Académique Documents
Professionnel Documents
Culture Documents
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 permissionof the copyright owner is unlawful. The code followingthis statement
indicates the copyright owners 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 efciencies and emit fewer pollutants than other major chemical-energy
conversion devices. In addition, the lowcapital 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.
Avariety 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 specic 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 SO
2
, is a major contributor to acid rain
This book focuses on a particularly serious difculty in low emissions gas tur-
bines: combustion-driven oscillations. These instabilities routinely constrain the
operatingenvelope andpower output of eldedmachines and, insome cases, leadto
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 difculties.
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
PREFACE xiv
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 fromevery contributor
to the eld, 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 specic experiences with combustion instabilities in the development
stage and in elded 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 ames 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 rst 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 Congurations 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
CombustionAcoustic 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 Modications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Chapter 5. Thermoacoustic Design Tools and Passive Control:
Siemens Power Generation Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Werner Krebs, Sven Bethke, JoachimLepers, Patrick Flohr, and Bernd Prade,
Siemens AG, M ulheim, 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 InjectorAir Swirler Dynamic Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Subscale Combustor Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
ix
Active-Control Demonstration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
Chapter 7. Monitoring of Combustion Instabilities:
Calpines 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
UKs 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
Benets 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
S ebastien Ducruix, Thierry Schuller, Daniel Durox, and S ebastien Candel,
CNRS and Ecole Centrale Paris, Ch atenay-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,
University Park, Pennsylvania
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
Cold Flow Characteristics of Swirl Injectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Flame Dynamics of Axial-Entry Swirl Injector . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
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
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 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
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
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,
University Park, Pennsylvania
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
Theoretical Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
Solution Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
Sample Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
Chapter 15. Implementation of Instability Prediction in Design:
ALSTOM Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
Christian Oliver Paschereit, Hermann-F ottinger-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
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524
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
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 570
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
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 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, M ulheim, 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
David S. Lewis Jr. Chair and Regents Professor, School of Aerospace Engineering.
t
p
(t)q
(t)dt > 0
T
p
(x, t ) q
(x, t ) dt dV
i
L
i
(x, t ) dt dV (1.1)
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 eld. 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 uctuation in the heat-
release rate, 2) the heat-release uctuation excites acoustic oscillations, and 3) the
acoustic oscillations generate the velocity and thermodynamic-state variable uc-
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 identied, as indicated in Fig. 1.7. Because most of these mechanisms
B. T. ZINN AND T. C. LIEUWEN 10
Heat Release
Oscillations
Acoustic
Oscillations
Flow and
Mixture
Perturbations
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 LineAcoustic 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 ame in which it
produces heat-release oscillations that drive the instability.
3) Oscillatory Atomization, Vaporization, and Mixing.
5, 16
Interactions of the
acoustic eld 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 ame and/or periodic variations of the equivalence ratio that produce
heat-release oscillations that drive the acoustic eld.
Unsteady mixing,
vaporization,
atomization
Fuel flow rate
oscillations
Vortex/flame
interactions
Fuel/air ratio
oscillations
Flame area and
reaction rate
oscillations
Combustion
Products
Flow rate
oscillations
Fig. 1.7 Flow and ame processes that can cause combustion instabilities in
gas turbines.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 11
Fig. 1.8 Computed image of swirling ame 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 ame cause periodic variation of the ame area and, thus, a periodic
heat-addition process that drives the acoustic eld.
5) Vortex Shedding.
19, 20
Large-scale, coherent vortical structures caused by ow
separation from ame holders and rapid expansions, as well as vortex break-
down in swirling ows, 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 ame and cause its surface
area to oscillate, thus producing an oscillatory heat-release rate process that can
also drive the acoustic eld 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 eld 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 fuelair mixing to a pressure pulse with period T. The uctuations in each of
the rates of the preceding four processes causes a uctuation in the heat-release
rate after a time delay , 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
1
depends on various geometric and operating
parameters of the fuel-delivery system; for example, the fuel-line length and fuel
gas temperature.
