by a factor of 2, corresponding to a change
in the Reynolds number from 9000 to 18,000, with no apparent decrease in the
overall chemiluminescence intensity.
Several studies involving detailed chemical kinetic calculations of lean premixed
laminar methane–air flames have been conducted to investigate the relationship
between the local rate of heat release, that is, the rate of heat release per unit flame
area, and the local chemiluminescence emission, that is, the rate of chemilumi-
nescence emission per unit flame area.41−44 These studies have shown that CH∗ ,
OH∗ , and CO∗2 occur within the reaction zone, which indicates that the location
of the chemiluminescence emission can be used as an indicator of the location
of the reaction zone. They have also shown that a correlation exists between the
chemiluminescence emission from both OH∗ and CO∗2 and the local rate of heat
release. An exception to both of these results occurs in extreme local strain or
flame curvature, for example, at cusps, where the calculations show that the local
chemiluminescence emission can effectively go to zero without local extinction of
the flame. These studies also indicate that most of the fuel goes through a reaction
path that includes the formation of CO∗2 , suggesting that CO∗2 chemiluminescence
should be a good indicator of the rate of heat release. Last, the studies show that
the local rate of heat release (HRlocal ) and the local chemiluminescence emission
(Ilocal ) are affected by unsteady strain and flame curvature and that they increase
exponentially with temperature, leading to a power-law relationship between the

local chemiluminescence emission and the local rate of heat release, that is,
Ilocal ∝ (HRlocal )α (16.1)
where the exponent α is a positive number and depends on the flame temperature
(as determined by the equivalence ratio, unburned gas temperature, dilution, and
radiation losses) and the effects of unsteady strain and flame curvature.41
To determine the relationship between the overall chemiluminescence emission
(Ioverall ) and the overall rate of heat release (HRoverall ), one must integrate the local
values over the flame area, that is,

Ioverall = Ilocal dAflame and HRoverall = HRlocal dAflame (16.2)
A A
If the flame temperature, that is, the equivalence ratio, unburned gas temperature,
dilution, and radiation losses, is constant and the effects of strain and flame cur-
vature are negligible or constant, then Ilocal , HRlocal , and α are constant over the
flame. These constant values result in the proportionality of the overall chemilu-
minescence emission and the overall rate of heat release, that is,
Ioverall = C HRoverall (16.3)
where the constant C depends on the flame temperature (i.e., equivalence ratio,
unburned gas temperature, dilution, and radiation losses) and the effects of strain
and curvature. This result is consistent with the experimental results presented in
Fig. 16.6, which show that for a fixed equivalence ratio and inlet temperature the
overall chemiluminescence emission increases linearly with fuel flow rate, that
is, the overall rate of heat release, and that the slope depends on the equivalence
ratio. Similarly, the results presented in Fig. 16.7, which show that the overall
chemiluminescence emission divided by the fuel flow rate increases exponentially
with the equivalence ratio, are also predicted by the detailed chemical kinetic
calculations. And last, the fact that the local chemiluminescence emission can be
affected by unsteady strain and flame curvature is consistent with the observations
that turbulence can reduce the overall chemiluminescence emission.
The relationship between the overall chemiluminescence emission and the over-
all rate of heat release is more complicated if the equivalence ratio and/or the effects
of strain and curvature vary over the flame surface. For example, a more compli-
cated relationship would occur in a partially premixed turbulent flame in which the
equivalence ratio is not constant over the flame surface. In this case, the exponent
α, in the equation relating the local chemiluminescence emission to the local rate
of heat release, varies with location on the flame surface, which in turn affects the
relationship between the overall chemiluminescence emission and the overall rate
of heat release. To some extent such variations are likely to average out such that
the relationship between the overall chemiluminescence and the overall rate of heat
release can be expressed in terms of the average equivalence ratio. Data supporting
this are shown in Fig. 16.9, which is a plot of the overall CO∗2 chemiluminescence
emission, for fixed overall equivalence ratio vs a parameter that is referred to as
Overall CO2Chemiluminescence Intensity (a.u.)

V = 84 m/s
2.0 inlet
T = 650 K
inlet
φ = 0.70
1.5
φ = 0.64
1.0
φ = 0.58
0.5 φ = 0.51
φ = 0.45
0.0
0 20 40 60 80 100
% Premixed
Fig. 16.9 The effect of incomplete fuel–air mixing on the overall CO∗2 chemilumines-
cence intensity.
“% premixed.”23 These measurements were made in the combustor illustrated in

Fig. 16.5 at a pressure of 100 kPa, an inlet temperature of 650 K, and an inlet
velocity of 84 m/s. In the 100% premixed case the fuel and air are perfectly mixed,
whereas in the “0% premixed” case there is a gradient in the equivalence ratio
across the annular mixing section that, for example, varies from 0.3 to 0.9 for an
overall equivalence ratio of 0.6. (Note that the fuel-distribution measurement was
made at the exit of the mixing section under cold flow, noncombusting conditions.)
Incomplete mixing increases the overall chemiluminescence emission, as would
be expected given the exponential dependence of chemiluminescence emission on
equivalence ratio; however, the effect of incomplete mixing is small, that is, there is
only a 10% increase in going from a 100% premixed to a 0% premixed condition.
The other factor affecting the overall chemiluminescence emission and the over-
all rate of heat release is the area of the flame. Any factors causing the flame area to
change, for example, flame–vortex interaction, will result in a change in the overall
chemiluminescence emission and the overall rate of heat release. Changes in the
flame area will not alter the relationship between the overall chemiliuminescence
emission and the overall rate of heat release, as long as the flame temperature and
the effects of strain and curvature are constant. In other words, as the flame area
changes, both the overall chemiluminescence emission and the overall rate of heat
release will change in proportion to the area change. On the other hand, if the
effects of stretch and/or curvature change over the flame surface as the flame area
changes, as might be expected during flame–vortex interaction, then the relation-
ship between the overall chemiluminescence emission and the overall rate of heat
release is likely to change as the flame area changes.
In general, the results of detailed chemical kinetic studies support the use of
chemiluminescence emission as a measure of the local and the overall rate of heat
release in lean premixed flames, but the studies also clearly indicate that such
measurements should be interpreted with caution. (Another technique that has
been proposed for making quantitative measurements of the rate of heat release is
HCO fluorescence.43,44 This technique is discussed in the Sec. V.)
In studies of unstable combustion in lean premixed combustors, chemilumines-
cence emission has been used by numerous researchers to characterize temporal
fluctuations in both the overall heat release12,20,30,31,35,38−40 and the spatial distri-
bution of the local heat release.7,12,20,30−32,34,35,37,39 As the preceding discussion
indicates, care must be taken when interpreting such measurements. For example,
when making overall chemiluminescence measurements, it is important to realize
that changes in the fuel flow rate and changes in the equivalence ratio indepen-
dently affect the overall chemiluminescence emission, whereas only changes in
the fuel flow rate affect the overall rate of heat release. This point can be illustrated
by considering two combustors in which the equivalence ratio at the inlet to the
combustor is fluctuating but for different reasons. In the first case, the equivalence-
ratio fluctuations are the result of fluctuations in the airflow rate, whereas the fuel
flow rate is constant. Under these conditions, fluctuations will occur in the overall
chemiluminescence emission; however, the overall rate of heat release will be con-
stant. (This is not to be confused with the fact that the local rate of heat release, i.e.,
the local flame speed, changes with the local equivalence ratio. This discrepancy
is compensated for by changes in the flame area such that the overall rate of heat
release remains constant.) In the second case, the equivalence ratio fluctuations are
the result of fluctuations in the fuel flow rate, whereas the airflow rate is constant.
Under these conditions, fluctuations will occur in the overall chemiluminescence
emission, that, in part, will be caused by equivalence ratio fluctuations and, in
part, by fuel flow rate fluctuations. In this situation, the chemiluminescence fluc-
tuations overestimate the fluctuations in the overall rate of heat release. The only
situation where the fluctuation in the overall chemiluminescence can be attributed
solely to fluctuations in the overall rate of heat release is when the equivalence
ratio is constant. In general, such conditions can only be achieved in a labora-
tory combustor, whereas in an actual combustor one would expect some degree of
feed-system coupling and, as a result, fluctuations in the equivalence ratio. Under
such conditions, measurements of the overall chemiluminescence-emission fluc-
tuations without simultaneous measurements of the equivalence ratio fluctuations
can potentially give misleading information about both the amplitude and phase
of the overall heat-release fluctuations.
There are also considerations when using chemiluminescence emission as a
measure of the local rate of heat release. The most obvious consideration is that
the chemiluminescence-emission measurement is a line-of-sight measurement,
that is, one measures the total emission integrated along the line of sight. This
effect can be significantly reduced by using an optical arrangement with a very
short depth of field; however, this reduction is at the expense of significantly re-
duced signal strength.37 Another approach, if the flame is axisymmetric, is to
use a deconvolution technique to reconstruct the two-dimensional emission field
from line-of-sight chemiluminescence images. (This approach is discussed and
illustrated later in this section) It is important to realize that the two-dimensional
chemiluminescence images obtained in this manner do not actually represent the
local chemiluminescence intensity, that is, on the scale of the flame thickness.
Because of the line-of-sight nature of the chemiluminescence measurement, the

two-dimensional chemiluminescence results effectively integrate over the local
three-dimensional flame structure. In other words, the intensity of the chemilumi-
nescence emission depicted in the two-dimensional chemiluminescence images
(I2D ) represents the product of the local chemiluminescence emission and the
local flame area, that is,
I2D = Ilocal · Ālocal (16.4)
where Ālocal is the flame area within a volume defined by the resolution of the
line-of-sight measurement.
Similarly, one can define a two-dimensional rate of heat release (HR2D ), which
represents the product of the local rate of heat release and the local flame area,
that is, HR2D = HRlocal · Ālocal . Of interest is the relationship between the two-
dimensional chemiluminescence emission and the two-dimensional rate of heat
release. Use of the preceding equations gives I2D = (Ilocal /HRlocal ) HR2D . Using
the power-law relationship between the local chemiluminescence emission and the
local rate of heat release, discussed previously, gives the following relationship:
I2D = (HRlocal )α−1 · HR2D = C2D · HR2D (16.5)
where C2D depends on the local flame temperature and any factors that affect
the flame temperature. Therefore, the intensity of the chemiluminescence emis-
sion shown in the two-dimensional chemiluminescence images is indicative of the
two-dimensional rate of heat release; however, it can also change independently
of the rate of heat release as a result of changes in the flame temperature and
any factors that affect the flame temperature. As with overall chemiluminescence
measurements, the most likely concern would be in a partially premixed flame in
which variations in the local equivalence ratio could lead to an inaccurate estimate
of the local rate of heat release.
An example of a measurement of the overall heat-release fluctuations during un-
stable combustion made in the optically accessible lean premixed combustor illus-
trated in Fig. 16.5 with CO∗2 chemiluminescence emission is shown in Fig. 16.10a.
The simultaneously recorded pressure fluctuation is shown in Fig. 16.10b. The
overall CO∗2 chemiluminescence emission plotted in Fig. 16.10a was measured
by imaging the entire flame onto a photomultiplier tube through an appropriate
bandpass filter as described previously. Care must be taken when making such
measurements to collect the chemiluminescence emission from the entire flame
so as to obtain an accurate indication of the total heat-release rate and to avoid
erroneous fluctuations caused by the flame moving in and out of the field of view.
Simultaneous measurements of the overall heat-release rate and the pressure, such
as shown in Fig. 16.10a and b, can be used to determine the phase difference be-
tween the heat-release and pressure fluctuations that is related to the overall system
damping and gain characteristics.3 Such measurements also provide information
on how the flame’s heat release responds to pressure fluctuations. For example,
Fig. 16.11 shows a plot of the rms overall heat-release fluctuation normalized
by the mean overall heat release (measured by using CO∗2 chemiluminescence)
vs the rms combustor pressure fluctuation during unstable combustion. These
a)
0.8
Overall Heat Release (a.u.)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.000 0.025 0.050 0.075
Time (sec)
b)
2
1
p´ (psi)
-1
-2
0.000 0.025 0.050 0.075
Time (sec)
Fig. 16.10 Simultaneous measurement of a) the overall heat-release fluctuations and
b) the combustor pressure fluctuations during unstable combustion.
measurements were made in an optically accessible single-nozzle test rig equipped

with a full-scale, industrial fuel nozzle (Solar Turbines Centaur 50) operating on
natural gas at an inlet temperature of 660 K and a pressure of 110 kPa over a
range of inlet velocities from 75 to 100 m/s and a range of equivalence ratios
from 0.575 to 0.7. This result shows that the normalized heat-release fluctuation
increases linearly as the pressure fluctuation increases until it becomes saturated at
high-pressure fluctuations, indicating that there is a nonlinear relationship between
the pressure and heat-release fluctuations during unstable combustion.45,46 A more
comprehensive assessment of the nonlinear response of lean premixed flames to
pressure fluctuations can be obtained from forced-response studies in which the
amplitude and relative phase of heat-release fluctuations resulting from imposed
pressure fluctuations over a range of frequencies and amplitudes are measured.40
Chemiluminescence emission can also be recorded by using an intensified
charge-coupled device (CCD) camera to obtain an image of the flame structure
during unstable combustion that represents the spatial distribution of the flame’s
heat release. An example of such a measurement is shown in Fig. 16.12a, which
35
30
25
qrms/qrms (%)
20
15
10
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Prms(psi)
Fig. 16.11 The rms heat-release fluctuation normalized by the average heat release vs
rms pressure fluctuation at various unstable operating conditions in a lean premixed
combustor.
presents a CO∗2 chemiluminescence image of the flame in a laboratory-scale lean

premixed dump combustor (Fig. 16.5) operating on natural gas. The gray scale
shown above the image indicates the magnitude of the chemiluminescence inten-
sity. Accompanying the chemiluminescence image is a line drawing that shows
the centerbody, the dump plane, and the location of the image within the 110-
mm-diam quartz combustor. The direction of flow in this and all subsequent
chemiluminescence images is from left to right. For this measurement, the im-
age acquisition is phase synchronized with the pressure oscillation and a total of
Low High
a) b)
Fig. 16.12 a) Line-of-sight integrated chemiluminescence image and b) correspond-

ing deconvoluted image.
30 individual images at that same phase angle are averaged to obtain the phase-
averaged image shown in Fig. 16.12a. This image is a record of the line-of-sight
integrated chemiluminescence intensity and therefore does not reveal the cross-
sectional structure of the flame. If the flame is assumed to be axisymmetric, one
can use a deconvolution procedure to reconstruct the two-dimensional flame struc-
ture, including “onion-peeling,” Abel transformation, and filtered backprojection
methods.47 The line-of-sight image shown in Fig. 16.12a was processed with
an Abel deconvolution procedure. The resulting image, which is shown in Fig.
16.12b, reveals the two-dimensional structure of the flame that was not apparent
in the original line-of-sight image.
A basic assumption of the deconvolution procedure is that the image is ax-
isymmetric. Because the line-of-sight image in Fig. 16.12a is not perfectly ax-
isymmetric, the upper and lower halves of the image were averaged to create an
axisymmetric image before applying the Abel inversion. This procedure of cre-
ating an axisymmetric line-of-sight image is usually necessary, and care must be
taken when interpreting the resulting reconstructed images. If the line-of-sight
images are reasonably axisymmetric, the insights gained from the reconstructed
two-dimensional images usually outweigh the uncertainty associated with the ax-
isymmetric approximation. Unfortunately there is no way to quantify this tradeoff;
therefore, the reconstructed two-dimensional images must always be interpreted
with care.
Figure 16.13 shows a sequence of 12 phase-averaged two-dimensional CO∗2
chemiluminescence images recorded in increments of 30 phase-angle degrees dur-
ing one period of a 235-Hz instability in the same laboratory-scale lean premixed
dump combustor mentioned previously. In this case the combustor was operating
at 100 kPa, with an inlet temperature of 673 K, an inlet velocity of 45 m/s, and an
equivalence ratio of 0.45. In addition, the exit of the combustor was not restricted,
Low High
#1 #2 #3 #4
Image Number
1 3 5 7 9 11 13
Pressure (psi)
1.0
0.5
#5 0.0 #6
-0.5
-1.0
0.0 1.0 2.0 3.0 4.0
#7 Time (msec) #8
#9 #10 #11 #12
Fig. 16.13 Flame-structure evolution during one period of unstable combustion with
a frequency of 235 Hz.
as it is in Fig. 16.5; hence, a marked change occurred in the frequency of the insta-
bility compared with the previously presented results. Note that only the upper half
of the flame is shown, because the reconstructed images are axisymmetric. Note
also, to the left of each image, a line drawing illustrates the location of the image
relative to the combustor. Figure 16.13 also contains a plot of the combustor pres-
sure vs time, measured at the dump plane, over one period of the instability with
markers indicating when each of the images was recorded. The two-dimensional
flame-structure image sequence reveals the temporal evolution of the flame struc-
ture during the instability and provides insight regarding the phenomenology of
the instability. For example, the images in Fig. 16.13 show a flame that is anchored
on the centerbody and extends outward into the recirculation zone and all the way
to the wall of the combustor. The overall flame shape remains very nearly the same
during the instability but there is a noticeable change in the overall intensity of the
flame’s heat release, indicating that minimum heat release, that is, images 9, 10
and 11, occurs when the pressure is minimal. There is also a periodic break in the
flame between where it is attached to the centerbody and the recirculation zone,
which also occurs when the pressure and overall heat release are at their minimum
levels.
Two flame-structure image sequences are shown in Fig. 16.14a and 16.14b,
which correspond to instabilities in the same lean premixed combustor (Fig. 16.5),
at the same operating conditions (Tinlet = 623 K, Vinlet = 59 m/s, p = 100 kPa, and
φ = 0.58), but with different inlet fuel distributions. In Fig. 16.14a, the fuel and
air are completely mixed before entering the combustor, whereas in Fig. 16.14b,
although the overall equivalence ratio is the same, the equivalence ratio increases
with increasing radius across the annular outlet of the mixing section. Again,
only the upper half of the image is shown because the reconstructed images are
axisymmetric. In both cases the instability frequency is approximately 350 Hz. The
images are phase synchronized relative to the pressure oscillation at the combustor
entrance and are acquired in increments of 24 phase-angle degrees, giving a total
of 15 images within one period, in which each image is an average of 30 individual
images acquired at a given phase angle. Both cases show evidence of flame–vortex
interaction, but the details of the interaction are noticeably different in the two
cases. The flame in Fig. 16.14a appears to be wrapped around the vortex, which
results in stretching and contraction of the flame zone, whereas the flame in Fig.
16.14b appears to be contained within the vortex, exhibiting periodic extinction
and reignition of the entire reaction zone.
Two-dimensional flame-structure images, such as those shown in Figs. 16.13 and
16.14, reveal the location and intensity of the flame’s heat release and its temporal
evolution during one period of the instability. Combining this information with the
measured pressure fluctuation, one can calculate the Rayleigh index distribution,
R(x, y), which is given by the following equation:

1
R(x, y) = p (x, y, t) q (x, y, t) dt (16.6)
T π
where q (x, y, t) is the local heat-release fluctuation determined from the

two-dimensional flame-structure images and p (x, y, t) is the local pressure
a)
#1 #2 #3 #4 #5
#6 #7 #8 #9 #10
#11 #12 #13 #14 #15
b)
#1 #2 #3 #4 #5
#6 #7 #8 #9 #10
#11 #12 #13 #14 #15
Fig. 16.14 Flame-structure evolution during one period of unstable combustion with
a frequency of 350 Hz at two different operating conditions in the same lean premixed
combustor.
fluctuation.7,20,30−32,34,35,39 Because the wavelength of the pressure oscillation is

often much greater than the length of the flame, the pressure in that case can be
assumed to be spatially uniform over the region of heat release and, therefore, only
a function of time.
The Rayleigh index is a measure of the correlation between the heat-release
fluctuation and the pressure fluctuation and, therefore, represents the strength of the
coupling between the two. A positive Rayleigh index indicates that the heat-release
and pressure fluctuations are in-phase, in which case the heat release fluctuation
amplifies the pressure fluctuation. A negative Rayleigh index indicates that the
heat-release and pressure fluctuations are out-of-phase, and therefore the heat-
release fluctuation acts to damp the pressure fluctuation.
The Rayleigh index distributions corresponding to the two-dimensional flame-
structure images shown in Fig. 16.14a and 16.14b are presented in Fig. 16.15a and
16.15b, where grey and white represent regions of negative and positive Rayleigh
index, respectively. The Rayleigh index distribution can be used to identify lo-
cations where the instability is amplified, that is, regions of positive Rayleigh
index, and locations where the instability is damped, that is, regions of negative
Rayleigh index, and thereby provides insight regarding the phenomenology of the
instability. Figure 16.15a shows two damping regions, one in the recirculation zone
a)
400
0
-200 200
400
0 600
b)
0 1000
800
600
400
200
Fig. 16.15 Rayleigh index distributions corresponding to the unstable flames shown
in Fig. 16.14a and 16.14b, respectively (white, positive Rayleigh index; gray, negative
Rayleigh index).
behind the centerbody and the other in the recirculation zone downstream of the
dump plane; whereas a larger region of positive Rayleigh index is located along
the shear layer between the dump plane and the centerbody recirculation zones.
Figure 16.15b shows a significantly different Rayleigh index distribution with a
large region of positive Rayleigh index centered in the dump-plane recirculation
zone and a smaller damping region immediately downstream of the dump plane. A
comparison of the locations of minimum and maximum Rayleigh index, that is, the
locations of damping and gain, with the flame-image sequence can provide insight
as to the phenomenology of the instability and the role of the instability-driving
mechanisms.
Chemiluminescence-emission measurements have also been shown to provide
information and insights that can be used to optimize active combustion-control
systems employing modulated secondary fuel injection for the suppression of un-
stable combustion. For example, it has been shown that effective suppression of an
instability can be achieved with less secondary fuel if the fuel is injected into the
region of maximum damping indicated in the Rayleigh index distribution.48 Chemi-
luminescence imaging can also be used to determine the optimum phase delay in
active combustion-control systems employing secondary fuel flow modulation.48
When a pulse of secondary fuel is injected into a combustor it produces a detectable
modulation of the chemiluminescence emission, referred to as the flame response,

which indicates when the secondary fuel reaches the flame front and burns. With
that signal as an indication of the heat release produced by the combustion of the
secondary fuel pulse, one can calculate what is called the flame-response Rayleigh
index by using the following equation:
to +T
1
Flame-response Rayleigh index = p qsecondary

dt (16.7)
T to
where T is the period of the secondary fuel flow modulation, to is the time delay
between the pressure signal zero crossing and the secondary fuel valve trigger

signal, p is the measured pressure signal, and qsecondary is the flame-response
function. An example of the flame-response function is shown in Fig. 16.16a.48
In this case, subharmonic secondary injection is used where the frequency of
secondary fuel injection is one-fourth that of the instability. This is illustrated in
Fig. 16.16b, in which the flame-response function is shown along with the pressure
oscillation. Also shown in Fig. 16.16b is the control signal to the secondary fuel-
control valve and the time delay between that signal and the zero crossing of the
pressure signal. The flame-response Rayleigh index is a measure of the effect of the
heat release caused by the secondary fuel on the instability. If its value is positive,
the secondary fuel acts to amplify the instability, whereas if it has a negative value,
it acts to damp the instability, where the optimum phase delay corresponds to the
case of maximum damping. The flame-response Rayleigh index as a function of
the time delay to , is plotted in Fig. 16.16c for the flame-response function and
pressure oscillation shown in Fig. 16.16b. According to this result, the time delay
between the zero crossing of the pressure oscillation and the valve trigger signal
for maximum damping is approximately 1.25 ms. This compares reasonably well
with the experimentally determined delay time for maximum suppression of this
instability, which is approximately 1 ms.
IV. Infrared-Absorption Measurements

Laser-absorption techniques have been used to measure various flowfield param-
eters such as gas concentration, temperature, pressure, and velocity.12,20,49−61 The
basic technique involves passing a laser light of a known wavelength and intensity
through the medium of interest and measuring the attenuation of the light due to
resonance absorption by specific atoms or molecules. The absorption process is
described by Beer–Lambert’s Law, that is,
I l
= 10− o ε cdx (16.8)
Io
where Io is the intensity of incident light, I is the intensity of transmitted light,

ε is the decadic molar absorption coefficient (square centimeters per mole), l
is the absorption path length (centimeters), and c is the molar concentration of
absorbing species (moles per cubic centimeters). The attenuation also depends on
the temperature and pressure through changes in the absorption coefficient. The
major limitation of this technique is the fact that it is a line-of-sight measurement,
a) 0.040
chemiluminescence)
0.035
0.030
q´ (CO2*
0.025
0.020
0.015
0.010
0.005
0.000
0 1 2 3 4 5 6 7 8 9 10 11
Time (msec)
b) 0.03
time delay (to) q´secondary
Heat release (a.u.)
0.02 Valve opening pulse
0.01
0
Period (T)
-0.01
-0.02
0 2 4 6 8 10 12
Time (msec)
c) 0.010
0.005
Global Response
Rayleigh Index
0.000
-0.005
-0.010
-0.015
-0.020
0 1 2 3 4
time delay (msec)
Fig. 16.16 a) Flame-response function, b) flame-response function superimposed
on the unstable pressure trace, and c) flame-response Rayleigh index vs time-delay
prediction.
a) 0.50
0.40
0.30
I/Io
0.20
0.10
0.00
250 300 350 400 450 500 550 600 650 700
Temperature (K)
b) 1
0.1
I/Io
0.01
T = 295 K
T = 473 K
T = 573 K
0.001 T = 673 K
100 200 300 400 500 600

Pressure (kPa)
Fig. 16.17 a) Temperature and b) pressure dependence of the normalized transmit-
tance (I/Io ) of a 3.39-µm He-Ne laser beam through a homogeneous methane–air
mixture.
that is, the measured attenuation is the result of the integrated absorption over the
entire beam path and, therefore, is a measure of the average flowfield properties
along the beam path.
The laser-absorption measurement that has proven most valuable in the study of
combustion dynamics is an infrared-absorption measurement of hydrocarbon fuel
concentration based on the fortuitous matchup between the 3.39-µm wavelength
of the infrared helium-neon laser and a vibrational-rotational energy level transi-
tion in hydrocarbon molecules.49,51 In this case, the absorbing molecule is a stable
species, therefore, the simplest procedure for making quantitative measurements of
concentration is to empirically determine the pressure and temperature dependence
of the absorption coefficient. An example of this empirical determination for
methane is shown in Fig. 16.17a and 16.17b, in which plots of the normalized
transmittance (I /Io ) vs temperature at a fixed pressure and vs pressure for values
of constant temperature, respectively, are shown.12 These measurements were
made in a flow cell with a homogeneous methane–air mixture at known conditions.
The 3.39-µm output from a He-Ne laser (3 mW) was used as a light source and
the transmitted light was detected by using a thermoelectrically cooled indium-
arsenide (InAs) detector. Once the normalized transmittance is known, the decadic
molar absorption coefficient ε can be obtained from Beer–Lambert’s Law, that is,

1 I
ε = − log10 (16.9)
cl Io
By using the data shown in Fig. 16.17, the following expression for the decadic
molar absorption coefficient for methane as a function of pressure and temperature
is obtained12 :

Po T P
ε = 84737 C1 + C2 −1 (16.10)
P 293 K Po
with
C1 = −0.1131 + 1.1875 (293 K/T )
and
C2 = 0.712 − 1.536 exp(−2.118 ∗ [293K/T ])
An important application of the infrared-absorption technique in combustion-

instability studies is for measuring temporal fluctuations in the equivalence ratio
caused by feed-system coupling. A drawing illustrating the experimental setup
for an infrared-absorption measurement in a single-nozzle test rig equipped with a
full-scale, industrial lean premixed nozzle (Solar Turbines Centaur 50) operating
on natural gas is presented in Fig. 16.18.20 As shown, the 3.39-µm laser beam
passes through the annular mixing section in the injector, just upstream of the en-
trance to the combustor. Beam passage through the annular mixing section required
a modification to the nozzle to provide two-sided optical access for the laser beam.
Note that sapphire windows are required for transmission of the infrared beam. The
actual measurement is of the normalized transmittance of the incident laser beam,
that is, the ratio of the transmitted to the incident laser power. To convert this ratio
to an equivalence ratio, an in situ calibration of the normalized transmittance vs the
overall equivalence ratio is required. This calibration involves making measure-
ments over a range of equivalence ratios, without combustion, at fixed pressure and
temperature. To use this calibration at other pressures and temperatures one must
account for the pressure and temperature dependence of the decadic molar absorp-
tion coefficient previously discussed. In addition, the fact that the density changes
with pressure and temperature must be accounted for because the absorption
measurement actually measures the fuel concentration not the equivalence ratio.
Figure 16.19a shows the equivalence ratio vs time measured with the infrared-
absorption technique over one period of a 465-Hz instability in the combustor
illustrated in Fig. 16.18. In this case the combustor was operating on natural gas,
a) Data Adapters
Acquisition InAs Detector
System
InAs Detector
HeNe Laser
(3.39 µm) Data Acquisition
System
Center bluff-body
b) 7 mm
Diffuser
Measurement Location
Preheated Air
Swirl vanes Fuel injection

Location
Fig. 16.18 Schematic drawing of the setup for equivalence-ratio measurements using
IR absorption: a) front view and b) side view.
at a pressure of 110 kPa, with an inlet temperature of 658 K and an inlet velocity of
100 m/s. It is assumed that the temperature of the mixture in the nozzle is constant
and, therefore, only fluctuations in the pressure are accounted for when converting
the measured transmittance to equivalence ratio. This result clearly shows that
feed-system coupling is playing a significant role in this instability, resulting in
peak-to-peak fluctuations in the equivalence ratio of approximately ±0.05 about a
mean of 0.65. The frequency spectrum corresponding to these fluctuations is shown
in Fig. 16.19d. The equivalence-ratio fluctuations show a dominant frequency
at 465 Hz, with weaker oscillations appearing at the higher harmonics. Shown
in Fig. 16.19b and 16.19c are the phase-synchronized pressure and heat-release
measurements for this instability, and shown in Fig. 16.19e and 16.19f are the
corresponding frequency spectra. The dominant frequencies of the equivalence-
ratio, pressure, and heat-release fluctuations are clearly the same, that is, 465 Hz,
although there are significant differences in the relative magnitude of the second
harmonic oscillations. Most noticeable is the 10-dB difference between the first and
second harmonics of the equivalence-ratio fluctuation vs 23-dB difference between
the first and second harmonics of the heat-release fluctuations. This suggests that
the flame, which is spatially distributed, acts to average out the higher-frequency
a) d)
0.75 -10
Equivalence
0.70 -20
φ)
dBVrms (φ
Ratio
0.65 -30
-40
0.60
-50
0.55 -60
b) e)
1 -10
dBVrms (Pc)
Pc (psi)
-30
0 -50
-70
-1 -90
c) f)
0.40 -10
dBVrms (q)
0.35
0.30 -30
q
0.25 -50
0.20 -70
0.15
0.0 0.2 0.4 0.6 0.8 1.0 0 400 800 1200 1600
t/T Frequency (Hz)
Fig. 16.19 a–c) Time traces and d–f) power spectra of equivalence ratio φ, combustor
pressure Pc , and heat release q during one period T of unstable combustion.
fluctuations in the equivalence ratio. Also note that the relative magnitudes of the
pressure, equivalence-ratio, and heat-release fluctuations for this instability are
3%, 5%, and 23% of the mean, respectively.
Simultaneous pressure, heat-release, and equivalence-ratio fluctuation measure-
ments can be used to determine the phase delay or time lag between these processes.
Of particular interest is the time lag between the equivalence-ratio fluctuation and
the heat-release fluctuation, because it is important in assessing the role of feed-
system coupling. To estimate whether the equivalence-ratio fluctuation produced
by feed-system coupling arrives at the flame front in-phase with the heat-release
fluctuation, one must estimate the convection time between the fuel-injection loca-
tion and the flame front. The most difficult part of the convection time to estimate
is the time required for the fuel to travel from the entrance of the combustor to
the flame front where the fuel burns. The phase delay or the time lag between the
equivalence ratio and the heat-release fluctuation shown in Fig. 16.19a and 16.19c
is a direct measurement of that quantity.
The 3.39-µm helium-neon laser-absorption technique has also been imple-
mented in a fiber-optic probe52,57 and in a fast-response extraction probe.59 The
main advantage of this approach is that spatially resolved measurements are pos-
sible, that is, with a spatial resolution on the order of 1 mm. Such probes have
been successfully used to measure spatial fuel distributions and equivalence-ratio
fluctuations in laboratory-scale and commercial single-nozzle lean premixed com-

bustors.
And last, note that the helium-neon laser-absorption technique can be used to
measure the fuel concentration in lean premixed combustors operating on either
gas or liquid hydrocarbon fuel; however, in combustors operating on liquid fuel,
the technique cannot be applied if liquid drops are present. In other words, all
the fuel must be vaporized, otherwise attenuation of the laser beam because of
Mie scattering from the drops will lead to erroneous results unless the amount of
light scattered from liquid droplets is accounted for.62 The infrared He-Ne laser-
absorption technique can also be applied to situations in which the fuel distribution
is not spatially uniform, for examples, in nonpremixed or partially premixed com-
bustors. Use of this laser-absorption technique, however, requires that absorption
measurements be made over a large number of beam paths and that tomographic
reconstruction techniques to be used.47,63 Such measurements are complex and
time consuming if the flow is steady; they are impractical if the flow is unsteady.
Laser-induced fluorescence, which is discussed in the next section, is a better
technique for measuring the spatial distribution of fuel concentration.
V. Laser-Induced Fluorescence Measurements

The basic principal of the laser-induced fluorescence technique is that laser ra-
diation is used to selectively excite an atomic or molecular species of interest to
an upper-electronic state via a laser-absorption process.64 The excitation process
is followed by the spontaneous emission of a photon when the excited atom or
molecule decays back to a lower-energy level. The spontaneous emission is re-
ferred to as fluorescence or as laser-induced fluorescence, and its intensity can be
related to the number density of the species of interest. To quantify the relation-
ship between the fluorescence intensity and the number density of the absorbing
species, one must account for the energy-level population distribution of the ab-
sorbing atom or molecule and for collisional quenching and redistribution effects.
For certain molecules, such as OH and CH, and with the selection of an appro-
priate excitation/detection scheme, it has been shown that the fluorescence signal
is directly proportional to the concentration of the absorbing species.65 Similarly,
excitation/detection schemes that allow for the determination of the temperature
have been developed for certain molecules.66 For stable species, the effects of col-
lisional quenching and redistribution can be accounted for by simply calibrating
the fluorescence intensity vs number density of the fluorescing species as a func-
tion of temperature and pressure. When calibrating a fluorescence measurement,
the overall composition should be approximately the same as in the actual mea-
surement because collisional quenching depends on composition. Keep in mind,
however, that quantitative fluorescence measurements are not always necessary
and that useful information can often be obtained from qualitative measurements
that provide a relative measure of the number density, or in some cases only indicate
the location, of the species of interest.
At low-laser excitation irradiance, the fluorescence signal, Sf (joules per square
centimeter), can be related to the mole fraction of the fluorescence species, χabs ,
by the following equation:

P
Sf ∝ (E/ hν) χabs σ (λ, T ) φ(λ, P, T ) (16.11)
T
where E is the laser irradiance (J/cm2 ), h is Planck’s constant, ν is the laser fre-
quency, χabs is the molar concentration of the fluorescence species, σ (λ, T ) is the
molecular absorption cross section of the fluorescence species, and φ(λ, P, T ) is
the fluorescence quantum yield. In this equation, the effect of gas composition on
the fluorescence yield is assumed to be constant. For a fixed excitation wavelength
and under isothermal and isobaric conditions, the fluorescence signal is only pro-
portional to the mole fraction of the fluorescence seed. However, in flows in which
the temperature and pressure are changing, the effect of the energy-level popula-
tion distribution and of collisional quenching and redistribution on the absorption
cross section and the fluorescence quantum yield, and therefore the fluorescence
signal, must be accounted for, as discussed before.
Laser-induced fluorescence can be used to make point measurements with sub-
millimeter spatial resolution or to make two-dimensional measurements, also re-
ferred to as planar laser-induced fluorescence (PLIF) measurements.67,68 In making
point measurements, the laser beam is focused with a spherical lens to a small-
diameter beam waist and the fluorescence signal is detected by imaging the beam
waist through an aperture onto a photomultiplier tube. In making two-dimensional
measurements, the laser beam passes through a combination of cylindrical and
spherical lenses to produce a thin laser sheet and the resulting fluorescence signal
is detected by imaging a portion of the laser sheet onto an intensified CCD camera.
In both cases, an appropriate interference filter is used to isolate and selectively
detect the desired fluorescence wavelength. The strength of the fluorescence sig-
nal for gas-phase fluorescence measurements is typically very low, requiring the
use of high-power pulsed lasers, where depending on the fluorescence species and
its concentration, single-pulse measurements are often possible. Unfortunately,
high-power pulsed lasers operate at relatively low pulse rates, that is, typically
10–20 Hz; therefore, this technique does not provide a continuous measure of the
fluorescence species concentration. Although periodic phenomena, such as unsta-
ble combustion, in which the measurements can be phase synchronized with the
instability and the periodic behavior reconstructed, present a special case.
A useful application of laser-induced fluorescence in the study of combustion
dynamics is in the characterization of fuel–air mixing. Measurement of fuel–air
mixing is important because both the temporal and spatial fuel distribution can have
a significant effect on the stability characteristics of the combustor.32 Many fuels
of interest, however, are not well suited for fluorescence measurements, and so a
common approach is to seed the fuel with a small amount of a fluorescence seed, in
which the concentration of the fluorescence seed, as measured by the fluorescence
technique, is assumed to be an indicator of the fuel concentration. An advantage
of using a fluorescence seed is that its concentration can be controlled. In addition,
a fluorescence seed can be selected that has optimum physical and spectroscopic
properties, making quantitative equivalence ratio measurements possible. Various
species have been used as fluorescence seeds for fuel–air-mixing studies and de-
tailed information on their fluorescence characteristics can be found in Refs. 69–74.
Several factors must be considered when selecting a fluorescence seed to charac-
terize mixing, including boiling point, autoignition temperature, absorption and
fluorescence characteristics, mass diffusion coefficient, cost, and toxicity.
For gaseous fuels, acetone is commonly used as a fluorescence seed because
of its low-boiling point (50◦ C at 1 atm) and high-vapor pressure (184 torr at
20◦ C), which allows for easy seeding and high-seed density. The spectroscopic
characteristics of acetone are also well known. It absorbs over a broad range of
wavelengths (225–320 nm) with maximum absorption between 270 and 280 nm.
The fluorescence emission is broadband in the blue (350–550 nm) and short lived
(τ ≈ 4 ns) with a fluorescence efficiency of 0.2%.71−74
Because acetone is a stable species, the effects of pressure and temperature
on the absorption cross section and the fluorescence yield are best accounted
for empirically. Figures 16.20a and 16.20b show the effect of temperature and
pressure, respectively, on the acetone fluorescence signal with 266-nm excita-
tion, that is, the fourth harmonic output of a Nd:YAG laser. Each panel shows
two curves. The filled circles are the actual measurements, and the filled squares
have been corrected to constant number density. The measurements were made in
a flow cell with the volume fraction of acetone fixed at 1%; therefore, changes
in pressure and temperature also affect the fluorescence signal as a result of
changes in the density. The actual measurements show that the fluorescence sig-
nal decreases with increasing temperature and increases with increasing pressure.
After correcting for changes in density, however, the fluorescence signal is in-
dependent of pressure, but it decreases with increasing temperature. Knowledge
of the effect of pressure and temperature on the fluorescence signal, indepen-
dent of their effect on density, is critical to the proper interpretation of fluores-
cence measurements. The temperature and pressure dependence of acetone-laser-
induced fluorescence with different excitation wavelengths can be found in the
Refs. 71–74.
Acetone PLIF was used to measure the fuel distribution in the natural-gas-fueled
laboratory-scale dump combustor that is shown schematically in Fig. 16.21a.75
This particular combustor was used in a study of the effect of combustor-inlet
fuel distribution on combustion stability and emissions and, hence, has the unique
capability of allowing for systematic variation of the fuel distribution. Systematic
variation is accomplished by injecting the fuel at one or more of three injection
locations, labeled (1), (2), and (3) in Fig. 16.21a. For the acetone-fluorescence
measurements, the fuel is replaced with air to which 0.5%, by volume, of acetone
has been added. (The measurements are made without combustion.) To ensure
complete vaporization of the acetone, the acetone is injected into the air by using a
spray nozzle and the air is preheated to 100◦ C. Note also that in injection locations
(2) and (3), the flow rate of the simulated fuel was set to match the momentum
flux of the actual fuel jet to properly simulate the mixing characteristics.
The excitation source is the fourth harmonic (266 nm) output of a pulsed
Nd:YAG laser, with a laser pulse energy and duration of 40 mJ/pulse and 7 ns,
respectively. The laser beam is formed into a 0.5-mm-thick by 40-mm-high sheet
that is positioned approximately 1 mm downstream of the dump plane across the
a) 2.5 800
2.0
600
Sfl • T
1.5
Sfl (•• )
400
1.0
200
0.5
0.0 0
300 400 500 600 700
Temperature (K)
b) 1.5 0.3
1.0 0.2
)
Sfl (•• )
Sfl/P (
0.5 0.1
0.0 0.0
0 2 4 6 8 10 12
P (atm)
Fig. 16.20 Acetone LIF signal with 266-nm excitation: a) temperature dependence
and b) pressure dependence.
exit of the annular mixing section. The fluorescence signal is recorded by using
an intensified CCD camera positioned downstream of the combustor and perpen-
dicular to the laser sheet. Subtraction of background noise and a uniform field
correction are applied to each of the acetone PLIF images. In addition, the images
are corrected for pulse-to-pulse fluctuations in laser energy. Figure 16.21b shows
the processed images for four different fuel distributions, in which only the fuel
distribution across the annular mixing section is shown and the equivalence-ratio
values are indicated by the accompanying gray scale. These results are averages
of 30 individual images and therefore represent the average fuel distribution. In
all four cases the overall equivalence ratio is the same, that is, 0.7, as are the
a) (3) Fused Silica Combustor
Center bluff-body
(2)
Swirl vanes ICCD Camera
Laser sheet
Air
(1) Premixed injection (PM)
(1) (2) Center body injection (CB)
(3) Downstream injection (DS)
b)
Equivalence Ratio Scale
PM CB 50% CB/50% DS DS
c) 1.05
0.95 PM
Equivalence Ratio
0.85 DS
0.75
0.65
50CB/50DS
0.55
0.45
0.35 CB
Vinlet = 50 m/s, T = 373K, φavg = 0.70
0.25
0.00 0.04 0.08 0.12 0.16 0.20 0.24 0.28 0.32 0.36
Radial Distance (in)
Fig. 16.21 a) Side view of the optically accessible axial dump combustor and
schematic diagram of PLIF setup; b) processed acetone PLIF images; and
c) equivalence-ratio distribution over the annular mixing section at the inlet of the
combustor.
combustor-inlet velocity (50 m/s), inlet temperature (373 K), and pressure (100
kPa). In the case labeled PM, the fuel and air are premixed well upstream of the
combustor. In this case the fuel distribution is expected to be perfectly uniform.
In the case labeled CB, all the fuel is injected through holes in the centerbody
at a location approximately 25 mm upstream of the dump plane, that is, location
(2). In this case the fuel penetrates to the outer wall of the mixing section, which
results in fuel-rich conditions along the outer wall and fuel-lean conditions along
the centerbody. In the case labeled DS, all the fuel is injected through holes in
the outer wall of the mixing section at a location approximately 25 mm upstream
of the dump plane, that is, location (3). In this case the fuel penetrates to the
centerbody, which results in fuel-rich conditions along the centerbody and fuel-
lean conditions along the outer wall of the mixing section. Note also the evidence
of the effect of the six-vane swirler on the fuel distribution in the “DS” image
shown in Fig. 16.21b. Last, in the case labeled 50%CB/50%DS, half of the fuel
is injected through the holes located in the centerbody, location (2), and half is
injected through the holes in the outer wall of the mixing section, location (3). The
resulting fuel distribution is very uniform. These results are further quantified by
calculating an average radial fuel distribution for each case. This calculation is an
average of the radial fuel distribution over 12 radial profiles spaced 30 deg apart
around the axis of the combustor. These results are presented in Fig. 16.21c.
Figure 16.22 shows the CO∗2 chemiluminescence flame-structure images cor-
responding to the PM, CB, and DS fuel distributions shown in Fig. 16.21. Note
that the velocity, temperature, and overall equivalence ratio were the same for all
three cases and the combustor was stable at these conditions. In the PM case the
flame is anchored on the centerbody and extends outward all the way to the wall of
combustor. For the CB case, a noticeable shift occurs in the most intense region of
the flame toward the outer wall of the combustor, which is consistent with the fact
that the fuel concentration is greatest away from the centerbody. For the DS case
the most intense region of the flame has moved closer to the centerbody, where
the fuel concentration is greatest. The fuel distribution also had an effect on the
stability characteristics. With premixed injection (PM) combustion was stable at
low-inlet velocities but became unstable as the inlet velocity increased, whereas
with centerbody (CB) injection the reverse was true, that is, combustion became
unstable as the velocity was decreased. Because velocity has little effect on the fuel
PM CB DS
Fig. 16.22 The effect of inlet fuel distribution on flame structure for the PM, CB, and
DS fuel distributions shown in Fig. 16.21.
distribution, it can be argued that differences in the susceptibility of the different

flame shapes to changes in velocity explain these results.
In combustors that do not have the optical access required for an acetone PLIF
measurement, it is possible to use a fiber-optic probe for making point mea-
surements of the equivalence ratio and thereby determine the fuel distribution.
A fiber-optic laser fluorescence equivalence ratio probe has been developed and
successfully used for this purpose.76 The overall diameter of the probe is 16 mm ( 58
in.), and the length can be made to accommodate different combustors. The probe
consists of a stainless steel, water-cooled jacket, inside of which are mounted two
fused silica optical fibers, one for transmitting the laser beam and the other for
transmitting the collected fluorescence signal. The measurement volume, which is
approximately 1 mm in diameter and 3 mm in length, is located 25 mm from the
side of the probe and faces upstream. This probe has been used to measure spatial
and temporal fuel–air distributions in several industrial and research gas-turbine
combustor facilities at combustor pressures up to 10 atm and inlet temperatures
up to 673 K, both with and without combustion.
Fuel concentration measurements using laser-induced fluorescence are most
successful in gas-fueled combustors. Such measurements can also be made in
liquid-fueled combustors, but many issues need to be considered. One issue is
the presence of liquid droplets in the measurement volume. In this case, the Mie
scattering from the liquid droplets will typically be orders of magnitude stronger
than the fluorescence intensity. This scattering necessitates the use of a band-
pass filter with very good stray-light rejection, which transmits the fluorescence
while rejecting the Mie scattering. Another issue related to the presence of liq-
uid droplets is the fact that the fluorescence signal strength, which is proportional
to the density of the fluorescence species, will be approximately three orders of
magnitude greater from fuel liquid than from fuel vapor. When making PLIF mea-
surements, this discrepancy actually allows one to differentiate, based on intensity,
between regions that primarily contain fuel liquid and regions that primarily contain
fuel vapor. One can also make simultaneous two-dimensional Mie scattering (see
Sec. VI) and fluorescence measurements, by using a split-image filter or two cam-
eras and using the Mie image to identify the location of liquid drops in the fluo-
rescence image. With either of these approaches, however, measurements made in
regions containing a mixture of vapor and liquid are very difficult to quantify. (Note
that the exciplex fluorescence technique was specifically developed to distinguish
between the liquid and vapor fluorescence, but this technique is significantly limited
because it cannot be used in the presence of oxygen because of quenching effects.77 )
Another concern when making measurements with liquid fuels pertains to mul-
ticomponent fuels, that is, most realistic liquid fuels, in which differences in the
vaporization characteristics and the fluorescence characteristics of the individ-
ual fuel components can make quantitative fuel-concentration measurements ex-
tremely difficult. One way to address this problem is to use a simulated fuel that
is composed of nonfluorescing fuel components, typically alkanes, which have
vaporization characteristics similar to the components in the actual fuel, and flu-
orescence seeds, typically ketones, with matching boiling points for tagging the
individual fuel components.78 An example of such a system for a three-component
fuel covering a range of boiling points from approximately 50◦ C to 150◦ C (at 1
atm) is given in Table 16.1. By using only one of the fluorescence tracers, any of
the individual fuel components can be followed.
Table 16.1 Fuel component simulants and matching fluorescence tracers
Fuel component TBP , ◦ C Fluorescence tracer TBP , ◦ C
2,3-Dimethyl butane 57.95 Acetone 56.15

Iso-octane 99.25 3-Pentanone 102.05
Nonane 150.85 4-Heptanone 144.05
Another application of laser-induced fluorescence is the use of HCO fluores-

cence to measure the rate of heat release. Although this technique has not been used
to measure either local or overall rates of heat release in lean premixed gas-turbine
combustors, it is often mentioned as an alternative to the chemiluminescence tech-
nique and therefore warrants discussion. Two advantages of the HCO fluorescence
technique, as compared with the chemiluminescence technique, are that the HCO
mole fraction has been shown to be an accurate indicator of the flame’s heat release,
even in the presence of unsteady strain and flame curvature, and that the fluores-
cence measurement is spatially resolved.43,44 The HCO fluorescence measurement,
however, also has many disadvantages in comparison with the chemiluminescence
technique. First, it requires both a laser and a detector, whereas the chemilumines-
cence technique requires only a detector. Second, it requires two-sided (90 deg)
optical access, whereas the chemiluminecence technique only requires optical ac-
cess from one direction. Third, the measurement rate is at best 20 Hz because of
the limited pulse rate of the required laser system, whereas chemiluminescence
measurements can be made continuously. Fourth, because the HCO fluorescence
signals are very weak, it is unlikely that two-dimensional fluorescence measure-
ments with adequate signal-to-noise ratios will be possible over typical combustor
dimensions. For these reasons, the chemiluminescence technique, although it only
provides a qualitative measure of the rate of heat release, is a more useful approach
for measuring heat-release rates in lean premixed combustor experiments.
Last, another application of the laser-induced fluorescence technique is the use
of OH planar laser-induced fluorescence to obtain detailed two-dimensional flame-
structure measurements during unstable combustion.79−81 The reaction zones in
lean premixed combustors can be expected to be predominantly in the so-called
wrinkled laminar-flame regime,82−84 with the exception of conditions that are
susceptible to local extinction, for example, regions of high strain and/or curvature,
particularly near the lean limit. In the wrinkled laminar-flame regime, the leading
edge of the reaction zone, that is, the flame front, is characterized by a steep gradient
in OH concentration. However, because OH is relatively long lived, it persists well
into the high-temperature products downstream of the flame front.65 The location
of the steep gradient in OH concentration can be used as an indicator of the location
of the reaction zone or flame front. Because OH PLIF signal strengths are typically
strong, single-shot measurements that provide a detailed space- and time-resolved
map of the two-dimensional flame structure are possible.
An OH PLIF image from the laboratory-scale dump combustor illustrated in
Fig. 16.5 is shown in Fig. 16.23a, where only the upper half of the combustor
is shown.79 This is an image of a single-shot measurement acquired at a partic-
ular phase angle during unstable combustion. It clearly shows that the flame is
anchored on the centerbody. It also shows a well-defined and highly wrinkled
flame front, which indicates that combustion is occurring in the so-called wrinkled
a)
b)
Fig. 16.23 Flame-surface area calculation procedure: a) Normalized OH PLIF image

and b) thresholded flame surface.
laminar-flame regime.82 Last, there is clear evidence of the interaction between the
flame front and the vortex that is shed from the dump-plane shoulder. The effect of
this interaction on the area of the flame, because the flame area is directly related
to the flame’s rate of heat release, is of particular interest.
The first step in calculating the flame area is to determine the location of the
flame front. This determination involves correcting the images for shot-to-shot
laser-energy variations, background noise, and laser-sheet nonuniformity. Because
there is a marked increase in OH concentration at the leading edge of the flame
front, its location can be readily determined by applying a threshold to the corrected
image. In this case, the threshold was not based on the magnitude of the OH
fluorescence intensity but rather on the magnitude of the local gradient of the OH
intensity.79 This magnitude was used to avoid mistakenly identifying as a flame
front the boundary between combustion products and unburned reactants that result
from the mixing of products and reactants in the recirculation zone of the dump
combustor. The flame front determined by this threshold procedure from the OH
PLIF image shown in Fig. 16.23a is shown in Fig. 16.23b. Once the flame front is
determined, the total flame area is calculated by revolving the flame front around
the centerline of the combustor. This calculation assumes that the flame front
determined from the OH PLIF image is representative of the flame front at other
cross sections of the flame. In addition, because the OH PLIF image is a single-shot
image, one must calculate the area for many such images from which an average
flame area can be determined.
To understand the role of flame-area changes during unstable combustion it
is necessary to obtain OH PLIF images at various times during one period of
the instability. The image-acquisition rate of the OH PLIF technique, however,
is considerably less than typical instability frequencies. Therefore, it is necessary
to reconstruct the image sequence by obtaining images at different phase angles
from different cycles. Figure 16.24 shows a sequence of flame fronts determined
from single-shot OH PLIF images over one period of a 378-Hz instability. These
measurements were made in the laboratory combustor referred to previously, op-
erating on natural gas at an equivalence ratio of 0.9, an inlet velocity of 59 m/s,
an inlet temperature of 623 K, and a pressure of 100 kPa. These results clearly
show the evolution of the interaction between the flame front and the vortex and
the resulting changes in the flame length. Five single-shot OH PLIF images were
acquired at each phase angle, the flame area was then calculated for each image
by using the procedure described earlier, and an average flame area at each phase
angle was calculated. The resulting flame area vs phase angle over one period of
the instability is plotted in Fig. 16.25, along with the measured overall heat-release
fluctuation, that is, the overall CO∗2 chemiluminescence-intensity fluctuation. In
this case the area and heat-release fluctuations are very nearly in phase, indicat-
ing that flame-area changes caused by flame-vortex interactions play an important
role in this instability. Results such as these provide valuable insight regarding the
phenomenology of unstable combustion and can be used to provide guidance for
the development of reduced-order models of unstable combustion.46
Detailed two-dimensional OH flame-structure measurements can also be used
to calculate the local flame-surface density, which is a measure of the local reac-
tion rate in turbulent flames. Such measurements have been made in a lean pre-
mixed combustor under stable and unstable operating conditions and have been in
t = 0ms t = 0.22ms t = 0.44ms t = 0.66ms
t = 0.88ms t = 1.10ms t = 1.32ms t = 1.54ms
t = 1.76ms t = 1.98ms t = 2.20ms t = 2.42ms

Fig. 16.24 Sequence of digitized flame surfaces over one period of unstable combus-
tion.
7 Heat release
24
ea Area
Ar 22
Flame area (sq. inches)
6
se
20
Heat release (a.u.)
ea
He
5
el
18
tr
at
ea
re
16
H
le
ea
4
as
14
Ar
e
12
3
10
8
2
0 0.5 1.0 1.5 2.0 2.5
Time (msec)
Fig. 16.25 Variation of flame area and heat release with time during one period of
unstable combustion.
good agreement with OH∗ chemiluminescence measurements of the rate of heat

release.80
VI. Laser Mie Scattering

In combustors operating on liquid fuels, the spray characteristics determine
the fuel distribution in the combustor and therefore have a significant effect on
the combustor’s stability characteristics. Of interest are the spray characteristics
under stable conditions, the response of the spray to acoustic fluctuations during
an instability, and the response of the spray to fuel flow modulation for the purpose
of active combustion control.
One technique that can be used to visualize the liquid-fuel spray is laser Mie
scattering. Mie scattering is an elastic scattering process, that is, the wavelength
of the scattered light is the same as the incident light, which occurs when the
dimension of the object from which the light is scattered is greater than the wave-
length of the incident light. One typically uses laser wavelengths near 0.5µm, for
example, an argon ion laser or a frequency-doubled Nd:YAG laser, when mak-
ing laser Mie-scattering measurements, whereas droplet diameters encountered in
typical gas-turbine sprays can range from 1µm to 100µm. In practice, it is dif-
ficult to detect drops much smaller than 5–10µm. The reason for this difficulty
is that the intensity of Mie scattering scales with drop diameter squared,85 there-
fore, in a spray with a broad distribution of drop diameters it is difficult to make
measurements with sufficient dynamic range to detect both large and small drops.
Laser Mie scattering is often implemented as a two-dimensional technique, in
which the laser beam is formed into a thin sheet that passes through the spray.
The Mie scattering from droplets that lie in the laser sheet is detected with a
digital camera located at an angle of 90 deg to the laser sheet. The advantage
of two-dimensional laser Mie scattering, compared with direct photography, is
that, because the measurement is made over a two-dimensional plane, the interior
structure of the spray is revealed.
When two-dimensional Mie scattering is used to characterize a spray, the field
of view is usually large enough to see the entire spray, in which case there is
typically more than one droplet in the field of view of each pixel on the digital
camera. Therefore, the signal strength detected by each pixel is the sum of the Mie
scattering from all the droplets in the pixel’s field of view. Because the intensity of
Mie scattering is proportional to the surface area of the droplet, the signal detected
by each pixel is a measure of the total surface area of the droplets in the pixel’s
field of view, not the total volume or mass of the droplets.
An example of a two-dimensional Mie-scattering image of a spray produced
by a liquid jet transversely injected into an air crossflow is shown in Fig. 16.26.86
For this measurement, the laser sheet was 3 mm thick and passed through the
centerline of the spray. The injector diameter was 1.27 mm, and the liquid was
water. By applying a threshold to the Mie-scattering intensity, the penetration of
the upper edge of the spray can be defined. Although this definition is somewhat
subjective, if the laser power, camera gain, and threshold level are kept constant,
useful information regarding the effect of operating conditions on penetration
can be obtained.86,87 In addition to the penetration, it is also possible to obtain
information on the dispersion and the rate of vaporization of the spray. For example,
Fig. 16.26 Two-dimensional Mie-scattering image of a liquid jet in a crossflow.
Fig. 16.27 shows two sequences of two-dimensional Mie-scattering images. For

these measurements, the laser sheet was positioned perpendicular to the centerline
of the spray and at various distances downstream of the injection location. In both
cases the liquid is acetone and the liquid-to-air momentum flux ratio is 18. The only
difference is in the temperature of the air, which is 18◦ C in the upper sequence and
250◦ C in the lower sequence. In the upper images, representing a nonvaporizing
case, the penetration of the spray and dispersion of the spray are clearly shown.
In the lower images, representing a vaporizing case, it is clearly shown that the
amount of liquid in the spray is decreasing with downstream distance because of
the effects of vaporization.
.25” .75” 1.25” 1.75”

Distance from injector
Fig. 16.27 Two-dimensional Mie-scattering images of nonvaporizing (upper) and

vaporizing (lower) sprays.
When interpreting Mie-scattering images, keep in mind that the Mie-scattering

intensity is indicative of the surface area, not the volume, of the droplets. For
example, in comparing Mie images for the same injector at two different operating
conditions, a difference in the intensity of the Mie signal at a particular location
indicates that the total surface area of the droplets at that location is less in the image
with the smaller intensity. This difference is not necessarily indicative of less liquid,
but could also indicate poorer atomization, in which the same amount of liquid
is in fewer and larger droplets. Therefore, care must be taken when interpreting
Mie-scattering spray images. Note also that this is not a concern when using
such data for the purposes of model validation, because spray models can readily
calculate the surface area of the droplets from the calculated size distribution and
number density.
Two-dimensional Mie scattering can also be used to characterize the behavior
of unsteady fuel sprays. Of particular interest for combustion dynamics research
is the modulated spray, in which the modulation is caused by the interaction of
the spray with the acoustic flowfield or by fuel flow modulation for the purpose
of active combustion control. An example of such a measurement is presented
in Fig. 16.28, which shows a sequence of phase-averaged Mie-scattering images
acquired at 36-deg phase-angle increments for a liquid jet that is modulated at
100 Hz with a 20% duty cycle. Each image is an average of 30 images acquired
at the same phase angle. Such image sequences, particularly when shown as a
movie, provide considerable understanding and insight regarding the behavior of
the modulated spray, for example, how the spray is swept across the width of the
mixing section during each cycle.
One aspect of a modulated spray that is of particular interest is what is referred
to as the fuel-transfer function. In general, a transfer function defines the functional
relationship between the input to a system and the resulting system output. In a
spray that is modulated by the acoustic flowfield, the system input might be the
pressure in the mixing section at the injector location as a function of time, whereas
t=0 ms t=1.0 ms t=2.0 ms t=3.0 ms
t=4.0 ms t=5.0 ms t=6.0 ms t=7.0 ms
t=8.0 ms t=9.0 ms
Fig. 16.28 Phase-averaged sequence of Mie-scattering image from a modulated jet
in crossflow (period of modulation, 10 ms).
in a spray that is modulated by fuel flow rate modulation, the system input might
be the mass flow rate of fuel at the exit of the injector as a function of time. In either
case, the system output could be the amount of fuel at the inlet to the combustor
as a function of time, that is, the time of arrival of the fuel at the combustor inlet.88
An example of a measurement of the fuel flow rate at the exit of an injector as a
function of time is shown in Fig. 16.29. This is a plot of the volume fraction of liquid
fuel exiting the injector as a function of time during one modulation period.88 In
this case the fuel flow rate was modulated at a frequency of 80 Hz with a duty cycle
of 50% by using an automotive fuel injector located upstream of the 0.27-mm-diam
injector. This measurement was made by using a rotating patternator technique.88
The output of the system in response to this input is the amount of fuel entering the
combustor as a function of time, which is measured by using a time-of-arrival Mie-
scattering technique, which is illustrated schematically in Fig. 16.30. As shown, a
laser sheet is positioned perpendicular to the crossflow, at a position corresponding
to the entrance to the combustor, and the Mie scattering from drops, as they pass
through the laser sheet, is detected by a photomultiplier tube (PMT) positioned 30
deg from the plane of the laser sheet. Time-of-arrival Mie-scattering measurements
corresponding to the system input shown in Fig. 16.29 are shown in Fig. 16.31.
The only difference between the operating conditions for the results shown in Fig.
16.31 is in the crossflow velocity, which is 50 m/s in the lower plot and 90 m/s
in the upper plot. The pronounced difference between these two results can be
explained in terms of improved atomization, that is, smaller droplet size, which
occurs with increased crossflow velocity, that is, an increased Weber number. For
example, smaller droplet size corresponds to a narrower droplet-size distribution,
as a result of which there is less dispersion of the droplets in the 90 m/s case, that
is, the output (Fig. 16.31) retains the shape of the input (Fig. 16.29). A detailed
0.14
0.12
0.10
Vliq 0.08
∑ V liq 0.06
0.04
0.02
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
t / Tinj
Fig. 16.29 The volume fraction of liquid exiting the injector as a function of time
during one modulation period for an 80-Hz modulation with a 50% duty cycle.
TEST SECTION
CROSSFLOW
LASER
PMT
Fig. 16.30 Experimental setup for time-of-arrival Mie-scattering measurement.
discussion of fuel-transfer function measurements, such as those in Figs. 16.29

and 16.31, can be found in Ref. 88.
In addition to visualizing fuel sprays, Mie scattering can also be used to visualize
a gas flow by seeding the gas with small particles, for example, smoke. This
technique was used in a study of active combustion control with multiple vortex
shedding.89 The flame in this case was a diffusion flame created by a central air jet
and surrounding ethylene jets. The air-flow and the fuel flow were independently
modulated, producing both fuel and air vortices. By seeding either the air or fuel
with smoke, it was possible to visualize the fuel and air vortices. A conventional
white-light source was used to illuminate the smoke and the Mie-scattering image
was recorded by using an intensified camera. By gating the intensified camera at
different phase angles during the modulation period, it was possible to monitor the
evolution and interaction of the fuel and air vortices and to determine the phasing
required for optimum fuel–air mixing.
VII. Phase Doppler Particle Analysis

The droplet-size distribution plays an important role in both steady86 and
modulated88 fuel sprays. Several techniques have been used to measure droplet
size in liquid-fuel sprays; however, the technique that is most widely used for
this purpose is phase Doppler particle analysis (PDPA).90 The PDPA technique
measures the size and velocity of individual droplets as they pass through a mea-
surement volume defined by the intersection of two laser beams. The main ad-
vantages of the PDPA technique are that it is a spatially resolved measurement
and that it measures droplet velocity and size simultaneously. The main limita-
tions of the PDPA technique are that it cannot be used in dense sprays, it typically
misses small droplets because of dynamic range limitations, and, because it is a
0.70
Mie Intensity (Volts)
0.60 Vcrossflow= 90 m/s
0.50
0.40
0.30
0.20
0.10
0.00
0.70
Mie Intensity (Volts)
0.60 Vcrossflow= 50 m/s
0.50
0.40
0.30
0.20
0.10
0.00
0.000 0.005 0.010 0.015 0.020 0.025 0.030
Time (s)
Fig. 16.31 Time-of-arrival Mie-scattering measurements for a crossflow velocity of
90 m/s (upper) and 50 m/s (lower).
point measurement, it is very time consuming to map an entire spray. There have
been many applications of PDPA to gas-turbine fuel sprays operating at steady
conditions, wherein the detailed spray characteristics as a function of operating
conditions have been determined.91 Such information is useful in studies of com-
bustion instabilities in that the spray characteristics determine the fuel distribution
in the combustor, which has a significant effect on the stability characteristics of
the combustor. However, no PDPA measurements of droplet size in fuel sprays
that are modulated have been reported, either for the purpose of active combus-
tion control or because of interaction with an acoustic field, for example, during
unstable combustion.
VIII. Conclusion
Chemiluminescence emission, infrared absorption, laser-induced fluorescence,
laser Mie scattering, and PDPA measurement techniques, in particular, when com-
bined with phase-synchronized pressure-fluctuation measurements, can be used
to obtain a detailed characterization of unstable combustion and the underlying
mechanisms of unstable combustion in gas-turbine combustors operating on gas

and liquid fuels.
Chemiluminescence-emission measurements can be used to monitor fluctua-
tions in the flame’s overall heat release and fluctuations in the flame’s structure
during unstable combustion. Simultaneous measurements of overall heat-release
and pressure fluctuations provide information related to the overall system gain
and damping. Phase-synchronized chemiluminescence flame-structure measure-
ments reveal the spatial and temporal evolution of the flame’s heat release and
provide insight regarding the phenomenology of an instability, such as showing
evidence of flame–vortex interaction or periodic extinction and reignition. Simul-
taneous flame-structure measurements and pressure measurements can be used to
calculate the Rayleigh index distribution from which regions of gain and damping
can be identified.
Chemiluminescence emission can also be used to characterize and optimize
the location and timing of the modulated heat release produced by primary or
secondary fuel flow modulation for the purpose of active combustion control.
Infrared absorption can be used to measure the frequency and magnitude of
equivalence ratio fluctuations at the entrance to the combustor during unstable
combustion for gaseous or vaporized liquid fuels. Such measurements, when com-
bined with simultaneous pressure and overall heat-release fluctuation measure-
ments, can be used to quantify the role of feed-system coupling and to assess the
effectiveness of control strategies for suppressing the instability.
Laser-induced fluorescence measurements can be used to characterize fuel–air
mixing and the resultant fuel distribution at the inlet to the combustor, whereas
phase-synchronized OH PLIF measurements can be used to obtain detailed in-
formation about the flame structure and its evolution during an instability. For
example, OH PLIF measurements can be used to calculate the flame area, which,
when combined with simultaneous heat-release and pressure fluctuation measure-
ments, provides quantitative information regarding the role of flame-area changes
during an instability.
Two-dimensional laser Mie scattering can be used to characterize the temporal
and spatial evolution of the liquid fuel in actual or simulated gas-turbine fuel
sprays, where the intensity of the Mie scattering is proportional to the total surface
area of the droplets in the measurement volume. This technique can also be used
to measure the time of arrival of the liquid fuel at a given downstream location
for modulated fuel injection. By combining this information with a measure of the
injected fuel flow rate vs time, the fuel-transfer function can be determined for use
in active combustion-control algorithms.
PDPA can be used to measure the droplet-size distribution in steady and modu-
lated gas-turbine fuel sprays. Such information is important because the droplet-
size distribution has a significant affect on the fuel distribution and, in turn, the
combustor’s stability characteristics.
The detailed information that can be obtained with these measurement tech-
niques is critical to improving our understanding of unstable combustion, to the
formulation and validation of reduced-order models of unstable combustion, and
to the identification and optimization of strategies for suppressing unstable com-
bustion.
Acknowledgments
We thank the many graduate and postdoctoral students who contributed to the
work presented in this chapter, including R. Bandaru, S. Berksoy, J. M. Deepe, E.
Gonzalez, K. Kim, S. Miller, L. Preston, J. Samperio, D. Simons, J. Stenzler, and
K. K. Venkataraman. We are also grateful for the financial support provided by the
Advanced Gas Turbine Systems Research Program of the Department of En-
ergy, the Air Force Office of Scientific Research, the Office of Naval Research,
NASA Glenn Research Center, General Electric, Pratt & Whitney, Siemens-
Westinghouse, Solar Turbines, and United Technologies Research Center.
References
1
Lieuwen, T., and McManus, K. (eds.), “Combustion Dynamics in Lean-premixed Pre-
vaporized (LPP) Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5, 2003,
pp. 721–829.
2
Yang, V., and Anderson, W. E. (eds.), Liquid Rocket Engine Combustion Instability,
Progress in Astronautics and Aeronautics, Vol. 169, AIAA, Washington, DC, 1992.
3
Oran, E. S., and Gardner, J. H., “Chemical-Acoustic Interactions in Combustion
Systems,” Progress in Energy and Combustion Science, Vol. 11, No. 4, 1985, pp. 253–
276.
4
Candel, S. M., and Poinsot, T. J., “Interactions Between Acoustics and Combustion,”
Proceedings of the Institute of Acoustics, Vol. 10, Inst. of Acoustics, Hertfordshire, U.K.,
1988, pp. 103–153.
5
6
7
Poinsot, T., Trouve, A., Veynante, D., Candel, S., and Esposito, E., “Vortex-Driven
1987, pp. 265–292.
8
Schadow, K. C., Gutmark, E., Parr, T. P., Parr, D. M., Wilson, K. J., and Crump,
J. E., “Large-Scale Coherent Structures as Drivers of Combustion Instability,” Combus-
tion Science and Technology, Vol. 64, 1989, pp. 167–186.
9
Straub, D., Richards, G., Yip, M. J., Rogers, W. A., and Robey, E. H., “Importance of
Axial Swirl Vane Location on Combustion Dynamics For Lean Premix Fuel Injectors,”
AIAA Paper 98-3909, July 1998.
10
Richards, G. A., and Janus, M. C., “Characterization of Oscillations During Premixed
Gas Turbine Combustion,” ASME Journal of Engineering for Gas Turbines and Power,
Vol. 120, No. 2, 1998, pp. 294–302.
11
12
tuation and Its Effect in Heat Release During Unstable Combustion,” Proceedings of the
13
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism of Combustion
Instability in Lean Premixed Gas Turbine Combustors,” ASME Transactions, Vol. 123,
No. 1, 2001, pp. 182–189.
14
Lieuwen, T. C., “Experimental Investigation of Limit-Cycle Oscillations in an Unstable
Gas Turbine Combustor,” Journal of Propulsion and Power, Vol.18, No. 1, 2002, pp. 61–67.
15
Lieuwen, T. C., “Statistical Characteristics of Pressure Oscillations in a Premixed Com-
bustor,” Journal of Sound and Vibration, Vol. 260, No. 1, 2003, pp. 3–17.
16
Harrje, D. T., and Reardon, F. H. (ed.), “Liquid Propellant Rocket Combustion Insta-
bility,” NASA SP-194, 1972, pp. 467–468.
17
Lieuwen, T., and Zinn, B. T., “On the Experimental Determination of Combus-
tion Process Driving in an Unstable Combustor,” Combustion Science and Technology,
Vol. 157, 2000, pp. 111–127.
18
Lee, D. H., and Lieuwen, T., “Acoustic Nearfield Characteristics of a Conical, Pre-
mixed Flame,” Journal of the Acoustic Society of America, Vol. 113, No. 1, 2003, pp. 167–
177.
19
Richards, G. A., Straub, D. L., and Robey, E. H., “Passive-Active Control of Combustion
Oscillations,” Spring Meeting of the Central States Section of the Combustion Institute,
Knoxville, TN, April 2002.
20
Lee, J. G., Kim, K., and Santavicca, D. A., “A Study of the Role of Equivalence Ratio
Fluctuation During Unstable Combustion in a Lean Premixed Combustor,” AIAA Paper
2002-4015, July 2002.
21
Gaydon, A. G., Spectroscopy of Flames, Chapman and Hall, London, 1974.
22
Gaydon, A. G., and Wolfhard, H. G., Flames, Their Structure, Radiation and Temper-
ature, Chapman and Hall, London, 1979.
23
Miller, S. A., “Development of a Flame Chemiluminescence Probe for Determination of
Primary Zone Equivalence Ratio in Gas Turbine Combustors,” M. S. Thesis, Pennsylvania
State Univ., University Park, PA, 1999.
24
John, R., and Summerfield, M., “ Effect of Turbulence on Radiation Intensity From
Propane-Air Flames,” Jet Propulsion, Vol. 27, 1957, pp. 169–178.
25
Clark, T., “Studies of OH, CO, CH and C2 Radiation From Laminar and Turbulent
Propane-Air and Ethylene-Air Flames,” NACA Technical Note 4266, 1958.
26
Diederichsen, J., and Gould, R. D., “Combustion Instability: Radiation From Premixed
Flames of Variable Burning Velocity,” Combustion and Flame, Vol. 9, 1965, pp. 25–31.
27
Hurle, I. R., Price, R. B., Sugden, T. M., Thomas, R. R. S., and Thomas, A., “Sound
Emission From Open Turbulent Premixed Flames,” Proceedings of the Royal Society of
London, Series A: mathematical and Physical Sciences, Vol. 303, 1968, pp. 409–427.
28
Price, R., Hurle, I., and Sugden, T., “Optical Studies of the Generation of Noise in
Turbulent Flames,” Proceedings of the Combustion Institute, Vol. 12, The Combustion
29
Hedge, E. G., Reuter, D., Daniel, B. R., and Zinn, B. T., “Flame Driving of Longitudinal
Vol. 55, 1987, pp. 125–138.
30
Langhorne, P. J., “Reheat Buzz: An Acoustically Coupled Combustion Instability. Part
1. Experiment,” Journal of Fluid Mechanics, Vol. 193, 1988, pp. 417–443.
31
Samaniego, J. M., Yip, B., Poinsot, T., and Candel, S., “Low-Frequency Combustion
Instability Mechanisms in a Side-Dump Combustor,” Combustion and Flame, Vol. 94, 1993,
pp. 363–380.
32
Shih, W.-P., Lee, J. G., and Santavicca, D. A., “Stability and Emissions Characteristics
of a Lean Premixed Gas Turbine Combustor,” Proceedings of the Combustion Institute,
33
Bandaru, R. V., Miller, S., Lee, J. G., and Santavicca, D. A., “Sensors for Measur-
ing Primary Zone Equivalence Ratio in Gas Turbine Combustors,” Proceedings of SPIE -
The International Symposium on Industrial and Environmental Monitors and Biosensors,
Vol. 3535, Society of Photo-Optical Instrumentation Engineers, Bellingham, WA, Nov.
1998, pp. 104–114.
34
Broda, J. C., Seo, S., Santoro, Shirhattakar, G., and Yang, V., “An Experimental Study
of Combustion Dynamics of a Premixed Swirl Injector,” Proceedings of the Combustion
35
Venkataraman, K. K., Preston, L. H., Simons, D. W., Lee, B. J., Lee, J. G., and
Santavicca, D. A., “Mechanism of Combustion Instability in a Lean Premixed Dump Com-
bustor,” Journal of Propulsion and Power, Vol. 15, No. 6, 1999, pp. 909–918.
36
Haber, L., Vandsberger, U., Saunders, W., and Khanna, V., “An Examination of the
Relationship Between Chemiluminescence Light Emissions and Heat Release Rate Under
Non-adiabatic Conditions,” Proceedings of the International Gas Turbine Institute, 2000-
GT-0121, 2000.
37
Ikeda, Y., Kojima, J., Nakajima, T., Akamatsu, F., and Katsuki, M., “Measurement of the
Local Flame-Front Structure of Turbulent Premixed Flames by Local Chemiluminescence,”
2000, pp. 343–350.
38
Lieuwen, T., and Neumeier, Y., “Nonlinear Pressure-Heat Release Transfer Func-
tion Measurements in a Premixed Combustor,” Proceedings of the Combustion Institute,
39
Paschereit, C. O., and Gutmark, E. J., “Enhanced Performance of a Gas-Turbine
Combustor Using Miniature Vortex Generators,”Proceedings of the Combustion Institute,
40
Bellows, B. D., Zhang, Q., Neumeier, Y., Lieuwen, T., and Zinn, B. T., “Forced Response
Studies of a Premixed Flame to Flow Disturbances in a Gas Turbine Combustor,” AIAA
Paper 2003-824, Jan. 2003.
41
Samaniego, J. M., Egolfopoulos, F. N., and Bowman, C. T., “CO∗2 Chemilumines-
cence in Premixed Flames,” Combustion Science and Technology, Vol. 109, No. 1, 1995,
pp. 183–203.
42
Dandy, D., and Vosen, S., “Numerical and Experimental Studies of Hydroxyl Rad-
ical Chemiluminescence in Methane-Air Flames,” Combustion Science and Technology,
Vol. 82, No. 1, 1992, pp. 131–150.
43
Najm, H. N., Paul P. H., Mueller, C. J., and Wyckoff, P. S., “On the Adequacy of
Certain Experimental Observables as Measurements of Flame Burning Rate,” Combustion
and Flame, Vol. 113, No. 3, 1998, pp. 312–332.
44
Najm, H. M., Knio, O. M., Paul, P. H., and Wyckoff, P. S., “A Study of Flame Observ-
ables in Premixed Methane-Air Flames,” Combustion Science and Technology, Vol. 140,
No. 3, 1998, pp. 369–403.
45
Dowling, A. P., “Nonlinear Self-Excited Oscillations of a Ducted Flame,” Journal of
46
Peracchio, A. A., and Proscia, W. M., “Nonlinear Heat-Release/Acoustic Model for
Thermoacoustic Instability in Lean Premixed Combustors,” American Society of Mechan-
ical Engineers, Paper 98-GT-269, 1998.
47
Dasch, C. J., “One-Dimensional Tomography: a Comparison of Abel, Onion-Peeling,
and Filtered Backprojection Methods,” Applied Optics, Vol. 31, No. 8, 1992, pp. 1146–1152.
48
Kim, K., Lee, J. G., and Santavicca, D. A., “Optimization of the Spatial and Tem-
poral Fuel Distribution for Active Control of Combustion Dynamics in Lean Premixed
Combustors,” AIAA Paper 2002-4024, July 2002.
49
Mallard, W. G., and Gardiner, W. C. Jr., “Absorption of the 3.39 µm He-Ne Laser
Line by Methane from 300 to 2400K,” Journal of Quantitative Spectroscopy and Radiative
Transfer, Vol. 20, No. 2, 1978, pp. 135–149.
50
Philippe, L. C., and Hanson, R. K., “Laser Diode Wavelength-Modulation Spectroscopy
for Simultaneous Measurement of Temperature, Pressure and Velocity in Shock-Heated
Oxygen Flows,” Applied Optics, Vol. 32, No. 30, 1993, pp. 6090–6103.
51
Yoshiyama, S., Hamamoto, Y., Tomita, E., and Minami, K., “Measurement of Hy-
drocarbon Fuel Concentration by Means of Infrared Absorption Technique with 3.39 µm
He-Ne Laser,” JSAE Review, Vol. 17, No. 4, 1996, pp. 339–345.
52
Mongia, R. K., Tomita, E., Hsu, F. K., Talbot, L., and Dibble, R. W., “Use of an Optical
Probe for Time-Resolved In situ Measurement of Local Air-to-Fuel Ratio and Extent of Fuel
Mixing with Applications to Low NOx Emissions in Premixed Gas Turbines,” Proceedings
2755.
53
Koenig, M., and Hall, M. J., “Measurements of Local In-Cylinder Fuel Concentration
Fluctuations in a Firing SI Engine,” Society of Automotive Engineers, Paper 971644, 1997.
54
Wehe, S. D., Baer, D. S., and Hanson, R. K., “Tunable Diode-Laser Absorption Mea-
surements of Temperature, Velocity and H2 O in Hypersonic Flows,” AIAA Paper 97-3267,
July 1997.
55
Mihalcea, R. M., Baer, D. S., and Hanson, R. K., “Diode-Laser Sensor for Measurements
of CO, CO2 and CH4 in Combustion Flows,” Applied Optics, Vol. 36, No. 33, 1997,
pp. 8745–8752.
56
Seitzman, J. M., Tamma, R., and Vijayan, R., “Infrared Absorption Based Sensor
Approaches for High Pressure Combustion,” AIAA Paper 97–0318, Jan. 1997.
57
Mongia, R. K., Dibble, R. W., and Lovett, J., “Measurement of Air-Fuel Ratio Fluctu-
ations Caused by Combustor Driven Oscillations,” American Society of Mechanical Engi-
neers, Paper 98-GT-304, 1998.
58
Mihalcea, R. M., Baer, D. S., and Hanson, R. K., “Advanced Diode Laser Absorption
Sensor for In-Situ Combustion Measurements of CO2 , H2 O and Gas Temperature,” Pro-
ceedings of the Combustion Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998,
pp. 95–101.
59
Mongia, R., Torres, J., Dibble, R., Lee, D., Anderson, T., and Sowa, W., “Fast Re-
sponse Extraction Probe for Measurement of Air-Fuel Ratio Fluctuations in Lean Premixed
Combustors,” American Society of Mechanical Engineers, Paper 99-GT-277, 1999.
60
Ebert, V., Fernholz, T., Giesemann, C., Pitz, H., Teichert, H., and Wolfrum, J.,
“Simultaneous Diode-Laser-Based In-Situ-Detection of Multiple Species and Tempera-
ture in a Gas-Fired Power-Plant,” Proceedings of the Combustion Institute, Vol. 28, The
61
Webber, M. E., Wang, J., Sanders, S. T., Baer, D. S., and Hanson, R. K., “In-situ
Combustion Measurements of CO, CO2, H2O and Temperature Using Diode Laser Ab-
sorption Sensors,” Proceedings of the Combustion Institute, Vol. 28, The Combustion Inst.,
Pittsburgh, PA, 2000, pp. 407–413.
62
Drallmeier, J., “Hydrocarbon-Vapor Measurements in Pulsed Fuel Sprays,” Applied
Optics, Vol. 33, No. 33, Nov. 1994, pp. 7781–7788.
63
Edwards, J. L., Gouldin, F. C., and MacDonald, M. A., “High Speed Absorption
Tomography with Advanced Reconstruction Algorithms,” AIAA Paper 2003-1013, Jan.
2003.
64
Eckbreth, A. C., Laser Diagnostics for Combustion Temperature and Species, 2nd ed.
Gordon and Breach, New York, 1996.
65
Nguyen, Q.-V., and Paul, P. H., “The Time Evolution of a Vortex-Flame Interaction
Observed via Planar Imaging of CH and OH,” Proceedings of the Combustion Institute,
66
Seitzman, J. M., Kychakoff, G., and Hanson, R. K., “Instantaneous Temperature Field
Measurements Using Planar Laser-Induced Fluorescence,” Optics Letters, Vol. 10, No. 9,
1985, pp. 439–441.
67
Hanson, R. K., “Combustion Diagnostics: Planar Flowfield Imaging,” Proceedings of
the Combustion Institute, Vol. 21, The Combustion Inst., Pittsburgh, PA, 1986, pp. 1677–
1691.
68
Hanson, R. K., “Planar Laser-Induced Fluorescence Imaging,” Journal of Quantitative
Spectroscopy and Radiative Transfer, Vol. 40, No. 3, 1988, pp. 343–362.
69
Pringsheim, P., Fluorescence and Phosphorescence, Interscience Publishers, New York,
1949.
70
Berlman, I. B., Handbook of Fluorescence Spectra of Aromatic Molecules, Academic
Press, New York, 1971.
71
Lozano, A., Yip., B., and Hanson, R. K., “Acetone: A Tracer for Concentration Mea-
surements in Gaseous Flows by Planar Laser-Induced Fluorescence,” Experiments in Fluids,
Vol. 13, 1992, pp. 369–376.
72
Thurber, M. C., and Hanson, R. K., “Simultaneous Imaging of Temperature and Mole
Fraction Using Acetone Planar Laser Induced Fluorescence,” Experiments in Fluids, Vol.
30, 2001, pp. 93–101.
73
Koch, J. D., and Hanson, R. K., “Ketone Photophysics for Quantitative PLIF Imaging,”
AIAA Paper 2000-0413, Jan. 2001.
74
Thurber, M. C., and Hanson, R. K., “Pressure and Composition Dependence of Acetone
Laser-Induced Fluorescence with Excitation at 248, 266 and 308 nm,” Applied Physics B:
Lasers and Optics, Vol. 69, 1999, pp. 229–240.
75
Samperio, J. L, Lee, J. G., and Santavicca, D. A., “Characterization of the Effect of
Inlet Operating Conditions on the Performance of Lean Premixed Gas Turbine Combustors,”
AIAA Paper 2003-0825, Jan. 2003.
76
Lee, J. G., and Santavicca, D. A. “Fiber-Optic Probe for Laser-Induced Fluorescence
Measurements of the Fuel-Air Distribution in Gas-Turbine Combustors,” Journal of Propul-
sion and Power, Vol. 13, No.3, 1997, pp. 384–387.
77
Melton, L. A., “Exciplex-Based Vapor/Liquid Visualization Systems Appropriate for
Automotive Gasolines,” Applied Spectroscopy, Vol. 47, No. 6, 1993, pp. 782–786.
78
Tong, K., Quay, B. D., Zello, J. V., and Santavicca, D. A., “Fuel Volatility Effects
on Mixture Preparatin and Performance in a GDI Engine During Cold Start,” Society of
Automotive Engineers, Paper 2001-01-3650, Sept. 2001.
79
Venkataraman, K. K., “An Investigation of the Instability Mechanism in Lean Premixed
Dump Combustors, Ph.D. Thesis, The Pennsylvania State University, University Park, PA,
2000.
80
Lee, S.-Y., Seo, S., Broda, J. C., Pal, S., and Santoro, R. J., “An Experimental Estimation
of Mean Reaction Rate and Flame Structure During Combustion Instability in a Lean
Premixed Gas Turbine Combustor,” Proceedings of the Combustion Institute, Vol. 28, The
81
Santhanam, V., Knopf, F. C., Acharya, S., and Gutmark, E., “Fluorescence and Tem-
perature Measurements in an Actively Forced Swirl-Stabilized Spray Combustor,” Journal
82
Peters, N., “Laminar Flamelet Concepts in Turbulent Combustion,” Proceedings of the
83
Buschmann, A., Dinkelacker, F., Schäfer, T., and Wolfrum, J., “Measurement of the
Instantaneous Detailed Flame Structure in Turbulent Premixed Combustion,” Proceedings
445.
84
Dunkelacker, F., Soika, A., Most, D., Hofman, D., Leipertz, A., Polifke, W., and
Döbbeling, K., “Structure of Locally Quenched Highly Turbulent Lean Premixed Flames,”
1998, pp. 857–865.
85
Grehan, G., Gouesbet, G., and Rabasse, C., “Monotonic Relationships Between Scat-
tered Powers and Diameters in Lorenz-Mie Theory for Simultaneous Velocimetry and Sizing
of Particles,” Applied Optics, Vol. 20, No. 5, March 1981, pp. 796–799.
86
Stenzler, J. N, Lee, J. G., and Santavicca, D. A., “Penetration and Dispersion of Liquid
Jets in a Heated Crossflow,” AIAA Paper 2003-1327, 2003.
87
Lin, K. C., Kennedy, P. J., and Jackson, T. A., “Penetration Heights of Liquid Jets in
High-Speed Crossflows,” AIAA Paper 2002-0873, Jan. 2002.
88
Stenzler, J. N, Lee, J. G., Deepe, J. M., Santavicca, D. A., and Lee, W., “Fuel Transfer
Function Measurements in Modulated Liquid Jets,” ASME International Congress, IMECE
2004-60673, 2004.
89
Yu, K. H., Wilson, K. J., Parr, T. P., and Schadow, K. C., “Active Combustion Control
Using Multiple Vortex Shedding,” AIAA Paper 96–2760, July 1996.
90
Bachalo, W. D., and Houser, M. J., “Development of the Phase/Doppler Spray Analyzer
for Liquid Drop Size and Velocity Characterizations,” AIAA Paper 84–1199, June 1984.
91
Wang, H., McDonnell, V. G., Sowa, W. A., and Samuelsen, S., “Experimental Study of
a Model Gas Turbine Combustor Swirl Cup. I - Two-Phase Characterization. II - Droplet
Dynamics,” Journal of Propulsion and Power, Vol. 10, No. 4, 1994, pp. 441–452.
V. Combustion Instability Control
Chapter 17
Passive Control of Combustion Instabilities in