B. T. ZINN AND T. C. LIEUWEN 12
time
T
TT
a. Pressure
at flame
b. Process 1
c. Process 2
d. Process 3
e. Process 4
4
Heat Release
Perturbation
due too:
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 eld 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 satises this criterion and that the heat
release by processes 24 does not. The heat release of process 1 and the pressure
pulses are positively correlated because the characteristic time of the process
1
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 systemcan 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 rst 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
1
= 2T and
2
= 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 showthat
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 signicantly differ from those that excite a
5000-Hz transverse-mode instability. Second, these examples showthat 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
COMBUSTION INSTABILITIES: BASIC CONCEPTS 13
0
0.02
0.04
0.06
0.08
15 20 25 30 35 40
Premixer Velocity (m/s)
p
'
(
a
r
b
i
t
r
a
r
y
u
n
i
t
s
)
430 Hz
630 Hz
Fig. 1.10 Measured
25
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 eld; instead,
these characteristic times must satisfy a relationship of the form T T/4 < n <
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 mechanismthat depends on the time required to trans-
port the reactants fromthe 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 nding
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 inuence their magnitude.
Energy can be dissipated/removed froman 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
B. T. ZINN AND T. C. LIEUWEN 14
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 ow-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 ow 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 uctuations. 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
uctuations. Consequently, the energy of the acoustic wave reecting from the
wall is smaller than that of the incident wave, because some of its energy is con-
verted into vorticity and/or entropy uctuations. These processes are analogous to
those occurring in a steady pipe ow in which heat transfer and viscosity result in
stagnation-pressure losses. The magnitudes of these dissipation mechanisms in-
crease with frequency as ( f
v
)
1/2
, where
v
is the viscous or thermal transport
timescale.
Flowseparation at sharp edges or during rapid owexpansions also damp acous-
tic waves by converting acoustic energy into vorticity. This damping mechanismis
analogous to the stagnation-pressure losses in steady, separated ow downstream
of sharp corners or edges, tabulated in many engineering handbooks and uid 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 uid 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 propagatingout throughits openend. Ingeneral, this dampingmech-
anismscales 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 uid motion with a magnitude that roughly scales with the mean ow
Mach number. Because the magnitude of acoustic damping in a typical system is
quite small, in general, the presence of low-velocity mean ow can have a signif-
icant impact on the systems damping level. To illustrate this point, consider the
dependence of the magnitude of a reected acoustic wave incident on the open
end of a pipe upon the ow Mach number in Fig. 1.11. It shows that the reection
coefcient equals 0.95 in the no-ow case, implying that the wave amplitude is re-
duced by 5%because of acoustic radiation. This reection coefcient is reduced to
COMBUSTION INSTABILITIES: BASIC CONCEPTS 15
0
0.2
0.4
0.6
0.8
1
0 0.1 0.2 0.3 0.4
Mach Number
R
e
f
l
e
c
t
i
o
n
C
o
e
f
f
i
c
i
e
n
t
Fig. 1.11 Dependence of the reection coefcient from an open-ended pipe on the
ow Mach number. Taken from Ingard and Singhal.
41
0.85 at M = 0.05, implying that the presence of a very-low-velocity owincreases
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 specic 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 amplied
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, uctuations at other frequencies or
to incoherent, broadband uctuations. Nonlinear combustor processes are a key
mechanismfor enabling the former; that is, they transfer energy froma certain fre-
quency f
o
, to higher harmonics (2 f
o
, 3 f
o
, . . .) or subharmonics ( f
o
/2, f
o
/3, . . .).
The energyat these higher frequencies is usuallydissipatedmore rapidlybythe 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 reection and scattering of a
sound wave from turbulent eddies
23
or a randomly apping ame front.
24
The
energy-transfer mechanism in this case can be purely linear and is essentially
caused by a random Doppler shift of the reected and transmitted waves. For ex-
ample, when a pure tone with a temporal dependence given by sin(t ) impinges on
a randomly moving ame front or a mediumwith a randomvelocity eld, reected
and transmitted waves with a random phase shift are generated. The temporal de-
pendence of the reected/transmitted waves is given by sin[t +(t )], where (t )
is a random phase shift. This mechanism is discussed further in Chapter 12.
B. T. ZINN AND T. C. LIEUWEN 16
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 innitesimally 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) =
H
A +H
n
(A) (1.2)
D(A) =
D
A +D
n
(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
A
D
r
i
v
i
n
g
/
D
a
m
p
i
n
g
D(A)
H(A)
D
1
H
1
A
LC
Fig. 1.12 Hypothetical dependence of the acoustic driving, H( A), and damping, D( A),
processes on the instability amplitude A.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 17
D(A), resulting in further growth of the disturbance. Because these two curves
diverge near the origin, their difference increases with amplitude, implying that
the amplitude growth rate increases with amplitude. When the amplitudes of the
oscillations are small, the driving and damping processes can be approximated
by linearized expressions; that is, H(A) =
H
A and D(A) =
D
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 e
t
of inherent combustor disturbances. It can be shown that
the initial growth rate of the instability is proportional to the difference between
the driving and damping processes in the linear regime; that is,
H
D
.
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
h
increases while
d
remains constant (see Fig. 1.12).
For
h
<
d
, A = 0 is the stable solution; that is, the system is linearly stable be-
cause all perturbations imposed on the systemwill decay. However, when
h
>
d
,
the solution A = 0 becomes unstable, causing the amplitude of the disturbance to
increase toward a new stable, limit-cycle equilibrium. The
h
=
d
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
h
d
.
Figure 1.13 shows that as
h
d
becomes positive, the system becomes linearly
h
-
d
A
m
p
l
i
t
u
d
e
Stable
Unstable
Fig. 1.13 An example of a supercritical bifurcation.
B. T. ZINN AND T. C. LIEUWEN 18
0
0.005
0.01
0.015
0.02
18 21 24 27 30
Mean Inlet Velocity (m/s)
N
o
r
m
a
l
i
z
e
d
P
r
e
s
s
u
r
e
(
p
'
/
p
)
Fig. 1.14 Measured data describing the occurrence of supercritical bifurcation at
combustion inlet velocity of 23.5 m/s.
25
unstable. Although
h
d
describes variation along the x-axis in Fig. 1.13, it
could be replaced in practice by any parameter that affects the systems 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 sufciently 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 ow 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 uctuate slightly. However, pressure
perturbations of sufciently 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.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 19
A
D
r
i
v
i
n
g
/
D
a
m
p
i
n
g
D(A)
H(A)
D
1
H
A
LC
A
T
1
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 ow,
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 difcult
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
equilibriumpoints at which the driving and damping curves intersect. Specically,
the damping exceeds the driving when A < A
T
, indicating that A = 0 is a stable
xed 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
ames 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
h
<
d
, the A = 0 and A = A
LC
are stable solutions, as noted
B. T. ZINN AND T. C. LIEUWEN 20
h
-
d
A
m
p
l
i
t
u
d
e
Stable
Unstable
A
T
Fig. 1.17 An example of a subcritical bifurcation.
before. For
h
>
d
, 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
h
d
from
a negative to a positive value, the systems amplitude will jump discontinuously
from A = 0 to A = A
LC
at
h
d
= 0. Hysteresis is also present in the system,
because if the systemparameter is subsequently decreased, the systems amplitude
decreases as it follows the stable branch on top, even for a range of
h
<
d
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 nite amplitude oscillations.
0
0.0025
0.005
0.0075
0.01
0.0125
0.015
13 13.5 14 14.5 15 15.5
Mean Inlet Velocity (m/s)
N
o
r
m
a
l
i
z
e
d
P
r
e
s
s
u
r
e
(
p
'
/
p
)
Fig. 1.18 Experimental data showing evidence of a subcritical bifurcation. Taken
from Lieuwen.