Stationary Gas Turbines
Geo A. Richards∗ and Douglas L. Straub†

U.S. Department of Energy, Morgantown, West Virginia
and
Edward H. Robey‡
Parsons Project Services, Morgantown, West Virginia
Nomenclature
G = transfer function relating relative heat release to relative
acoustic pressure (-)
H = transfer function relating relative acoustic pressure to relative
heat release (-)
L = length, m
M = Mach number, µ/c (-)
P = time-average pressure, Pa
Q = time-average heat-release rate, W
R = acoustic transfer matrix for a cylindrical element
S = acoustic transfer matrix for a step expansion
T = gas temperature (K), or the acoustic transfer matrix for a damper
Z = acoustic impedance, p/ν, (m s)−1
c = speed of sound (m/s)
f = frequency, Hz
k = stagnation pressure loss coefficient (-)
p = complex acoustic pressure, Pa
q = complex amplitude of heat-release variation, W
s = cross-sectional area, m2
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
∗ Energy Systems Dynamics Focus Area Leader, National Energy Technology Laboratory.
† Mechanical Engineer.
‡ Scientist.
533
534 G. A. RICHARDS ET AL.
γ = ratio of specific heats

ζ = the ratio of the speed of sound to area, c/s (m s)−1
µ = bulk gas velocity, m/s
ν = acoustic mass velocity, kg/s
ρ = gas density, kg/m3
ψ = the transfer function relating acoustic pressure to velocity source, Pa∗ s/kg
τ = bulk time lag, s
ω = circular frequency, 2π f , rad/s
I. Introduction
C OMBUSTION dynamics has become a significant operational concern for

low-emission engines now in service. Although engine developers and oper-
ators have learned how to achieve stable combustion with very low emissions, this
performance is often restricted to a tight operating window. The restrictions on the
operating range lead to other issues, such as placing a cap on the peak power that
can be produced,1 more stringent requirements on fuel composition,2 and routine
retuning of the fuel splits.3 These complications have been the motive to develop
successful active-control systems described elsewhere in this book. Active control
offers the potential to readjust the combustion dynamics to accommodate prob-
lems like changing ambient conditions, fuel composition, or engine wear. Although
active-control concepts have significant potential and may become a preferred sta-
bilization strategy in the future, at the present time, most engine developers are
using passive methods to stabilize combustion.
This chapter summarizes common passive methods used to improve the stability
of low-emission combustors in stationary-power gas turbines. Most of the content
of this chapter has been presented in a previous article,4 but some notable revisions
have been made. For example, the discussion of flame-transfer functions has been
condensed significantly. Chapter 4 of this book describes this topic in more detail.
Another notable addition to this chapter is the detailed examples of acoustic damper
designs and how these dampers can be used to stabilize combustion.
Simple control-model concepts that have become common in the literature are
introduced in Sec. II. In Sec. III, the application of time-lag modifications for
solving dynamics problems is discussed. Section III also describes the similarities
between the control concepts introduced in Sec. II and time-lag models. Techniques
to enhance stability, such as using multiple time lags or adding a pilot flame are
discussed in Sec. III, as well. In Sec. IV, a review of recent applications of acoustic
dampers to stabilize combustors is presented, and example calculations of acoustic
damper design are discussed.
II. Control-System Models

Combustion dynamics are the result of an interaction between acoustic pressure
( p ) and heat-release perturbations (q ). This interaction can be described as a
closed-loop feedback, shown schematically in Fig. 17.1. Acoustic pressure p
interacts with the flame and can produce a variation in the heat-release rate q .
The heat-release perturbation can generate acoustic waves as described by Chu.5
In a physically closed volume, such as a combustor, the boundary conditions
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 535
Combustion
+ Σ
Processes
G
-
q'
p'
Acoustic
Processes
H
Fig. 17.1 Block diagram of a dynamic thermoacoustic system.
will establish standing waves that can produce a periodic disturbance in the heat-
release rate, q . The system will be unstable if the timing (phase) and the amplitude
(gain) of these variations in pressure and heat-release rate produce constructive
feedback. This feedback process is analogous to conventional feedback-control
systems, where the processes shown in Fig. 17.1 would correspond to control-
system components. The G and H nomenclature and the summation circle shown
on the left side of Fig. 17.1 follow directly from the control-system literature and
will be discussed in more detail later.
Because of the feedback analogy, control-system models have become popu-
lar tools to both represent and diagnose combustion instabilities. Various levels
of detail can be included in control-system models, ranging from reduced-order
models,6 to computational fluid mechanics,7 to complete engine models, by us-
ing a combination of approaches.8 Practical application of these models has been
demonstrated by many authors.1,6,8,9,10 Even though a full model is not sought
to solve a particular problem, it is helpful to understand the concepts, because
many experimental efforts to develop passive control can be explained by control-
system ideas. Thus, in what follows, a simple representation of a control model
for combustion dynamics is presented as a framework for subsequent discussions.
Sections II.A and II.B provide background information, so that no prior training
in control theory is necessary. In Sec. II.C an example calculation is discussed
to demonstrate the use of feedback models. Section II.D reviews the physical
processes that contribute to the flame response, and Sec. II.E reviews various
computational approaches used to predict combustion-system stability.
A. Operational Block Diagrams: General Overview

A schematic of a typical fuel–air premixer and combustor is shown in Fig. 17.2.
Acoustic waves are generated at the flame by variations in heat-release rate, q . The
creation of sound by unsteady heat release is a complex process,5 but, in simple
terms, q perturbations create expansion and contraction of the gas, generating
pressure waves. These pressure waves are reflected and continue to interact with
the flame, such that standing waves are established in the combustor.
In Fig. 17.1, the feedback element (H) represents the conversion of heat-release
variations into a pressure disturbance. If the heat release could be manipulated
at a periodic rate, the output signal from block H would represent the pressure
Fuel Acoustic waves

Generated Reflected
τ
Air Flame
Fuel/air ratio disturbance,

with convection time, τ
Fig. 17.2 Schematic illustration of an unstable premixed combustion process.
produced in the flame region, as well as the pressure produced by the acoustic
characteristics of the system.
In Fig. 17.1, the system element G represents the conversion of a pressure vari-
ation to a variation in the heat-release rate. Many mechanisms can contribute to a
variable heat-release rate. These mechanisms may include periodic changes in the
flame-surface area,11 changes in equivalence ratio,1,12−17 vortex shedding,18−23
changes in the bulk flow,24 and changes in flame anchoring.25−27 Which of these
mechanisms contributes to oscillations in a given problem is an important prac-
tical question and is discussed later. However, attention is often focused on the
equivalence-ratio variation, because it will usually accompany all the other mech-
anisms. The pressure drop across the premixer air passage is typically a few percent
of the operating pressure. Therefore, modest perturbations in the combustor pres-
sure will create significant variations in premixer airflow, and subsequent variations
in fuel–air ratio in the premixer. These variations in fuel–air mixture are transported
to the flame after a convection time lag τ , creating a heat-release perturbation that
may add to perturbations produced by other mechanisms, such as a variable flame
area.
B. Operational Transfer Functions: General Overview

To perform a stability analysis on the system shown in Fig. 17.1, a model of
the physical processes must be known in sufficient detail. An operational transfer
function is simply a mathematical model that relates the output from an operational
block (i.e., G or H in Fig. 17.1) to the input. The next few paragraphs review the
basic ideas connected with transfer functions so that the subsequent discussion can
be understood without prior knowledge of control theory.
Figure 17.3a shows a schematic illustration of a transfer function. An input
signal A cos(ωt) enters at the left, and a resulting signal B cos(ωt + φ) exits at
the right. Considering a range of frequencies ω, the ratio B/A is the gain of the
transfer function, and φ is the phase angle. It is algebraically simpler to consider the
complex counterpart to these real quantities (Fig. 17.3). In this case, the input and
output are A e jωt and B e j(ωt+φ , respectively. By using this notation, the constant
B can be redefined as a complex quantity including the phase angle (B = Be jφ ).
In this manner, the time dependence is not needed in a feedback-loop analysis
because all the blocks have the same time dependence e jωt . Thus, the transfer
function is the complex ratio of the output to the input, or B /A.
A cos(ωt) B cos(ωt+φ)
Transfer
jω t function jω t+φ jω t
Ae Be = B' e
a) Transfer function nomenclature
connect
B'/A D'/C
equivalent to
B' D'
AC
b) Series connection of transfer functions
B'/A
+
Σ B'/A +D'/C
+
D'/C
equivalent to
c) Parallel connection of transfer functions

Fig. 17.3 Illustration of transfer-function nomenclature and block diagram
manipulations.
Sequential processes, in which the output of one transfer function supplies the
input to a second process, are analyzed by multiplying the transfer functions in or-
der. As shown in Fig. 17.3b, if the B /A process described before connects to a D /C
process, the net transfer function is (B /A)∗ (D /C). Signals can also be added alge-
braically because attention is restricted to a linear system. For example, a given in-
put may supply both the A-B transfer function and the C-D transfer function, with
the outputs combined (see Fig. 17.3c). The combined system response is the com-
plex sum (B /A) + (D /C). By adding or multiplying individual transfer functions,
it is possible to reduce more complex physical processes into a forward transfer
function (usually denoted G) and a feedback transfer function (usually denoted H).
C. Example Problem: Linear Stability Analysis

Once the various processes have been described and the problem is reduced
to the form shown in Fig. 17.1, the response of the system to disturbances can
be considered. As shown in Fig. 17.1, a disturbance is added to the signal at the
Closed acoustic boundaries
Flame Position
10 cm
Fig. 17.4 Schematic illustration of combustor geometries used in the example

problems.
summing point. In this paper, the feedback that emerges from H is subtracted from
the disturbance. This operation is defined as negative feedback in control theory.
With this nomenclature, disturbances that originate at the summing point will pass
through G and H, with modifications to both amplitude and phase. Note that a phase
of 180 deg(π ) corresponds to multiplication by −1 [i.e., cos(π )]. Intuitively, if the
disturbance is returned from H with a larger absolute magnitude and a negative
sign, the original disturbance will have a larger amplitude after passing through
the summing point. Note that as the signal from the feedback block passes through
the negative branch of the summing point, the signal is multiplied by −1. Under
this idealized condition, the disturbance will grow in amplitude each time it passes
around the loop, and the system will become unstable.
This intuitive understanding can be matched by formal analysis that leads to
a criterion for stability. The output of a signal passed through G and H but not
returned through the summation point is known as the open-loop frequency re-
sponse. The open-loop frequency response can be used to evaluate stability from
both Bode and Nyquist plots described subsequently.
Two combustor examples shown in Fig. 17.4 will be analyzed. The fuel–air
premixer at the left supplies a step expansion into the region where the flame is
stabilized. The flame is treated as a thin disk located just downstream of the step
expansion. The remainder of the combustor is a long tube, including a second step
expansion, and is then terminated at a closed acoustic boundary. These examples
approximate the conditions typically encountered in combustion test rigs, in which
the downstream boundary may represent a backpressure-control valve. For the cal-
culated results presented next, parameters such as pressure, temperature, flow rate,
and fuel–air ratio are all selected to be representative of gas-turbine combustors.
Treating the flame as a discontinuity that interacts with acoustic waves, a one-
dimensional acoustic analysis is used to determine the acoustic pressure produced
by imposed heat-release perturbations. For the geometry shown, the flame–acoustic
relations presented by Chu5 are used, and the acoustics are modeled using a transfer
matrix method.28 The transfer matrix method can account for mean flow effects
and acoustic losses at abrupt area changes. An outline of these calculations is
presented in the Appendix.
1. Bode Stability Analysis

Let P and Q represent the steady-state or average values of pressure and heat-
release rate in the combustor. With the methods shown in the Appendix, the
a) p′ / P
=H
q′ / Q
0.3
Phase (deg)
200
Magnitude 100
0.2
0
0.1 -100
0.0 -200
0 100 200 300 400 0 100 200 300 400
b) q ′ / Q
= G = 6e−jωτ
p′ / P
200
Phase (deg)
Magnitude
6.0
100
4.0 0
2.0 -100
0.0 -200
0 100 200 300 400 0 100 200 300 400
c) GH
2.0 200
Phase (deg)
Magnitude
1.5 100
1.0 0
0.5 -100
0.0 -200
0 100 200 300 400 0 100 200 300 400
Fig. 17.5 Frequency response (Bode plots) of transfer functions used in the first
example problem with a single acoustic mode (Fig. 17.4, top).
normalized pressure response ( p /P) to a normalized heat-release perturbation

(q /Q) is shown in Fig. 17.5a. These results correspond to the combustor geome-
try in Fig. 17.4. The plot shows that the acoustics of the system produce a strong
response to heat-release perturbations at ∼240 Hz. This maximum amplitude cor-
responds to the natural frequency of the system. Figure 17.5a also indicates that
the system has a pressure node near the flame at ∼140 Hz. At this frequency, there
is no pressure response at the flame to perturbations in the heat-release rate. Near
140 Hz, the phase abruptly changes from −90 to +90 deg, which is expected be-
havior for acoustic nodes. As the frequency approaches 240 Hz, the phase begins
to decrease as the amplitude rises. The phase angle exhibits another transition from
+90 to −90 deg, which is typical of a resonant frequency. The magnitude of the
amplitude peak and the width of the phase transition (phase-roll) are both related
to the acoustic losses or damping in the system. In this example, acoustic losses
arise from the step expansion, mean flow, and mean heat release. For this example,
Fig. 17.5a represents the H transfer function depicted in Fig. 17.1.
The combustion response to acoustic pressure perturbation must also be ana-
lyzed. In actual applications, this response must account for the various mech-
anisms that will be described later (i.e., variable fuel–air ratio, variable flame
area). For the purposes of this example, a simple flame-transfer function will be
considered. Again referring to the normalized perturbations ( p /P and q /Q), the
flame will be treated as having a constant gain of magnitude 6.0, but with a time
delay τ = 2 ms relative to the acoustic pressure at the flame. This transfer function
is 6.0e− j2πτ f . Thus, a normalized pressure perturbation produces a normalized
heat-release rate perturbation six times larger and 2 ms later, which is easily re-
alized in practical systems. This transfer function for the combustion response is
shown in Fig. 17.5b. The phase plot in Fig. 17.5b is representative of all time-delay
systems. The phase angle decreases in a linear fashion with frequency because the
phase angle θ = −2π τ f . In this plot, the phase angle is wrapped into the range
−180 to +180 deg. The same information can be plotted from 0 to 360 deg as
well, avoiding the abrupt discontinuity at ±180 deg.
The open-loop response of this example is the series connection of both the G
and H transfer functions. As explained earlier (Fig. 17.3), this series connection
is computed as the product of the individual gain functions and the sum of the
individual phase angles. The resulting frequency response is shown in Fig. 17.5c.
Note that the magnitude is greater than unity at ∼240 Hz and the phase angle is
±180 deg. If this GH output is connected to the summation point in Fig. 17.1
(i.e., closed loop), the system would be unstable for the reason explained earlier,
that is, the disturbance would grow in amplitude each time around the loop. If
the open-loop gain is less than unity at a phase angle of ±180 deg, the system
would be stable because the signal returning from the summing junction would be
smaller than the original disturbance each time around the loop. This reasoning
is not entirely complete because of other complications such as whether the gain
plot crosses magnitude 1.0 more than once. This complication can be addressed by
using a Nyquist analysis, which is discussed next. The presentation that follows is
an adaptation of analysis discussed by Fannin et al.29 Although a brief description
of the Nyquist stability criterion is presented in the following paragraphs, it is
not intended to be a complete tutorial. A more complete description of Nyquist
analysis is found in control textbooks.30
2. Nyquist Stability Analysis

The Nyquist analysis requires plotting the same information as the Bode plots,
but the information is plotted in polar form. In polar form, the radius is equal to
the magnitude, and the angle with respect to the positive x axis is equal to the
phase angle. For example, magnitude 1.0 at 0 deg of phase is point (1, 0 j) on
the positive, real x axis. Magnitude 1.0 at 90 deg of phase is point (0, 1 j) on the
positive, imaginary axis.
Figure 17.6 shows Nyquist plots for the example problem presented in Fig. 17.5
at three different values of the time lag. Figure 17.6b shows a Nyquist diagram
a) Nyquist Plot
2
1 258 Hz
Imaginary
τ = 1.6 ms
0
-1
-2
-2 -1 0 1 2
Real
b) Nyquist Plot
2
251 Hz 263 Hz
1
Imaginary
τ = 2 ms
0
-1
241 Hz
-2
-2 -1 0 1 2
Real
c) Nyquist Plot
2
1
Imaginary
-1 241 Hz τ = 2.6 ms
-2
-2 -1 0 1 2
Real
Fig. 17.6 Nyquist plots for the single acoustic mode example (Fig. 17.4, top) at three
different time lags (τ = 1.6, 2.0, and 2.6 ms).
corresponding to Fig. 17.5c, where the time lag is 2 ms. The circular lobe corre-
sponds to frequencies between 200 and 300 Hz, in which appreciable magnitude
exists from the open-loop response. For clarity, three of the frequencies are indi-
cated on the lobe. The corresponding points (phase, magnitude, and frequency)
can be found from a close inspection of the Bode plot (Fig. 17.5c). Note that the
lobe represents a small range of frequencies in this example problem.
As explained in control-theory textbooks, the system stability can be evaluated
by counting how many times the Nyquist plot encircles the point −1 on the x axis.
The definition of what constitutes encirclement is fairly involved, and one must
refer to control textbooks for complete details.30 In brief, encirclement direction
(clockwise or counterclockwise) must be counted as positive or negative encir-
clement, and the sum of all the positive and negative encirclements are added to
arrive at a net number of encirclements. The plot also requires considering infor-
mation at negative frequencies, essentially a reflection of the Bode plot into the
negative-frequency axis, and the open-loop system must itself be stable. In this ex-
ample problem, these details do not enter the discussion but should be considered
before using Nyquist analysis on more complex problems. The complete Nyquist
analysis predicts that the system will be unstable if the net number of encirclements
is greater than zero. Figure 17.6b shows that the Nyquist plot does indeed encircle
−1 on the real axis and would therefore be unstable.
The benefit of the Nyquist analysis becomes very apparent when assessing how
different time lags affect system stability. Figure 17.6 shows the open-loop system,
GH, at three different values of the time lag: τ = 1.6, 2.0, and 2.6 ms. Notice that
increasing the time lag rotates the lobe clockwise. This behavior can be understood
by noting that over the small-frequency range in which the amplitude is significant
(230–270 Hz), changes to the phase angle θ = −2πτ f are dominated by changes
in τ . Changes in τ appear to rotate each point on the lobe approximately the same
angle, producing a rotation of the lobe.
The Nyquist plots can also be used to investigate stability boundaries for the
system. For example, both the short and the long time lags shown in Fig. 21.6
almost encircle the −1 point on the real axis. The values of τ that almost produce
encirclement of the −1 point are stability boundaries for the system. Each of these
stability boundaries, or values of τ , has a corresponding frequency at which the lobe
crosses the negative real axis. In this example, the system would be unstable for
frequencies of 258 and 241 Hz at τ = 1.6 and 2.6 ms, respectively. The frequency
range and the size of the lobe depend on the rolloff in the phase-angle function.
For problems that have larger acoustic losses, the phase rolloff near the resonant
frequency may cover a larger frequency range, and the lobe in the Nyquist plot
would also cover a wider range of frequencies.
If the system in this example were actually operated in the closed-loop mode, the
limit-cycle frequencies could be estimated from the so-called describing-function
theory.30 Under the assumption of real-valued describing functions, the limit-cycle
frequency would correspond to the frequency at which the Nyquist plot crosses
the negative real axis. Considering the sequence of time lags in Fig. 17.6, the
implication is that as the time lag increases from 1.6 to 2.6 ms, the frequency
would change from 258 to 241 Hz. This frequency shift as a function of time lag
is a general feature of the Nyquist analysis, and the range of frequencies depends
on the specific case, as noted earlier. An experimental example of the frequency

shift will be shown later.
3. Multiple Resonant Frequencies

In Figs. 17.5 and 17.6, the combustor acoustic response has been characterized
by a single acoustic mode. The second example geometry shown in Fig. 17.4 is
slightly longer and has an additional step expansion. Calculations for Fig. 17.4 are
now conducted in exactly the same manner as in the previous example. However,
the combustion gain is reduced from 6.0 to 1.0 for convenience. Therefore, the
G portion of the transfer function is simply e− j2π τ f . The resulting Bode plot
with τ = 1.5 ms is shown in Fig. 17.7. This second example has strong acoustic
responses near 185 and 410 Hz. Note that the magnitude plot is greater than unity
at both of these resonant frequencies, and so it is more difficult to visualize the
stability limits based on the Bode plot alone. This example is a case where the
Nyquist analysis is much easier to use.
Figure 17.8 shows the Nyquist analysis for three different values of the time lag.
As before, increasing the time lag has the effect of rotating the lobes clockwise.
Starting at τ = 0.9 ms, the system is stable but near a high-frequency stability
boundary (411 Hz); the −1 point on the real axis is almost encircled. Increasing
5.0
4.0
Magnitude
3.0
2.0
1.0
0.0
0 100 200 300 400 500
Frequency (Hz)
200
Phase (degrees)
150
100
50
0
-50
-100
-150
-200
0 100 200 300 400 500
Frequency (Hz)
Fig. 17.7 The open-loop frequency response (Bode plot) for the example with two
resonant acoustic modes (Fig. 17.4, bottom), τ = 1.5 ms.
a) Nyquist Plot
3
2 τ = 0.9 ms
411 Hz
Imaginary 1
0
-1
-2
-3
-3 -2 -1 0 1 2 3
Real
b) Nyquist Plot
3
2 τ = 1.5 ms
408 Hz
1
Imaginary
0
-1 186 Hz
-2
-3
-3 -2 -1 0 1 2 3
Real
c) Nyquist Plot
3
2 τ = 2.0 ms
184
1
Imaginary
0
-1
-2
-3
-3 -2 -1 0 1 2 3
Real
Fig. 17.8 Nyquist plots for the example with two resonant acoustic modes (Fig. 17.4,
bottom) at three different time lags (τ = 0.9, 1.5, and 2.0 ms).
the time lag to τ = 1.5 ms makes the system unstable to the high-frequency mode
(408 Hz). Note that the high-frequency lobe encircles the −1 point on the real axis,
but the low-frequency mode does not. A further increase to τ = 2.0 ms causes the
instability to shift frequency modes and become unstable at 184 Hz.
The rotation of these lobes underscores the fundamental problem with achieving
stability from changes to the combustion time lag (i.e., by changing the phase).
Changes in time lag can simply change the oscillating frequency rather than pro-
duce stability. Even when a combustor has just a single dominant acoustic mode,
the width of the phase roll can produce oscillations over a range of frequencies
corresponding to the range of selected time lags as shown in Fig. 17.6. In summary,
careful consideration must be given to the acoustic modes before attempting to
solve a dynamics problem by adjusting the time lag.
The preceding example assumed that the flame response is a constant magnitude.
In real applications, the transfer-function magnitude and phase are governed by
the flame response to acoustic perturbations. As noted earlier, this response can
involve multiple physical processes that are often difficult to differentiate. The
next section reviews the physical processes associated with the flame response and
various approaches used to describe a combustion-transfer function.
D. Physical Processes Contributing to the Combustion-Transfer Function

The heat release from a premixed flame is the product of the reactants consumed
by the flame and the heat of reaction. The heat release can be written:
Q = ρY f SA f H (17.1)
where ρ is the density of the reactants, Y f is the mass fraction of fuel in the
premixed gases, S is the flame speed, A f is the area of the flame, and H is the
heat of reaction per unit mass of fuel. Based on this equation, the heat release
clearly can vary with perturbations in density, fuel mass fraction, flame speed,
and flame area. In gas-turbine combustion, the density perturbations arising from
acoustic pressure are typically much smaller than the other terms and are often
neglected. On the other hand, factors such as unsteady aerodynamics may produce
a significant modulation in flame area and may not be neglected. Likewise, changes
in the flow of either fuel and/or air will change the fuel mass fraction and the flame
speed.
In short, numerous mechanisms can generate perturbations in the heat release
at the flame. In most practical applications, it is difficult to separate and con-
trol these mechanisms to achieve stable combustion. Nevertheless, this section
is aimed at discussing some of these physical processes and attempts to model
the combustion-transfer function. In the following paragraphs, previous efforts to
understand simple premixed flames with constant fuel–air ratio will be discussed.
Following this discussion, the combined problem of fuel–air variation and flame-
area response will be considered.
1. Simple Premixed Flames

The transfer function of fully premixed flames has been investigated by various
authors. Blackshear31 proposed one of the earliest models for the response of a
premixed burner to acoustic perturbations. This work showed that variations in the
flame area were responsible for the driving, or damping, acoustic waves imposed on
the burner. Companion experiments demonstrated that the flame response depends
significantly on the mean flow velocity and the fuel–air ratio. Merk32 presented an
improved analysis of a premixed burner flame and was able to derive an explicit
expression for the flame-transfer function [see Eq. (17.2)].
1
A = u (17.2)
1 + jωτ1
Here, the perturbation quantities are normalized by their corresponding steady-

state values. This transfer function represents a first-order response between the
dimensionless flame area A and the dimensionless supply velocity u . The analysis
identifies a characteristic time τ1 that represents the average time for gas exiting
the burner to be consumed by the flame cone. For clarity, the notation τ1 is used
to make a distinction with the convective time lag τ identified in Fig. 17.2. Note
that the convective time lag τ includes the premixing process.
Note that for large values of ωτ1 , Eq. (17.2) predicts that the flame-area response
magnitude will be much less than one, and approach a phase of −90 deg. This is
different than the pure time-lag response e jωτ described in the example problem,
which will have arbitrarily large phase angles for large ωτ . This distinction will
be noted in a comparison to experiments discussed later.
Further investigations of the flame-transfer function have been carried out, which
include both analytical work and experimental data.11,33−37 Mugridge35 reports
preliminary attempts to measure the flame-transfer function, using techniques de-
scribed by Hadvig.38 Although few experimental details are given, the measured
transfer function showed considerable variation in the phase of the response over
relatively small-frequency ranges, in contrast to model expectations.
Matsui36 investigated experimental data from three multiport premixed burner
configurations and again identified a characteristic time lag in the flame-transfer
function. Matsui compared the various transfer functions that had been published
with that time and noted that the magnitudes were similar, but the phase angles were
considerably different. Note that Matsui’s transfer function includes a multiplier
with a pure time-lag term (e jωτ ). As noted previously, this pure time-lag term
allows the transfer function to reach large phase angles for large ωτ .
More recently, Fliefil et al.11 developed an analytical model for the flame-transfer
function that describes the distortion of the flame-surface area in response to im-
posed velocity perturbations at the base of the flame. Unlike earlier work,31,32,36
this model accounted for the distortion of the flame surface by tracking the kine-
matics of the flame movement in the oscillating flow. The variation in heat release
again results from the variable flame-surface area. The predicted transfer function
is qualitatively similar to earlier models and can be approximated as a first-order
system. However, no experimental data were used to verify the predictions.
Using a theoretical model very similar to Fliefil et al.,11 Ducruix et al.37 com-
pared the measured and predicted flame-transfer function of a bunsen flame with
an oscillating supply of premixed gases. Compared with the theoretical model, the
magnitude of the transfer function was predicted reasonably well for two different
burner configurations and at several operating conditions. In contrast, the phase
of the transfer function was poorly predicted for frequencies beyond ∼30 Hz. For
frequencies higher than ∼30 Hz, the phase of the response depended significantly
on the mean flow velocity and the burner dimensions. Additional experiments (not
reported) showed that the spatial velocity distribution at the base of the flame must
be accounted for to improve model predictions.
Although the preceding discussion is limited to relatively simple premixed Bun-
sen or jet flames, the difficulty of accurately predicting a flame-transfer function
is clear. In bluff or step-stabilized flames (e.g., afterburners or dump combustors),
flame-area variations may originate from oscillations in the shear layer and from
vortex merging.18−24,39 In swirl-stabilized flames, these aerodynamic phenomenon
are even more complicated. The swirl angle, the size of the combustor-step expan-
sion, and the length of the combustor can all affect the flow dynamics. Thus, for
swirl-stabilized flames, there is no general approach to estimate the contribution
of the flame-area variation to the flame response. Some recent attempts to measure
or predict transfer functions for swirl flames are discussed below.
2. Practical Swirl-Stabilized Flames

In addition to aerodynamic processes, perturbations in the fuel–air ratio that
occur at the fuel injector (Fig. 17.2) will result in variations in the heat release
q after a convective time lag τ . Although the supply-system dynamics are not
formally a part of a flame-transfer function, it is convenient to include the feed-
system dynamics as part of the combustion response. A considerable body of
literature is available for rockets, industrial burners, and gas turbines,1,14,40−44
which describe the importance of the supply system dynamics. Several recent
papers45,46 suggest that it is also necessary to account for the dispersion of fuel–air
perturbations in the premixer when describing the supply-system dynamics.
For some combustor configurations, flame-area variations occur simultaneously
with the feed-system dynamics. For example, Peracchio and Proscia6 have pro-
posed a model that includes both the variation in flame area and the simultaneous
variation in fuel–air ratio. With an appropriate choice of empirical parameters, this
approach compared favorably with measurements of the heat-release response to
pressure perturbations. In addition, Peracchio and Proscia commented that the role
of flame-area variations is very significant and probably larger than expected from
a simple flame model. These observations were made based on visual observations
of the flame and unpublished computations of fluid dynamics. In contrast, other
experiments47 show only modest structural change in practical turbine flames dur-
ing oscillations. This result is not general, however, because the same combustor
exhibited significant flame-area variation when tested at atmospheric pressure.47
In addition to the flame-area effects, the flame-anchoring method also plays an
important role in the combustion-transfer function. Kendrick et al.26 compared the
dynamics of two different fuel–air premixers and showed that flame-anchoring
methods will significantly affect the flame response. One premixer used aerody-
namic stabilization of the center recirculation zone, and the second premixer used
a bluff-body stabilization (with swirl). The aerodynamic stabilization was noted
to provide weaker flame anchoring, allowing the flame-reaction zone to oscillate
axially in the flow throughout the pressure cycle. This axial movement may add
another complexity to the flame-transfer function. Schuermans and Polifke25 also
-15
(dB)
Magnitude (dB)
-25
Magnitude
-35
Φ = 0.55
Φ = 0.60
Φ = 0.65
-45
10 100 1000
Frequency (Hz)
Frequency (Hz)
0
-200
(degres)
Phase (degrees)
-400
Phase
-600
-800 Φ = 0.55
Φ = 0.60
Φ = 0.65
-1000
10 100 1000
Frequency (Hz)
Frequency (Hz)
Fig. 17.9 Measured flame response to perturbations in acoustic velocity equivalence

ratios (Φ) from Khanna.51
noted that it is necessary to include flame translation in their analytical model of

the flame response.
Several recent papers have attempted to measure the dynamic response of
premixed, swirl-stabilized flames that are characteristic of low-emission gas
turbines.48−52 As an example, Khanna51 and Khanna et al.52 measured the flame-
transfer function of a swirl-stabilized flame with fuel–air premixing similar to what
is used in gas-turbine applications (Fig. 21.9). Data are shown for three different
equivalence ratios that range from 0.55 to 0.65. Note that the magnitude curve is
very complicated, having several peaks and minima. It is suggested that part of
the response is associated with near-field acoustics that are not accounted for in a
simple feedback model between the local acoustic pressure p and the heat release.
The large values for the phase angle and the drop in frequency are indicative of a
time-lag response, which is noted in a corresponding analysis.51,52 The phase plot
also demonstrates a significant change in the phase with equivalence ratio. Note
that the change from = 0.6 to 0.55 produces an abrupt change in the phase. This
remarkable change has the effect of changing the orientation of the lobes in the
Nyquist plots discussed earlier and is an example of how the flame dynamics can
change appreciably with relatively minor changes in operating conditions.
E. Stability Analysis Using Computational Fluid Dynamic (CFD)

Simulations and Finite-Element Analysis (FEA)
Many investigations have used computational fluid dynamics (CFD) to predict
the flame response.7,8,53−57 However, because CFD solutions converge in the time
domain, the results must be transformed into the frequency domain for use in a
stability analysis like those mentioned previously. Some authors8,54,55 have used
CFD to model the transient response of a flame to a step change in the reactant
flow. A Fourier transform is used to acquire the desired frequency-domain flame
response. Compared with limited experimental data, this approach requires some
empirical filtering to produce reasonable agreement with experiments.55 The same
technique has been extended to include the frequency response of a burner supply
system by Krueger et al.,8 providing a stability analysis for an entire engine. In
a slightly different approach Zhu et al.57 used CFD to predict a flame response
to several different types of input-flow-perturbation signals: sinusoidal, random
binary, and sum of sinusoidal. With Fourier analysis, the frequency-domain re-
sponse was calculated for the input signals. Compared with a direct time-domain
response at a single frequency, the different input signals provide various advan-
tages in accuracy or computational speed to predict the transfer function. Although
the results were not compared with experimental data, the predicted transfer func-
tion is a first-order response similar to Eq. (17.2). The analysis in Zhu et al.57
applies to a spray-flame combustor and is limited to relatively low frequencies
(<120 Hz).
To use detailed models in a stability analysis, both the G and H transfer functions
must be accurately known. To calculate H, it is necessary to know the acoustic
response of the combustor. Finite-element methods have been used to predict
the acoustic response,58,59 but it is difficult to combine these acoustic analyses
with predictions of the flame response, especially in combustors with multiple
burners. Complications arise in situations that include coupled acoustic modes. For
example, in an annular combustor, both longitudinal and circumferential acoustic
modes may be present, and the modes may be coupled. This coupling is not
typically accounted for in linear acoustic models, but recent low-order models have
been developed to capture this detail.60,61 Accounting for this coupling allows an
improved description of the acoustic pressures and velocities at individual burners.
It will be noted in Sec. III.B that response of individual burners is another passive
control approach that should be exploited to improve stability.
Pankiewitz and Sattelmayer60 developed an acoustic model that is solved in the
time domain and then specified a model for the heat release (i.e, the flame response)
at the individual burner elements. Instead of using Nyquist analysis, Pankiewitz
and Sattelmayer assessed stability from the growth or decay of finite disturbances,
similar to what has been done in rocket-engine analysis.9 An interesting result
is the prediction of spinning acoustic modes in the combustion chamber; that is,
the pressure nodes rotate around the annular geometry. These spinning modes
have been observed in commercial annular combustors62 and may deserve more
consideration in future analysis.
In summary, Sec. II has shown how control models can be used to evaluate
the stability of combustion systems. The combustion process can be treated as
the forward transfer function G and the system acoustics can be represented in the
feedback path H. The physical processes that contribute to the combustion response
have been reviewed along with measurements and models that have been used to
describe the transfer functions. Nyquist analysis demonstrates both the potential
benefits and potential problems associated with adjusting the combustion time lag
to produce stability. In particular, frequency shifts and mode changes can frustrate
attempts to solve a combustion-dynamics problem. This will be discussed in more
detail with reference to experimental data in the next section.
III. Methods to Improve Combustion Stability

Although various approaches have been used to improve the stability of com-
bustion systems, the scope of this chapter focuses on passive control techniques.
Passive control approaches are typically a part of the hardware design or operating
envelope. Some examples of passive control approaches include: 1) changing the
average convective time lag12,13,63−65 ; 2) changing the time-lag distribution by
introducing multiple time lags14,63,66 or combining fuel injectors with different
dynamic response67 ; 3) changing the flame location and the use of pilot-flame
stabilization27,64 ; and 4) modifying the dynamic response of the feed system.68,69
As already noted, several different mechanisms may simultaneously contribute
to the flame dynamics, and the magnitude of each contribution is usually unknown
in practical problems. However, both experiments and engine tests have demon-
strated that fuel–air variations play a significant role in combustion dynamics, in
particular, for premixed combustion systems.12,13,63−65 Thus, practical solutions
to dynamics problems often focus on the convective time lag τ .
Relatively simple models have been successfully used to describe the effect
of the convective time lag. These models are a partial description of what has
been shown in the Nyquist analysis. Section III.A will describe time-lag models
and discuss the connection to the Nyquist analysis. Section III.B will review the
effect of increasing the distribution of time lags and using multiple time lags.
The remainder of this section will describe other passive methods to improve
combustion stability, such as changing the flame geometry, introducing diffusion
pilot fuel, and modifying the dynamic response of the supply system.
A. Convective Time-Lag Models and Approaches to Modify

the Average Time Lag
In combustion dynamics, the importance of various time lags is well documented
in the literature, dating back to the 1950s. Putnam41 summarized many of these
ideas and used simple arguments based on the Rayleigh criterion to establish when
a system will oscillate or be stable. In simple terms, the Rayleigh criterion states that
the heat-release perturbations q should be in phase with pressure perturbations p
to strongly drive oscillations. Likewise, when p and q are out of phase, oscillations
are strongly damped.
Consider a situation in which the fuel injector responds to sinusoidal pres-
sure perturbations in the combustor. Assume that the local fuel–air ratio at the
fuel-injection location is slightly higher when combustor pressure sinusoid is at

a maximum. Slightly richer perturbations are then transported downstream to the
flame with a convection time lag τ . Additional time lags may characterize the com-
bustion process (e.g., the time needed to burn the reactants) and could be included
in an overall time lag. For the purposes of this example, it is assumed that richer fuel
perturbations arriving at the flame create an immediate increase in the heat-release
rate. For the heat-release perturbation to coincide with the next pressure peak, the
convective time lag should be equal to one acoustic period, T = 1/ f , where f
is the frequency of the sinusoidal pressure perturbation. It is also possible that
the convective delay could equal two periods, three periods, etc., and the pressure
and heat-release rate perturbations would still be in phase. Thus, oscillations may
occur when
τ/T = τ f = 1, 2, 3, . . . (17.3)
The criterion for instability expressed by Eq. (17.3) is an analog to the Nyquist
stability criterion described in Sec. II. That is to say, if the time delay (phase
angle) between the richer pockets of fuel–air ratio produced in the premixer and a
subsequent increase in the heat release is such that the combustor becomes unstable,
then the Nyquist plot should produce an encirclement of the −1 point on the real
axis. The concept introduced by Eq. (17.3) expresses the same idea. For example,
if p produces a heat-release rate variation q that will amplify the next cycle of p
(i.e., τ f = 1), or subsequent cycles (i.e., τ f = 2, 3, . . . ), then an instability will
occur. The sequence 1, 2, 3, . . . is specific to the simple example described. As
noted by Putnam,41 it is not necessary that the p and q perturbations be precisely
in phase to meet the Rayleigh criterion. In principle, oscillations can occur for a
range of τ f (i.e., ± 0.25) centered on the indices shown before.
Depending on the acoustic response of both the fuel system and the air passage
in the premixer, rich and lean pockets of mixture may be produced at various phase
angles relative to the pressure p in the combustor. For example, if the premixer
response produces richer pockets at the minimum of the combustor pressure, then
only 1/2 acoustic period is needed to align q with the next maxima in pressure.
Thus, the sequence would be
τ/T = τ f = 0.5, 1.5, 2.5, . . . (17.4)
Lacking details of the premixer response, it is not known a priori what sequence of
numbers will describe oscillating regions in a given application. To further com-
plicate matters, τ is an average representation of the time from fuel injection to the
time of combustion. As discussed by Lieuwen et al.,65 this time lag depends on
the flame location and the flame shape. In practical systems, neither flame location
nor flame shape is easily measured or predicted. For these reasons, the sequences
described in Eq. (17.3) and Eq. (17.4) are typically determined experimentally, al-
though some applications of computational fluid dynamics have shown promising
predictions of both the time lag and the associated stability regions.8,12,50
Experimental evaluation of the time-lag model described before has been
demonstrated by Richards and Janus13 and Straub and Richards.63 A can-style
combustor test rig (Fig. 17.10) has been used to record the pressure dynamics from
Optical Access
(OH* Signal)
Preheated Water Spray

Air Removable
Plug
Gas Turbine
Premixer
1 meter
Fig. 17.10 Schematic illustration of experimental combustion test rig from Ref. 13.
a premixed gas-turbine fuel injector. The modular premixer design (Fig. 17.11)
allows the position of fuel injection to be changed from three positions A, B, C,
or simultaneously from two of the three positions. Thus, changes to the time lag
could be studied by changing the bulk flow velocity and the physical distance L.
Data are collected over a range of operating conditions. The time lag for each
condition has been estimated by using a fixed-flame standoff.
Figure 17.12 shows the observed rms pressure as a function of τ f . These data
indicate that oscillations are confined to a band 0.45 < τ f < 0.7. The edges of
this region are referred to as the stability boundaries. This plot can be used to
understand how proposed changes in the nozzle geometry, such as increasing
the premixer length, would affect stability. For a stable combustor operating at
τ f = 0.4, Fig. 17.12 indicates that proposals to move the fuel injector upstream
at a fixed velocity will increase τ f such that oscillations will occur. This increase
has been demonstrated experimentally by Richards and Janus.13 It would also be
possible (but not necessarily advisable) to increase the fuel time lag enough to
get to the upper-stability boundary. This increase in fuel time lag has been shown
experimentally by Straub and Richards63 and will be discussed in more detail later.
It is useful to understand how the τ f plot and the control-model analysis are
related. In the Nyquist analysis, the lobes of the plot rotate as the time lag is
changed. As previously discussed (Fig. 17.6a–c), if encirclement occurs at a par-
ticular frequency (i.e., f 1 ), small increases in the time lag will continue the rotation
of the Nyquist plot, and the frequency of the instability will decrease until the lobe
does not encircle −1 at a lower frequency (i.e., f 2 ). Attempts to solve instability
problems by adjusting the time lag must recognize that the frequency will change
in response to changes in the time lag.
Although the stability boundary in the τ f plot is clearly recognized (Fig. 17.12),
increasing the time lag will result in lower-frequency oscillations, depending on
the bandwidth of the resonant frequency. The bandwidth is often overlooked, or
misunderstood, when using plots similar to Fig. 17.12, but the Nyquist analysis
indicates this very clearly (Fig. 17.6a–c). Note also that the left stability boundary
shown in Fig. 17.12 is typically accompanied by the higher-frequency oscillations,
assuming the combustion system has a single resonant frequency.
This discussion suggests that the bandwidth of the acoustic response is an impor-
tant consideration when applying passive-control techniques to control combustion
Geometry 1
Geometry 2
Fig. 17.11 Schematic illustration of fuel–air premixer configurations used to inves-

tigate passive-control approaches via changes to the time lag. Fuel injection points A,
B, and C are shown.
5.0
RMS Pressure (%)
4.0
3.0
2.0
1.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0
(Time Lag)*(Frequency)
Fig. 17.12 Data showing the experimentally determined stability boundaries using a
time-lag model from Ref. 13.
τf = τf = τf =
0.50-0.75 1.50-1.75 2.50-2.75
1000
Natural
800 freq's
Frequency (Hz)
f1 f1 =
600 640 Hz
f1 or f2?
f1 or f3?
f2 f2 =
400
410 HZ
f2 or f3?
200 f3 f3 =
190 HZ
0
0 1 2 3 4 5
Time lag τ (msec)
Fig. 17.13 Stability boundaries of the time-lag model plotted as frequency verses time
lag. Also shown are three different frequency modes that can complicate approaches
to control combustion oscillations by changing the time lag.
dynamics. If the bandwidth of the acoustic response is large, the lobe of the Nyquist
plot covers a wide frequency range, and significant changes in the value of τ are re-
quired to cross the stability boundaries. Fannin et al.29 showed a more detailed com-
parison between the Nyquist stability boundaries and the simple time-lag model.
Multiple acoustic modes introduce another complication to the time-lag model.
When the system acoustics produce strong response at several frequencies, at-
tempts to modify the time lag could result in a jump between frequencies rather
than produce stability. This point has been demonstrated in the example Nyquist
plots shown in Fig. 17.8, in which subsequent lobes of the Nyquist diagram rotate
and encircle −1 as the time lag is increased (or decreased). The concept can also
be described less formally by a plot like the one shown in Fig. 17.13. In Fig. 17.13,
frequency is plotted on the vertical axis, and the time lag is plotted on the horizon-
tal axis. The indices at which instability occurs are plotted as regions having an
assumed width (i.e., 0.5 < τ f < 0.75, etc.). Horizontal lines are drawn through
combustor natural acoustic frequencies as shown. Combinations of τ f within these
shaded regions have the correct phase to meet the Rayleigh criterion and may pro-
duce an unstable condition. If a given combustor exhibits oscillations at frequency
f 1 with τ = 1 ms, a proposed solution might be to increase the time lag. However,
because of the multiple acoustic modes, this solution has the potential to get to
the edge of the stability boundary τ f 1 = 0.75, and then drop to τ f 2 = 0.50. It is
conceivable that continued increases in time lag to ∼2 ms could produce stability,
but slightly greater time lag could then jump back to τ f 1 = 1.50, or τ f 3 = 0.50.
Thus, adjustment of the time lag must be pursued with a full understanding of the
frequency spacing of acoustic modes. A similar conclusion can be drawn from the
formal Nyquist analysis, in which multiple natural frequencies produce multiple
lobes on the Nyquist diagram.
Single Fuel Injector Tests

(Locations A, B, and C)
10.0%
8.0%
RMS Pressure
6.0%
4.0%
2.0%
0.0%
0 100 200 300 400 500
Frequency (Hz)
Fig. 17.14 Experimental data showing multiple frequency modes.
Experimental demonstration of the frequency-switching behavior described here

has been presented by Straub and Richards.63 The modular fuel injector shown
in Fig. 17.11 has been studied over a range of bulk velocities (30–60 m/s) and
with different fuel-injection positions A, B, or C. This produced a net variation
in time lag from 1.8 to 7.3 ms. Figure 17.14 is a plot of all the experimental data
and shows that this combustor has two strongly responsive frequencies near 160
and 220 Hz. Based on calculations, these frequencies are close to (but not equal
to) the natural frequencies expected for the combustor. The τ f plot is shown in
Fig. 17.15. As expected, oscillations are confined to relatively small regions of
the τ f plot, despite the multiple acoustic modes and the wide range of time lags
investigated. Note that few data points are in the stable region between the τ f peaks.
The absence of data points in the stable region is an example of the frequency
switching shown schematically in Fig. 17.13. To clarify this point, Fig. 17.16
shows the dominant frequency of oscillation plotted as a function of the time
lag for nozzle configuration C and a range of equivalence-ratio conditions. As
the time lag increases, the frequency drops slightly, and the stability boundary
for τ f = 0.80 is approached. At this condition, the rms pressure levels strongly
depend on the equivalence ratio. Weak oscillations near 260 Hz are exchanged
with strong oscillations at 160 Hz as the equivalence ratio is changed. Further
increases in time lag cause an increase in frequency and move the system to the
next τ f band, at which the amplitude is large again. Still further increases in time
lag are accompanied by a reduction in oscillating frequency. All the experimental
observations are consistent with both the time-lag model and the Nyquist analysis
presented in Sec. II. However, contrary to the time-lag model that only considers
the timing (phase) of the system, the Nyquist analysis incorporates both the gain
and the phase of the system.