25
COMBUSTION INSTABILITIES: BASIC CONCEPTS 21
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,
p(t )
p
=
(t )
(1.3)
Assume that the density oscillates harmonically with an amplitude A and angular
frequency , that is,
(t )
= A sin t (1.4)
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 )
p
=
( 1)
4
A
2
+
A +
( 1)( 2)
8
A
3
sin t
( 1)
4
A
2
cos 2t
( 1)( 2)
24
A
3
sin 3t + (1.5)
Equation (1.5) shows that the expansion for the pressure consists of a linear term
that is proportional to A (the second termon 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 (rst term on the right) that is proportional to A
2
and
represents the change in the mean pressure caused by the purely oscillatory density
disturbance. Furthermore, the expansion includes harmonics that are proportional
to A
2
, oscillating at twice the disturbance frequency, and terms proportional to
A
3
, oscillating with frequencies of 3 and . 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 signicant shifts in mean pressure because
of these instabilities. However, changes in mean ame 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
B. T. ZINN AND T. C. LIEUWEN 22
by ow 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 NavierStokes and energy equations,
for example, convective terms such as u u in the NavierStokes equation
or the previously discussed nonlinear pressuredensity relationship. Such terms,
in general, become signicant when the amplitudes of the uctuating 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
/ p < 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, ow
oscillations may control the response of the combustion process, resulting (in
these cases) in nonlinearities that become important when the ratio u
/ u 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, ow 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 | u|
2
, resulting in an unsteady damping process that is proportional
to (| u
|
2
+2| u
| |
u|) +
(|
u|
2
+| u
|
2
+2| u
| |
u|).
35
This expression indicates
that the presence of mean ow 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 ow (i.e., |
T. J. Held,
G. C. Hsiao,
and R. P. Pandalai
Lead Engineer.
Staff Engineer.
43
H. C. MONGIA ET AL. 44
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
combustionacoustic 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 combustionacoustic 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 combustionacoustic 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. Amore 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 ight-
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 reected 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 rst category of coupling mechanisms is the pressure-disturbance inter-
action with the instantaneous ame position and shape. As the ame 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 identied with high-frequency (>1 kHz) acoustic ac-
tivity. Lieuwen
1
(see also Chap. 12) gives an excellent reviewon the modeling pro-
cesses of premixed combustionacoustic wave interaction. Based on analytical and
experimental observations, it appears that the combustion process is increasingly
sensitive to perturbation in the equivalence ratio under lean operating conditions.
The second category of coupling mechanisms is often termed fuelair wave
coupling. Within this category are several submechanisms, which share a com-
mon interaction between a pressure oscillation and the local fuelair ratio of the
combustor. The physical interaction can take place through effects on local airow
rate, fuel ow rate, or fuel-spray characteristics. This type of interaction is often
identied with midrange frequencies (1001000 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 fuelair ratio to extinguish.
The energy reaching the turbine immediately reduces, causing the rotor speed to
decrease. The airow through the engine is reduced, causing the fuelair 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 classied as a
combustionacoustic 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 signicant impact on combustion-
dynamics behavior.
2, 3
These changes can inuence either or both the fuel and
airow paths
4
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 ow (and thus ame temperature)
H. C. MONGIA ET AL. 46
distribution,
5
inaxial fuel stagingadjustments,
6
or inasymmetric fuel distribution.
7
Such fuel-distribution strategies generally deteriorate the nitrous oxide (NOx)
emissions performance of the combustor, because they shift the ame 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 resonators
8
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 congura-
tion of the devices.
Active controls also take multiple forms, the most common being modulation
of the fuel ow with a frequency and phase relationship designed to destructively
interact with the combustion-dynamic oscillation. This modulation can take the
formof the main fuel ow,
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 signicant 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 from1 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 newdiffusion-ame 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 modications as summarized in Fig. 3.2. The basic
concept in these xes is to change the interaction between the fuel-injection system
and the combustor to eliminate coupling between the two systems.