(Locations A, B, and C)
10.0%
8.0%
RMS Pressure
6.0%
4.0%
2.0%
0.0%
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80
(Time Lag) * (Frequency)
Fig. 17.15 Data showing experimentally determined stability boundaries for
multiple-frequency mode combustor by using the time-lag model.
This discussion has shown that solving dynamics problems by adjusting the time
lag may be complicated by the multiple τ f bands and multiple acoustic modes. If
the frequency spacing f between adjacent acoustic modes is such that mode tran-
sitions are possible (as shown in Fig. 17.13), there is a good chance that changing
the time lag will simply produce a frequency shift rather than stable combustion.
This conclusion depends on the gain of the various modes, but it does suggest the
following rule of thumb: where the mode spacing f is such that f τ ∼ < 0.5,

(Location C)
500
Tau*f=0.45
Tau*f=0.80
400 Tau*f=1.25
Tau*f=1.35
Frequency (Hz)
300 Weak Oscillations
200
100
0
0 2 4 6 8 10
Time Lag (msec)
Fig. 17.16 Experimental data showing transition between τ f bands and different
frequencies.
changes in the time lag can lead to mode switching. In these instances, rather than
modifying τ to produce stability, consideration should be given to increasing the
time-lag distribution or modifying the acoustic response of the combustor.
B. Approaches to Increase the Time-Lag Distribution

The preceding discussion shows how discrete changes in time lag affect system
stability, and some potential issues associated with this approach have also been
discussed. In the preceding discussion, a single (i.e., average) value for the con-
vective time lag is used. In practice, however, a distribution of time lags will be
produced by various mechanisms (i.e., turbulent diffusion, flame location, flame
shape, etc.). This time-lag distribution can also be changed by the design of the
supply system (i.e., multiple fuel-injection locations, adjusting flow splits, and mul-
tiple injectors with different fuel-injection locations). This section will address the
concept of modifying the distribution of time lags to enhance combustion stability
with particular attention given to design approaches for the supply system.
The concept of using multiple time lags to address combustion oscillations
has been noted by Keller14 and is the subject of a commercial patent.70 Multiple
time lags could be produced using two (or more) points of axial fuel injection
simultaneously (i.e., positions A and B or A and C, shown in Fig. 17.11). Both
fuel-injection points produce richer and leaner packets of fuel–air mixture in re-
sponse to variations in the premix airflow. However, only one of these packets
can produce heat-release rate perturbations that are in phase with the pressure
perturbations; the other packet may be arranged to produce perturbations that
are out of phase with the acoustic pressure. The out-of-phase perturbations add
a considerable amount of damping to the system. The benefit of this approach
has been demonstrated experimentally in Straub and Richards,63 in which single-
point injection is compared with multiple-point injection by using the geometry
shown in Fig. 17.11. The resulting stability maps are shown in Fig. 17.17. The
horizontal axes are the equivalence ratio and bulk premixer velocity. The verti-
cal height represents the rms level of the pressure oscillation. Note that injection
from points A, B, or C had almost universally higher amplitudes than simulta-
neous injection from points A and B or A and C. Fannin et al.29 investigated
these experimental results by using Nyquist analysis to demonstrate that the mul-
tiple time lags can produce a significant reduction in oscillating pressure, as seen
in the data. The Nyquist analysis shows that multiple time delays produce a re-
duction in the amplitude of the fuel–air mixture perturbations and adjustment to
the phase.
An interesting practical application of the multiple time-lag concept has been
demonstrated by Berenbrink and Hoffman.67 These authors recognized that the
time lags can be adjusted on individual fuel injectors so that the combined com-
bustion response is more stable. This approach has an advantage compared with
placing multiple fuel ports in a single injector, because it is easier to optimize the
mixing from a single fuel port. On a combustor with 24 fuel injectors, a total of
20 injectors were outfitted with extension tubes that produced different time lags.
The resulting engine test demonstrated that it was possible to operate the engine
at conditions that had previously been limited by dynamic oscillations.
In addition, a slightly different approach using asymmetric burners67 has also
produced comparable stabilization. The asymmetric burner concept simply placed
Location A Location A
10%
10%
RMS Pressure
8%
RMS Pressure
8%
6%
6%
4%
4%
2%
2%
0%
0.59
0% Equiv.
0.59
30 40
0.67
Equiv.
30 40
0.67
50 60 Ratio
0.77
50 60 Ratio Velocity
0.77
Velocity
(m/s)
(m/s)
Location B Location C
10% 10%
RMS Pressure
RMS Pressure
8% 8%
6% 6%
4%
4%
2%
2%
0%
0.59
Equiv. 0%
0.59
Velocity 30 40 50
0.67
Ratio 30 40 Equiv.
0.67
0.77
(m/s) 60 Velocity 50 Ratio

0.77
(m/s) 60
Location A&B Location A&C
10% 10%
RMS Pressure
RMS Pressure
8% 8%
6% 6%
4% 4%
2% 2%
0% 0%
0.59
Equiv.
0.59
30
Equiv.
0.67
40
30 40
0.67
50
Velocity Ratio
60
50
0.77
Velocity Ratio
0.77
(m/s) 60
(m/s)
Fig. 17.17 Experimental data showing the effect of changing the distribution of time
lag by using fuel injection from two axial locations, A and B and A and C in Fig. 17.11.
different fuel injectors at different angles relative to the combustor flow axis. The
resulting asymmetry in the flame shape prevents uniform coupling to the acoustics
and also has the effect of shifting the flame position on different oriented injectors.
The net effect is again a distribution of time lags among the different injectors.
Lovett and Uznanski66 used a different concept to produce multiple time lags
in a single fuel injector. Their approach allowed a split between fuel supplied
to a main and a secondary premixer passage. A different time lag is associated
with each of these passages. Therefore, a composite response could be achieved
by adjusting the flow splits. This approach produced reduced dynamics at select
values of the secondary fuel flow.
As described in the preceding paragraphs, deliberate introduction of multiple
time lags is one approach to improve combustion stability. Another approach is
to simply take advantage of the mixing processes that occur during fuel injec-
tion. Because the fuel–air perturbations are mixed by turbulent processes in the
premixer, a single time lag does not completely describe the premixer response
to flow perturbations. To account for turbulent mixing, a distribution of time lags
will better represent the response of even a single point of fuel injection. Scarinci
and Freeman45 showed that turbulence in the premixer can significantly disperse
fuel–air perturbations and suggested that dispersion may reduce the oscillating
amplitude of combustors using longer premixer barrels.
Sattelmayer46 has also shown that the time-lag distribution can play a key role
in reducing the overall magnitude of the heat-release oscillation. Thus, attempts
to improve combustion stability should recognize the contribution of time-lag
distribution in the premixing process and in the flame geometry. These flame-
geometry effects will be described next.
C. Flame-Geometry Effects
Another consideration that has an impact on the time-lag distribution is the
geometry of the flame. This effect can be understood by noting that the arrival of
fuel–air perturbations is distributed over the surface of what is typically a conical-
flame geometry. The conical geometry means that the delivery of a fuel perturbation
will produce a combustion response that is distributed over the surface of the flame
and subsequently over a range of time lags. Putnam41 recognized the importance
of these geometric features and developed a correction to a time lag by integrating
over the surface of the flame. More recently, Lieuwen et al.65 developed a similar
integral analysis and applied it directly to flame geometries that are of interest to
gas-turbine combustors. This analysis showed that correction factors to the time
lag can be as large as a factor of 1.5, and the correction is sensitive to the shape of
the flame.
Because of the uncertainties connected with describing the position and shape
of turbine combustion flames, rig measurements of the time lag may be necessary
to describe the flame response. Straub et al.71 have made preliminary attempts to
measure the flame time lag in a practical scale combustor, but more work is needed
on this topic. Krebs et al.50 measured the time lag in an atmospheric-pressure model
of a gas-turbine combustor and showed that a time-lag distribution was needed to
describe the measured response. The time-lag distribution is needed to account for
the flame shape.
D. Flame Anchoring and Pilot-Flame Stabilization Effects

Flame anchoring, which establishes the instantaneous location of the flame, is
another important consideration for reducing the flame dynamics. As mentioned
in Sec. II.A, Kendrick et al.26 suggest that poor flame anchoring can allow the
flame location to change such that the time lag will favorably drive oscillations.
Paschereit et al.27 showed that an oscillating flame anchor was responsible for driv-
ing dynamics in their combustion system. This problem was solved by physically
locating a pilot flame in the region where positive flame anchoring was needed.
The addition of a pilot flame is a common approach to reducing flame
dynamics,26 but the stabilizing mechanism is often uncertain. Both Kendrick et al.26
and Paschereit et al.27 attribute pilot-enhanced stability to an improvement in flame
anchoring, but there are additional reasons why pilot flames may improve stability.
The pilot flame is typically operated as a diffusion flame or partially premixed,
and it may have a lower dynamic response than a purely premixed flame. In a
diffusion flame, the local reaction rate is controlled by local mixing processes
instead of the overall fuel–air ratio. Thus, momentary perturbations in the fuel or
air supply cannot produce an appreciable change in the reaction rate. Control by
local mixing processes reduces the likelihood of oscillations in diffusion flames
and may explain why diffusion pilots can be used to silence oscillations. Although
more work is needed to understand the stabilizing effect of pilot flames, diffusion-
pilot flames are undesirable, because they contribute significantly to NO and NO2
emissions.
E. Approaches to Modify the Dynamic Response of the Supply System

Although the time-lag distribution is beneficial in reducing the impact of the
fuel–air perturbations, this feature is only a guaranteed benefit when perturbations
are the sole cause of instability. Perturbations as the sole cause of instability may
not be the case in many instances, because even fully premixed flames can exhibit
dynamic response to variations in mixture flow. As discussed in Sec. II.D.1, un-
steady aerodynamics can produce a variation in heat-release caused by changes in
flame area or vortex shedding/merging. These flame dynamics can occur without
any perturbations in fuel–air ratio. In these instances, a Nyquist analysis would
show that adding fuel–air perturbations could silence the oscillation that is pro-
duced by the flame-area variation. That perturbations could silence oscillations is
the basis for active combustion control using input-fuel perturbations.
The active-control system changes the magnitude and phase of the fuel-system
feedback, producing stability by controlling the arrival of mechanically actuated
fuel perturbations. A similar idea has been demonstrated by making passive ad-
justments to the feed-system dynamics, producing stability at select conditions by
controlling the arrival of fuel–air perturbations.43,68,69
IV. Acoustic Dampers

In the turbine-combustion literature, passive-control methods are strongly
weighted toward stabilizing the combustion process. Although acoustic dampers
are commonly used to stabilize rocket and afterburner applications, less empha-
sis has been given to the use of acoustic dampers in gas-turbine applications. In
stationary engines, the disparity of damper use may be related to the relatively low
frequencies encountered in most turbine applications (hundreds of hertz) vs those
typically encountered in rocket engines (kilohertz range). The lower frequencies
require physically larger dampers, which complicates engine packaging, and may
be a potential drawback. Nevertheless, dampers should not be overlooked in a
strategy to stabilize combustion.
Given the difficulty of proposing changes to the combustion response (the G
transfer function, in Fig. 17.1), a damper design can be proposed that will very
likely reduce acoustic feedback (the H transfer function, in Fig. 17.1). This proposal
is not to imply that damper design is easy or without uncertainties. However, given
the cost of reengineering a combustion system to produce a desired combustion
response, addition of dampers may be worth consideration. These acoustic dampers
are described next.
A. Damper Description
The simplest damper of all is a hole, releasing acoustic energy from the com-
bustion chamber that would otherwise return to the feedback loop. The efficacy
of this method is well known to practicing combustion engineers. Putnam41 noted
the following advice for practitioners faced with stubborn oscillation problems in
industrial burners, “To solve an oscillating combustion problem, drill a hole. If that
doesn’t work, drill two holes.” Although this anecdote is a humorous, it represents
genuine experience that reducing the acoustic gain can stabilize oscillating systems.
Conversely, eliminating holes can lead to combustion instabilities. Modern pre-
mixed combustors are designed specifically to avoid dilution holes, removing a
source of acoustic damping. The avoidance of holes is yet another reason why
premixed combustors tend to have problems with dynamics. Earlier diffusion-
flame combustors used numerous dilution holes around the perimeter of the liner,
providing a source of acoustic damping that is absent in premixed combustors.
Although drilling holes may be an acceptable control strategy in industrial burn-
ers, it is not an option for gas-turbine combustors in which flow splits must be ac-
curately controlled to meet performance targets. Alternatively, closed resonators
could be used to absorb acoustic energy. Because the resonators are closed, they
do not compromise the designed flow splits. These types of resonators have been
used extensively in rockets9,72 and afterburners.73 Figure 17.18 shows two com-
mon resonator geometries that have been used in gas-turbine applications. Both
types (i.e., the Helmholtz resonator and the quarter-wave resonator) are shown in
Fig. 17.18 for convenience. These two geometries are not necessarily combined in
practice, however. Figure 17.18 also shows the formulas and nomenclature used
to calculate the natural frequency f 0 of the resonator.
When pressure oscillations occur, flow enters and exits the resonator mouth.
The energy dissipated at the entrance–exit provides damping to the system. It can
be shown9 that the greatest losses are generated by maximizing the magnitude of
the oscillating velocity; this maximization is achieved by tuning the resonator so
that the natural frequency f 0 is close to the frequency that is to be damped in the
combustor. For a given f 0 , the Helmholtz neck area S can be optimized with the
length L and volume V to meet packaging space requirements. In the quarter-wave
design, f 0 can only be established by the length.
Helmholtz
c S
Resonator f0 ⋅
2⋅ π ( L + 1.7⋅ a ) ⋅ V
Fuel Volume, V
Radius, a
Area, S
Combustion
Air Flame products
Quarter L
wave 1 c
f0 ⋅
resonator 4 L
Fig. 17.18 Schematic illustration of resonator concepts to dampen the acoustic feed-
back of the combustion system.
The actual performance of a resonator depends on the resonator geometry and the
operating conditions. Because the acoustic dissipation occurs at the entrance/exit
flow near the mouth, the geometry of this region is very important. Laudien et al.72
show that rounded versus square corners at the resonator–combustor connection
produce a significant difference in the acoustic response. Furthermore, because the
resonator gas may include combustor products and purge cooling gas, the speed of
sound is uncertain. This uncertainty makes the design the resonator for a specific
natural frequency more complex. The correct resonator design and tuning is not
usually achieved from analysis alone. In most rocket applications and in the few
turbine applications cited subsequently, the resonator properties are finalized by
experimental testing.
Selecting the position and number of resonators is also an important considera-
tion. It does little good to place the resonator at an acoustic node. For example, in
annular combustors, circumferential acoustic modes (with waves traveling around
the annulus) are often characterized by standing nodes at specific positions. Res-
onators added at these node positions will not provide damping, because there is
no acoustic pressure to drive the oscillator. Attempts to position resonators at the
pressure antinodes may be frustrated by a repositioning of the node to the new res-
onator location. Thus, it may be essential to position multiple resonators by using
the number of resonators that are guaranteed to produce an assymetry relative to
the acoustic mode shape. For example, three resonators cannot all be aligned with
a wave structure having only two nodes. Decisions about the number of resonators
are made based on an analysis of acoustic waveform and the required damping in
specific situations. Some examples are described in the next section.
B. Application of Acoustic Dampers

Gysling et al.74 examined the use of Helmholtz resonators on a sector rig combus-
tor. Their analysis identified many of the important design variables for installing
a resonator. For example, the resonator-to-combustor volume ratio, the resonator
frequency, and the loss coefficient at the resonator mouth are all important consid-
erations in a practical acoustic damper design. The final resonator performance was
characterized by measuring the performance in situ. This characterization was ac-
complished by measuring the transfer function between the combustion-chamber
pressure and the resonator pressure and then fitting the theoretical parameters to
the resulting data. Note that this fit was made at each operating condition of inter-
est. As noted earlier, the gas temperature and flow conditions in the combustor will
affect the resonator natural frequency and damping, which needs to be accounted
for in resonator performance.
Five resonator configurations and three test conditions were reported by Gysling
et al.74 In the best cases, appropriately tuned resonators reduced the oscillating
pressure by almost an order of magnitude. Appropriate tuning required resonator
volumes that were approximately 12% of the combustion chamber volume. Based
on both theoretical modeling and experimental data, Gysling et al. also identify
many of the important considerations for resonator use.
Because a given engine may experience a frequency shift with operating condi-
tions, the resonator system must provide damping over a range of frequencies. This
damping can be accomplished by mistuning the resonator or by using multiple res-
onators tuned to different frequencies. This approach provides less than maximum
damping at a single frequency, but it allows good performance over the range of
conditions. In Gysling et al.,74 this approach was successfully demonstrated with
two resonators, which were tuned to two different frequencies. Good attenuation
was observed on this combustor application where oscillating frequencies ranged
from 232 Hz to 278 Hz.
Belluci et al.75 used a Helmholtz resonator model to design dampers that were
added to the silo combustor of a stationary gas turbine. The resonator model
included more physical detail for loss mechanisms than in Gysling et al.74 Experi-
mental testing was again used to establish the model parameters and then to design
resonators for the actual combustion system. Seven resonators were installed at
the inlet end of the combustor. The pressure-oscillation amplitude was reduced
by about 60% compared with the baseline without resonators. This paper did not
report the resonator performance at other frequencies or operating conditions. The
authors also used a supply of purge gas to keep the resonator temperature low.
As a final example of the use of resonators, Pandalai and Mongia3 report on the
use of acoustic dampers in aeroderivative engine applications. Unlike the previous
citations, these authors installed the damper tubes upstream of the combustor,
just before the fuel–air mixer. The damper tubes were constructed from 25.4-mm
tubes. A perforated plate at the resonator mouth was used to control the resonator
impedance. Although these authors do not compare the oscillating pressure with or
without the damper tubes, note that these dampers are now in wide commercial use,
having logged more than 100,000 hours of operation in various engine installations.
The wide commercial use is an excellent example of how dampers can play a
significant role in reducing pressure oscillations in stationary engines.
A few words of clarification should be given before concluding this discussion

of acoustic dampers. It is sometimes overlooked that dampers themselves partic-
ipate in the overall acoustic response of the combustion chamber. When dampers
with sufficient mouth area represent a considerable portion of the combustor vol-
ume, their presence can lead to a change in the natural frequency of the entire
combustion system, and they may actually destabilize an otherwise stable system.
This possibility was noted by Gysling et al.74 and should be carefully considered
before sizing and tuning resonators. Mitchell76 notes that in rocket-damper appli-
cations, sizing the resonators apart from understanding the acoustic structure in
the combustor can lead to less than optimal results. Mitchell provides an example
in which the greatest stability was achieved with mistuning the resonator from the
oscillating frequency; this stability was attributed to the change in waveform that
accompanied addition of the resonators.
C. Example of Acoustic Damper Design

This section describes a specific example to demonstrate both the method of
design and the potential benefits of acoustic dampers. However, this analysis is
theoretical, and as previously mentioned, experimental rig tests should be used to
finalize a real design application.
The example combustor geometry shown in Fig. 17.4 (lower) is used as a base-
line for this example. Recall from the discussion of Figs. 17.7 and 17.8 that this
combustor is characterized by strong resonances near 185 Hz and 410 Hz and that
both of these modes can be driven by combustion feedback over a wide range of
combustion time lags τ . As noted earlier, this wide range implies that changes to
the time lag would merely shift the oscillating frequency and not produce stability.
For this reason, acoustic dampers are worth consideration in this problem.
A starting point is to decide which modes are sought for attenuation by the
resonator. It is possible to design multiple resonators for multiple frequencies.
Likewise, it is possible, and may be advisable, to use multiple resonators for
a single frequency just to minimize the size of the resonator. These decisions
are tradeoffs that must be determined by specific applications, and the approach
described next can be easily extended to these situations. However, only a single
resonator addition is considered in the example.
For purposes of illustration, the single resonator will be designed to attenuate the
higher frequency (410 Hz) of the combustor shown in Fig. 17.4 (lower). Attenuating
the lower frequency necessarily requires a longer resonator, which may be an issue
for packaging the resonator assembly in a practical application.
The first step in adding the resonator is to decide what type of resonator to use.
As explained by Laudien et al.,72 the quarter-wave resonator has some advantages
in damping performance at high acoustic pressures because the loss mechanism
is more predictable for the quarter-wave design. However, the length required for
the quarter-wave frequency can be prohibitive for typical frequencies encountered
in stationary engines. A shorter-resonator package can often be designed by using
the Helmholtz configuration. The geometry and natural frequency of this resonator
are shown in Fig. 17.18. Note that neck cross section S, length L, volume V , and
the neck radius a all contribute to the natural frequency. These parameters can be
selected to optimize the package length and resonator performance.
The resonator position in the combustor is another design consideration. As

noted earlier, when selecting the resonator position the designer must be cognizant
of the potential for acoustic nodes in the system. A resonator placed at an acoustic
node will do little good, because acoustic pressure is required to drive flow in
the resonator. In this example, analyses of the combustor acoustics show that the
region near the dump plane is a pressure antinode. Thus, the resonator is located
in the combustion chamber, downstream of assumed flame location. This location
will allow hot combustion products to be swept into the resonator, and it is ex-
pected that a cooling purge would be required as in Bellucci et al.75 Alternatively,
the resonator could be added to the combustor dome, where (ideally) cooler com-
pressor discharge would enter the resonator. However, cooling precautions may
still be required, because it is likely that small oscillations may cause combustion
products to enter the resonator neck.
These considerations point out another difficulty in resonator design. Because
the resonator contents may include a mix of purge gas and combustion products,
it is difficult to predict the as-built natural frequency. Thus, design calculations
should be carried out for a range of natural frequencies, ensuring that adequate
damping can be achieved over the possible operating conditions.
For this example, the resonator design is based initially on gas properties of
the combustion products. The Helmholtz resonator volume V must be selected to
begin the design. This volume must be a reasonable fraction of the combustion
volume so that acoustic energy generated by the flame can be exchanged into
the resonator. For example, it has been shown that appreciable attenuation can be
achieved if the volume of the resonator is approximately 7% of the combustion
volume.74 Therefore, this value is used as a starting point. The combustion volume
here is defined as the length of the cylindrical combustor needed to achieve a plug
flow residence time of 30 ms, which is typical of the time needed for complete CO
oxidation in premix combustors. Note that the specific geometry of the resonator
volume does not need to be precisely defined. In this example, the resonator volume
is formed from a cylindrical geometry with aspect ratio of unity (e.g., length =
diameter).
The resonator is designed with a natural frequency equal to the combustor
frequency that needs attenuation (410 Hz). The natural frequency formula is shown
here:

c S
f0 = (17.5)
2π (L + 1.7a)V
By using the approximate volume ratio described previously and the natural fre-
quency, only the neck length L and cross section S must be chosen. Assuming the
neck radius a is a direct function of S, a range of values for S can be studied. Note
that S must be a significant fraction of the combustor cross section to allow the
transfer of acoustic energy between the combustor and resonator. In this example,
S is chosen as 10% of the combustor cross section. This choice allows the length
L to be calculated directly from Eq. (17.5). The resulting resonator geometry is
shown to scale relative to the overall combustion system in Fig. 17.19. Note that
space and packaging considerations may lead to variations on this initial design.
Resonator
Flame
D
C 10 cm
B
A
Fig. 17.19 Geometry of Helmholtz resonator design, shown approximately to scale,
on the example combustor of Fig. 17.4 (bottom). Positions A to D represent different
resonator locations.
For example, the same natural frequency could be achieved with a shorter neck
but larger volume.
After the resonator geometry is defined, the resonator location can be assessed
by using the system model described in the example problem. Experimental ver-
ification is also warranted for actual applications, but, for the purposes of this
example, the effects of resonator location will be pursued by using the model
outlined in the Appendix. Four resonator locations A thru D will be analyzed to
demonstrate the effect of axial position on the system response.
As noted in the Appendix, the resonator can be included in the transfer-matrix
analysis by the addition of a transfer matrix containing the resonator impedance
Z r . Laudien et al.72 provide an expression for the resonator impedance:

2π f 2 4 L µvis π f 2π f (L + 1.7a) f 02
Zr = + ε+ +j· 1− 2 (17.6)
c S 2a ρ S f
In this expression, f is the frequency and f 0 is the natural frequency of the resonator
calculated from Eq. (17.5). The speed of sound c and the dynamic viscosity µvis
are evaluated at the (assumed) gas conditions in the resonator. The dimensionless
term ε is a resistance factor that depends on the quality of the orifice and nonlinear
processes at the mouth of the resonator. This term is difficult to evaluate analytically
and depends on the magnitude of the oscillating pressure. The resistance factor ε
has been measured, and the value typically varies from 10 to more than 30.72 A
value of 20 is selected for this study. Again, a parametric study of this parameter on
resonator performance is suggested to define the range of possible behavior before
conducting experimental testing. For this example, investigation of the effect of ε
(not reported) showed only a modest effect on resonator attenuation.
Given the geometry and parameter selections described in preceding text,
Fig. 17.20 presents the open-loop frequency response of the example combus-
tor without the resonator. Note that the baseline case without the resonator is
shown in the top of Fig. 17.20. Recall that the baseline corresponds to the spec-
tra of Fig. 17.7. With the resonator tuned to 410 Hz, the amplitude of the higher
combustor mode is replaced by two much smaller adjacent modes. Gysling et al.74
noted that the addition of a resonator produced two new eigenvalues, representing
5.0
4.0
Magnitude
3.0
2.0
1.0
0.0
0 100 200 300 400 500
Frequency (Hz)
5.0
4.0
Magnitude
3.0
2.0
1.0
0.0
0 100 200 300 400 500
Frequency (Hz)
Fig. 17.20 Comparison of baseline open-loop frequency spectra without resonator

(top) to with resonator in location A (bottom). With the resonator, the high-frequency
410-Hz mode is replaced by two much smaller modes.
the interaction of the resonator with the existing acoustic modes. The appearance
of additional modes is not a complication in this example, because their amplitude
is very low. However, the possibility of creating oscillations at different frequen-
cies should not be overlooked when adding a resonator. If the combustion gain
were much larger at the new frequencies, oscillations might occur. Note that the
resonator has little effect on the lower-frequency combustor mode, as expected.
Figure 17.21 shows the Nyquist plot for the combustor and resonator. Fig-
ure 17.21 can be directly compared with Fig. 17.8a. Note that the resonator has
eliminated the higher-frequency lobe that encircled the −1 point, and so, accord-
ing to this analysis, the combustor would be stable. However, note that the lower
frequency would still be unstable at longer time lags, because the low-frequency
lobe would rotate, as in Fig. 17.8, at longer time lags.
As noted, several parameters are assumed in the resonator design. If the gas
temperature is different than assumed temperature, the resonator natural frequency
will differ from the planned value. To explore this issue, the same calculations are
carried out to compare resonators with natural frequencies of 390, 410, and 430
Hz. The resulting Bode plots are shown in Fig. 17.22. For clarity, the scale is
increased relative to Fig. 17.20, and only the high-frequency range is shown. Note
that all cases have a magnitude less than unity, meaning that the combustion would
be stable. However, if the assumed combustor gain is larger, some new frequencies
Nyquist Plot
3
2
1
Imaginary
0
-1
-2
-3
-3 -2 -1 0 1 2 3
Real
Fig. 17.21 Nyquist plot of open-loop frequency response with the resonator. These
results compare with plots in Fig. 17.8a.
could become unstable. For example, adding a resonator tuned to 390 Hz produces
a peak response of 0.5 at 420 Hz, implying that if the combustor gain rises by a
factor of 2, the system could oscillate at 420 Hz.
Figure. 17.23 shows the effect of moving the axial position of the 410-Hz res-
onator to the axial positions A thru D in Fig. 17.19. Because the resonator ideally
must interact with the pressure antinode near the flame, moving the damper along
the axis reduces the effectiveness. All the gains are still less than unity, implying
stability, but this stability again depends on the assumed combustion gain and other
parameters.
In summary, these example calculations have shown that the design of a damp-
ing resonator is possible by selecting a resonator with a natural frequency equal to
the mode to be damped. The system response is analyzed by inserting the resonator
impedance Z r into the acoustic analysis presented in the Appendix. These calcula-
tions can be repeated to assess the effect of various design parameters, such as the
natural frequency, loss mechanisms, and resonator position. Although predictions
in this example show that the system is stabilized by the damper, care must be
1.0
0.8
Magnitude
0.6
390
0.4
410
0.2 430
0.0
300 400 500
Frequency (Hz)
Fig. 17.22 Effect of resonator tuning on the Bode plot for the higher-frequency
range. Resonators with natural frequencies 390 and 430 Hz are compared with
the baseline (410 Hz).
1.0
0.8
Magnitude
D
C
0.6
B
0.4
A
0.2
0.0
300 400 500
Frequency (Hz)
Fig. 17.23 Effect of resonator position on the Bode plot for the higher-frequency
range. Axial positions A to D are shown in Fig. 17.19.
exercised to assess the uncertainty in combustion response at new frequencies that

may be excited by the addition of a resonator.
V. Conclusion
This chapter described various passive-control methods for stabilizing premixed
combustion in gas turbines. Feedback-control models reviewed in Sec. II showed
how both the flame-transfer function and the feedback acoustics are linked in a
dynamics problem. Nyquist analysis illustrated how multiple acoustic modes can
confound approaches to eliminate combustion instabilities by changing the con-
vective time lag. The physical processes affecting the flame response have been
reviewed, and some examples of the complex details of practical flame-transfer
functions were provided. In Sec. III, applications of time-lag modifications were
presented, and the limitations of these approaches were discussed. Experimen-
tal results demonstrating frequency shifts predicted by the Nyquist analysis were
shown. Furthermore, the stabilizing effect of time-lag distribution was discussed,
along with some field applications that confirmed the beneficial effect of increas-
ing the time-lag distributions. Improvements to stability from flame-anchoring
modifications, and pilot flames were also reviewed. In Sec. IV, a review of acous-
tic damping suggested that dampers can be used effectively to stabilize premixed
combustion. A few laboratory-scale and fielded-engine studies were reviewed,
showing excellent attenuation from the dampers. Example calculations showed
the various factors to consider in planning a damper addition and demonstrated
how damper performance can be predicted from Nyquist analysis.
It is interesting to speculate on the potential situations in which dynamics may
be a concern in future gas-turbine engine applications. It is often assumed that
dynamics problems are limited to stationary-gas turbines using premixed combus-
tion. However, two different integrated gasification combined-cycle power plants
recently suffered dynamics problems well into engine commissioning.77 The trend
for most advanced-power generators and aeroengine applications is to raise the op-
erating pressure to enhance efficiency. At the same time, it is desirable to reduce
the cooling and dilution flows to enhance performance or to reduce emissions. The
higher operating pressures release more heat in the same volume, increasing the
magnitude of heat-release rate perturbations. The reduction in dilution or cooling
flows also reduces the acoustic losses from the combustion liner. These combined
features raise the potential for dynamics. Thus, it seems likely that the prominence
of this problem will continue to be an issue. A combination of passive-control
strategies outlined here and emerging active-control strategies offer an opportu-
nity to mitigate these problems as new systems are designed.
Acknowledgments
This work was supported by the U.S. Department of Energy National En-
ergy Technology Laboratory. We acknowledge many helpful discussions with
colleagues at Virginia Polytechnic Institute and State University (Will Saunders,
Bill Baumann, and Uri Vandsburger) who encouraged the use of feedback-control
modeling.
Appendix
This Appendix documents the calculations used in the generation of Figs.
17.5–17.8. The acoustic computations are described first, followed by a deriva-
tion of the response of the system to a velocity source located at the flame and
incorporation of a flame model to yield the H transfer function. Note that
AB, C, D, E, F, G = Combustor sections and the acoustic transfer matrices that

represent these sections. Elements of these 2 × 2 matrices are
identified by using subscripts Ai, j where i = 0,1 and j = 0,1.
A “generic” combustor geometry is shown in Fig. 17.A1. The premix injector is

depicted by regions A and B, with air entering at the upstream side of region A and
fuel entering in the plane between regions A and B. The computations described
subsequently do not include the dynamics of air and fuel mixing explicitly, so
the injector region will be referred to jointly as AB. The fuel–air mixture enters
the combustion chamber at region D where it continues to flow downstream a
short distance before burning in the plane between regions D and E. In the two-
mode system described in this chapter, a step expansion is placed at the interface
between regions E and G. Computation stations, numbered 0 to 9, are also shown in
Fuel Flame
A B D E G
Combustion
Air products
0 12 34 56 9
Station #'s
Fig. 17.A1 Generic combustor geometry (not to scale).
Table 17.A1 Geometry and gas conditions used in the two-mode combustor model
Geometric region
Dual-mode combustor
simulation parameters A+B D E G
Diameter, m 0.06858 0.13335 0.13335 0.17780

Length, m 0.06350 0.01270 1.39700 0.63500
Area, m2 0.003694 0.013966 0.013966 0.024829
Temperature, K 533.3 533.3 1811.1 1811.3
Pressure, Pa 2020000 2020000 2020000 2020895
Density, kg/m3 13.21 13.21 3.89 3.89
Gamma (-) 1.4 1.4 1.3 1.3
MW, kg/kg-mol 29 29 29 29
c, m/s 462.7 462.7 821.6 821.6
U, m/s 28.87 7.64 25.93 14.58
Mach No. (-) 0.06240 0.01651 0.03156 0.01775
M-dot, kg/s 1.409 1.409 1.409 1.409
Std. Vol. Flow, scfh 150000 150000 150000 150000
Boundary conditions
pin p4 = p3
νin ν0 = 0
ρ4
pout p3 = p4 ν4 = ν3 + νs
ρ3
νout ν9 = 0
Fig. 17.A1. To simplify the calculations, “hard” acoustic boundaries were assumed
to exist at stations 0 and 9 (i.e., the acoustic velocity is zero). The specific geometry
and other conditions used in the calculations for the two-mode system are listed
in Table 17.A1.
The acoustic properties of the combustor model are described by the acoustic
impedances. Impedance is the complex ratio of acoustic pressure to mass velocity
at each station. The impedance at each station is computed through the use of
acoustic transfer matrices. These matrices of complex numbers relate the acoustic
pressure and velocity at one station to the next station. These relations are a function
of the geometry and the local-gas conditions. For example,

p1 p
=C 2 (17.A1)
ν1 ν2

p0 p1
= AB (17.A2)
ν0 ν1
and

p2 p3
=D (17.A3)
ν2 ν3
The acoustic pressures p and acoustic velocities ν are represented by complex

phasors. The 2 × 2 matrices AB, C, and D are acoustic transfer matrices for their
respective regions. The matrices C and F do not correspond to specific regions but
represent the acoustic transfer matrices for the step expansions that occur between
regions AB and D and regions E and G, respectively. The acoustic pressure and
velocity at a particular station are a linear function of the acoustic pressure and
velocity at the next station downstream. A similar linear relationship does not exist
between stations 3 and 4 because of the presence of the flame, which acts as an
acoustic source.
The 2 × 2 acoustic transfer matrices for various geometries are defined in
Munjal.28 For simple cylindrical elements:
 
cos 2π f L jc
· sin 2πcf L
 c s  µ 2π f
R(s, L , µ, f, c) =   · exp − j L
js
· sin 2π f L
cos 2πcf L c c
c c
(17.A4)
For step expansions such as the region between stations 1 and 2, the transfer matrix
is
−1  k Ma2 k Ma ζa

1 M b ζb 1− 1−Ma2 1−Ma2
S(ζa , ζb , Ma , Mb , k, γ ) := Mb
 
1 (γ −1)k Ma3 (γ −1)k Ma2
ζb
(1−Ma2 )ζa
1− 1−Ma2

1 Ma ζa
× Ma (17.A5)
ζa
1
where k, the loss coefficient for a step expansion, is

2
sb
k= −1 (17.A6)
sa
The acoustic impedances at each station in the model are calculated starting from
known boundary conditions. At station 0, ν0 = 0 such that
Z 0 = p0 /ν0 = ∞ (17.A7)
At station 1, ν0 = 0 = AB1,0 ∗ p1 + AB1,1 ∗ ν1 , which implies
Z 1 = p1 /ν1 = −AB1,1 /AB1,0 (17.A8)
At station 2,
1
p1 C0,0 p2 + C0,1 ν2 ν2 C0,0 Z 2 + C0,1
Z1 = = · = (17.A9)
ν1 C1,0 p2 + C1,1 ν2 1
ν2
C1,0 Z 2 + C1,1
therefore
−(Z 1 C1,1 − C0,1 )
Z2 = (17.A10)
(Z 1 C1,0 − C0,0 )
And, similarly at station 3,
−(Z 2 D1,1 − D0,1 )

Z3 = (17.A11)
(Z 2 D1,0 − D0,0 )
At station 9, ν9 = 0 such that:
Z 9 = p9 /ν9 = ∞ (17.A12)
At station 6,
p6 G0,0 p9 + G0,1 ν9 G0,0

Z6 = = = (17.A13)
ν6 G1,0 p9 + G1,1 ν9 G1,0
At station 5,
p5 F0,0 p6 + F0,1 ν6 F0,0 Z 6 + F0,1

Z5 = = = (17.A14)
ν5 F1,0 p6 + F1,1 ν6 F1,0 Z 6 + F1,1
And at station 4,
E0,0 Z 5 + E0,1
Z4 = (17.A15)
E1,0 Z 5 + E1,1
The flame is modeled as an acoustic velocity source located in the plane be-
tween stations 3 and 4. The boundary conditions at this interface are p3 = p4 and
ν4 = ν3 ∗ (ρ4 /ρ3 ) + νs . The transfer function relating the acoustic pressure to the
velocity source, that is,
p3
ψ= (17.A16)
νs
is derived in terms of system impedances, starting with the definition of the

impedance at station 4:
1
p4 p3 p3 p3 1
Z4 = = ρ4 = ρ4 · = ρ4 (17.A17)
ν4 ν3 ρ3 + νs ν3 ρ3 + νs 1
p3
1
Z3
· ρ3
+ 1
ψ
Therefore,
Z3 Z4
ψ= (17.A18)
Z 3 − ρρ43 · Z 4
The transfer function relating acoustic pressure to variations in heat release is

derived by relating heat release and an acoustic velocity source.25

p3 T4 q T4 p3
νs = = ρ4 − 1 µ3 s 3 + ρ4 − 1 µ3 s 3 (17.A19)
ψ T3 Q T3 P3
Equation 17.A19 provides the last link needed to derive H, the relative pressure
response to oscillations in heat release. Derivation of the H transfer function started
with equation A19 by separating terms involving p3 from ν3 and dividing through
by P3 to yield

p3 1 1 T4 q T4 µ3 s3
− ρ4 − 1 µ3 s 3 = ρ4 −1 (17.A20)
P3 ψ P3 T3 Q T3 P3
Therefore,
p3 /P3 {ρ4 [(T4 /T3 ) − 1] (µ3 s3 /P3 )}

H= = (17.A21)
q /Q ((1/ψ) − (1/P3 ) {ρ4 [(T4 /T3 ) − 1] µ3 s3 })
A simple form for transfer function G, the relative heat-release response to pres-
sure fluctuations, was assumed to compute the open-loop response function. The
function G is assumed to be a simple time lag. The time-lag model is shown here
with a gain of one half:
q /Q 1
G= = e−1 jwτ (17.A22)
p3 /P3 2
The gain and time lag of function G can be varied according to the situation being
modeled. The equations for G and H are used to compute the open-loop response
functions, G∗H, for the example Bode and Nyquist plots discussed in the body of
the paper.
Finally, for use in the discussion of acoustic dampers, the addition of a side-
branch resonator can be handled28 by adding one more acoustic matrix T that
includes the impedance of the resonator mouth Z r :

1 0
T = (17.A23)
1/Z r 1
This matrix is simply inserted at the interface between the cylindrical matrices
(A4) that comprise the geometry of the combustor. In this manner, the damper can
be easily added, or repositioned by including the matrix T in the multiplication of
matrices that define any of the secs. A, B, D, E, or G in Fig. 17.A1. In the example
calculations presented in the text, the resonator is added to the matrix defining
sec. E.
References
1
Hobson, D. E., Fackrell, J. E., and Hewitt, G., “Combustion Instabilities in Indus-
trial Gas Turbines - Measurements on Operating Plant and Thermoacoustic Modeling,”
Journal of Engineering for Gas Turbines and Power, Vol. 122, No. 3, 2000, pp. 420–
428.
2
Sholz, M. H., and Depietro, S. M., “Field Experience on DLN Typhoon Industrial Gas
Turbines,” American Society of Mechanical Engineers, Paper 97-GT-61, 1997.
3
Pandalai, R. P., and Mongia, H. C., “Combustion Instability Characteristics of Industrial
Engine Dry Low Emission Combustion Systems,” AIAA Paper 98-3379, 1998.
4
Richards, G. A., Straub, D. L., and Robey, E. H., “Passive Control of Combustion
Dynamics in Stationary Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5,
2003, pp. 795–810.
5
Chu, B. T. “On the Generation of Pressure Waves at a Plane Flame Front,” Proceedings
of the Combustion Institute, Pittsburgh, PA, Vol. 4., 1953, pp. 603–612.
6
Peracchio, A. A., and Proscia, W. M., “Nonlinear Heat-Release/Acoustic Model for
Thermoacoustic Instability in Lean Premixed Combustors,” Journal of Engineering for
Gas Turbines and Power, Vol. 121, No. 3, 1999, pp. 415–421.
7
Brookes, S. J., Cant, R. S., Dupere, I. D., and Dowling, A. P., “Computational Modeling
of Self-Excited Combustion Instabilities,” American Society of Mechanical Engineers,
Paper 2000-GT-0104, 2000.
8
Kruger, U., Huren, J., Hoffinan, S., Krebs, W., and Bohn. D., “Prediction of Thermoa-
coustic Instabilities with Focus on the Dynamic Flame Behavior for the 3A-Series Gas
Turbine of Siemens KWU,” American Society of Mechanical Engineers, Paper 99-GT-11,
1999.
9
Harrje, D. T., and Reardon, F. H., “Liquid Propellant Rocket Combustion Instability,”
NASA SP-194, 1971.
10
Arana, C. A., Sekar, B., and Mawid, M. A., “Determination of Thermoacoustic Re-
sponse in a Demonstrator Gas Turbine Engine,” Journal of Engineering for Gas Turbines
11
Fleifil, M., Annaswamy, A. M., Ghoneim, Z. A., and Ghoniem, A. F., “Response of a
Laminar Flame to Flow Oscillations: A Kinematic Model and Thermoacoustic Instability
Results,” Combustion and Flame, Vol. 106, 1996, pp. 487–510.
12
Combustion Instability in Lean Premixed Combustors,” Journal of Engineering for Gas
13
Richards, G. A., and Janus, M. C., “Characterization of Oscillations During Premix
Gas Turbine Combustion,” Journal of Engineering for Gas Turbines and Power, Vol. 120,
No. 2, 1998, pp. 294–302.
14
15
16
tuation and Its Effect on Heat-release During Unstable Combustion,” Proceedings of the
17
Mongia, R. K., Tomita, E., Hsu, F. K., Talbot, L., and Dibble, R. W., “Use of an Optical
Probe for Time-Resolved In Situ Measurement of Local Air-to-Fuel Ratio and Extent of Fuel
Mixing with Applications to Low NOx Emission in Premixed Gas Turbines,” Proceedings
18
Schadow, K. C., and Gutmark, E., “Combustion Instabilities Related to Vortex Shedding
Science, Vol. 18, 1992, pp. 117–132.
19
Smith, D. A., “An Experimental Study of Acoustically Excited, Vortex Driven Combus-
tion Instability Within a Rearward Facing Step Combustor,” Ph.D. Dissertation, California
Inst. of Technology, Pasadena, CA, 1985.
20
Sterling, J. D., and Zukoski, E. E., “Longitudinal Mode Combustion Instabilities in a
Dump Combustor,” AIAA Paper 87-0220, 1987.
21
Gutmark, E., Schadow, K. C., Sivasegaram, S., and Whitelaw, J. H., “Interaction Be-
tween Fluid-Dynamic and Acoustic Instabilities in Combusting Flows Within Ducts,” Com-
bustion Science and Technology, Vol. 79, 1991, pp. 161–166.
22
Poinsot, T. J., Trouve, A. C., Veynante, D. P., Candel, S. M., and Esposito, E. J., “Vortex-
Driven Acoustically Coupled Combustion Instabilities,” Journal of Fluid Mechanics,
Vol. 177, 1987, pp. 265–292.
23
Langhorne, P. J., “Reheat Buzz: An Acoustically Coupled Combustion Instability,
Part 1, Experiment,” Journal of Fluid Mechanics, Vol. 193, 1988, pp. 417–443.
24
Yu, K. H., Trouve, A., and Daily, J. W., “Low-Frequency Pressure Oscillations in a
Model Ramjet Combustor,” Journal of Fluid Mechanics, Vol. 232, 1991, pp. 47–72.
25
Schuermans, B. H., and Polifke, W., “Modeling Transfer Matrices of Premixed Flames
and Comparison with Experimental Results,” American Society of Mechanical Engineers,
Paper 99-GT-132, 1999.
26
Kendrick, D. W., Anderson, T. J., Sowa, W. A., and Snyder, T. S., “Acoustic Sensitivities
of Lean-Premixed Fuel Injectors in a Single Nozzle Rig,” Journal of Engineering for Gas
27
Pascheriet, C. O., Flohr, P., Knopfel, H., Geng, W., Steinbach, C., Stuber, P., Bengtsson,
K., and Gutmark, E., “Combustion Control by Extended EV Burner Fuel Lance,” American
Society of Mechanical Engineers, Paper GT-2002-30462, 2002.
28
Munjal, M. L., Acoustics of Ducts and Mufflers, Wiley, New York, 1987.
29
Fannin, C. A., Baumann, W. T., Saunders, W. R., Richards, G. A., and Straub, D. L.,
“Thermoacoustic Stability Analysis for Mutli-Port Fuel Injection in a Lean Premixed Com-
bustor,” AIAA Paper 2000-0711, 2000.
30
Phillips, C. L., and Harbor, R. D., Feedback Control Systems, 4th ed., Prentice–Hall,
Upper Saddle River, NJ, 2000.
31
Blackshear, P. L. “Driving Standing Waves by Heat Addition,” Proceedings of the
32
Merk, H. J. “An Analysis of Unstable Combustion of Premixed Gases,” Proceedings
of the Combustion Institute, Vol. 6, 1956, pp. 501–512.
33
Becker, R., and Gunther, R., “The Transfer Function of Premixed Turbulent Jet Flames,”
34
Baade, P. K., “Design Criteria and Models for Preventing Combustion Oscillations,”
Transactions of the American Society of Heating, Refrigerating and Air-Conditioning En-
gineers, ASHRAE 84, Part 1, 1978.
35
Mugridge, B. D., “Combustion Driven Oscillations,” Journal of Sound and Vibrations,
Vol. 70, No. 3, 1980, pp. 437–452.
36
Matsui, Y., “An Experimental Study on Pyro-Acoustic Amplification of Premixed
Laminar Flames,” Combustion and Flame, Vol. 43, 1981, pp. 199–209.
37
Ducruix, S., Durox, D., and Candel, S., “Theoretical and Experimental Determination
38
Hadvig, S., “Combustion Instability: System Analysis,” Journal of the Institute of Fuel,
Vol. 44, 1971, pp. 550–558.
39
Hegde, U. G., Reuter, D., Daniel, B. R., and Zinn, B. T., “Flame Driving of Longitudinal
Vol. 55, 1987, pp. 125–138.
40
Fieler, C. E., and Heidmann, M. F., “Dynamic Response of Gaseous Hydrogen
Flow Systems and Its Application to High-Frequency Combustion Instability,” NASA TN
D-4040, 1967.
41
Putnam, A. A., Combustion Driven Oscillations in Industry, Elsevier, New York, 1971.
42
Janardan, B. A., Daniel, B. R., and Zinn, B. T., “Driving of Combustion Oscillations
by Gaseous Propellant Injectors,” Proceedings of the Combustion Institute, Pittsburgh, PA,
Vol. 17, 1977, pp. 1353–1361.
43
Lieuwen, T., and Zinn, B. T., “Theoretical Investigation of Premixed Combustion In-
stability Mechanisms,” AIAA Paper 98-0641, 1998.
44
Richards, G. A., Straub, D. L., and Robey, E. H., “Dynamic Response of A Premix
Fuel Injector,” American Society of Mechanical Engineers, Paper 2001-GT-036, 2001.
45
Scarinci, T., and Freeman, C., “The Propagation of a Fuel-Air Ratio Disturbance in a
Simple Premixer and Its Influence on Pressure Wave Amplification,” American Society of
46
Sattlelmayer, T., “Influence of the Combustor Aerodynamics on Combustion Instabil-
ities From Equivalence Ratio Fluctuations,” American Society of Mechanical Engineers,
Paper 2000-GT-0082, 2000.
47
Anderson, T. J., Sowa, W. A., and Morford, S. A., “Dynamic Flame Structure in a Low
NOx Premix Combustor,” American Society of Mechanical Engineers, Paper 98-GT-568,
1998.
48
Paschereit, C. O., Polifke, W., Schuermns, B., and Mattson, O., “Measurement of
Transfer Matrices and Source Terms of Premixed Flames,” Journal of Engineering for Gas
49
Lawn, C. J., “Thermo-Acoustic Frequency Selection by Swirled Premixed Flames,”
50
Krebs, W., Hoffman, S., Prade, B., Lohrmann, M., and Buchner, H., “Thermoacoustic
Flame Response of Swirl-Stabilized Flames,” American Society of Mechanical Engineers,
Paper GT-2002-30065, 2002.
51
Khanna, V. K., “A Study of the Dynamics of Laminar and Turbulent Fully and Par-
tially Premixed Flames,” Ph.D. Dissertation, Virginia Polytechnic Inst. and State Univ.,
Blacksburg VA, 2001.
52
Khanna, V., Vandsburger, U., Saunders, W. R., and Baumann, W. T., “Dynamic Analysis
of Swirl Stabilized Turbulent Gaseous Flame,” American Society of Mechanical Engineers,
Paper GT-2002-30061, 2002.
53
Bohn, D., Deutsch, G., and Kruger, U., “Numerical Prediction of the Dynamic Behav-
ior of Turbulent Diffusion Flames,” Journal of Engineering for Gas Turbines and Power,
Vol. 120, No. 4, 1998, pp. 713–720.
54
Bohn, D., Li, Y., Matouschek, G., and Kruger, W., “Numerical Prediction of the Dy-
namic Behavior of Premixed Flames Using Systematically Reduced Multi-Step Reaction
Mechanisms,” American Society of Mechanical Engineers, Paper 97-GT-265, 1997.
55
Kruger, U., Hoffman, S., Krebs, W., Judith, H., Bohn, D., and Matouschek, G., “Influ-
ence of Turbulence on the Dynamic Behavior of Premixed Flames,” American Society of
56
Smith, C. E., and Leonard, A. D., “CFD Modeling of Combustion Instability in Pre-
mixed Axisymmetric Combustors,” American Society of Mechanical Engineers, Paper 97-
GT-305, 1997.
57
Zhu, M., Dowling, A. P., and Bray, K. N. C., “Flame Transfer Calculations for Com-
bustion Oscillations,” American Society of Mechanical Engineers, Paper 2001-GT-0374,
2001.
58
Walz, G., Krebs, W., Hoffmann, S., and Judith, H., “Detailed Analysis of the Acous-
tic Mode Shapes of an Annular Combustion Chamber,” American Society of Mechanical
Engineers, Paper 99-GT-113, 1999.
59
Cronemyr, P. J. M., Hulme, C. J., and Troger, C., “Coupled Acoustic-Structure Analysis
of an Annular DLE Combustor,” American Society of Mechanical Engineers, Paper 98-GT-
502, 1998.
60
Pankiewitz, C., and Sattelmayer, T., “Time Domain Simulation of Combustion Instabil-
ities in Annular Combustor,” American Society of Mechanical Engineers, Paper GT-2002-
30063, 2002.
61
Evesque, S., and Polifke, W., “Low-Order Acoustic Modeling for Annular Combustors:
Validation and Inclusion of Modal Coupling,” American Society of Mechanical Engineers,
Paper GT-2002-30064, 2002.
62
Krebs, W., Walz, G., Flohr, P., and Hoffmann, S., “Modal Analysis of Annular Com-
bustors: Effect of Burner Impedance,” American Society of Mechanical Engineers, Paper
2001-GT-0042, 2001.
63
Straub, D. L., and Richards, G. A., “Effect of Fuel Nozzle Configuration on Premix
Combustion Dynamics,” American Society of Mechanical Engineers, Paper 98-GT-492,
1998.
64
Hermsmeyer, H., Prade, B., Gruschka, U., Schmitz, U., Hoffmann, S., and Krebs, W.,
“V64.3A Gas Turbine Natural Gas Burner Development,” American Society of Mechanical
Engineers, Paper GT-2002-30106, 2002.
65
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism of Combustion
Instability in Lean Premixed Gas Turbine Combustors,” Journal of Engineering for Gas
66
Lovett, J. A., and Uznanski, K. T., “Prediction of Combustion Dynamics in a Staged
Premixed Combustor,” American Society of Mechanical Engineers, Paper GT-2002-30646,
2002.
67
Berenbrink, P., and Hoffmann, S. “Suppression of Dynamics Combustion Instabilities
GT-0079, 2000.
68
Richards, G. A., Straub, D. L., and Robey, E. H. “Control of Combustion Dynamics
Using Fuel System Impedance,” American Society of Mechanical Engineers, Paper GT-
2003-38521, 2003.
69
Lee, J. G., Kim, K., and Santavicca, D. A., “A Study of the Role of Equivalence Ratio
Fluctuations During Unstable Combustion in a Lean Premixed Combustor,” AIAA Paper
2002-4015, 2002.
70
Lovett, J. A. , “Bluffbody Flameholders for Low-Emission Gas Turbine Combustors,”
U.S. Patent No. 5,471,840, 1995.
71
Straub, D. L., Richards, G. A., Baumann, W. T., and Saunders, W. R., “Measurement of
Dynamics Flame Response in a Lean Premixed Single-Can Combustor,” American Society
of Mechanical Engineers, Paper 2001-GT-0038, 2001.
72
Laudien, E., Pongratz, R., Pierro, R., and Preclik, D., “Experimental Procedures Aiding
the Design of Acoustic Cavities,” Liquid Rocket Engine Combustion Instability, Progress
in Astronautics and Aeronautics, AIAA, Washington, DC, Vol. 169, 1995, pp. 377–399.
73
Lewis, G. D., and Garrison, G. D., “The Role of Acoustic Absorbers in Preventing
Combustion Instability,” AIAA Paper 71-699, 1971.
74
Gysling, D. L., Copeland, G. S., McCormick, D. C., and Proscia, W. M., “Combus-
tion System Damping Augmentation with Helmholtz Resonators,” American Society of
75
Bellucci, V., Paschereit, C. O., Flohr, P., and Magni, F., “On the Use of Helmholtz
Resonators for Damping Acoustic Pulsations in Industrial Gas Turbines,” American Society
of Mechanical Engineers, Paper 2001-0039, 2001.
76
Mitchell, C. E., “Analytical Models for Combustion Instability in Liquid Rocket En-
gine Combustion Instability,” Liquid Rocket Engine Combustion Instability, Progress in
Astronautics and Aeronautics, AIAA, Washington, DC, Vol. 169, 1995, pp. 403–430.
77
DeBiasi, V., “Gasification on Track to Turn Problem Fuels into Electricity and Power,”
Gas Turbine World, Nov–Dec, 1999, pp. 12–20.
Chapter 18
Factors Affecting the Control of Unstable Combustors
Jeffrey M. Cohen∗
Pratt & Whitney, East Hartford, Connecticut
and
Andrzej Banaszuk†
United Technologies Research Center, East Hartford, Connecticut
Nomenclature
G0 = transfer function representing combustor, fuel line, and valve
dynamics
Gc = controller transfer function
N = random-input describing function
S = sensitivity function
b = “on” level for solenoid valve
k = exponent determining rolloff of open-loop transfer function
outside control bandwidth
ω1 = performance bandwidth
ω2 = control bandwidth
ii = power-spectral density function of the input disturbance
pp = power-spectral density function of the combustor pressure
ε = required attenuation level for sensitivity function over
performance bandwidth
φ = fuel–air equivalence ratio
µ = mean fuel–air ratio
ω = frequency
ωb = lower-boundary control bandwidth
ωc = higher-boundary control bandwidth
σ = standard deviation of valve command
σr = real part of unstable pole
τ = delay
Copyright c 2005 by United Technologies Corporation. Published by the American Institute of

Aeronautics and Astronautics, Inc., with permission.
∗ Manager, Combustor and Augmentor Aerodynamics.
† Project Leader, Combustion Dynamics.
581
582 J. M. COHEN AND A. BANASZUK
I. Introduction
T HE lean, premixed combustor designs used in low-emissions industrial gas

turbines are often prone to combustion instabilities. Significant efforts to
suppress these instabilities by using active-control techniques have been reported.
Researchers at the United Technologies Research Center1,2 have demonstrated
up to 16-dB suppression of combustion instabilities in a full-scale single com-
bustor and a 6.5-dB attenuation in a three-nozzle sector combustor. Engineers at
Siemens/KWU3,4 have deployed an active instability-control system on a full-scale
engine, resulting in reductions in fluctuating pressure of as much as 17 dB. ABB
Alstom investigators5 have performed considerable work on a laboratory-scale
combustor, which yielded suppression levels of up to 12 dB. Other organizations
have demonstrated similar levels of success in other premixed combustors, in-
cluding the U.S. Department of Energy,6 Honeywell,7 and Westinghouse/Georgia
Institute of Technology.8
Based on these successes, this technology clearly holds promise as a means for
attenuating combustion instabilities. However, the factors that affect (and possibly
limit) the performance of active instability-control systems have not been fully
investigated. In particular, the achieved reduction of pressure oscillation varied
between experiments from 6 dB to 20 dB. Moreover, in some cases, the attenu-
ation of oscillations at the primary frequency was accompanied by excitation of
oscillations at other frequencies.1,9−11
A significant number of factors determine the effectiveness of an active
instability-control system. These factors can be categorized as follows: combustor
dynamics, actuation system, sensing, and controller/algorithm. A more detailed
discussion of these categories follows.
A. Combustor Dynamics
The character of the pressure oscillations to be controlled is important to the
effectiveness of the control system. This effectiveness includes the fundamental
issue of whether the system is dynamically unstable (i.e., in a limit cycle) or whether
it is linearly stable and driven by noise. Large, coherent pressure oscillations are
possible in either case. Another important factor is the mechanism (or mechanisms)
through which the unsteady heat release and pressure couple with each other. To
arrive at an effective control system, it is important to understand and prioritize
these mechanisms to be able to devise actuation schemes to interfere with them. If
multiple mechanisms are at play, it may be difficult to deal with more than one at
once. Some mechanisms may not lend themselves to practical actuation schemes or
may be better suited to passive-control approaches, such as acoustic resonators.12
Multiple-instability modes and the interaction between those modes (especially
nonlinear interactions) may also compromise the performance of a control system.
Further issues with how the combustor pressure oscillations behave as a function
of engine-operating conditions and engine transients will also affect controller
performance.
B. Actuation System
For the actuation system to be truly effective, it must interfere with the root-
cause physics that lie behind the instability. The most important consideration in
actuation-system design is developing an understanding of what these physics are.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 583
It is often difficult to sort out cause-and-effect relationships from measurements

made on an unstable system. A large number of potential coupling mechanisms
could be at work in any one system. Modeling and detailed experimentation are
often required to acquire this information. Once the intent of the actuator has
been specified (i.e., how it will interfere with the causal physics), the practical
considerations of the actuator and the actuation system must be addressed. In most
systems, the fuel flow rate has been controlled via some sort of fast-response valve.
This valve must be fast enough to modulate fuel flow at the required frequency
and, potentially, over a range of frequencies. This bandwidth requirement will
set specifications for the valve’s bandwidth and natural frequency bandwidth. In
addition, it must have a flow capacity that is consistent with the goals for the supply
pressure and flow rate of the system, but it still must represent the metering orifice
of the system. The interaction of the actuator with the controller is also an important
factor. In valves with hysteresis or nonlinear response, position feedback may be
required. Proportional control systems will require proportional valves with a wide
dynamic range of flow capacity. Although the valve is a crucial component of the
actuation system, the remainder of the system cannot be overlooked. The remainder
includes the plumbing system and the fuel-injection system. The plumbing system
acoustics can be important and can even be exploited6,13 for both liquid and gaseous
fuels. The capacitance of the fuel-delivery system can also have a severe effect on
the time variations in the delivery of fuel to the combustor. This severe effect is
especially true in gaseous-fuel systems but can also be a factor when air is trapped in
liquid-fuel systems. The fuel-injection technique may also affect actuator authority,
which includes atomization, mixing, and transport effects. These phenomena can
act to attenuate and/or delay the fuel pulse delivered by the actuator, which causes
the response at the flame front to be minimal. Multiple actuators may be required
in some systems. The placement of these actuators3 and their coordination is also
a significant element to consider.
C. Sensing
Although the technology for sensing combustion instabilities may be fairly
mature (e.g., high-response pressure transducers or optical chemiluminescence
measurements), the implementation of these sensors in a control system can be
critical to the system’s performance. Note that, Whereas it is relatively simple
to measure pressure or heat-release oscillations in an unstable system, it may
be much more difficult when a control system is successfully minimizing those
oscillations. Once again, a knowledge of the causal physics is important. Sensors
must be placed at the proper locations relative to the acoustic mode shape to
properly identify phase information and to maximize the signal-to-noise ratio.
Multiple sensors may also be required to identify or accommodate changes in
mode or mode shape. These sensors must be well matched in terms of both
amplitude and phase. Filtering, signal processing, and averaging may also be
necessary, depending on the nature of the sensed signal. Secondary sensing of the
actuation system or external parameters (fuel flow rate, inlet temperature, etc.)
may also be required to set control-system parameters.
D. Controller/Algorithm
Designing a control algorithm that processes a pressure sensor signal(s) and
sends a command signal a fuel valve(s) to quench the oscillations would be a routine
task if the combustor’s operating conditions were fixed; the transport delay were
small (much less than the acoustic period); the actuator authority and bandwidth
were adequate; and a model of pressure response to the fuel valve command
were available. In such case, the control algorithm would need to provide an
appropriate phase shift of the pressure signal. An appropriate phase shift could be
provided by using several approaches,2,3,9,11,14−16 including the time delay, lead
lag, linear quadratic regulator, H infinity (minimizing gain from disturbance energy
to pressure energy), H 2 (minimizing gain from disturbance energy to maximum
value of pressure), and observer-based controllers.
The first obstacle to model-based control design is lack of accurate physics-
based predictive models for combustion-system response. Therefore, the control
design typically utilizes either a model obtained from or calibrated with experimen-
tal data2,3,9,11,14,15 or an adaptive scheme17−20 to automatically tune the controller
parameters in a way that reduces pressure oscillations. In particular, an adaptive
scheme is needed as a preliminary control algorithm in the case in which obtaining
a data-based model is not practical. This situation would be true of an unstable in-
dustrial combustor operating in a regime where hardware damage is likely or when
operating conditions vary (as in power transients for industrial gas turbines) and
are subject to unknown disturbances (external temperature, power load changes).
Both fixed-parameter and adaptive-control approaches have their own limitations,
and will be discussed next.
As mentioned earlier, the level of suppression achieved with a fixed-parameter
controller varied between various experiments. We will explain, using methods of
control theory, how large transport delay (comparable with the acoustic period)
and limited actuator bandwidth reduce the achievable attenuation level of pressure
oscillations. In essence, in the presence of a large delay, attenuation of pressure os-
cillations at certain bands of frequencies is accompanied by the excitation of oscil-
lations in adjacent bands.15,21,22 Limited actuator bandwidth prevents the possibil-
ity of compensation for this problem in the control algorithm. The tradeoff between
the attenuation of oscillations in certain frequency bands and excitation in adjacent
bands is expressed in terms of a controller-independent lower bound15,21,23,24
on the function that shows maximum-pressure-magnitude magnification over the
excitation band caused by the controller. The lower bound is an increasing function
of the transport delay and a decreasing function of the actuator bandwidth.15,21
The limitations of adaptive-control algorithm performance include the limita-
tions of the fixed-parameter controllers with few extra limitations introduced by the
adaptation. First, unless the combustor transient timescale is an order of magnitude
slower than the adaptation timescale, the stability of any adaptive scheme cannot
be guaranteed.25−27 (For some ad hoc adaptive schemes, no stability guarantees
exist even under the assumption of the timescale separation.) Unfortunately, one
cannot arbitrarily decrease the control-parameter adaptation timescale to achieve
the timescale separation. One factor that limits the speed of adaptation, especially
in industrial applications, is the noise present in the pressure time traces,17 which
can be attributed to the response of the acoustic modes to random disturbances
(such as turbulence). This noise needs to be filtered out so that the control algo-
rithm can distinguish the pressure reaction to the control input from a response to
random disturbances. The presence of the noise filters necessarily slows down the
speed of adaptation because the time required to average out the effect of noise
is proportional to the noise-to-signal ratio.17,28 The reduced speed explains why
Burst disk
Bypass leg
Choked orifice
Choked plate
Heated, high venturi
pressure air
Main fuel Gas sampling

Pilot fuel
probe (6)
Premixing Combustor
fuel nozzle
Fig. 18.1 Schematic illustration of single-nozzle combustor flame tube. Test section
diameter = 15.2 cm. For clarity, only one of the six sampling probes is shown.
adaptive controllers demonstrated in laboratory-scale combustors (low noise) may

not perform as well in industrial (high-noise) settings. There are also tradeoffs be-
tween the performance, stability, and speed of adaptation of adaptive algorithms.28
A more in-depth examination of several of these factors is presented here, includ-
ing actuated fuel mixing, actuation time delay, and fundamental control limitations.
These factors were chosen based on an estimation of their criticality in the active-
control system development process. This assessment was made based on the
results of diagnostic experiments, detailed physical modeling, and reduced-order
dynamical modeling.
II. Description of the Combustor

Results from two different experimental combustors are addressed in subsequent
sections (Figs. 18.1 and 18.2): single-premixer flame tube with natural gas fuel
(4 MW) and three-premixer sector rig with liquid diesel fuel (12 MW).
The experiments were performed across a wide range of equivalence ratios
and at inlet temperatures and pressures corresponding with real engine-operating
Emissions Probe
Field of View
Air Supply
Premixer (3) Combustion
Products
Fiber Optic
Optical Emissions
Probe (3)
Pressure Sensors (2)

Actuator
Pressure Case
Fuel
Solenoid Valve (3)
Fig. 18.2 Cross section of a three-premixer sector combustor test facility with an
instrumentation and actuation system.
Liquid Fuel TangentialAir Inlet Slot

Manifold
Liquid Fuel Axial
Liquid Fuel Spokes (6)
Supply
Centerbody Premixed
Fuel / Air
TangentialAir Inlet Slot
Fig. 18.3 Schematic illustration of premixing fuel nozzle. Top picture shows tangen-
tial air scrolls and gaseous-fuel-injection scheme. Bottom picture shows nozzle cross
section with a liquid-fuel-injection scheme.
conditions (nominally 710 K and 1.5 MPa, respectively). The experiments used
similar embodiments of the same engine-scale premixing nozzles. The three-nozzle
sector rig used a 60-deg-arc sector of the engine combustor liner with convec-
tively cooled sidewalls. The combustor rigs are discussed in detail in Refs. 1, 2,
and 29.
The premixing fuel injector used in these combustors has been described in
detail by Stufflebeam et al.30 The fuel nozzle is shown schematically in Fig. 18.3.
Air was delivered into the premixing chamber through two tangentially oriented air
slots that ran the entire axial length of the chamber. Natural gas fuel was injected
through a row of orifices in the inlet section to each of these air slots. Fuel–air
mixing was measured and optimized for a low-emissions operation, as described
by Stufflebeam et al.30 The liquid-fuel version of the injector used a series of six
axial spokes to atomize and inject fuel in the interior of the premixer.
The instability mode (∼200 Hz) to which control systems were applied was a
Helmholtz mode (n = 0), in which the fluctuating pressure was uniform within
the combustor. Pressure fluctuations were coupled with the heat-release process
through their effect on the flow rate of air delivered through the premixer. The
time-varying air-flow rate produced a time-varying equivalence ratio at the fuel
nozzle exit and, therefore, a time-varying heat-release rate. This conceptual model
of the instability was discussed in more detail by Peracchio and Proscia31 and the
phenomenon has been described by many other authors.5,6,32 Fuel-concentration
measurements performed by Lee and Anderson33 in this combustor confirmed this
link between equivalence-ratio fluctuations and pressure fluctuations. Figure 18.4
shows the spectrum of the fluctuating pressure for this instability, as observed in
the single-nozzle (flame-tube) version of the combustor.
The fundamental problem of control-system design for low-emissions combus-
tors is to maximize the system’s authority over the relevant dynamic processes
while minimizing the combustor’s emissions and ensuring this performance over a
range of operating conditions. The active-control system consisted of three parts: a
pressure sensor, a control algorithm, and an actuator. Because of the uniform spatial
distribution of the fluctuating pressure within the combustor, only one combus-
tor pressure measurement was required to describe the unsteady pressure field as
input to the control system. In both experiments, on/off actuation of a portion of
the fuel flow to the premixer was performed. A closed-loop control algorithm was
developed to use the actuators’ authority to damp combustor pressure oscillations.
14
12
10
Amplitude (psi)
0
0 200 400 600 800 1000
Frequency (Hz)
Fig. 18.4 Spectrum of uncontrolled combustion instability, showing high-amplitude

pressure fluctuations at approximately 200 Hz.
The control algorithm chosen consisted of a frequency-tracking observer imple-

mented in software, which identified the frequency and in-phase and quadrature
components of the combustor pressure oscillations from the high-response pres-
sure signal. The phase-shifted pressure-oscillation signal was then fed back to the
on/off control valve.
III. Actuated Fuel Mixing

Actuation technology is often identified as a critical-path item for the product
deployment of active instability-control systems in real engines. Most successful
efforts in this area have used modulation of some sort of fuel flow as an actua-
tion technique. These techniques have used existing fuel-system components or
have added secondary fuel injectors. The obvious actuation technology barrier is
represented by the ability to modulate large fuel flows at the high frequencies at
which combustion instabilities occur (normally hundreds of hertz in gas turbines).
In some cases, the acoustics of the fuel system have been tuned to compensate
for poor actuator performance in the frequency range of interest.13 A second, less
obvious barrier that relates to actuation is the physics of the actuation process. For
an actuator to be effective, it must have the ability to interfere with the coupling
process between the acoustic pressure field and the combustion heat-release rate.
An actuator that modulates fuel flow may not have significant authority (effec-
tiveness) if the fuel is not injected in a manner that allows it to interfere with the
coupling process.
An illustration of the intent of the actuation scheme used in this effort is shown
in Fig. 18.5. The controlled fuel-injection system was used to reduce variations in
the fuel–air ratio at the exit of the premixer by introducing a modulated fuel flow
into it. The goal of the system was to keep the fuel–air ratio being delivered to
the combustor uniform in both time and space. Temporal uniformity inhibits the
development of combustion instabilities, and spatial uniformity inhibits the pro-
duction of nitrousoxide (NOx). Complete cancellation of fuel–air variations would
Uncontrolled Controlled
Combustor
pressure
Air flow rate
Fuel flow rate
Premixed
equivalence
ratio
Time
Fig. 18.5 Illustration of the intent of the actuation technique, utilizing the reduction
of equivalence-ratio fluctuations by pulsed fuel injection.
likely require on-line adaptation of the amplitude of the fuel flow modulation. Only
fixed-amplitude (on/off) actuation was used in this study.
The gas–fuel premixer was modified to incorporate four different actuated fuel-
injection configurations, as shown in Figs. 18.6 and 18.7. These injection config-
urations were designed specifically to provide different levels of mixing between
the actuated fuel flow and the remainder of the premixed reactants. Three of the
configurations used an axially oriented spoke mounted on the premixer centerbody.
The length of the spoke and the number of injection sites were varied to modify
mixing. The fourth configuration modulated the flow through two of the injection
orifices in the main fuel-injection array at the inlet to the air scroll. In all cases,
the level of actuated fuel flow was held constant.
A. Nonreacting Injector Evaluation

A nonreacting acetone PLIF (planar laser-induced fluorescence) technique15
was used to assess the steady-state-mixing features of each injection concept.
The fuel flow was not modulated during these tests. In each of the tests, 10% of
the total fuel simulant injected into the premixer was passed through the control
fuel-injection system. In all cases, controlled fuel flow was injected inside the
premixing nozzle, though in different fashions. The results of these tests are shown
in Fig. 18.8. The concentration at the exit of the premixing nozzle is shown for
each configuration. For three injectors, the nonuniformity created by the localized
injection can be observed in the concentration profiles as a locally richer spot.
Figure 18.8 also shows the spatially averaged unmixedness, as represented by the
ratio of the standard deviation of the concentration distribution (σ ) to the mean
concentration value (µ), for each configuration.
Three of the injection concepts (configurations 1–3) demonstrated poorer
steady-state mixing than the baseline premixer. For those configurations, mix-
ing improved as the fuel simulant was injected through a larger number of sites
over a larger area. Injection through the original orifice array gave the best mix-
ing, which is consistent with the fact that this array had been optimized for good
mixing.
Phase-locked PLIF measurements29 were made for the best-mixing configura-
tion, which used the main orifice array (configuration 4). A high-speed on/off valve
in the control fuel system was driven at 200 Hz. The PLIF-imaging camera was syn-
chronized with the valve actuation, and five different time–phase delays were used
over the full cycle of injection at 200 Hz. At each delay setting, 1800 images were
acquired and averaged together to generate a representative, ensemble-averaged
image of the concentration distribution at the exit of the premixer through a cy-
cle of actuation. Figure 18.9 shows these results. These results demonstrate that
the spatial concentration distribution changed only slightly over the period of one
cycle. Figure 18.10 shows how the spatially averaged concentration (fuel–air ra-
tio) changed over one cycle. The unmixedness (σ/µ = standard deviation/mean)
changed over the range of 3.6% to 7.9% during the cycle, indicating that reasonably
good mixing was being maintained throughout. The average acetone concentra-
tion changed by ±7% during a cycle, compared with the command variation of
±10%. These tests demonstrated the ability of injector configuration 4 to control
Fig. 18.6 Cross section schematic illustration of the premixing fuel nozzle with an
actuated fuel-injection spoke (configurations 1–3, top) and actuated portion of the
main injection array (configuration 4, bottom) for introduction of controlled fuel flow.
Fig. 18.7 Detailed view of three-spoke injection configurations.
the temporal character of the fuel–air ratio at the exit of the premixer without
excessively degrading the spatial fuel–air mixing at any point in time.
B. Combusting Injector Evaluation

Each of the control fuel-injection concepts was tested in the gas-fueled, single-
nozzle, flame-tube combustor under controlled and uncontrolled conditions. Dur-
ing uncontrolled operation, fuel was delivered through the control fuel circuit at
the same mean flow rate as used in the controlled tests. Use of the same mean
Fig. 18.8 Steady-state fuel–air concentration profiles at premixer exit for four fuel-
injection concepts. Concentration values have been normalized by the mean value for
each case.
Fig. 18.9 Phase-averaged fuel–air concentration profiles and unmixedness at pre-

mixer exit for 200-Hz modulation of fuel injection by using configuration 4.
flow rate was necessary to truly isolate the effect of the unsteady aspect of the fuel
injection.
Three of the poorer-mixing injection concepts showed increased NOx and CO
levels (relative to baseline levels) during operation with steady, nonmodulated flow
through the control fuel system. These levels did not change appreciably when
control was applied (for the same mean flow through the control fuel system). In
general, the level of NOx increase that was observed correlated with the degree
of unmixedness that was observed. Emissions were measured by using a ganged
array of six water-cooled sampling probes near the exit of the combustor.
For each of these injection configurations (1–3), a piloting effect on the instabil-
ity was also observed. When a constant fuel flow was delivered through the control
fuel system, the local enriching resulted in an attenuation of the pressure fluctua-
tions. This behavior is similar to that observed when a diffusion-flame pilot was
Fig. 18.10 Variation of spatially averaged fuel–air ratio at a premixer exit over one
period (5 ms) of 200-Hz fuel flow modulation for configuration 4. Bars indicate spatial
variance with about average value at each time.
applied to the system. When piloted, the pressure-fluctuation levels were reduced
but at the expense of higher NOx emissions.
Configuration 4, in which the control fuel was delivered through a portion of
the main fuel-injection array, showed no difference in either pressure fluctuations
or emissions between the baseline case and the case in which steady fuel flow was
delivered through the control fuel system (as expected).
Figure 18.11 shows that the level of reduction in the amplitude of the dominant
instability mode correlated with the mixing performance when amplitude was
18
Reduction in amplitude of instability (dB)
16 Configuration 4
14
12
Configuration 2
10
Configuration 3
8
4
Configuration 1
2
0 0.05 0.1 0.15 0.2
Unmixedness
Fig. 18.11 The effect of steady-state fuel–air mixing on actuator authority in com-
bustion tests for four different actuated fuel-injection configurations.
measured at the optimal control-phase delay for each configuration. Better-mixing

injection schemes provided significantly better actuation authority, with configu-
ration 4 producing a 16-dB (6.3 times) reduction. The broadband rms pressure-
fluctuation level in the 0- to 2-kHz band was reduced by a factor of 2.4 with this
control.
Both NOx and CO emissions were improved under controlled conditions for
configuration 4. NOx emissions were reduced by 27% and CO emissions were
reduced by 54%. This trend is consistent with observations made by Cohen et al.2
It is most likely that the control system, by reducing equivalence-ratio fluctuations,
also reduced temporal hot spots and cold spots, thus decreasing the production of
NOx and CO, respectively. These effects are substantial because of the highly
nonlinear relationship between the pollutant-emission production rates and flame
temperature at these low equivalence ratios.
IV. Actuation Time Delay

Another consideration that affects the authority of actuation systems is the time
delay between when the actuator moves and when the effect of that actuation
is observed at the flame front. This time delay was investigated by using both
the liquid-fueled three-nozzle sector combustor1 and the single-nozzle flame-tube
combustor. Semiempirical dynamical models of the system were developed to
explain the origin of these effects.
A. Experimental Observations
Speculating that the effectiveness of the controller could be increased with the
additional actuator authority produced by actuating more nozzles, the control sys-
tem was tested with multiple, simultaneously actuated fuel nozzles. Figure 18.12
shows power-spectral density (PSD) plots for phase-optimized (minimum pressure
oscillations) single, dual, and triple closed-loop, controlled fuel nozzles. Com-
bustor pressure oscillations were reduced by going from single- to dual-nozzle
actuation, but no further reduction was obtained by actuating all three fuel noz-
zles, despite the open-loop forcing results. The best control was achieved with
dual-nozzle actuation, yielding a 6.5 dB (2.1 times or 53%) reduction in the bulk
mode pressures and a 25% reduction in broadband rms pressure. These reductions
in combustor pressure oscillations via active control were accompanied with no
penalties to emissions compared with the uncontrolled operation. The magnitude
of the reduction was limited by the splitting of the spectral peak into two smaller
peaks. This splitting behavior was evident for both two- and three-nozzle actu-
ation, but the amplitude of the secondary peaks was larger for the three-nozzle
case.
In later experiments with the gas-fueled, single-nozzle flame-tube combustor,
the length of tubing between the actuation fuel valve and the fuel-injection location
was changed from 2.8 cm (1.1 in.) to 15.2 cm (6 in.) and 45.7 cm (18 in.). The ef-
fect of these variations on the control-system performance is shown in Fig. 18.13.
The negative effects of peak splitting increased as the fuel line was made longer.
Although no direct measurements of time delay were made for the different con-
figurations, it is reasonable to assume that increasing the fuel-line length increased
1.6
uncontrolled
1.4 single nozzle
no control
dual nozzle
Combustor Pressure (psi)
1.2
triple nozzle
1
single nozzle triple nozzle
0.8 dual nozzle
0.6 triple nozzle
dual nozzle
0.4
0.2
0 50 100 150 200 250 300 350 400 450 500
Frequency, Hz
Fig. 18.12 Multiple-nozzle, closed-loop actuation led to relatively small incremental
reductions in pressure-fluctuation levels because of the peak-splitting phenomenon.
the time delay of the actuation system. This phenomenon has been experimentally
observed by other investigators, as well.7
B. Dynamic Interpretation
The explanation for this phenomenon traces back to the fundamental nature
of the pressure oscillations observed in the combustor. Pressure oscillations in a
combustor dominated by a narrow-frequency band can be interpreted by using
a limit-cycling model or a stable, noise-driven model. Most references attribute
pressure oscillations in combustors to self-excitation of coupled acoustics and
heat-release systems, resulting in a limit-cycling behavior. However, it will be
shown that the uncontrolled-sector combustor behavior and the splitting of the
bulk-mode peak observed during controlled operation can be better explained
with a model of the combustor as a lightly damped, linearly stable system driven
by noise attributed to turbulence. Note that important differences exist between
behavior observed in small laboratory combustion-control experiments and full-
scale industrial combustors. Laboratory combustors typically have no liner and
thus have lower damping than industrial combustors. At the same time, laboratory
combustors may have lower turbulence levels than larger, more complex devices.
With low-damping and low-noise levels, it is likely that significant pressure oscil-
lations will only occur because of self-excited limit-cycle oscillations.14 Industrial
combustors can exhibit noticeable pressure oscillations in a stable, noise-driven
regime; hence, a self-excited model is, in many cases, not necessary. In this sense
the term “combustion instability” is less appropriate for this analysis, because it
will use a stable model of the sector rig and add a driving broadband stochastic
2.5
1-in.; 3.6x
in.;3.6x
1.11-in.; 3.6X
Pressure (psi)
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
2.5
6 6-in.;
in.; 2.6X
6-in.; 2.6x
2.6x
Pressure (psi)
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
2.5
1818-in.;
in.; 1.7X
18-in.; 1.7x
1.7x
Pressure (psi)
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 18.13 Combustor pressure spectra showing the reductions in peak amplitude of
optimum-phase control for three different lengths of fuel manifold. Results indicate
a reduction in control system authority because of the appearance of dual peaks at
longer tubing lengths. Solid lines represent controlled cases. Dashed lines represent
uncontrolled cases.
Bode Plot of Combustor Pressure over Valve Command Signal

0.025
Measured Response
Response
0.02 Overall Fit
Fit Response dominated
dominatedby
by
Magnitude
nd
0.015 ì 2nd
“2 Order”
Orderî system
system
0.01
0.005
0
100 120 140 160 180 200 220 240 260 280 300
Frequency, Hz Measured Response:
Response:
500
time delay
time delay factored
factoredout
out
nd
ì“22nd Order”
Orderî fit
Phase
-500
Phase Response
dominated by delay
-1000
100 120 140 160 180 200 220 240 260 280 300
Frequency, Hz
Fig. 18.14 Bode plot of combustor pressure over valve-command signal with no
control.
disturbance to account for the observed pressure oscillations. Of course, in some

cases, a self-excited model of pressure oscillations in industrial combustors will be
more appropriate than a stable-driven model.2,3 Note, though, that large combustor
pressure oscillations are possible in both scenarios.
Experimentally determined frequency responses of the combustor pressure to the
valve-actuation voltage (Fig. 18.14) closely resembled responses typical of linear
systems with time delay. A second-order model with delay was used to fit these
dynamics with good agreement in the frequency range of 150–400 Hz. In principle,
these empirical fits are necessary but not sufficient to conclude that the combustor
dynamics were in fact linearly stable (as opposed to perhaps limit-cycling behavior)
because a limit-cycling system may produce a frequency response resembling that
of a stable driven system. However, there are several additional arguments that a
driven stable system is indeed a good model of the observed behavior. First, the
uncontrolled pressure PSD can be closely matched by using a noise-driven stable
model. Second, Fig. 18.15 shows that the probability distribution of pressure from
experiments is well approximated by Gaussian distribution, which is a typical
distribution of an output of a linear system driven by Gaussian input. A noise-driven
limit-cycling system would show a double-hump distribution22,34,35 of pressure.
Last, it will be shown that the stable, driven model reproduces the peak-splitting
effect of the controller in simulations with encouraging fidelity to data.
The transfer function of combustor pressure to valve-command voltage was
measured via open-loop swept-sine tests actuating one of the three nozzles. The
first step in fitting a measured transfer function was to identify the time delay from
the slope of the phase in the frequency range of 220–260 Hz. Next, the phase
800
700
Number of samples
600
500
400
300
200
100
0
-6 -4 -2 0 2 4 6
Pressure (psi)
Fig. 18.15 Distribution of 20,000 samples of uncontrolled, unsteady combustor pres-
sure (thin line) and a fit with a Gaussian distribution (thick line).
lag caused by delay was subtracted from the experimental phase lag, yielding a
nearly classical second-order response with the phase dropping 180 deg through
the magnitude-response peak. A stable second-order transfer function with two
poles and one zero was fitted numerically.
A schematic of the closed-loop simulation block diagram is shown in Fig. 18.16.
The plant G 0 ( jω) is the empirical second-order system with delay representing
the combustor dynamics. With the controller off, the standard deviation of the
white Gaussian noise was adjusted in the simulation to match the PSD of the
experimental pressure data. A likely physical source of the driving disturbance
was the turbulent flow fluctuations driving the acoustic mode directly or through
the heat-release process.
The effect of multiple-nozzle actuation was simulated by linearly scaling the
controller output by the number of nozzles, which was consistent with experimental
open-loop forcing results. Figure 18.17 shows that the simulation exhibited a peak-
splitting phenomenon similar to the phenomena observed in the experiments. The
amplitude at the dominant oscillation frequency was attenuated, whereas secondary
Plant
Noise + Pressure
Go((jω)
jω)
Fuel flowl
-
modulation
b Gc((jω)
jω)
-b
On/off valve Controller
Fig. 18.16 Schematic illustration of closed-loop combustor simulation block diagram.

Noise is represented here as random fluctuations of fuel–air ratio.
Fig. 18.17 Second-order model of combustor with delay reproduces peak-splitting

phenomenon (Fig. 18.12) in closed-loop simulation.
peaks were amplified by actuating more nozzles and therefore more authority. The
slight asymmetry of peaks after the third valve was turned on in the experiment
can be attributed to a different phase lag of the third actuation system relative to
the first two systems. In fact, this asymmetry was reproduced in the model by
assigning a larger delay to valve 3 than to valves 1 and 2 in the model.
Even though a linear, stable system driven by a white Gaussian disturbance
was a good model of the sector combustor during the experiments, the use of
on/off valves for control made the closed-loop system strongly nonlinear. Thus,
an analysis of peak splitting by using linear control-theory tools14 may not seem
immediately relevant. However, Banaszuk and colleagues21,22 have argued that
a quasi-linear analysis using random input-describing functions is appropriate to
study the nonlinear dynamics of the combustion model with on/off valves in the
presence of large-amplitude noise, as with this closed-loop model. In this tech-
nique, the signals in the model were approximated as sums of constant, sinusoidal,
and random components with Gaussian distribution. The static nonlinear elements
were replaced with equivalent gains called random input-describing functions.36
The values of constant components, amplitudes, and frequencies of sinusoidal
components and standard deviations of Gaussian components in the system can
be found by solving a system of nonlinear equations.
It can be shown that the system of Fig. 18.16 with the identified standard devia-
tion of Gaussian input has low effective gain of the on/off valve for the sinusoidal
signal so that a limit-cycle oscillation cannot be sustained. Therefore, only the
balance of Gaussian processes in the loop has to be carried out to approximately

predict the PSD of the combustor pressure under closed-loop control. The random
input-describing function for the on/off valve with the “on” level denoted by b is

2 b
N (σ ) = (18.1)
πσ
where σ is the standard deviation of the Gaussian process at the input of the valve.
One can shown that, given the noise-input PSD ii ( jω), σ can be found from the
Gaussian process balance equation:

2
1 ∞ G 0 ( jω)G c ( jω)

σ = 2π 1 + √(2/π )(b/σ ) G 0 ( jω)G c ( jω) ii ( jω) dω (18.2)
−∞
which can be solved numerically. Once the value of σ is known, the pressure PSD
pp ( jω) can be obtained from the formula
pp ( jω) = G 0 ( jω)/1 + N (σ )G 0 ( jω)G c ( jω)ii ( jω) (18.3)
The closed-loop transfer from noise to pressure is G 0 ( jω)/[1 + G 0 ( jω)G C

( jω)N (σ )], indicating that for |1 + G 0 ( jω)G C ( jω)N (σ )| < 1, pressure oscil-
lations are amplified by the controller, and for |1 + G 0 ( jω)G C ( jω)N (σ )| > 1,
pressure oscillations are attenuated by the controller.
Figure 18.18 shows the Nyquist plot of G 0 ( jω)G C ( jω)N (σ ) in the complex
plane for the value of b corresponding to three valves. [For this plot G 0 ( jω)G C ( jω)
and σ were obtained from model simulation rather than calculation, mainly because
Fig. 18.18 Nyquist diagram for single-nozzle, closed-loop control near the optimum-
control phase, showing that the controller excited secondary peaks (B and C) and
attenuated the primary peak (A).
the controller G C ( jω) used in the experiments and in the simulation was a nonlinear
phase-shifting controller based on a frequency-tracking extended kalman filter,
which does not have a simple linear-transfer function, even though for a fixed
central frequency of oscillations it can be closely approximated by a linear-transfer
function.]
By analyzing the Nyquist diagram shown in Fig. 18.18, we conclude that the
controller amplifies certain frequency bands while attenuating the pressure oscil-
lations at the frequency band centered at about 208 Hz (peak A in Fig. 18.17). The
two nearly symmetric branches of the Nyquist plot that cross into the unit-radius
circle, centered at (−1, 0), for frequencies greater than 225 Hz and less than 195
Hz were the root cause of the secondary peaks B and C in Fig. 18.17. They arise
because of presence of large delay in the combustor-transfer function causes signif-
icant rolloff of the phase of the open-loop transfer function G 0 ( jω)G C ( jω)N (σ ).
Although adding more nozzles and therefore more actuator authority increased
the control gain in the attenuation band, it also increased the control gain in the
excitation band, imposing a limit on the phase-shifting controller’s effectiveness.
Note, however, that increasing the number of actuated nozzles does not correspond
to a proportional gain increase, because the standard deviation σ of the Gaussian
process at the input of the valve is a function of b.
V. Fundamental Limitations of Achievable Performance

In the preceding section, we showed that increasing delay between the fuel-
control command and its effect on the combustion process reduces attenuation of
the pressure oscillations in a combustor. A natural question arises about whether
the increase of the delay could be compensated by a choice of the control algorithm.
Banaszuk and colleagues21,22 showed (by using the approach of Freudenberg and
Iooze23 ) that one cannot arbitrarily decrease the level of pressure oscillations by
using linear controllers. In this section, we review these results. The derivation
of fundamental limitation is provided for the case of a linear combustor response
and linear controller-transfer function. Extension to the nonlinear actuator case is
discussed at the end of this section.
Recall that the combustor pressure PSD pp ( jω) can be obtained from the
formula
pp ( jω) = |G 0 ( jω)S( jω)|2 ii ( jω) (18.4)
where
1
S( jω) := (18.5)
1 + G 0 ( jω)G c ( jω)
is called the sensitivity function. Note that the square of the sentitivity function is
the factor by which the pressure PSD in the combustor is reduced (or amplified)
at any given frequency. The objective for active control of combustion is to shape
the sensitivity function so that it is small at and near the resonant frequency ωr of
the combustor. This requirement can be stated as
|S( jω)| < ε for ω ∈ ω1 (18.6)
where ω1 is the so-called performance bandwidth, that is, the interval containing
the resonant frequency ωr over which reduction of pressure oscillations by the fac-
tor of ε relative to uncontrolled level is enforced. The fundamental limitations23,24
yield controller-independent lower bounds on the maximum of the sensitivity func-
tion. Assume that the combustor response-transfer function G 0 ( jω) has, at most,
one unstable complex conjugate pole pair with the real part denoted by σr . If
the combustor model is stable, we define σr = 0. An example of fundamental
limitations is the Bode integral formula for the sensitivity function:
∞
ln |S( jω)|dω = 2πσr (18.7)
0
This equation shows that the negative area under the logarithm of the abso-
lute value of the sensitivity function (corresponding to attenuation of pressure
oscillations relative to an uncontrolled combustor) in one frequency band must be
accompanied by a positive area (amplification of pressure oscillations) in some
other band. If the control bandwidth is infinite, the positive area may be distributed
over a wide frequency range so that amplification at any given frequency may be
designed to be arbitrarily small. However, if the control bandwidth is finite be-
cause of factors such as actuator bandwidth [so that G 0 ( jω)G C ( jω) is close to
zero beyond certain low and high frequencies], the positive area would have to be
accommodated in a smaller band (where loop gain is high), and this accomoda-
tion would necessarily result in peaking of the sensitivity function. If the peaking
occurs in the region in which the combustor response-transfer function has a non-
vanishing gain, the peaking in the sensitivity function will result in a peak splitting
in the closed-loop response. Figure 18.19 and 18.20 illustrate this phenomenon.
2.5
Combustor Pressure (psi)
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 18.19 Peak-splitting phenomenon, showing sidebands on either side of uncon-

trolled peak during controlled operation. Solid line represents controlled case. Dashed
line represents uncontrolled case.
8
6
4
2
+ +
dB 0
-
-2
-4
-6
-8
-10
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 18.20 A typical sensitivity function, showing the sensitivity tradeoffs caused by
finite-controller performance bandwidth.
Assume that the combustor-transfer function G 0 ( jω) is a relative degree of

at least two (i.e., it has at least two more poles than zeros). This assumption is
typically satisfied if actuator and sensor dynamics are included in the combustor
response-transfer function. To model the effect due to finite control bandwidth, we
require the open-loop gain to satisfy the inequality:
ω 1+k
c
|G 0 ( jω)G c ( jω)| ≤ δ for ω > ωc (18.8)
ω
Here, it is assumed that δ < 1/2 and k > 0 (relative degree of at least two). We
impose a similar constraint on the loop gain:
1+k
ω
|G 0 ( jω)G c ( jω)| ≤ δ for ω < ωb (18.9)
ωb
Let us define the control bandwidth ω2 := ωc − ωb . Figure 18.21 illustrates

the finite-bandwidth performance specification with the performance and control
bandwidth. The restrictions of the loop gain at high and low frequencies impose
additional constraints on the sensitivity function. Now, in addition to the perfor-
mance specification |S( jω)| < ε for ω ∈ ω1 , a control bandwidth specification
must also be met.
We now compute the performance limitations as peaking in the sensitivity-
function magnitude. Let S∞ := Supω |S( jω)| denote the so-called H infinity
norm of the sensitivity function. Note that S∞ is the supremum over all frequen-
cies of the amplification of the pressure oscillations by the control system relative
to the uncontrolled response, that is, a measure of control-induced peaking. For
the finite-control bandwidth case, the area formula together with the constraints
shown in Fig. 21 and high-frequency rolloff characteristics can be manipulated
(Control bandwidth)
|S(j
|S(jω)| ∆ω2
1
Loop gain small at Loop gain rolls off
low frequencies at high frequencies
ε
ωb ωr ωc Frequency
0
∆ω 1
(Performance bandwidth)
Fig. 18.21 Illustration of performance and control bandwidth.
(as in Freudenberg and Iooze23 ) to show that

1
log S∞ ≥
ω2 − ω1

1 1 3δωc 1
2π σr + ω1 log − ωb log − 1− (18.10)
ε 1−δ 2k (1 + π/τ ωc )
The preceding formula shows that the factors that bound the supremum of the
sensitivity function from below are as follows:
1) The desired performance. This is represented by the product ω1 log(1/ε).

2) The limitation on the actuator bandwidth relative to required performance
bandwidth, which is represented by the amplifying term 1/(ω2 − ω1 ).
3) The real part of the unstable combustor pole. This influence is represented
by the term 2π σr . The larger the growth rate of the pole, the larger the peak of the
sensitivity function.
5
4.5 ∆ω1 = 40 Η z
||S|| 4 δ = 0.1 (20 dB)
8
3.5 σr= 0
3 20 dB
2.5
2
15 dB
1.5
1 10 dB
2 3 4 5 6 7 8 9 10
∆ω 2 /∆ω1
Fig. 18.22 Lower bounds on sensitivity-function norm as function of control band-
width for three values of ε.
2
∆ω1 = 40Η z
1.8
ε = .3 (10 d B)
1.6
δ = 0.1 (20 dB)
1.4 τ = 43 ms
||S|| σr= 0
τ = 4.3 ms
8
1.2
1 τ = 0.6 ms
0.8 τ = 0 (no delay)
0.6
2 3 4 5 6 7 8 9 10
∆ω 2/ ∆ω1
Fig. 18.23 Lower bounds on sensitivity-function norm as function of control band-
width for four values of τ .
4) The combustion response delay τ . One can verify that the lower bound on the
sensitivity peak is an increasing function of the delay (assuming other parameters
are fixed).
Using the inequality on the sensitivity peaking, Figs. 18.22, 18.23, and 18.24
plot lower bounds on sensitivity-function norms. The plots show that as the ratio
of the control bandwidth to the performance bandwidth decreases, the sensitivity
peaking becomes more and more severe. Further, the peaking is accentuated by an
increase in the delay τ , an increase in the performance requirements (lower ε), or
an increase in the real part of the unstable pole of the open-loop plant σr .
Extension of fundamental limitations to the case in which either plant or con-
troller has nonlinear characteristics is possible by using the concept of random
input-describing functions introduced in the preceding section. For example, in
the sector combustor controlled with on/off valves the fundamental limitations in
terms of lower bounds on the logarithm of sensitivity function as presented in this
section apply to G C ( jω) replaced with G C ( jω)N (σ ), where N (σ ) is the random
7
6 ∆ω1 = 40Η z
ζ = −4 % ε = .3 (10 d B)
5 δ = 0.1 (20 dB)
||S|| 4 ζ = − 2% τ = 4.3ms
8
3 ζ = − 0.4%
2
1
2 3 4 5 6 7 8 9 10
stablel
∆ω 2 / ∆ω1
Fig. 18.24 Lower bounds on sensitivity function norm as function of control band-
width for four values of combustor pole-damping ratio.
input-describing function of the on/off valve and σ is the standard deviation of the
Gaussian component of the valve command. Fundamental limitation analysis also
extends to the cases in which more general nonlinearities are present in the model
of combustor or controller, Gaussian noise sources are present, and the combustor
feedback loop operates at a limit cycle. In each particular case one has to prove
existence of the Gaussian and periodic signals in the feedback loop that provide
balance and examine stability of solution. More details are presented in Ref. 21.
VI. Conclusion
We have discussed some of the factors that may limit the performance of active
control systems for the attenuation of combustion instabilities. A broad range of
factors was identified and discussed. This list is not complete, by any measure.
Many effects can be system dependent; they are critical to one system but irrelevant
for another. System dependence may vary depending on the nature of the control-
system architecture and on the combustion dynamics that are being controlled.
Three of the critical factors limiting the control of the lean, premixed combustor
design considered in this paper have been examined in more detail.
The ability of the fuel actuator to affect the root-cause physics behind pressure–
heat-release coupling was found to be tied strongly to the mixing of the actuated
fuel flow with the remainder of the premixed reactants. In effect, the actuated fuel
flow must act to achieve a high degree of premixedness, both in time and space.
Another limiting factor relates to actuation time delay, as represented by the time
between movement of the fuel valve and realization of that fuel flow modulation
in the unsteady heat release or combustor pressure. Large values of time delay
were found to shrink the frequency band over which the control system could
attenuate pressure oscillations and to provide a mechanism through which pressure
oscillations outside this bandwidth could be amplified. This finding led to the peak-
splitting phenomenon that limited the degree to which pressure oscillations could
be suppressed.
Attempts to deal with these issues led to a sensitivity-function analysis of the
fundamental limits of combustor pressure oscillation control. It was shown that one
cannot arbitrarily decrease the level of pressure oscillations using linear controllers.
This limit is strongly influenced by system time delays, control bandwidth, and
performance bandwidth.
Although these issues represent current limitations, they are certainly not the
only factors that will affect performance of instability-control systems. Other sys-
tems (not of this type) will face different issues and different factors may control
their performance. Active combustion-instability control has been well demon-
strated as a technology with significant potential. For the technology to mature to
the point of being practically applicable, future efforts must focus on these limiting
factors, quantify them, and devise methods for dealing with them.
Acknowledgments
The work presented in this paper was supported by Contract MDA972-95-C-
0009 from Defense Advanced Research Projects Agency and Contract F49620-
01-C-0021 from the Air Force Office of Scientific Research and by United
Technologies Corporation. We would like to acknowledge our coworkers at United

Technologies Research Center for their contributions to this work, in particular,
Clas Jacobson, William Proscia, Prashant Mehta, Alex Khibnik, John Stuffle-
beam, Torger Anderson, Thomas Rosfjord, Nancy Rey, Randy Hibshman, Matthew
Maciolek, Jason Wegge, and Jeffrey Walker.
References
1
Hibshman, J. R., Cohen, J. M., Banaszuk, A., Anderson, T. J., and Alholm, H. A., “Active
Control of Combustion Instability in a Liquid-Fueled Sector Combustor,” American Society
of Mechanical Engineers, Paper 99-GT-215, June 1999.
2
for Gas Turbines and Power, Vol. 121, No. 2, 1999, pp. 281–284.
3
Seume, J. R., Vortmeyer, N., Krause, W., Hermann, J., Hantschk, C.-C., Zangl, P.,
Gleis, S., Vortmeyer, D., and Orthmann, A., “Application of Active Combustion Instability
Control to a Heavy Duty Gas Turbine,” Journal of Engineering for Gas Turbines and Power,
Vol. 120, No. 4, 1998, pp. 721–726.
4
Hoffmann, S., Weber, G., Judith, H., Hermann, J., and Orthmann, A., “Application
of Active Combustion Instability Control to Siemens Heavy Duty Gas Turbines,” Sym-
posium of the AVT Panel on Gas Turbine Engine Combustion, Emissions and Alternative
Fuels, NATO Research and Technology Organization, Neuilly-Sur-Seine Cedex, France,
Oct. 1998.
5
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Control of Combustion Driven
Oscillations by Equivalence Ratio Modulations,” American Society of Mechanical Engi-
neers, Paper 99-GT-118, June 1999.
6
Richards, G. A., Yip, M. J., Robey, E., Cowell, L., and Rawlins, D., “Combustion
Oscillation Control by Cyclic Fuel Injection,” American Society of Mechanical Engineers,
7
Anson, B., Critchley, I., Schumacher, J., and Scott, M., “Active Control of Combustion
Dynamics For Lean Premixed Gas Fired Systems,” American Society of Mechanical Engi-
neers, Paper GT-2002-30068, June 2002.
8
Sattinger, S. S., Neumeier, Y., Nabi, A., Zinn, B. T., Amos, D. J., and Darling, D. D.,
“Sub-scale Demonstration of the Active Feedback Control of Gas-Turbine Combustion
Instabilities,” Journal of Engineering for Gas Turbines and Power, Vol. 122, April 2000,
pp. 262–270.
9
Bloxsidge, G. J., Dowling, A. P., Hooper, N., and Langhorne, P. J., “Active Control of
Reheat Buzz,” AIAA Journal, Vol. 26, 1988, pp. 783–790.
10
Langhorne, P. J., Dowling, A. P., and Hooper, N., “Practical Active Control Sys-
tem of Combustion Oscillations,” Journal of Propulsion and Power, Vol. 6, pp. 324–333,
1981.
11
Fleifil, M., Annaswamy, A. M., Hathout, J. P., and Ghoniem, A. F., “The Origin of
Secondary Peaks with Active Control of Thermoacoustic Instability,” Combustion Science
and Technology, Vol. 133, June 1998, pp. 227–260.
12
Gysling, D. L., Copeland, G. S., McCormick, D. C., and Proscia, W. M., “Combus-
tion System Damping Augmentation With Helmholtz Resonators,” American Society of
Mechanical Engineers, Paper 98-GT-268, June 1998.
13
Hermann, J., Gleis, S., and Vortmeyer, D., “Active Instability Control (AIC) of Spray
Combustors by Modulation of the Liquid Fuel Flow Rate,” Combustion Science and
Technology, Vol. 118, Taylor & Francis, Philadelphia, 1996, pp. 1–25.
14
Saunders, W. R., Vaudrey, M. A., Eisenhower, B. A., Vandsburger, U., and Fannin, C.
A., “Perspectives on Linear Compensator Designs for Active Combustion Control,” AIAA
Paper 99-0717, Jan. 1999.
15
Banaszuk, A., Jacobson, C. A., Khibnik, A. I., and Mehta, P. G., “Linear and Nonlinear
Analysis of Controlled Combustion Processes. Part I: Linear Analysis,” Proceedings of
Conference on Control Applications, IEEE Publications, Piscataway, NJ, Aug. 1999.
16
Hathout, J. P., Annaswamy, A. M., Fleifil, M., and Ghoniem, A. F., “A Model-Based Ac-
tive Control Design for Thermoacoustic Instability,” Combustion Science and Technology,
Vol. 132, 1998, pp. 99–105.
17
Banaszuk, A., Aryiur, K. B., Krstic, M., and Jacobson, C. A., “An Adaptive Algo-
rithm for Control of Combustion Instability,” Automatica Vol. 40, No. 11, Nov. 2004,
pp. 1965–1972.
18
Johnson, C. E., Neumeier, Y., Lubarsky, E., Lee, Y. J., Neumaier, M., and Zinn, B. T.,
“Suppression of Combustion Instabilities in a Liquid Fuel Combustor Using a Fast Adaptive
Algorithm,” AIAA Paper 2000-0476, Jan. 2000.
19
Murugappan, S., Gutmark, E. J., and Acharya, S., “Application of Extremum-Seeking
Controller to Suppression of Combustion Instabilities in Spray Combustion,” AIAA Paper
2000-1025, Jan. 2000.
20
Evesque, S., “Adaptive Control of Combustion Oscillations,” Ph.D. Dissertation,
Cambridge Univ., Cambridge, England, U.K., 2000.
21
Banaszuk, A., Mehta, P. G., Jacobson, C. A., and Khibnik, A. I., “Limits of Achievable
Performance of Controlled Combustion Processes,” submitted to IEEE Transaction on
Automatic Control (submitted for publication, 2005).
22
Banaszuk, A., Jacobson, C. A., Khibnik, A. I., and Mehta, P. G., “Linear and Nonlinear
Analysis of Controlled Combustion Processes. Part II: Nonlinear Analysis,” Proceedings
of Conference on Control Applications, IEEE Publications, Piscataway, NJ, Aug. 1999.
23
Freudenberg, J. S., and Iooze, D. P., “A Sensitivity Tradeoff for Plants With Time
Delay,” IEEE Transactions on Automatic Control, Vol. AC-32, Feb. 1987, pp. 99–104.
24
Seron, M. M., Braslavsky, J. H., and Goodwin, G. C., Fundamental Limitations in
Filtering and Control, Springer, New York, 1997.
25
Ariyur, K. B., “Multivariable Extremum-Seeking Adaptive Control,” Ph.D. Disserta-
tion, University of California, San Diego, 2002.
26
Krstic, M., and Wang, H. H., “Stability of Extremum Seeking Feedback for General
Nonlinear Dynamic Systems,” Automatica, Vol. 36, April 2000, pp. 595–601.
27
Krstic, M., “Performance Improvement and Limitations in Extremum Seeking Control,”
Systems and Control Letters, Vol. 39, April 2000, pp. 313–326.
28
Zhang, Y., “Stability and Performance Tradeoff with Discrete Time Triangular Search
Minimum Seeking,” Proceedings of the American Control Conference, IEEE Publications,
Piscataway, NJ, June 2000.
29
Cohen, J. M., Stufflebeam, J. H., and Proscia, W., “The Effect of Fuel/Air Mixing
On Actuation Authority In An Active Combustion Instability Control System,” Journal of
Engineering for Gas Turbines and Power, Vol. 123, No. 3, 2001, pp. 537–542.
30
Stufflebeam, J. H., Kendrick, D. W., Sowa, W. A., and Snyder, T. S., “Quantifying
Fuel/Air Unmixedness in Premixing Nozzles Using an Acetone Fluorescence Technique,”
American Society of Mechanical Engineers, Paper 99-GT-399, June 1999.
31
Peracchio, A. A., and Proscia, W., “Nonlinear Heat-Release/Acoustic Model for Ther-
moacoustic Instability in Lean Premixed Combustors,” Journal of Engineering for Gas
32
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B., “A Mechanism for Combustion
33
Lee, D. S., and Anderson, T. J., “Measurements of Fuel/Air – Acoustic Coupling in
Lean Premixed Combustion Systems,” AIAA Paper 99-0450, Jan. 1999.
34
Lieuwen, T. C., “Experimental Investigation of Limit Cycle Oscillations in an Unstable
Gas Turbine Combustor,” Journal of Propulsion and Power, Vol. 18, No. 1, 2002, pp. 61–67.
35
Mezic, I., and Banaszuk, A., “Comparison of Systems with Complex Behavior: Spectral
Methods,” Physica D, Nonlinear Phenomena, Vol. 197, No. 1–2, October 2004, pp. 101–
133.
36
Gelb, A., and Vander Velde, W. E., Multiple-Input Describing Functions and Nonlinear
System Design, McGraw-Hill, New York, 1968.
Chapter 19
Implementation of Active Control in a Full-Scale

Gas-Turbine Combustor
Jakob Hermann∗
If TA GmbH, Groebenzell, Germany
and
Stefan Hoffmann†
I. Introduction
I N 1995, Siemens started its first tests with the new gas-turbine family Vx4.3A in
its test facility in Berlin. In contrast to the former gas-turbine family with silo-
combustion chambers, annular combustion chambers were implemented to achieve
an increased power density and efficiency and lower emission values. However,
during the tests, combustion instabilities detrimental to a reliable operation of the
gas turbines were observed. Already very early in the process of working out
countermeasures against this phenomenon, it was decided that besides passive
measures, such as operational modifications or an acoustic detuning by design
modifications of the burner,1 also active measures for avoiding these instabilities
should be developed and tested.
At that time, active combustion control was still in a research stage, as sum-
marized in several review papers.2−6 Various authors described successful tests
based on laboratory-scale burners with a thermal power of between 1 kW and
250 kW.7−9 In all these publications, attenuation of combustion oscillations was
achieved by phase-shifted sound signals generated via loudspeakers. In addition
to this method, other types of intervention and control strategies were researched
for various combustion systems. However, all tests were performed on labora-
tory scale.10−17 Based on these experiences a system with an adequate sensor,
controller, and actuator had to be developed to make it possible to damp com-
bustion instabilities in a heavy-duty gas turbine. The first full-scale tests with this
∗ Managing and Research Director.
† Director, Gas Turbine Component Engineering, Power Generation.
611
612 J. HERMANN AND S. HOFFMANN
Fig. 19.1 Half-section drawing of a Vx4.3A series gas turbine.
technique, called active instability control (AIC), were made in 1996 in the test
facility of Siemens on the V84.3A gas-turbine delivering 160 MW of electrical
power.18 Because of the positive results, this technique was also applied to the
largest type of this family of gas turbines, the Siemens V94.3A with an electric
power output of 267 MW.19,20
The implementation of the AIC technology in Siemens Vx4.3A gas turbines and
the problems encountered are the subjects of Sec. II. Results achieved with AIC
are discussed in Sec. III. Sections IV and V present long-term experiences and
a short evaluation of the advantages offered by AIC in comparison with passive
measures.
Fig. 19.2 Three-dimensional drawing of the annular combustion chamber of the

Vx4.3A with a total of 24 burners.
ACTIVE CONTROL IN A FULL-SCALE GAS-TURBINE COMBUSTOR 613
II. Implementation of AIC on Siemens-Type Vx4.3A

Land-Based Gas Turbines
Figure 19.1 shows a longitudinal section through the upper-half of the gas tur-
bine. In contrast to former types of Siemens gas turbines that were fitted with
silo-combustion chambers, this family of gas turbines features annular combus-
tion chambers (see illustration in Fig. 19.2). In total, this gas turbine comprises
24 Siemens hybrid burners spread uniformly over the circumference of its annular
combustion chamber.
Figure 19.3 shows a simplified drawing of the Siemens hybrid burner. This
burner may be operated both on liquid and gaseous fuels. At startup and with
gaseous fuel, the burner is operated in diffusion mode (see engine-operating map
in Fig. 19.12). When a specific turbine power is reached, a switchover will occur
from pure diffusion to so-called mixed operation. For this purpose, a combination of
diffusion and premix flames is used. When the temperature within the combustion
chamber is sufficiently high to stabilize a pure premix flame, operation is switched
Fig. 19.3 The standard Siemens Hybrid Burner.

from this mixed mode to a purely premixed operation. To stabilize the premix
flame, every Siemens hybrid burner features an additional pilot burner, which is a
diffusion burner delivering approximately 10% of the thermal power provided by
the complete burner unit.
A. Instability Problem
During the test phase of the prototype and the commissioning of the type of
gas turbine described in preceding text, problems occurred with self-excited com-
bustion oscillations in various power ranges, depending on the modifications per-
formed on the hybrid burner used. Because of the low frequency of the instabilities,
this phenomenon is sometimes called “humming.”
Research on the instabilities by Seume et al.18 showed that standing sound waves
are generated within the annular combustion chamber by self-excited combustion
oscillations. According to the direction of propagation of these waves along the
circumference of the annular combustor, they are designated “azimuthal modes.”
For example, Fig. 19.4 shows the azimuthal modes of the second harmonic, char-
acterized by a total of four nodes and four antinodes. For the largest version of
this gas-turbine type, V94.3A, this eigenmode is excited at a frequency of approx-
imately 170 Hz.
The equation f = n · c/(π · d) provides a theoretical estimate for the charac-
teristic acoustical frequencies of any annular combustion chamber. At an average
combustion chamber diameter of d = 3 m and a speed of sound of c = 844 m/s
(assuming an average combustion chamber temperature of 1500◦ C), the resulting
frequency for a V94.3A is 179 Hz for the second harmonic (n = 2), a value agree-
ing well with experimentally determined characteristic frequencies. For smaller
Fig. 19.4 Excited azimuthal modes for the second harmonic in the annular combus-
tion chamber of the V94.3A.
versions of this turbine family, this equation returns higher frequencies for the
occurrence of combustion oscillations, because of the reduced diameter of their
combustion chambers (see Ref. 18).
B. AIC Installation
To avoid the oscillation problem, an AIC system for this type of gas turbine was
developed in addition to passive measures such as burner-design modifications.1
Figure 19.5 provides a simplified schematic diagram of the basic design of this
AIC.
1. Sensor
As an input quantity, this AIC system uses the sound pressure measured at the
burner flanges. The wall temperatures are substantially lower at the burner flanges
than within the combustion chamber so that high-temperature piezo-pressure trans-
ducers can be used without requiring any additional cooling. Several tests verified
that the sound-pressure signals measured at the burner flanges correlate suffi-
ciently well in amplitude and phase with the pressure signals within the combustion
chamber.
During some initial tests, potential uses of optical probes for measuring AIC
input signals were researched in addition to sound-pressure measurements at the
burner flanges. However, because of limited viewing angles and thermal problems
with probe installations, this method was soon given up (see Ref. 18 for some more
detail).
Fig. 19.5 Schematic AIC diagram of the Siemens model Vx4.3A heavy-duty gas
turbine.
Fig. 19.6 Schematic diagram of the DDV valve used.
2. Actuator
The crucial problem for implementing an AIC system is actively influenc-
ing the combustion. Owing to the elevated air and fuel volume flows through
Vx4.3A gas turbines, these flows cannot be sufficiently modulated in full, nei-
ther by means of acoustical actuators nor by valves. Detailed research showed
that the main premix flame of the Siemens hybrid burner, controlled by much
smaller pilot-diffusion flames comprising no more than approximately 10% of
the entire mass flow, will respond very precisely to fluctuations in the conversion
rates of those pilot flames. Therefore, modulating the pilot-gas mass flow con-
trols not only the pilot flames themselves but also the premixed main flame to a
significant extent. Pilot flames are supplied via their own fuel line. The actuator,
a direct-drive valve (DDV) (shown in Fig. 19.6) developed specifically for this
application by Moog Germany, is integrated in this fuel line. To obtain maximum
control of the main flames, and thus of the combustion oscillations arising within
the gas turbine, every burner of the turbine was equipped with its own valve.
Thus, a total of 24 valves was installed around the annular combustor of the gas
turbine.
In the absence of any AIC signal, DDV valves will be 50% open, so that the
pilot-gas mass flows required for normal premixed operation will reach the burners.
To modulate individual pilot-gas mass flows and thus the pilot flames themselves,
valve spools will be moved around their static opening value of 50%, that is,
opened further or closed down at the frequency of the combustion oscillation to be
damped. The level of fuel flow modulation achieved can be determined by valve
spool stroke. According to the frequency response of the used valve type (shown in
Fig. 19.7) the valves allow the control of combustion oscillations with frequencies
of up to 400 Hz.
Fig. 19.7 Frequency response of the DDV valve used.
By moving the valve spools at frequencies within this range, sound waves will
be generated within the pilot-gas systems in addition to the modulation of the
gas mass flow. The resulting effects must be taken into account and the pilot-gas
system must be tuned as described in Hermann et al.16 and Hantschk et al.17 to
optimize the modulation amplitude.
3. Controller
As described in Seume et al.18 and shown in Fig. 19.4, azimuthal modes are ex-
cited within annular combustion chambers by combustion oscillations. Excitation
by combustion oscillation means that the pressure fluctuations excited along the
combustion-chamber circumference are characterized by both different amplitudes
and different phases and that the burners placed evenly along the circumference of
the annular combustion chamber are accordingly located at positions characterized
by differing pressure amplitude and phase values. Fluctuations of the heat-release
rate are coupled characteristically with pressure oscillations (self-excitation), thus
heat released at the various burners will fluctuate at differing amplitudes and phases.
Because AIC requires both antiphase and in-amplitude influencing of the unsteady
flame, an individual control of each burner is necessary, which means that it must
be possible to control every burner individually. The least complicated case would
therefore require one sensor and one feedback loop for every burner, so that a
24-channel AIC system would have to be installed.
To minimize the number of feedback loops, the symmetry of azimuthal modes
arising within the combustion chamber was exploited. As shown by the az-
imuthal mode corresponding to the first harmonic in Fig. 19.8, the sound-pressure
fluctuations for two burners lying precisely opposite each other are characterized
by identical amplitudes p̂ and a phase shift ϕ of 180 deg (Fig. 19.9). As indicated in
Fig. 19.8 Exploiting the symmetry of azimuthal modes, for example, for the first
harmonic. One sensor and one controller provide the input signals for two DDVs.
Fig. 19.9, for an AIC system, burners placed precisely opposite each other means
that a signal measured at a specific position by one sensor can be used to control not
only the actuator for this position, but also the one located at the precisely opposite
point of the combustion chamber by merely inverting the controller output signal
for the second actuator.
In second or even higher harmonics, the number of feedback loops can be
reduced even further, as shown in Fig. 19.10. Here, four actuators are con-
trolled by one input signal and one feedback loop. For the control system of
V94.3A burners, an installation as in Fig. 19.8 was chosen, because for this
type of gas turbine, the first harmonic will likewise appear due to self-excited
Fig. 19.9 Sound pressure at the two valve positions of Fig. 19.8.
Fig. 19.10 Exploiting the symmetry of azimuthal modes, for example, for the second
harmonic. One sensor and one controller provide the input signals for four DDVs.
oscillations. In total, this type of gas turbine was fitted with 12 sensors and 12
feedback loops.
Those 12 loops were realized by means of six signal processors, every processor
handling two input signals and generating four output signals for a total of 24 ac-
tuators. The algorithm used works within the frequency range and can, simplified,
be characterized as a phase shifter and amplifier. Because of its implementation
in the frequency domain the controller works extremely rapidly. In addition, this
algorithm allows the specific control of individual frequencies and is suitable for
controlling two frequencies, a restriction entailed by the limited computing capac-
ity of the type of signal processor used. The necessary control parameters depend
on the operation parameters of the gas turbine and are set automatically during con-
tinuous operation. Hermann et al.20 provides a detailed description of this control
system.
4. Technical Realization
The hardware and software setup is optimized with respect to time-saving im-
plementation and commissioning as well as facilitated maintenance. Figure 19.11
shows a complete AIC setup with a front view of the electronic hardware con-
sisting of a control and a power cabinet. The cabinets are self-contained and only
need a minimum number of data lines to interface with the gas-turbine control
system. Thus they can easily be integrated into already existing control systems.
The following hardware and software features have been included: modular ar-
chitecture (easy to extend and service), hardware monitoring for automatic error
diagnosis, actuator monitoring (valve seizing, deviation from the desired set point,
marginal check), sensor monitoring (cable and sensor defects), redundant power
supply (interchangeability during operation), integrated liquid-crystal displays and
Fig. 19.11 The complete AIC setup for the V94.3A gas turbine.
keyboard for monitoring and user access, data and control interface to the gas-
turbine control system, failure- and data-recording system, and special tools for
fast commissioning.
III. Results and Experiences with AIC during Gas-Turbine Operation

The first tests with a prototype of the AIC system described in the preceding
section were run on a type V84.3A machine delivering 170 MW of electric power
at the Siemens test facility in Berlin. Detailed specifications of the prototype AIC
structure and of the tests performed are described in Seume et al.18 Owing to
these successful tests, AIC systems were subsequently built into type V94.3A
machines for field-testing purposes. These machines deliver between 233 MW and
267 MW of electrical power. For this purpose, the electronic system and the control
strategy used for V84.3A were completely redesigned and further developed into
an industrial-grade system as described in Sec. II.B.4.
During field tests, AIC systems were applied to V94.3A machines with various
burner configurations showing oscillation problems at various operating points.
The following sections present results in which AIC systems damped instabilities at
four different burner configurations A, B, C, and D. The burners were all prototypes,
which differ mainly in swirl and outlet geometry and in fuel-nozzle configuration.
Figure 19.12 shows the corresponding instability points in a gas-turbine op-
erating map for an easier understanding of the different results. A summarized
description of the operational mode of the engine combustor at the different AIC
applications is given in Sec. II and illustrated in Fig. 19.12.
A. Active Control at Switch-Over Process with Burner Configuration A

Burner configuration A mainly showed combustion oscillations during swit-
chover from mixed to premixed operation, that is, when transitioning from a com-
bination of diffusion and premixed flame to a purely premixed flame. To stabilize
Fig. 19.12 Operating map of the gas turbine with active control of combustion in-
stabilities at several operating points observed in combination with different burner
types.
the premixed flame, the pilot burner is switched on while switchover is in progress.
Figure 19.13 shows this process without an activated AIC system and Fig. 19.14
shows it with an activated AIC system. From top to bottom, the following quanti-
ties are plotted: maximum sound-pressure spectrum calculated based on the signals
delivered by the 12 AIC sensors at three different points of time; maximum rms
value of the sound pressure versus time, calculated based on the signals delivered
by the 12 AIC sensors; opening cross section vs time for the gas valves actuated
during switchover, as a percentage of the maximum opening cross section for the
corresponding valve.
As shown in Fig. 19.13 strong combustion oscillations occur approximately 3.5 s
after opening the pilot-gas main valve (start of switchover, marked in the bot-
tom diagram) as the pilot-gas flows into the combustion chamber. After another
2 s, the diffusion-gas valve is closed and the premix-gas valve opens, with the
thermal power within the combustion chamber being kept constant during that
process. About 10.5 s after initiating the switchover (t = 15 s), that is, as the
diffusion-gas valve closes, the combustion-oscillation amplitude slightly weakens
before it returns, after another 5 s, to the value characterizing constant opera-
tion (see rms signal, Fig. 19.13). At t = 20 s, the gas turbine is running in its
fully stabilized premix mode. Considering the frequency spectrum of the sound
pressure measured at various times during switchover, it becomes obvious that,
in the beginning, the first harmonic dominates at about 90 Hz (t = 9.7 s). At
t = 13.7 s, this frequency will have dropped to approximately 80 Hz. Because of a
Fig. 19.13 Switchover from mixed to premixed operation without AIC. The inlet
guide vane remained permanently closed.
Fig. 19.14 Switchover from mixed to premixed operation with AIC. The inlet guide
vane remained permanently closed.
nonlinear effect, high amplitudes of the first harmonic will, produce higher-order
harmonics.
Figure 19.14 shows the same switchover process with activated AIC. As the
sound-pressure history for the rms value indicates, AIC achieves an almost com-
plete attenuation of combustion oscillations. Only one specific position of the
diffusion-gas valve (t = 15.2 s) produces a short peak lasting approximately 0.5 s,
its rms value already substantially reduced. The corresponding frequency spectrum
reveals that the harmonic excited is no longer the first but the third one, at approx-
imately 250 Hz. To achieve even further damping of this oscillation by means of
AIC, the system was preset for both frequencies (first and third harmonics). As
shown in the spectrum of another switchover recorded 10 min later, the amplitude
of this oscillation was further attenuated with the first harmonic being damped at
the same time. It must be stated that the damping achieved by AIC was more than
sufficient for safe gas-turbine operation.
B. Active Control at Load Change during Premixed Operation

with Burner Configuration B
During premixed operation over a power range of 60% to 100% of base load,
burner configuration B was characterized by self-excited combustion oscillations.
If this gas turbine is operated without AIC, increased amplitudes for the third
harmonic at a frequency of 270 Hz may appear as early as at 60% of base load. At
approximately 80% of base load, this frequency shifted to 170 Hz, corresponding
to the second harmonic of the combustion chamber. A load increase higher than
80% was not possible because the allowed oscillation level of the gas turbine was
already reached.
By activating the AIC system, the oscillations were attenuated enough to allow
safe turbine operation up to 100% of base-load power. Figure 19.15 demonstrates
the AIC operation for a load increase, and Fig. 19.16 demonstrates the AIC op-
eration for a load decrease. From top to bottom, the following oscillation and
operating parameters of the turbine are shown in both figures, all of them plotted
versus time: maximum rms-sound pressure value calculated on the basis of the
signals of 12 AIC sensors; maximum sound pressure amplitude of the two char-
acteristic frequencies for which the AIC is tuned; electrical power output of the
gas turbine. First, the AIC is tuned to the third harmonic at approximately 270 Hz
and is automatically activated at 60% of base-load power. Immediately after AIC
activation, the amplitudes characterizing the oscillations are damped by approxi-
mately 65% and remain approximately at this level, as the load is further increased
up to 80% of base-load power. At this load, the second harmonic is excited at
170 Hz, whereas the amplitude of the third harmonic continues to decrease fur-
ther. By converting the AIC target frequency to the second harmonic, its amplitude
is reduced immediately by about 30%. Switching over AIC from the third to the
second harmonic was required because, at that point of the test runs, simultane-
ously damping even- and odd-numbered modes had not yet been implemented for
the AIC system.
The same damping efficiency as shown for load increase was likewise achieved
when the load level was decreased, even though, in this case, switching from
Fig. 19.15 Load increase to base load in premixed operation with activated AIC
system. While running up the gas turbine, the system must be activated earlier than
when the gas turbine is being run down (see Fig. 19.16). The AIC system was also
always activated at base load to suppress combustion instabilities.
Fig. 19.16 Load decrease to part load in premixed operation with activated AIC
system.
the second to the third harmonic did not produce any significant amplitudes. The
diagrams in Fig. 19.15 demonstrate this AIC operation.
C. Active Control at Base-Load Operation of the Gas Turbine

with Burner Configuration C
Because of a burner modification, a significant improvement of the gas-turbine
performance concerning self-excited combustion instabilities could be obtained.
However, problems still occurred at loads greater than 97% of the base load, mak-
ing long-term operation at this load level difficult. The spectrum of the observed
humming showed two dominant peaks at 170 Hz and 340 Hz with significant
amplitudes.
With the AIC system activated, the load could be increased to 102% of the
base load without any combustion instabilities. At even higher loads, a strong
resurgence of the humming took place, which could not be successfully suppressed
by the AIC system because of the limited hydraulic cross sections of the pilot-gas
system and the actuator. Examples of the typical system behavior of this combustor
modification with and without AIC are given in Figs. 17 and 18.
The prototype of this combustor setup has been operating in a V94.3A base-
load machine in Argentina since January 1999, giving a higher power output
at decreased nitrous oxide (NOx) emissions (well below 30 ppm) without any
reported problems or degradation so far.
D. Active Control at Startup and at Lower Part-Load Operation of the

Gas Turbine with Burner Configuration D
A gas turbine with burner configuration D is started, and subsequently operated
at its lower part-load range, exclusively in diffusion mode, in which flames are
easier to stabilize despite the low level of combustion-chamber temperatures still
prevailing. Once a certain temperature limit has been reached, there is a switchover
to mixed operation. In this lower part-load range, from start to switchover into
mixed operation, burner configuration D will be subject to problems caused by
self-excited combustion oscillations, with the second and the fourth harmonics
being excited simultaneously within the annular combustion chamber.
Even though the AIC application presented relies on modulating pilot-gas mass
flow and pilot flames, which are, in principle, provided only for premixed opera-
tions, it proved possible to use the AIC system successfully even during part-load
operations. For this purpose, the pilot burners were used together with diffusion
flames, and the AIC system was activated. The system was tuned for the two
excited frequencies.
As shown in Fig. 19.19, which displays the maximum-frequency spectrum of
the 12 AIC sensors with and without AIC for this point within the operational
range, it was possible to damp both excited frequencies almost completely. For
the second and fourth harmonics, damping levels amounted to 20 dB and 14 dB,
respectively.
The separate damping of two eigenmodes by AIC, as described previously, is
exemplified again in Fig. 19.20. The diagram plots sound pressure as a spectrogram
Fig. 19.17 Base-load operation without AIC during the long-term field demonstration
in a V94.3A.
Fig. 19.18 Base-load operation with AIC during the long-term field demonstration in
a V94.3A. The system has been operating continuously since January 1999, enhancing
the base-load limit by 5% points.
Fig. 19.19 Suppression of two frequency peaks by AIC during part-load operation.
Fig. 19.20 Separate damping of two dominant eigenmodes of a combustion instability

by AIC.
(maximum values for the 12 sensors) versus time. Dark areas indicate high am-
plitudes; lighter areas indicate the general noise level. The two horizontal bars
at completely deactivated AIC mark the two eigenmodes excited, in this case,
at 145 Hz and 290 Hz. Starting at t = 70 s, by slowly reducing the AIC output
signal to suppress the fourth harmonic at 290 Hz within the 70 s< t < 85 s range,
the amplitudes of this harmonic increase. During the next 23 s, AIC only damps
oscillations at 145 Hz. By completely switching off the AIC system at t = 108
s, the second harmonic is likewise excited at 145 Hz. By reactivating the AIC
for both frequencies at t = 127 s, the two oscillations are damped out completely
once again. This example demonstrates that both oscillations can be self-excited
independently of each other and that, in general, if combustion oscillations are to
be avoided completely, both will have to be damped by using AIC.
IV. AIC Fault Tolerance and Long-Term Experiences

To research the influence of a failure of any feedback loop on AIC’s damping
characteristics, for instance, because of defective pressure transducers or valves,
specific loops were switched off on purpose during test runs. It was shown that
reducing the number of feedback loops to 10, corresponding to no more than 20
active valves, had no significant influence on damping characteristics.
Figure 19.21 shows operating hours of several AIC systems that were used for
damping instabilities, in part, and base load during a time span of about three
years. In both cases operating hours of up to 18,000 h were reached, whereas in
base load the AIC system had to act much more than in part load. For example,
damping an instability with a frequency of 170 Hz means in this case that the
valve had to perform 11 billion movements of the valve spool in a time span of
18,000 h. Nevertheless, no control and valve failure and no decrease in the damping
efficiency occurred. Furthermore, the abrasion of the moving parts of actuators was
negligible. During these long-term tests, AIC systems were operated automatically
in conjunction with the corresponding gas-turbine control units and had to control
Fig. 19.21 AIC operating hours for 14 commercially operating V94.3A gas turbines
named A through N.
automatically a wide range of operating conditions and situations to damp any

arising combustion oscillations safely.
The monitoring functions of the system allow for scheduled, condition-
based maintenance. Because of the above-mentioned robustness and the modular
stand-alone design, maintenance activities could be performed during the usual
scheduled engine-maintenance cycles.
V. Advantages of Active Measures

Compared with passive measures, AIC systems have many advantages. Whereas
the damping effect of passive measures frequently remains restricted to narrow-
frequency or operating ranges of the combustion system concerned, the AIC is
distinguished by its high degree of flexibility and its wide range of operation.
The examples describing how different types of oscillation problems occurring
in a variety of operational situations were solved successfully make this obvi-
ous. For damping low-frequency oscillations numerous passive measures such as
Helmholtz resonators or sound absorbers require a great deal of space and thus
cause many unwanted design complications, whereas the AIC system needs little
space and can be installed in a comparatively simple manner. Moreover, engines
equipped with AIC are generally protected against the sudden development of
combustion instabilities caused by changes in environmental conditions such as
ambient temperature or fuel composition.
Another AIC advantage is that—thanks to its degree of development reached at
present—it takes significantly less time to install and requires fewer tests. Accord-
ingly, novel combustion systems can be commissioned faster and take less time
to market. By comparison, developing suitable passive measures is still mostly a
trial-and-error process and takes much experimenting, that means time and money.
Thus, higher expenses required for AIC systems soon pay off.
In addition to the already mentioned benefits of AIC, this system may render
it possible to operate certain combustion systems at lower emission levels. In the
tests run in the premixed mode presented earlier, it was possible with an activated
AIC to operate the gas turbine at lower levels of pilot-gas mass flows than without
the system. Because pilot flames are important for influencing the NOx emissions
of gas turbines, the AIC system allowed the reduction of NOx emissions by more
than 60% in certain cases.
VI. Conclusion
AIC has made it possible to successfully suppress self-excited combustion oscil-
lations in a Siemens V94.3A heavy-duty gas turbine with an electric power output
of 267 MW. This suppression was achieved by developing a multichannel AIC
system with a maximum of 12 control loops, each with an individual sensor and
a total number of 24 actuators in the pilot-gas system. Counteracting the oscilla-
tions is realized by a modulation of the pilot-gas mass flow out of phase to the
heat-release fluctuation of the flame. Because self-excited oscillations are typically
caused by a coupling between the heat-release rate of the flame and the acoustics,
eliminating the heat-release fluctuations also eliminates the acoustic oscillations.
During the development process, AIC demonstrated that, irrespective of the

burner configuration, the combustion instabilities could be suppressed even with
relatively minor modulations of the fuel flow in a wide range of operation modes:
1) In cases in which the maximum achievable load was limited by dynamics,

a load increase of > 5% could be reached, allowing the engines to operate at the
rated power output.
2) In another case the NOx emission values could be reduced by 60% down to
20 ppm because AIC allowed for more favorable fuel splits to the pilot burner.
3) Pressure fluctuations arising during switchover processes and at part load
were eliminated successfully in any case, thus allowing for smoother changeovers
and significantly reduced commissioning times.
The results presented of long-term tests also show the high reliability of the system
even under demanding operating conditions. The fail-safe and the condition-based
maintenance features worked well over time. Results of test series at steady and un-
steady operation of the gas turbine demonstrate sustainable damping performance
for longer operation periods. Even after 18,000 h of operation, the actuators and
controllers demonstrated excellent performance, proving that AIC has reached
maturity.
The modular and compact design of the stand-alone system with a minimized
number of interfaces to the controller allows for rapid implementation into existing
systems and makes the overall arrangement very service friendly. By foreseeing
the interfaces in the design of a new engine, retrofitting an AIC system can be
performed within the time frame of a regular engine outage.
Despite the costs connected with the use of an AIC system the features men-
tioned earlier make AIC a very powerful risk-mitigation feature for the market
introduction of new or revised combustion systems, thus allowing for a smoother
market introduction of new products or power upgrades. The market-introduction
delays which every original equipment manufacturer (OEM) encountered in the
mid-1990s with their highly efficient frames could have been significantly reduced
by the use of AIC.
References
1
Berenbrink, P., and Hoffmann, S., “Suppression of Dynamic Combustion Instabilities
GT0079, May 2000.
2
McManus, K. R., Poinsot, T., and Candel, S. M., “A Review of Active Control of Com-
bustion Instability,” Progress in Energy and Combustion Science, Vol. 19, 1993, pp. 1–29.
3
Yang, V., and Schadwo, K. C., “AGARD Workshop on Active Combustion Control
for Propulsion Systems,” Proceedings of NATO-RTO Symposium on Gas Turbine Engines
Combustion, Emissions and Alternative Fuels, RTO-MP-14, 1998, pp. 36/1–36/20.
4
Dowling, A. P., “Active Control of Instabilities in Gas Turbines,” RTO NATO
Conference, Braunschweig, Germany, May 2000.
5
Annaswamy, A. M. and Ghoniem, A. F., “Active Control of Combustion Instability:
Theory and Practice,” IEEE Control Systems Magazine, Vol. 22, No. 6, Dec. 2002, pp. 37–54.
6
Candel, S., “Combustion Dynamics and Control: Progress and Challenges. (Hottel
Lecture).” 29th Proceedings of the Combustion Institute, Vol. 29, Sapporo, Japan, 2002,
pp. 1–28.
7
Lang, W., Poinsot, T., and Candel, S., “Active Control of Combustion Instability,”
Combustion and Flame, Vol. 70, 1987, pp. 281–289.
8
Gulati, A., and Mani, R., “Active Control of Unsteady Combustion-Induced Oscilla-
tions,” Journal of Propulsion and Power, Vol. 8, No. 5, 1992, pp. 1109–1115.
9
Poinsot, T., Veynante, D., Bourienne, F., Candel, S., and Esposito, E., “Initiation and
Suppression of Combustion Instabilities by Active Control,” 22nd Symposium (Interna-
tional) on Combustion, The Combustion Inst., Pittsburgh, PA, 1988, pp. 1363–1370.
10
Bloxsidge, G., Dowling, A., Hooper, N., and Langhorne, P., “Active Control of Reheat
Buzz,” AIAA Journal, Vol. 26, No. 7, 1988, pp. 783–790.
11
Langhorn, P. J., “Reheat Buzz an Acoustically Coupled Combustion Instability,”
Journal of Fluid Mechanics, Vol. 193, 1988, pp. 417–443.
12
Poinsot, T., Candel, S., Esposito, E., Lang, W., and Bourienne, F., “Suppression of
Combustion Instabilities by Active Control,” Journal of Propulsion and Power, Vol. 5,
No. 1, Jan.–Feb. 1989, pp. 14–20.
13
Wilson, K. J., Gutmark, E., and Schadwo, K. C., “Flame-Kernel Pulse Actuator
for Active Combustion Control,” ASME 1992, Active Control of Noise and Vibration,
DSC-Vol. 38, 1992, pp. 75–81.
14
Yu, K. H., Parr, T. P., Wilson, K. J., Schadow, K. C., and Gutmark, E. J., “Active
Control of Liquid-Fueled Combustion Using Periodic Vortex-Droplet Interaction,” 26th
Symposium (International) on Combustion, Naples, Italy, 1996, pp. 2843–2850.
15
Yu, K., Wilson, K. J., and Schadow, K. C., “Scale-Up Experiments on Liquid-Fueled
Active Combustion Control,” AIAA Paper 98-3211, July 12–15 1998.
16
Hermann, J., Gleis, S., and Vortmeyer, D., “Active Instability Control (AIC) of Spray
Combustors by Modulation of the Liquid Fuel Flow Rate,” Combustion Science and
17
Hantschk, C., Hermann, J., and Vortmeyer, D., “Active Instability Control with
Direct Drive Servo Valves in Liquid-Fuelled Combustion Systems,” 26th Symposium
(International) on Combustion, The Combustion Inst., Pittsburgh, PA, 1996.
18
Seume, J. R., Vortmeyer, N., Krause, W., Hermann, J., Hantschk, C.-C., Zangl, P.,
Gleis, S., Vortmeyer, D., and Orthmann, A., “Application of Active Combustion Instability
Control to a Heavy Duty Gas Turbine,” American Society of Mechanical Engineers, Paper
97-AA-119, September 1997.
19
Hoffmann, S., Weber, G., Judith, H., Hermann, J., and Orthmann, A., “Application
of Active Combustion Control to Siemens Heavy Duty Gas Turbines,” Presented at
the Symposium of the AVT Panel on Gas Turbine Engine Combustion, Emissions and
Alternative Fuels, Lisbon 12–16 Oct., RTO Meeting Proceedings 14, 1998, pp. 40-1–40-13.
20
Hermann, J., Orthmann, A., and Hoffmann, S., “Application of Active Instability Con-
trol to a Heavy Duty Gas Turbine,“ XIV ISABE, 5–10 Sept., Florence, Italy, A99-34186,
1999.
SUBJECT INDEX
Index Terms Links
ABAL. See Acoustics and blowout avoidance

logic
Abel deconvolution procedure 496
Abel transformation 496
ACC. See Active combustion control
Acetone 508
Acetone-fluorescence measurements 508
Acoustic absorbers 403
Acoustic access 149 150
Acoustic admittance at injector exit 240
Acoustic admittance function 431
radial distributions of 241
Acoustic analysis
engine 116
Euler 129
of gas-turbine combustion systems 369
one-dimensional 93
Acoustic approximation 382
Acoustic coupling
of combustor 157
fuel feed line 10
Index Terms Links
Acoustic dampers 560

application of 562
description of 561
designs 534 564
Acoustic design 534 560
Acoustic disturbances 319
propagation of 320
Acoustic energy
convection of 14
radiation of 14
transfers of 15
Acoustic field
characterization of 285
flame’s effect on 350
interactions 353
in step duct 433
in straight duct with temperature jump 435
Acoustic fluctuations 277
Acoustic frequencies
natural 15
Acoustic interactions
with flame 265
with flame response 97
with inherent flame instabilities 353
with laminar and turbulent flames 317
oscillatory flame dynamics and 265
Acoustic isolation 127
Acoustic losses 469
Index Terms Links
Acoustic modeling
combustion and 51
damping processes and 59
deviations in accuracy of 57 59
difficulty of 54
passive damping devices in 59
results of 57
Acoustic modes energy transfer 8
Acoustic nodes 565
Acoustic oscillations 5 433
Acoustic perturbation 317
Acoustic phenomena
of can-annular combustion system 102
Acoustic pressure 439
in swirl-stabilized combustor 439
Acoustic pressure perturbations 344
Acoustic reflection coefficient 431
Acoustic resonators 85 404
Acoustic transfer-matrix 98
Acoustic velocity 186
Acoustic wave frequency 185
Acoustic wave generation 223
Acoustic wave interactions 315
Acoustic waves 78
Acoustic-flame coupling 181
Acoustic-flame interactions
spatial and temporal scales’ relation with 319
Index Terms Links
Acoustic-impedance
definition of 119
experimental rig for 119
measurements of 119
Acoustic-pressure distribution 101
Acoustics and blowout avoidance logic (ABAL) 51
Acoustics for reacting flows 183
compact flame in duct for 185
heat-release fluctuation for 183
Acoustic-vortex-flame (AVF)
field composition interpretation using 283
gas-turbine combustion systems and 277
interactions
droplet-vortex interactions and 296
droplet-vortex-flame interactions and 302
factors affecting 287
in gas turbines 277
occurrence of 278
swirl and 287
time-evolving flow data for understanding
of 308
length and time scales of 278
swirl and 287
theoretical considerations for 280
Active combustion control (ACC) 100 135
Active control
at base-load operation 627
at load change during premixed operation 624
Index Terms Links
Active control (Cont.)

at startup/lower part-load operation 627
at switch-over process 620
Active instability control (AIC) 100
advantages of 632
fault tolerance of 631
in full-scale industrial gas turbines 135
gas-turbine operation results and experiences
with 620
implementation of 612 613
instability problem in 614
installation of 615
actuator for 616
controller for 617
sensor for 615
technical realization for 619
Active solutions 114
Active-combustion control systems 499
Active-control demonstration 135
actuator characterization in 135
control methods of, development and
demonstration for 139
Active-control system 587
Actively tuned passive damping 405
Actuated fuel mixing 588
combusting injector-evaluation 591
nonreacting injector-evaluation and 589
Actuated fuel-injection spoke 590
Index Terms Links
Actuated portion 590

Actuation system 582
controller’s relation to 583
intent of 583
valve in 583
Actuation technology 588
barriers in 588
intent illustration of 588
Actuation time delay 594
dynamic interpretation of 595
experimental observations for 594
Actuator
authority 593
bandwidth 584
characterization 135
Adaptive phase-shifting control method 142
Adaptive scheme 584
Adaptive Sliding Phasor Averaged Control
(ASPAC) 139
Admittance function 431
Advanced combustor 113
Aeroderivative industrial engine combustors 43
Aerodynamic damping devices 86
materials for 87
Aeroengine combustor 135 388
Aeroengine gas-turbine
emission reduction for 113
Index Terms Links
Aeroengine gas-turbine (Cont.)

experiments for emission
reduction/performance enhancement 143
AFR. See Airflow ratio
AIC. See Active instability control
Air supply system 469
Air-blast swirl injector 224
Aircraft gas-turbine combustors 143
Aircraft propulsion 43
Airflow
distribution 128
oscillations 82
pulsed 124
Airflow ratio (AFR) 109 110
Alstrom approaches 445
Ambient conditions 166 174
Analytical solution 471
Annular combustion system 99 474
Annular combustor 31 399 474
acoustic transfer-matrix of 98
acoustic-pressure distribution 101
azimuthal mode shape in 96
complex eigenfrequencies of 100
geometry of 98
intermediate-frequency Helmholtz resonators
for 106
mode-shape analysis for 101
Index Terms Links
Annular combustor (Cont.)

operation envelope of 109
resonator types in 108
Annular duct 389 391
eigenfunction of 422
frequency response of 451 471
interconnection of 473
modal expansion and 469
modal solutions and 391
state-space representation of 469
time sequence of pressure distribution in thin 393
Annular duct interconnection 473
Annular mixing section 508
ANSYS 434
Arrhenius term 358
ASPAC. See Adaptive Sliding Phasor Averaged
Control
Asymmetric burners 557
Asymptotics 204
Automated data screening 165
Automatic tuning systems 100
AVF. See Acoustic-vortex-flame
Axial heat-release distribution 384
Axial swirlers 32 213
Axial variation 424
Axial velocity
contours 307
contours of, as function of swirl number 293
Index Terms Links
Axial velocity (Cont.)

contours of normalized mean 245
fluctuation 425
radial distributions of mean-flow 249
time evolution of 237
time-average contours of 292
Axial-entry swirl injector
flame dynamics of 250
flow dynamics of 243
mean flow properties of 244
precessing vortex and 245
swirler orientation effects on flow
development and 248
unsteady flow evolution and 245
Axisymmetric
chamber 250
conical linear transfer function 335
extension 304
wedge linear transfer function 336
Azimuthal direction 221
Azimuthal mode shape 96
Azimuthal modes 614
annular combustor and 96
symmetry exploited in 619
Azimuthal velocities 229
instantaneous isosurfaces of 227
isosurfaces snapshots 239
radial distributions of mean-flow 249
Index Terms Links
Bandpass filtering 122 126

Bandwidth
actuator 584
control 603
finite-controller performance 603
performance 602 605
Baroclinic mechanism 323 328
Baroclinic torque 283
Base recirculation bubble (BRB) 294
Baseline open-loop frequencies 567
Base-load operation 627 628 629
Beer-Lambert’s Law 500 503
Bernoulli equation 455 474
Bifurcations
subcritical 18 19 20
supercritical 17
Blade-path variance 158
Bode integral formula 602
Bode plot 122 538 542
of combustor pressure 597
resonator’s effect on 568
Bode stability analysis 538
Bolt flange 159
Borescope pictures 157
Borghi diagram 280
Index Terms Links
Boundary conditions 338

at combustor exit 431
inlet 379
for linear equations of motion 373
in the model problem 375
reflecting/nonreflecting 452
for three-dimensional linear stability analysis 420 430
Brayton cycle 8
BRB. See Base recirculation bubble
Broad lean blowout (LBO) 243
Broadband white-noise forcing 117
Bulk flow 536
Burner
nozzle 172 455
Burner assembly failure 170
dynamics of 171 173
high-frequency dynamics and 174
NOx levels/EGT prior to 173
premix tubes failure resulting 173
Burner transfer matrix
CFD analysis of time delays and 457
flame model for 455
lossy flow through burner nozzle 455
modeling 454
Burner-exit configurations 104
Index Terms Links
Calpine Turbine Maintenance Group 147

general instability characteristics of fleet of 151
Can combustor 36 472 473
Can-annular combustion system
acoustic phenomena of 102
axial mode shape in 102
can-can interaction in 102 103
mode-shape analysis for 101
CB. See Centerbody (CB) injection
CBOs. See Cylindrical Burner Outlets
CCD. See Charge-coupled device
CDM. See Combustion-dynamics monitoring
Centaur Type H engine 33
CPO reduction in 37
testing of 33
Centerbody (CB) injection 511
Central recirculating flow 244
Central toroidal recirculation zones (CTRZs) 213 219 252 258
Centrifugal instabilities 215 230 270
CFD. See Computational fluid dynamics
CFD modeling
of combustion dynamics 56
Charge-coupled device (CCD) 494
Chemiluminescence. See also Overall
chemiluminescence emission
in active-combustion control systems 499
Index Terms Links
Chemiluminescence. See also Overall (Cont.)

definition of 485
emission from 486 523
as measure of overall rate of heat release 491
emission record by CCD camera 494
emission spectrum 486
flame-structure images corresponding to 511
as heat release indicator 489
images of 493 495
lean premixed (LP) combustor measured by 487
line-of-sight integrated 495
measurements 485 523
oscillations 323
varieties of 487
Choked nozzles 250 426 438
Circumferential modes 402
Circumferential waves 389
Closed-loop simulation block diagram 598
Closed-loop suppression 139
Coaxial swirl injector
injector response to external excitation and 232
spectral characteristics of 231
vortical flow evolution 217
vortico-acoustic interaction 220
Coherence 121 125 126
Cold flow characteristics 216
Cold tones 153
Index Terms Links
Colocated inputs/outputs 470

Combusting injector-evaluation 591
Combustion 180 453
See also Oscillatory
combustion; Unstable combustion
acoustic modeling and 51
annular 99 474
can-annular 101
chamber of 421
compact region occurrence 280
computational configurations for analysis of 96
emission reduction in systems of 35
failures in 163 168
flow perturbations’ relation to 181
heat release and 420 426
liquid spray 215
LP 29 30
network representation of 447
nonlinear unstable 18
oscillations 6 90 92
multiple time lags for 557
oscillatory 66 68
PM 511
premixed 296 315
process nonlinearities 22
PVC affected by 294
regimes 318
Index Terms Links
Combustion (Cont.)
semiempirical 53
spontaneous oscillatory 68
spray 302 304
time-lag 51
turbulent 319
unstable 187
as velocity source term 186
waves/flow perturbations’ relation to 181
Combustion dynamics
in aircraft propulsion 43
CFD modeling of 56
concerns with 534
control of in gas-turbine combustors 47
control strategies for 45
controlling difficulty of 61
coupling mechanics 45
detrimental impacts of 153
engine tests for 47
examples of 46
framework for modeling of 44
frequency response and 168
as function of load and ambient temperature 167
as function of load and frequency response 168
fundamental causes of 45
gas-turbine 415
laboratory scale experiments of 127
Index Terms Links
Combustion dynamics (Cont.)

lean premixed combustor inhibited by 49
monitoring of 147
monitoring questions for 163
on-line monitoring system schematic for 164
passive-control 550 552 553
predicting 61
problems of 44
single-nozzle test rig for 127
uniformity in 154
Combustion instabilities 433
See also
Gas-turbine combustion systems;
Stability analysis
acoustics/combustion time delay for 180
amplitudes of 5
analytical models for treating 417
basic interactions leading to 180
causes 3 8
characterization of 3
closed-loop suppression achieved for 139
conditions for 4
control methods for 139
coupled nature of 378
damping processes’ relation to 13
dangers of 417
development of 180
Index Terms Links
Combustion instabilities (Cont.)

driving mechanisms for 75
elimination of 61
engine 115
experimental diagnostics of 481
frequencies of 4 91
fuel staging for control of 75
in gas turbines 9
gas-turbine 433
governing equations in 281
growth and saturation of 16
hardware damaged by 153
heat-release oscillations’ relation to 12
historical overview of 5
infrared-absorption technique for 500
laboratory replication of 127
laser-induced fluorescence measurements for 506
lean premixed 251 252
occurrences of 3
passive control of 403 533
prediction difficulty of 23
in premixed combustors 179
pressure-fluctuation levels and 84
process for reduction of 114
in ramjet-powered missiles 8
Richards-Lieuwen mechanism in 74
Rolls-Royce and 66
Index Terms Links
Combustion instabilities (Cont.)

sensing 583
solutions
active 114
passive 114
in a swirl-stabilized combustor 436
three-dimensional 401
time delays in 76 180
uncontrolled 587
Combustion liner failure 168
exhaust-temperature during 170
NOx/dynamics levels in weeks leading up to 169
Combustion stability
methods to improve 550
multiple time lags for 557
passive control for 550
Combustion system
simplified representation of 450
Combustion-driven oscillation feedback cycle 92
Combustion-dynamics monitoring (CDM) 147
benefits of 147 165
case studies of 166
for combustor health monitoring 154
combustor monitored by 148
data-acquisition of 150
engine tuning by 152
flow obstructions and 155
Index Terms Links
Combustion-dynamics monitoring (CDM) (Cont.)

instability characteristics and 151
integration of, with other systems 166
on-line system for 164
pilot-nozzle cracking and 157
pressure instrumentation of 148
spectral analysis of 150
system of 148
transition-piece failure 160
tuning issues with 153
Combustion-process nonlinearities 22
Combustion-transfer function 545
Combustor 85 131 288
See also Annular
combustor; Engines; Gas-turbine
combustion systems
acoustic coupling of 157
acoustic isolation of 127
advanced 113
aeroderivative industrial engine 43
aeroengine 135 388
aircraft gas-turbine 143
can 36 472
CDM monitoring of 148
conventional 46
cylindrical 5
damper tubes in 60
Index Terms Links
Combustor (Cont.)
description of 585
double annular 46
dry low-emissions 48 68 213
dump 284
dynamics 582
engines’ rig pressure compared to 134
exit 431
fuel-air ratio 131
gas-turbine 465 475 611
geometry 570
health monitoring 154
heat release in 76
high mach-number 469
inlet 431
lean-premixed 49 487
linear processes of 17
linear waves and 397
liquid-fueled three-nozzle sector 594
low-emissions 587
LPP gas-turbine 369 370 378
modes 476
multiple-flame-holder dump 188
multiple-frequency mode 556
nonlinear processes of 15 17 21
operating conditions of, in single-nozzle test rig 131
physical domain of model 438
Index Terms Links
Combustor (Cont.)
premixed 179
premixer 13
pressure spectra of 596
second-order 599
silo 475
simple 400 406 409
geometry and flow conditions for 398
simple quasi-one-dimensional 395
single-nozzle research 485
stability map of 446
stability modeling of 446
subscale, experiment of 127
swirl-stabilized 251 436
three-stage premix 76
of Trent 60 68
two-mode 571
unstable 15 581
unsteady pressure of 132
Combustor pressure 130
ASPAC and 139
of Bode plot 597
experimentally determined frequency
responses of 597
fluctuations 483 494
measurements 484
spectra of 596
Index Terms Links
Combustor pressure (Cont.)

transfer function of 597
unsteady 132
Combustor pressure oscillations (CPOs) 13 30
causes of 36
Centaur and 37
control strategy for 35
future challenges with 40
high amplitude 37
injectory screening for 36
LP combustion and 30
Mars engine reduction of 38
model for 37
pervasiveness of 40
reduction of 30 35
Compact choked inlet 374
Compact flame
in a duct 185
Complex thermoacoustic systems
annular combustion system 474
annular duct interconnection 473
can-type combustor 472
examples of 472
frequency domain approach 462
frequency domain stability analysis of
gas-turbine combustor and 465
Gas-turbine combustor application to 475
Index Terms Links
Complex thermoacoustic systems (Cont.)

modal expansion of 468
modeling of sources and nonlinearities 472
network interconnections of 462
reduced-order modeling of 461
stability of 463 464
time-domain approach 466
Composition inhomogeneities 202
Compressibility 280
Computational fluid dynamics (CFD) 52 98 397 535
analysis of time delays by 457
analysis of results 459
numerical set-up for 458
Combustion dynamics and 56
fuel injectors and 458
stability analysis using 549
time delay analyzed by 457
Confinement geometry 287
Conical flames 329
geometry of in perturbed case 199
geometry of in steady state 199
modulated 197
transfer function of 201 334
Conservation equations 396
Control algorithm 584
Control bandwidth 603
illustration of 604
Index Terms Links
Control system models 534

linear stability analysis for 537
operational block diagrams 535
operational transfer functions 536
Control-model analysis 552
Convected character 322
Convected disturbance 320
Convection
of acoustic energy 14
shear-wave, velocity 322
Convective backside cooling 33
Convective time-lag models
for heat-release perturbation 551
to modify average time-lag 550
Conventional combustor 46
Cooled wall 192
Cooler compressor discharge 565
Cooling
convective backside 33
effusion 33
liner 33
Corner recirculation zone (CRZ) 217
formation of 252
Corresponding deconvoluted image 495
Counterflow flames 357
Coupling mechanisms 45
forms of 207
Coupling processes 206
Index Terms Links
CPOs. See Combustor pressure oscillations

Crack initiation 153
Cross spectra 451
CRZ. See Corner recirculation zone
CS. See Large-scale coherent structures
CTRZs. See Central toroidal recirculation
zones
Cylindrical burner outlets (CBOs) 104 105
Cylindrical combustor 5
Cylindrical duct 389
modal solutions and 389
DAC. See Double annular combustors

Damper
acoustic design of 534 560
simple 561
Damper resonators
design of 564
Damper tubes 50
in combustor 60
Damping processes 13 469
actively tuned passive 405
aerodynamic 86
combustion instabilities’ relation to 13
Index Terms Links
Damping processes (Cont.)

heat-transfer 14
Helmholtz 477 478
modal 469
nonlinear 22
overall system 493
passive 59
resonators for 564
tubes for 50
viscous 14
DARCS. See Double annular conterrotating
swirler
Darrius-Landau instability 354
Data reduction 164
Data-acquisition system 150
DDV. See Direct drive valve
Decadic molar absorption 503
Deconvolution 496
Density fluctuation 419
Diffuser air pressure 131
Diffuser air temperature 131
Dimensionless overall activation energy 343
Dirac delta function 284
Direct drive valve (DDV) 616
frequency of, used 617
schematic diagram of use of 616
Direct numerical simulations (DNSs) 296
Index Terms Links
Dispersion. See also Droplet dispersion

spray 303
transverse droplet 301
turbulent 79
Disturbance field
features of 319
flame’s effect on 326
DLE. See Dry low-emissions
DNSs. See Direct numerical simulations
Doppler particle analysis (PDPA) 521
Double annular combustors (DAC) 46
Double annular conterrotating swirler (DACRS) 49
Driving processes 206
Droplet atomization 482
Droplet concentration fluctuation 300
Droplet dispersion 296
behavior of 298
in mixing layers as a function of Stokes
number 297 299 300
quantifying 298
swirl’s increase of 302
transverse 301
Droplet motion 296
Droplet vaporization 482
Droplet-vortex interaction 296
Droplet-vortex-flame interactions 302
Dry low-emissions (DLE) 48 163
Richards-Lieuwen mechanism in 74
Index Terms Links
Dry low-emissions (DLE) (Cont.)

Trent 60 and 68
Dry low-emissions combustor 48 67 213
Dual-annular counter-rotating swirl (DACRS) 215
Duct 185
annular 389 391 393 422
469 508
compact flame in 185
cylindrical 389
frequencies/growth rates of with choked
inlet/outlet with entropy/vorticity waves 392
long 185
mixing technology of 83
premixed 395
step 433
straight 435
Duct-cutoff frequency 185
Dump combustors 284
Dynamic pressure transducers 164
Dynamic response 143
ambient conditions on 166
to fuel injector 118
operating regime’s effect on 167
Dynamic-pressure spectrum 92
Effusion cooling 33
Index Terms Links
EGT. See Exhaust gas temperature

Elementary processes 182
Emissions 48 143 213
See also Dry
low-emissions; Overall
Chemiluminescence emission
from chemiluminescence 486 494 523
low 548 587
reduction of 35 113
Encirclement direction 542
Energy
acoustic 14
dimensionless overall activation 343
dissipation/removal processes of 13
flux of 428
kinetic 246 259 260
normalized turbulent kinetic 246
transfer of, out of natural acoustic
frequencies 15
turbulent kinetic 259
Engine combustion instability 115
Engines
acoustic analysis of 116
CDM tuning of 152
centaur type H 33 37
combustion dynamics tests for 47
combustion instability of 115
Index Terms Links
Engines (Cont.)
combustor’s rig pressure compared to 134
design for testing operating conditions of 128
F-1 6
GE Aircraft 243
Mars 33 38
tuning issues with 152
Entropy
equation of 281
unsteady generation of 387
Entropy disturbances 319 323
propagation of 320
separable solutions for 390
Entropy fluctuations 387
of combustor inlet 431
mean-flow gradients’ relation to 426
Entropy generation 323
Entropy perturbations 280
Entropy waves 388 392
Equivalence ratio
flame-speed sensitivity to 345
increasing 488
mean 349 350
overall Chemiluminescence emission effected
by 488 492
Equivalence-ratio
oscillations 10
Index Terms Links
Equivalence-ratio disturbance 347

Equivalence-ratio perturbations 202 203
as convected with mean flow velocity 347
Equivalence-ratio transfer function 347
Euler analysis 116 129
Euler code 134
prediction by 133
Euler model 116
Eulerian-Lagrangian approach 296
Exhaust gas temperature (EGT) 170 173
Exit combustor 431
Exit velocity 121
External excitation 232 472
Extinction 356
F-1 engine 6
Facility air piping 127
FAR. See Fuel Concentration; Fuel-air ratio
Far-field radiated pressure 194
Fast flame-ionization detector (FID) 77
Fast Fourier transform (FFT) 150
Fast-response extraction probe 505
Favre-filtered conversation equations 216
FEA. See Finite element analysis
Feedback acoustics 569
Feedback element 535
Index Terms Links
Feedback loop 9
Feedback transfer function 537
Feed-system coupling 481 504
FEM. See Finite element method
FFT. See Fast Fourier transform
Fiber-optic laser fluorescence equivalence ratio
probe 512
FID. See Fast flame-ionization detector
Field decomposition
interpretation using 283
Field equations
sources and sinks in 285
Filter
bandpass 122 126
frequency-tracking extended kalman 601
low-pass 198
Filtered backprojection methods 496
Finite element analysis (FEA) 549
Finite element method (FEM) 475
Finite-controller performance bandwidth 603
First radial (1R) 224
Flame 181 183 185 189
319 488
See also Heat release
acoustic field effected by 350
acoustic interactions with 265
acoustic/vortical perturbation responded to by 317
Index Terms Links
Flame (Cont.)
bifurcation of structure of 255
brush thickness of 352 353
computation of, disturbed by vortical
structure 321
conical 197 334
corrugations 337
counterflow 357
disturbance field effected by 326
geometries 329 559
global-heat release rate of 325
heat release monitored by chemiluminescence
and 523
heat release response to flame speed
perturbations and 346
hysteresis 257
impinging on plate by 192
incoming equivalence-ratio perturbations
responded to 203
inherent stabilities of 353
instabilities of 317
acoustic interactions with 353
during instability cycle 193
interacting with plate 194
interacting with wall 193
interactions of, with heat release boundaries 189
intrinsic premixed instabilities 353
Index Terms Links
Flame (Cont.)
laminar 317
lean hydrocarbon 486
as low-pass filter 198 204
modulated conical 197
mutual annihilations of 181 195
as nonlinear flame-transfer function 407 408 409
nonlinearities’ explicit form disappearing in 338
as not compact 384
overall shape of 497
perturbed, interactions with boundaries of 181
perturbed oblique 201
photographic images of stable/unstable 253
premixed 318 395
premixed swirl stabilized 548
response of, in general, nonlinear case 336
simple premixed 545
speed of, response to perturbations 342
stabilization 294
stable 253 254
step-stabilized 547
stoichiometric methane-air 344
strained 205
swirl-stabilized 547
transfer matrices of 446
turbulent 317
unstable 253
unsteady velocity jump across 327
Index Terms Links
Flame (Cont.)
vortex interactions with 267
wedge 329 334 335
wrinkled 352
wrinkled effects of 350
zero response of area of 349
Flame anchoring 356 536 560
Flame area 515 516
Flame aspect ratio 339
Flame bifurcation 255
vortex-flashback dictating 257
Flame blowoff 4
Flame disturbances 326
Flame dynamics 215
of axial-entry swirl injector 250
flame sheets and 324
of lean premixed swirl injectors 213
oscillatory 258
stable 251
Flame front 515
Flame geometries 559
conical 329
wedge-shaped 329
Flame instabilities 317
Flame interactions 183 265 277
Flame location 557
Flame model 455
time-delay spread on 456
Index Terms Links
Flame response 500

acoustic interaction with 97
analysis of 97
CBOs’ affect on 104
companion analytical analysis of 265
to composition inhomogeneities 202
design method for changing 102
dynamic properties of 98
heat-release fluctuations and 202
to incident composition inhomogeneities 181
measurements of 97
Rayleigh index and 500
Flame shape 557
Flame sheets 324
acoustic field interactions with inherent flame
stabilities and 353
basic concepts/analytical framework of 324
disturbance field effected by 326
flame anchoring/flashback/extinction and 356
flame-area response to flow perturbations and 329
flame-speed response to perturbations and 342
wrinkled flame effects and 350
Flame stabilization 294
Flame strain 344
Flame structure 511
bifurcation of 255
as function of swirl number 293
high swirl’s effect on 290
Index Terms Links
Flame structure (Cont.)

recirculation bubble’s effect on 290
as typical isocontour 294
Flame surface
digitized 516
evolution of 263
threshold 514
Flame transfer function 204 457 569
investigations of 546
mean equivalence ratio’s effect on 349 350
Strouhal number depended on by 348 349
for time-delay distribution 461
Flame transfer matrix 461
Flame zone 373
Flame-anchoring method 547
Flame-area fluctuations 325
Flame-area linear dynamics 330
Flame-area nonlinearities
dominant factors affecting 338
Flame-area response
to flow perturbations 329
Flame-area velocity
Strouhal number’s relation to 338
Flame-burning velocity 324
Flame-driven instability 6
Flamed-structure evolution 496
Flame-flame interaction 196
Index Terms Links
Flame-front motion
Mach number of 351
Flame-geometry effects 559
Flame-response function
pressure trace superimposed on 501
Flame-response Rayleigh index 500 501
Flame-speed
sensitivity of, to equivalence ratio 345
Flame-speed perturbations 342
flame’s heat release’s responded to by 346
Flame-speed response 342
Flame-structure evolution 496
Flame-structure image sequences
two-dimensional 497
Flame-surface
heat-release fluctuations evaluations by 200
mutual flame annihilations control of
area of 197
Flame-trapping process 255
Flame-vortex dynamics 187
Flame-vortex interactions 181 481
in unstable combustions systems 187
Flashback 356
unsteady phenomena of 356
Flashback thermocouple 155
Flight propulsion engines 44
Index Terms Links
Flow. See also Mean flow

bulk 536
central recirculating 244
of fuel injector-air swirler 121
gas 521
inviscid 372
laminar Poiseuille 19
lossy 455
mass rate of 236
mean 249 385 426
reacting 183
shear 280
swirling 228
turbulent 459
unsteady evolution of 245
wake 291
zero mean 386
Flow coupling 230
Flow development 248
Flow disturbances 318
Flow dynamics 216
of coaxial swirl injector 216
of lean-premixed swirl injectors 213
of radial-entry swirl injector 224
vortical flow evolution and 217
vortico-acoustic interaction 220
Flow evolution 215
Index Terms Links
Flow obstructions 155

Flow perturbations 181
combustion’s relation to 181
flame-area response to 329
heat-release response to 323
linear 387
WSR’s response to 358
Flow uniformity 338
Fluctuating pressure field
normalized 248
time evolution of 222
Fluctuations 84
See also Entropy fluctuations;
Heat-release fluctuations; Pressure
fluctuations
acoustic 277
axial velocity 425
combustor pressure 483 494
density 419
droplet concentration 300
FAR 76
flame-area 325
fuel mass flow 123
overall chemiluminescence-emission 492
pressure 221 222
strain-rate 344
surface-area 191
Index Terms Links
Fluctuations (Cont.)
temporal 492
velocity 185
vortex-driven 183
Fluid mechanic response measurements 120
Fluorescence 123
OH planar laser-induced 513
Fluorescence seeds 508
Fluorescence signal 506 508
Fluorescence tracers 513
Fluorescence yield 508
Forcing frequency 241 321
Forcing signals 117
Forward transfer function 537
Fourier transform 83
Frequencies 106 174 321 392
393 440 539 556
568
See also
High-frequency dynamics;
Intermediate-frequency dynamics;
Low-frequency dynamics; Midfrequency
dynamics
acoustic wave 185
annular duct and 451 471
baseline open-loop 567
characteristic 233
combustion dynamics and 168
Index Terms Links
Frequencies (Cont.)
of combustion instabilities 4 91
combustor pressure and 597
complex thermoacoustic systems and 462
of DDV 617
discrete sinusoidal 232
duct-cutoff 185
forcing 241 321
high, resonators 105
instability 23
intermediate 106
multiple resonant 543
natural acoustic 15
natural shedding 321
oscillation 22 440 552
pressure fluctuations and 221 222
resonant 539 543
sinusoidal 232
time lag’s relation to 555
transfer functions and 539
Frequency domain stability
analysis of gas-turbine combustor and 465
operating point effect 465
time-lag spread effect 466
Frequency dynamics
high 91
intermediate 91
low 91
Index Terms Links
Frequency modes
experimental data for 555
Frequency response 597
of annular duct 451
Frequency-domain approach
for complex thermoacoustic systems 462
Frequency-switching behavior 555
Frequency-tracking extended kalman filter 601
Fuel actuator 135 606
Fuel component stimulants 513
Fuel composition 534
Fuel concentration 77
Fuel concentration measurements 512
Fuel feed line-acoustic coupling 10
Fuel flow actuator 448
Fuel flow distribution 45
Fuel flow rate
measurement of, at injector exit 520
overall chemiluminescence emission
related to 490
Fuel injection 213
Fuel injector
CFD and 458
importance of 118
LP gaseous 32
multiple time lags in 557
Index Terms Links
Fuel injector-air swirler

acoustic-impedance measurements of 119
bandpass filter’s similarity to 122
coherence measurements for 125
combinations of 124
dynamic response of 118
exit velocity of 121
flow of 121
fluid mechanic response measurements of 120
fuel-spray response measurements of 122
under study 119
Fuel mass flow fluctuation 123
Fuel mixing 213
Fuel nozzles 127 586 590
Fuel penetration 460
Fuel spray 124
steady/modulated 521
Fuel staging 75
Fuel valve 135
dynamic characterizations in 136
transfer function of 136
Fuel-air mixing
measurement of 507
Fuel-air premixer
schematic illustration of 553
Fuel-air ratio (FAR) 73
decoupling of 82
fluctuations of 76
Index Terms Links
Fuel-air ratio (FAR) (Cont.)

oscillations in 74 78 82
perturbation 79 83
Fuel-air ratio wave damping 74 76
in Trent combustor 85
Fuel-air wave coupling 45
Fuel-split variations 84
Fuel-spray response measurements 122
Full width at half-maximum (FWHM) 489
Full-scale fuel preparation subcomponents 127
Full-stability analysis 98
Fundamental limitations 602
Fused silica optical fibers 512
FWHM. See Full width at half-maximum
Gain characteristics 493

Gain curves 341
Gain-transfer function 339
Galerkin
approximation 380
expansion 381
method 407 417 423
series 381
Gas
perfect 371
phase 419
Index Terms Links
Gas (Cont.)
sampling 129
Gas flow 521
Gas turbine combustion dynamics
three-dimensional linear stability analysis 415
Gas turbine swirl injector 224
Gas-dynamic nonlinearities 22
Gas-phase fluorescence measurements 507
Gas-turbine combustion dynamics
three-dimensional linear stability analysis for 415
boundary conditions for 430
Greek symbols for 415
matching conditions 428
overscripts and superscripts for 415
spatial averaging and 421
subscripts for 415
symbols for 415
system equations for 432
theoretical formulation for 418
treatment of inhomogeneous terms for 426
Gas-turbine combustion instability
three dimensional acoustic analysis
solution procedure for 433
Gas-turbine combustion systems 135 174
acoustic analysis of 369
aeroengine 113 143
AIC in 135 620
Index Terms Links
Gas-turbine combustion systems (Cont.)

ambient conditions’ effect on 166
annular combustor and 399
annular geometry of 391
application of 394
AVF interactions in 277
azimuthal mode shape in 96
Centaur Type H 33
combustion instabilities of 9
complex thermoacoustic systems’ application 475
design options for safety with 101
experiments for emission
reduction/performance enhancement 143
frequency domain stability analysis of 465
geometry of 394
linear failure in 168
linear waves and 397
low emission 548
Mars 33
network representation of 447
one-dimensional acoustic analysis for 93
operating regime and 167
plenum and 394
premixed duct and flame and 395
schematic of 418
Siemens 89
Index Terms Links
Gas-turbine combustion systems (Cont.)

stationary 533
thermoacoustic properties in, methods for 93
thermoacoustic stability for 100
three-dimensional acoustic analysis for 94
Trent 60 66
unique features of 277
Gas-turbine combustor 611
Gas-turbine operation 620
Gaussian component
of valve command 606
Gaussian distribution 597
Gaussian process 600
Gaussian white-noise process 459
GE Aircraft Engines 243
GE Dual-Annular Counter-Rotating Swirling
(GE-DACRS) 301 303
GE-DACRS. See GE Dual-Annular
Counter-Rotating Swirling
General Electric engines
acoustic access of 150
LFDs in 151
LM6000 combustor 288
Governing equations 281
Green’s function 417 468
Green’s theorem 423
Index Terms Links
H infinity norm 603

Hardware monitoring 619
HCF. See High-cycle fatigue
HCO fluorescence 492
heat release measure by 513
HCO mole fraction 513
Heat 376
Heat conductivity 354
Heat release 8 182 419 505
523
chemiluminescence as indicator for 489 491
combustion and 426
in combustor 76
evolution of 263
flame interactions with boundaries of 189
flame speed perturbation response by 346
flame-vortex interactions and 187
HCO fluorescence measured by 513
mutual flame annihilation and 195
oscillations 11 325
overall Chemiluminescence emission related
to 490
as pressure source 187
rate of 493
Index Terms Links
Heat release (Cont.)

surface-area fluctuations and 191
time derivative of signal of 194
unstable combustion and 516
unsteady 323 336 377 399
Heating systems 3
Heat-release distribution 384
Heat-release fluctuations 183 207 494
acoustics for reacting flows for 183
dominant source terms associated with 184
flame response to composition
inhomogeneities 202
flames’ interactions with solid walls and 189
flame-surface evaluations of 200
modulated conical flames 197
role of 183
unsteady strain rate effects and 203
waves driving 197
Heat-release perturbations 551
Heat-release response 323
Heat-transfer damping mechanisms 14
Helical 215 270
Helmholtz dampers 477 478
Helmholtz equation 421
Helmholtz resonator 36 46 105 192
370 427 474 561
in acoustic resonator 404
Index Terms Links
Helmholtz resonator (Cont.)

configuration of 106
dangers with 404
design of 403
drawbacks of 405
eiqenfrequency of 105
geometry of 566
intermediate frequency 106
mean flow through 404 405 406
placement of 405
Helmholtz-type oscillations 4
He-Ne laser 503
Hertz mode 471
Hewlett-Packard signal analyzer 122
HFD. See High-frequency dynamics
High Mach-number combustor 469
High swirl 290
High swirl-number (HSN) 225
High-cycle fatigue (HCF) 153
High-frequency dynamics
burner assembly failure and 174
High-frequency dynamics (HFD) 91 152
ambient temperature/gas turbine load’s effect
on 174
dangers of 174
High-frequency resonators 105 107
High-frequency valve 136
High-response pressure transducers 129
Index Terms Links
High-shear nozzle/swirler (HSNS) 225

High-temperature-mixture filling process 255
Homogenous wave equation 379 384
Honeywell 582
Hot tones 153
HSN. See High swirl-number
HSNS. See High-shear nozzle/swirler
Hydrodynamic stretch 345
Hysteresis 257
IFD. See Intermediate-frequency dynamics

IGV. See Inlet guide vane
InAs. See Indium-arsenide detector
Incident composition inhomogeneities 181
Incipient blowout coupling 45
Incomplete mixing 491
Increasing equivalence ratio 488
Indium-arsenide (InAs) detector 503
Infrared-absorption measurements 500 523
Infrared-absorption technique 503
Inherent flame instabilities 353
Inhomogeneities
composition 202
incident composition 181
Index Terms Links
Inhomogeneous terms
combustion heat release and 426
mean flow’s effect on 426
surface condition effect of 427
treatment of 426
Inhomogeneous wave equation 381
Injector exit 240 520
Injector response
to external excitation 232
acoustic admittance at injector exit 240
instantaneous flow structures and 233
mass transfer function and 242
mean flow properties and 238
Injectors 520 589
See also Axial-entry
swirl injector; Coaxial swirl injector;
Fuel injector; Fuel injector-air swirler;
Radial-entry swirl injector; Swirl
injectors
air-blast swirl 224
fuel 118 458
gas turbine swirl 224
laboratory swirl 214
Lean premixed (LP) gaseous fuel 32
Lean-premixed swirl 213 215
liquid-fueled swirl 215
Mars 38
Index Terms Links
Injectors (Cont.)
operational 240
VFI 135
Inlet boundary conditions 379
Inlet combustor 431
Inlet guide vane (IGV) 152
Inlet swirl-vane geometry 287
Inlet velocity 287
Input-output relation 470
Instability characteristics 151
Instability driving mechanism 9
Instability frequencies 23
Instability prediction 445
instability-amplitude prediction 24
Instantaneous axial velocity fields 219 236
Instantaneous flow structures 233
Instantaneous flowfield 260
Instantaneous fluctuation pressure field 220
Instantaneous Karlovitz number 346
Instantaneous pressure field 352
Instantaneous total mass flow rate 243
Integrated liquid-crystal displays 619
Intermediate-frequency dynamics (IFD) 91
Intermediate-frequency resonators 106
International Organization for Standardization
(ISO) 89
Index Terms Links
Intrinsic premixed-flame instabilities

body force 353
diffusive thermal 353
hydrodynamic 353
Inviscid flow 372
IR absorption 504
ISO. See International Organization for
Standardization
Jet fuel
fluorescence of 123
Jet A 123
Jets
instabilities of 188
Jump condition 380 387 456
Kelvin–Helmholtz 215 270

Kinematic restoration processes 337
Kinetic energy
normalized turbulent 246
turbulent 259 260
Kirchhoff solution 477
Kolmogorov scale 279
Kronecker delta function 422
Index Terms Links
Laboratory swirl injectors 214

Laminar flames 317
Laminar Poiseuille flow 19
Laminar-burning velocity 354
Land-based power-generation applications 481
Landfill gas flares 7
Laplacian operator 422
Large-amplitude oscillations 189
Large-eddy simulation (LES) 56 98 181 214
215 288 438
Large-pressure-oscillation amplitude 70
Large-scale coherent structures (CS) 284
Large-velocity oscillations 354
Laser mie scattering 517 523
Laser-absorption measurements 502
Laser-absorption techniques 500 505
Laser-Doppler velocimetry (LDV) 243
Laser-induced fluorescence measurements 506 523
Laser-induced fluorescence technique 506
LDV. See Laser-Doppler velocimetry
Lean hydrocarbon flames 486
Lean premixed (LP) combustion instabilities 251 252
Lean premixed (LP) combustion systems 29
combustion dynamics, inhibition of 49
configurations and operating conditions of 30
Index Terms Links
Lean premixed (LP) combustor

chemiluminescence measurements of 487
combustion dynamics, inhibition of 49 487
Lean premixed (LP) gaseous fuel injector 32
Lean premixed, prevaporized (LPP) combustion 369
Lean-premixed swirl injectors
dominant processes associated with 215
flow and flame dynamics of 213
LEM. See Linear-eddy mixing
Length and time scales 278
Length-correction factor 474
LES. See Large-eddy simulation
LFD. See Low-frequency dynamics
Lighthill’s Reynolds stress sound-generation
term 286
Limit-cycle
oscillations of 20 407
predictions with 406
Limiting reactant diffusivity 354
Linear combustor processes 17
Linear flow perturbations 387
Linear fractional transform 467
Linear gain 342
Linear gasdynamic effect 420
Linear internal splash plate 34
Linear stability 16
Index Terms Links
Linear stability analysis 262

bode stability analysis and 538
for control system models 537
example problem with 537
multiple resonant frequencies and 543
Nyquist stability analysis and 540
Linear transfer function 337 339 601
Linear wave-equation 5
Linear-eddy mixing (LEM) 305
Linearized equations 371
Liner cooling 33
Ling acoustic driver 123
Ling Electro-Pneumatic driver 120
Liquid spray combustion 215
Liquid-fuel spray 517
Liquid-fueled swirl injectors 215
Liquid-fueled three-nozzle sector combustor 594
Load change 624
Load decrease 626
Load increase 625
Local shear-layer instability 238
Long duct 185
Longitudinal acoustic behavior 128
Lossy flow 455
Low swirl-number (LSN) 225
Low-amplitude oscillations 5
Low-cycle fatigue (LCF) 153
Low-emissions combustor 587
Index Terms Links
Lower part-load operation 627

Lower-frequency oscillations 552
Low-frequency dynamics (LFD) 91 151
Low-laser excitation irradiance 506
Low-pass filter 198 204
LP combustion
CPOs and 30
development of 29
LPP. See Lean premixed, prevaporized
combustion
LPP gas-turbine combustors 369 370
rate of heat release’s effect on 378
LSN. See Low swirl-number
Lumped-element representation 461
Mach number 351

Markstein number 345
Mars engine 33
CPO reduction in 38
fuel spoke location in 39
testing of 33
Mars injector 38
Mars injector optimization 37
Mars Pathfinder descent motor 8
Mass burning-rate 344
Mass flow rate 236
Index Terms Links
Mass fluctuation transfer function 127

Mass transfer function 242
Mass-balance condition 429
Matching conditions 428
Mean dilation field 308 309
Mean equivalence ratio 349 350
Mean equivalence transfer functions 349 350
Mean flow 249 385
additional consequences of 388
basic consequences of 385
effect of 426
entropy fluctuations related to gradients of 426
gradients of 426
through Helmholtz resonator 404
one-dimensional disturbances and 385
oscillatory flame dynamics structures by 258
velocity 347
zero 386
Mean flow properties 238
of oscillatory flame dynamics 258
Mean flow velocity 347
Mean flowfields 228
Mean pressure 308 309
Mean temperature contours 439
Measured pressure signal 484
Measured transfer function 464
Message Passing Interface (MPI) library 216
Index Terms Links
MFD. See Midfrequency dynamics

Midfrequency dynamics (MFD) 152
Mie scattering images 518 519
Mie scattering spray images 519
Mie-scattering 512
definition of 517
gas flow visualized by 521
laser 517
two-dimensional 518
MIMO. See Multiple inputs and multiple outputs
Minuteman intercontinental ballistic missile 8
Mixed first tangential (1T) 224
Mixture perturbations 323
Modal coupling 401 435
Modal damping 469
Modal expansion 421 468
state-space representation 468
of an annular duct 469
Modal solutions 389
annular duct and 391
cylindrical duct and 389
illustration of 391
narrow annular gap and 391
Mode transition 436
Model-based control method 142
Mode-shape analysis 101
Modular architecture 619
Modulated conical flames 197
Index Terms Links
Modulated spray 519

Momentum flux 428
Motion 371
MPI. See Message Passing Interface library
MSEK. See Multiscale extended Kalman
Multiburner configuration 472
Multiple fuel-injection locations 557
Multiple inputs and multiple outputs (MIMO) 468
Multiple resonant frequencies 543
Multiple-flame-holder dump combustor 188
Multiple-frequency mode combustor 556
Multiple-nozzle actuation 598
Multiple-nozzle, closed-loop actuation 595
Multiscale extended Kalman (MSEK) 140
Multistep chemistry effects 353
Mutual flame annihilations 181
flame-surface area controlled by 197
heat release and 195
outcome of 196
Mutual flame interactions 183
Narrow annular gap 391 432

Natural acoustic frequencies 15
Natural shedding frequency 321
Navier-Stokes equations 22 371
Newton Rhapson algorithm 464
Index Terms Links
Nitrous oxide (NOx) 46 173

Nonlinear combustor processes 17
causes of 21
phenomena caused by 21
Non-linear combustors 15
Nonlinear cycle limitation 454
Nonlinear damping processes 22
Nonlinear flame-transfer function 407
example of 408
variation of 409
Nonlinear saturation 342
Nonlinear stability 16
Nonlinear transfer function 337 339
Nonlinear unstable combustion system 18
Nonlinearities 472
Nonreacting acetone PLIF technique 589
Nonreacting injector-evaluation 589
Nonuniformity field velocity 337
Normalized fluctuating pressure field 248
Normalized mass burning-rate 344
Normalized mean axial velocity 245
Normalized transmittance 502 503
Normalized turbulent kinetic energy 246
Numerical root-finding procedure 463
Nyquist analysis 540 541 557 569
benefit of 542
for different values of time lag 543
Index Terms Links
Nyquist analysis (Cont.)

potential benefits/problems associated with
combustion time 550
Nyquist diagram 448 540 600
Nyquist plot 538 549 552 600
definition of 465
for the example with two resonant acoustic
modes 544
of open-loop frequency with resonator 568
for single acoustic mode 541
system stability and 542
Nyquist stability analysis 540 554
OEM. See Original equipment manufacturers

OH intensity 515
OH planar laser-induced fluorescence 513
OH PLIF image 513 514 515
OH PLIF measurements 523
One-dimensional acoustic analysis 93
One-dimensional disturbances 374
Galerkin series and 381
linear 383
mean flow and 385
plane wave solutions for 374
temperature gradients and 382
unsteady heat addition of 376
Index Terms Links
On-line combustion-monitoring system

automated data screening in 165
data reduction in 164
description of 164
dynamic pressure transducers in 164
remote data analysis in 165
signal conditioners in 164
signal processors in 164
Open-loop response 540
Open-loop transfer function 448
Operating regime 167
Operational block diagrams
general overview of 535
Operational injector 240
Operational transfer functions
general overview of 536
Organ-pipe resonances 392
Original equipment manufacturers (OEM) 148
Orr-Sommerfield operator 286
Oscillations 4
See also Combustor pressure
oscillations
acoustic 5
sample studies for 432
airflow 82
chemiluminescence 323
combustion 6 90 92 557
Index Terms Links
Oscillations (Cont.)
conditions of 22
equivalence-ratio 10
FAR 74 78 82
frequency of 22 440
heat-release 11 325
large-amplitude 189
large-velocity 354
limit-cycle 20 407
low-amplitude 5
lower-frequency 552
pressure 154 584
thermoacoustically induced 90 110
transverse 434
in unstable combustor 15
velocity 439
Oscillatory atomization 10
Oscillatory combustion 66
basic characteristics of 69
FAR-wave in 74
fuel splits changed in 73
instability control for 73
in Trent 60 DLE 68
Oscillatory flame dynamics 258
acoustic and flame interaction and 265
flame surface/heat release evolution and 263
instantaneous flowfield of 260
mean flow structures of 258
Index Terms Links
Oscillatory flame-area variation 11

Outer shear-layer instability 229
Overall Chemiluminescence emission 488 491
equivalence ratio’s effect on 488 492
flame temperature’s effect on 488
fuel flow rate related to 490
overall rate of heat release related to 490
turbulence’s reduction of 489
Overall Chemiluminescence-emission
fluctuations 492
Overall flame shape 497
Overall system damping and gain
characteristics 493
Padé approximation 472 473

Parametric instabilities 354
parametric oscillator equation 355
Particle trajectories 460
Particle-imaging velocimetry (PIV) 198
Passive control 403 533 553
Passive damping devices 59
Passive solutions 114
PCB piezoelectric pressure transducers 132
PDF. See probability density function
PDPA. See Doppler particle analysis
Peak-splitting phenomenon 599 602
Index Terms Links
Perfect gas 371

Perforated plates 370
Performance bandwidth 602 605
Periodic pressure disturbance 12
Perturbations 342
acoustic 317
acoustic pressure 344
entropy 280
FAR 79
flame-speed 342
flow 181 323 387
heat-release 551
Incoming equivalence-ratio 203
linear flow 387
mixture 323
pressure 377 387
small residence-time 359
velocity 195
vortical 317
Perturbed oblique flame 201
Phase delay 505
Phase doppler particle analysis 521
Phase transition 540
Phase-averaged fuel-air concentration profiles 592
Phase-locked PLIF measurements 589
Phase-roll 540
Phase-synchronized chemiluminescence
flame-structure measurements 523
Index Terms Links
Photomultiplier tube (PMT) 124 520

Piezoelectric pressure transducers 482
Pilot-flame stabilization 560
Pilot-nozzle cracking 157
in bolt flange 159
CDM and 157
failure spectrum of 158
PIV. See Particle-imaging velocimetry
Planar laser-induced fluorescence (PLIF)
measurements 507 589
Planck’s constant 507
Plane sound waves 386
Plenum 86 394
Plenum chamber 475
PLIF. See Planar laser-induced fluorescence
measurements
PM. See Premixed injection (PM) combustion
PMT. See Photomultiplier tube
POD. See Proper orthogonal decomposition
POD mode shape 232
Poorer-mixing injection 592
Power generation 3
Power Technology 163
Power-spectral density (PSD) 594
Practical swirl-stabilized flame 547
Precessing vortex 245
Index Terms Links
Precessing vortex core (PVC) 231 270 290

combustion’s effect on 294
origin of 250
Premix burner
sketch of 455
Premixed combustion
acoustic wave interactions 315
background of 318
disturbance field features and 319
heat-release response to flow and mixture
perturbations and 323
swirl for 296
Premixed combustion-acoustic wave
interactions 315
goals for 361
Greek for 316
nomenclature for 315
subscripts for 317
superscripts for 316
Premixed combustors 179
Premixed duct 395
Premixed flame 318 395
Premixed injection (PM) combustion 511
Premixed operation 624
Premixed swirl stabilized flame 548
Premixing fuel nozzle
cross section schematic illustration of 590
Index Terms Links
Pressure 12 70 84 126
130 132 134 220
247 352 393 482
494 501 508
See also Combustor pressure; Combustor
pressure oscillations
acoustic 439
acoustic, distribution 101
acoustic, perturbations 344
combustor spectra of 596
diffuser air 131
disturbance interaction 45
drops 128
dynamic spectrum of 92
dynamic, transducers 164
far-field radiated 194
field 352
fluctuating field of 248
heat-release as source for 187
high-response, transducers 129
instrumentation 148
mean 308 309
measured signal of 484
measurements 482
oscillations 154 584
perturbations 377 387
signal zero crossing 500
Index Terms Links
Pressure (Cont.)
simultaneous 505
source 187
spray mass flow-acoustic, transfer functions 126
transducer 483
unsteady 132
waves 195
Pressure fluctuations
frequency fluctuations of 221 222
in fuel line 484
in nozzle 484
predicted spectrum of, in combustion test rig
with a nonreflecting/reflecting exit 453
random noise generating 450
Pressure oscillations
analysis of 154
transport delay/limited actuator bandwidth’s
effect 584
Pressure-oscillation amplitude 70
Pressure-wave attenuator 86
Primary-zone temperature 72
Probability density distribution function 475
Probability density function (PDF) 251
Proper orthogonal decomposition (POD) 224
Propulsion 3
PSD. See Power-spectral density
Pulsed airflow 124
PVC. See Precessing vortex core
Index Terms Links
Quarter-wave resonator 561

Quarter-wave tubes 46 427
Radial modes 401

Radial swirlers 213
Radial variation 402
Radial velocity 249
Radial-entry swirl injector
vortical flow evolution of 225
Random input-describing functions 605
Rankine-Hugoniot jump conditions 456
Rayleigh conductivity 404
Rayleigh Criterion 4 45
definition of 8
heat-release oscillations and 12
illustration of 182
Rayleigh index 46 499
definition of 498
distribution 497
flame response and 500
minimum/maximum 499
positive/negative 498
Rayleigh parameter 295
Index Terms Links
Rayleigh’s integral 4
Reacting flows 183
Reacting vortices 189
Recirculation bubble 290
Redheffer star products 467 472
Reduced order models 535
Redundant power supply 619
Reflection coefficients 452
Relative velocity 207
Remote data analysis 165
Resonant frequency 539 543
Resonators 105
See also Helmholtz
resonator
acoustic 85
AFR increased by 110
Bode plot effected by 568
configurations of 563
damper 564
for damping 564
geometry of 562
high-frequency 105 107
installing 563
intermediate-frequency 106
modeling of 109
parameters assumed in 567
placement of 106
positioning of 562 565
Index Terms Links
Reynolds numbers 19 287

Reynolds-averaged Navier-Stokes (RANS)
equations 214 438
RHS. See Right-hand side
Richards-Lieuwen mechanism 74 76 77 85
Riemann invariants 447 451
Right-hand side (RHS) 286
RMS. See Root mean square
Rocket-engine instability 187
Rolls-Royce 66
Rollup 288
Root mean square (RMS) 298
Rotating patternator technique 520
Saffman-type instabilities 353

Scalar transport 305
Scattering matrix 448
accurate measurements of 449
four elements of 451 452
source term found from 451
Scondary fuel flow modulation 499
Screech liners 114
Screech tones 154
Secondary fuel valve trigger 500
Semiempirical combustion 53
Index Terms Links
Sensitivity function 601

factors that bound the supremum of 604
lower bounds of 604
typical 603
Sensor monitoring 619
SGS. See Subgrid-scale model
Shear flow 280
Shear-layer evolution 223
Shear-wave convection velocity 322
Siemens gas turbines 89 90
annular combustion systems of 99
component testing for 94
one-dimensional acoustic analysis for 93
prediction methods for 94
three dimensional acoustic analysis for 94
Siemens hybrid burner 613
Siemens Powergeneration 89
Siemens-Westinghouse engines
acoustic access of 149
general layout of 149
waveguide of 148
Signal conditioners 164
Signal processors 164
Silo combustor 475
Simple combustor
geometry and flow conditions for 398
limit-cycle mode shape for 409
Index Terms Links
Simple combustor (Cont.)

mode shapes for 400
resonant modes of 406
Simple premixed flames 545
Simple quasi-one-dimensional combustor 395
Simultaneous pressure 505
Single input single output (SISO) 468
Single-nozzle combustor flame tube 585
Single-nozzle industrial nozzle 485
Single-nozzle research combustor 485
Single-nozzle test rig
for combustion dynamics 127
combustor operating conditions of 131
computed power spectrum of combustor
pressure of 130
computed pressure mode shape of 130
controllers for 139
design of 129
Euler code’s prediction and 134
finalization/installation of 131
guidelines for 127
Sinusoidal frequencies 232
SISO. See Single input single output
Small residence-time perturbations 359
Solar Turbines Inc. 29 32
SoLoNOx technology 33
Sound generation 323
Source terms 449
Index Terms Links
Space Shuttle 8
Spanwise vorticity 300
Spatial averaging 421 423
Spatially uniform velocity disturbance 330
Spectral analysis 150
Spectral characteristics 231
Spiral vortex 261
Spontaneous oscillatory combustion 68
Spray 122
combustion 304
combustions simulations 302
dispersion 303
fuel 124
liquid 215
liquid-fuel 517
modulated 519
Spray and reaction rate contours 306
Spray mass flow-acoustic pressure transfer
function 126
Spray-flow interactions 481 482
Spray-vortex interactions 482
Stability 463
borders of 446 464 466
of complex thermoacoustic systems 463
linear analysis of 262
linear/nonlinear 16
nonlinear un- 18
Index Terms Links
Stability (Cont.)
thermoacoustic 100
of transfer matrix 463
Stability analysis. See also Linear stability
analysis; Three-dimensional linear
stability analysis
Bode 538
CFD/FEA used by 549
Full 98
of linearized one-dimensional conservation 283
Nyquist 540
Stability boundaries 552
experimentally determined 553
experimentally determined for
multiple-frequency mode combustor 556
Stable flame 253 254
Stable flame dynamics 251
Stagnation enthalpy 386
State-space representation 462 467
of an annular duct 469
Stationary gas turbines 533
Step duct 433
Step-stabilized flames 547
Stoichiometric methane-air flame 344
Stokes number 125 297 299 300
Straight chamber 436
Straight duct 435
Index Terms Links
Strained flames 205

Strain-rate fluctuations 344
Streamlines and temperature contours 307
Strouhal number 120 348 349
flame area’s velocity relation to 338
flame transfer function dependent on 348 349
scaling of 122
Subcritical bifurcations 18
Subgrid-scale (SGS) model 251
Subscale combustor experiment 127
Supercritical bifurcation point 17
Supply system 557 560
Surface condition 427
Surface-area fluctuations 191
Swept-sine data acquisition 122
Swept-sine forcing 117
Swirl 225
AVF interactions and 287
droplet dispersion increased by 302
high 290
for premixed combustion 296
Swirl injectors. See also Coaxial swirl injector
axial-entry
flame dynamics of 250
mean flow properties 244
precessing vortex and 245
Index Terms Links
Swirl injectors. See also Coaxial swirl injector (Cont.)

swirler orientation effects on flow
development and 248
unsteady flow evolution and 245
CFM56 243
characteristic frequencies in high number 233
cold flow characteristics of 216
discrete sinusoidal frequencies responded to
by 232
flow passage of 248
gas turbine 224
laboratory 214
lean-premixed 213
liquid-fueled 215
radial entry 224
Swirl number 293
Swirler orientation 248
Swirlers
air 118
axial 32
radial 213
Swirling coaxial air 291
Swirling flow 228
Swirl-stabilized combustor 251 436
acoustic pressure in 439
calculated oscillation frequencies and
damping coefficients for 440
Index Terms Links
Swirl-stabilized combustor (Cont.)

combustion instabilities in 436
schematic of 437
Swirl-stabilized flame 547
practical 547
premixed 548
transfer matrices of 446
Switch-over process 620
Sysnoise 471 473
TCA. See Tunable combustion acoustics

Temperature 508
gradients 382
jump 435 437
Temporal fluctuations 492
Temporal uniformity 588
Thermoacoustic design 93
Thermoacoustic induced oscillations 90 110
Thermoacoustic instabilities 5
Thermoacoustic network 477
Thermoacoustic properties 93
Thermoacoustic stability 100 111
Thermoacoustic systems
block diagram of dynamic 535
complex
reduced-order modeling of 461
Index Terms Links
Thermoacoustic systems (Cont.)

network representations of 447
Three-dimensional acoustic analysis 94
computational configurations for 96
of gas turbine combustion instability
solution procedure for 433
Three-dimensional combustion instabilities 401
Three-dimensional linear stability analysis
boundary conditions for 420
fluctuating quantities decomposed for 423
for gas turbine combustion dynamics 415
Three-stage premix combustor 76
Threshold flame surface 514
Time delay 79 457
actuation 594
CFD analysis of 457
combustion instabilities and 76 180
distribution of 459
flame model and 456
Time evolution
of axial velocity 237
of fluctuating pressure field 222
of streamlines and pressure field 247
Time lag 466 505 550
adjusting 550
frequency’s relation to 555
lower-frequency oscillations from increasing 552
Index Terms Links
Time lag (Cont.)

models to modify 550
Nyquist analysis for different values of 543
Nyquist lobes and 554
Time trace 483
Time-averaged axial velocity contours 292
Time-delay prediction 501
Time-domain approach 466
Time-lag combustion
premixer exit plane of 52
Time-lag distribution
approaches to increase 557
changing by using 558
for different mechanisms 557
flame-geometry effects 559
for supply systems 557
Tomographic techniques 506
Transfer functions 198
axisymmetric conical linear 335
axisymmetric wedge linear 336
combustion
physical processes contributing to 545
of combustor pressure 597
conical-flame 201 334
definition of, between relative velocity and
heat-release fluctuations 207
Index Terms Links
Transfer functions (Cont.)

equivalence-ratio 347
feedback 537
flame 204 348 457 461
546 569
forward 537
frequency response of 539
of fuel valve 136
gain 339
linear 337 339 601
mass 242
mass fluctuation 127
mean equivalence 349 350
measured 464
nonlinear 337 339
nonlinear flame 407 408 409
open-loop 448
operational 536
sequential processes for 537
spray mass flow-acoustic pressure 126
wedge-flame 332
Transfer matrix 395
of burner 448
experimental determination of 449
experimental validation of 452
nonreflecting boundary conditions 452
reflecting boundary conditions 452
Index Terms Links
Transfer matrix (Cont.)

linearity in 454
modeling of burner and 454
stability of 463
Transfer matrix method 538
Transfer-matrix approach 94
Transition-piece failure 160
Transport equation 425
Transverse droplet dispersion 301
Transverse oscillations 434
Trent 60 Aeroderivative engine
acoustic resonators in 85
aerodynamic damping devices for 86
combustor of 68
cross section of 67
DLE and 68
mixing duct technology for 83
overview of 66
potential structural damage for 69
premixers of 84
time delay in 79
turbulent dispersion in 79
Tunable combustion acoustics (TCA) test rig 54
benefits of 56
results of 55
Turbulence 489
Turbulent combustion 319
Turbulent diffusion 459 557
Index Terms Links
Turbulent dispersion 79
Turbulent flame 317
Turbulent flow 459
Turbulent kinetic energy 259
Turbulent mixing 460
Turbulent spreading 461
Two-dimensional Mie-scattering 518
Two-mode combustor model 571
Two-phase interactions 419 420
Uncontrolled combustion instability 587

United Technologies Research center 582
Unstable combustion 481
actuation system 582
Chemiluminescence measurements for 485
combustor dynamics of 582
controller/algorithm 583
flame area/heat release variation with time
during 516
flamed-structure evolution during 496 498
flame-vortex interactions 187
nonlinear 18
overall heat-release fluctuations/combustor
pressure fluctuations’ measurements of 494
premixed, schematic illustration of 536
Index Terms Links
Unstable combustion (Cont.)

pressure measurements for 482
sensing 583
time trace/power spectrum during 483
Unstable combustor
factors affecting control of 581
nomenclature for 581
summary for 606
fundamental limitations of achievable
performance and 601
oscillations in 15
Unstable flame 253
Unstable pressure trace 501
Unsteady combustion heat release 277
Unsteady combustor pressure 132
Unsteady dilation 308 309
Unsteady flashback phenomena 356
Unsteady flow evolution 245
Unsteady heat addition 376
Unsteady heat release 399
concentration of in a single axial plane 378
entropy/vorticity disturbances generated by 323
in phase with pressure perturbation 377
sound/vorticity/entropy generation by 323
velocity’s relation to 336
Unsteady strain rate effects 181 203
Unsteady stretch effects 345
Unsteady velocity 327
Index Terms Links
Unsteady vorticity 323
Valve authority 136

Valve/feed-line/injector (VFI) 135
VBB. See Vortex breakdown bubble
Velocity 292 330 338 439
acoustic 186
axial 237 245 249 292
293 307 425
azimuthal 227 229 239 249
disturbance of 330
exit 121
flame-burning 324
fluctuations 185
axial 425
circumferential 425
radial 425
inlet 287
Instantaneous axial, fields 219 236
jump condition 380
laminar-burning 354
mean flow 347
nonuniformity field 337
normalized mean axial 245
oscillations 439
Index Terms Links
Velocity (Cont.)
perturbations 195
radial 249
relative 207
shear-wave convection 322
unsteady 327
unsteady heat release’s relation to 336
Venturi 127 250
VFI. See Valve/feed-line/injector
Viscous damping mechanisms 14
Viscous dissipation 419 420
Volumetric dilation 282
Vortex
breakdown 226 270
burning 189
flame interactions with 267
merging 547
motion 277
rollup 187
shedding 11 23 189 536
periodic 288
spiral 261
structure 303
structure of 303
Vortex breakdown bubble (VBB) 290
Vortex-driven fluctuations 183
Vortex-droplet coupling 301
Vortex-flashback 257
Index Terms Links
Vortex-flashback process 255 257

Vortical disturbances 319 390
Vortical flow evolution 217
of coaxial swirl injector 217
interaction and competition of instability
modes and 230
outer shear-layer instability and 229
of radial-entry swirl injector 225
vortex breakdown and 226
Vortical perturbation 317
Vortical structures 321
Vortices
Large-amplitude oscillations driven by 189
reacting 189
Vorticity
equation for 282
generation 323
spanwise 300
thickness 223
unsteady 323
waves 392
Vorticity disturbances 323
propagation of 320
Vortico-acoustic interaction 220
Index Terms Links
Wake flows 291

Wake recirculation zone (WRZ) 258
Wave equation 418
solution of 421
Wave-propagation mechanism 235
Waves 74 76 148 181
392
acoustic 78 185 223
acoustic interactions of 315
circumferential 389
combustion’s relation to 181
entropy 388 392
fuel-air, coupling 45
heat-release fluctuations driven by 197
homogenous, equation 379 384
inhomogenous equation of 381
linear 5 397
plane solutions of 374
plane sound 386
pressure 195
pressure, attenuator 86
quarter, resonator 561
quarter, tubes 46
vorticity 392
well-established vortical 234
Weber number 520
Index Terms Links
Wedge flames 329 334 335

Wedge-flame transfer function 332
Well-established vortical wave 234
Well-stirred reactor (WSR) 357
basic analytical framework of 357
conceptual problems with 360
flow perturbations responded to by 358
reactor residence time of 359
Westinghouse/Georgia Institute of technology 582
White Gaussian disturbance 599
Wrinkled flame 350
WRZ. See Wake recirculation zone
WSR. See Well-stirred reactor
Zero mean flow 386

Zero-mean-flow jump condition 387
Zeroth-order approximation 382
AUTHOR INDEX
Index Terms Links
Banaszuk, A. 581
Bellucci, V. 445
Bethke, S. 89
Blust, J. 29
Candel, S. 179
Cohen, J. M. 113 581
DeLaat J. 113
Dowling, A. P. 369
Ducruix, S. 179
Durox, D. 179
Flohr, P. . 89
Flohr, P. 445
Index Terms Links
Goy, C. J. 163
Held, T. J. 43
Hermann, J. 611
Hoffmann, S. 611
Hsiao, G. C. 43
Huang, Y. 213
James, S. R. 163
Johnson, C. 89
Krebs, W. 89
Lee, J. G. 481
Lepers, J. 89
Lieuwen T. C. 3 315
Menon, S. 277
Mongia, H. C. 43
Index Terms Links
Pandalai, R. P. 43
Paschereit, C. O. 445
Prade, B. 89
Proscia, W. 113
Rea, S. 163
Richards, G. A. 533
Robey, E. H. 533
Santavicca, D. A. 481
Sattinger, S. 89
Scarinci, T. 65
Schuermans, B. 445
Schuller, T. 179
Sewell, J. B. 147
Smith, K. O. 29
Sobieski, P. A. 147
Stow, S. R. 369
Straub, D. L. 533
Wang, S. 213
Index Terms Links

Docslide - Us Combustionj PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Docslide - Us Combustionj PDF

Uploaded by

Copyright:

Available Formats

Combustion Instabilities in Gas

Frank K. Lu, Editor-in-Chief

1-56347-669-X/05 $2.50 + .50

John D. Binder Richard M. Lloyd

Steven A. Brandt Ahmed K. Noor

Fred R. DeJarnette Albert C. Piccirillo

Philip D. Hattis Ben T. Zinn

Abdollah Khodadoust Peter H. Zipfel

Chapter 1. Combustion Instabilities: Basic Concepts . . . . . . . . . . . . . . . . . 3

II. Case Studies

Chapter 2. Combustion Instabilities in Industrial Gas Turbines:

Chapter 3. Incorporation of Combustion Instability Issues into

Chapter 4. Combustion Instability and Its Passive Control:

Chapter 5. Thermoacoustic Design Tools and Passive Control:

Chapter 6. Characterization and Control of Aeroengine Combustion

Active-Control Demonstration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

Chapter 7. Monitoring of Combustion Instabilities:

Chapter 8. Monitoring Combustion Instabilities: E.ON

III. Fundamental Processes and Mechanisms

Chapter 9. Combustion Instability Mechanisms in

Chapter 10. Flow and Flame Dynamics of Lean Premixed

Chapter 11. Acoustic-Vortex-Flame Interactions in Gas Turbines . . . . 277

Chapter 12. Physics of Premixed Combustion-Acoustic

IV. Modeling and Diagnostics

Chapter 13. Acoustic Analysis of Gas-Turbine Combustors . . . . . . . . . . . 369

Limit-Cycle Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406

Chapter 14. Three-Dimensional Linear Stability Analysis of Gas

Chapter 15. Implementation of Instability Prediction in Design:

Chapter 16. Experimental Diagnostics of Combustion Instabilities . . . 481

V. Combustion Instability Control

Chapter 17. Passive Control of Combustion Instabilities in

Chapter 18. Factors Affecting the Control of Unstable

Chapter 19. Implementation of Active Control in a Full-Scale

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635

Combustion Instabilities: Basic Concepts

Ben T. Zinn∗ and Timothy C. Lieuwen†

C OMBUSTION instabilities are characterized by large-amplitude oscillations

What Causes Combustion Instabilities?

Driving Flame adds energy to acoustic field when

phase between heat

Damping Oscillations damped by viscosity, heat transfer,

Condition for Driving of Damping of

Transverse Radial Mode

Transverse Azimuthal Mode

II. Historical Overview

With a focus on combustion-driven oscillations for the remainder of the chapter,

Fig. 1.4 Landﬁll gas ﬂares that experienced low-frequency, combustion-driven

Combustion instabilities have also been encountered in numerous solid-

III. Causes of Instabilities

an instantaneous “heat-addition pulse” when its pressure is maximum and cooling

1) Fuel Feed Line–Acoustic Coupling.5,16 Pressure oscillations in the combustor

4) Oscillatory Flame-Area Variation.18 Interactions of acoustic velocity oscilla-

Heat-release oscillations add energy to an acoustic ﬁeld if the magnitude of

Fig. 1.9 Hypothetical responses of various combustor processes to a periodic pressure

According to the Rayleigh criterion, the heat-release oscillations shown in

p' (arbitrary units)

Fig. 1.10 Measured25 dependence of the instability amplitude on the premixer

1. Viscous and Heat-Transfer Damping Mechanisms

2. Convection and/or Radiation of Acoustic Energy

0.85 at M = 0.05, implying that the presence of a very-low-velocity ﬂow increases

3. Transfer of Energy Out of Natural Acoustic Frequencies

IV. Growth and Saturation of Instabilities

A. Linear and Nonlinear Stability