Two recent examples of postcertication 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 rst 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 identied combustorfuel
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 47
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 ame 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 red near its maximum fuel ow. Because
reliable analytical design tools were not available, an iterative empirical approach
was used to x the instability problem. Several candidate approaches were identi-
ed and screened for ease of implementation, including alternate fueling modes,
Helmholtz resonators, fuel-nozzle cavity modications, 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 rst 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 inuence of pilot
fuelair 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 ying
test bed was subsequently conducted to get additional data from an altitude en-
gine environment. Data from the ying 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 identied a new fueling mode as the optimal conguration
H. C. MONGIA ET AL. 48
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 rigtests were followedbynal combustion-systemrenements 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.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 49
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 ow to each of the three domes is used to operate within the combined
constraints of emissions, combustion acoustics, and demanded total fuel ow rate
(shaft power). In addition, compressor air is bled from the engine at part-power
operation to extend the range over which the ame temperatures can be maintained.
As power is further reduced, radial and circumferential staging modes are also used
to bring ame 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
H. C. MONGIA ET AL. 50
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 ame 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 ame temperatures are affected
by variation in fuel property, ambient conditions, and load changes; thus, unstable
operation of the combustor can develop from changes in these parameters.
Fig. 3.6 The LM6000 DLE premixer showing ELBO fuel injection for controlling
dynamics.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 51
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 ame 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 ows for the operating
conditions based on a cycle model calibrated for each engine. The ABAL control
system increases the ame 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 eld 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 ight-
engine combustors will likely follow the path set by the DLE combustor toward
increasing levels of premixing and lower ame temperatures.
16
Thus, a strong need
exists to drive the design process toward a more predictive capability of identi-
cation and mitigation of combustion-dynamics phenomena in such combustors.
V. CombustionAcoustic 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 eld is established
through the use of either empirical means or by computational uid dynamics. The
remainder of the modeling problem consists of determining boundary conditions
and dening 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 CombustionAcoustic 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 ight of individual
fuel parcels from the point of injection to the ame front. The primary underly-
ing assumption here is that the turbulent ame brush can be modeled as a steady,
contiguous thin reaction zone anchored at xed 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 airow rates.
Even when such uctuations are ignored, the single convective delay-time ap-
proximation assumes that the ame sheet is positioned in such a manner that the
time of ight for each fuel parcel is equivalent. The local owvelocity and turbulent
H. C. MONGIA ET AL. 52
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.
ame speed determine the location of the ame sheet. Given the complexities of
the oweld 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 conned to a narrowrange. In addition, variation between premixers,
whether by design or chance, can result in further disparity in convective times.
A steady-state computational uid dynamics (CFD) simulation
17
of a 24-deg
sector of the LM6000 DLE combustor provides an illustrative example of the
variation in convective timescales. Plots of axial-ow 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 ame 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 ame 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 ame front, a threshold value
of unity is chosen for the reaction progress variable G(x, t ). This choice denes the
ame 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. Fuelair mixing is not modeled; instead, experimentally
derived proles of velocity and mixture fraction are used as inlet boundary condi-
tions. As such, the convective delay times represent only the time of ight fromthe
premixer exit planes to the ame sheet. The time of ight 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 sufcient to illustrate the complications of choosing a single timescale.
As seen in Fig. 3.8, convective times between the premixer exit and the ame
front range from0.1tomore than0.7ms. The median, the mass-weightedmean, and
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 53
Fig. 3.8 Distribution of convective times from the premixer exit plane to the ame
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 fuelair mixture in the premixer is
approximately 0.45 ms. Statistical differences between individual premixers are
signicant because median times for individual premixers range from 0.25 to
0.47 ms. The overall distribution is not well t 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 0.10.7 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 associatedwiththe 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 CombustionAcoustic Model
Rather than utilize a fully analytical description of the combustionacoustic
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 injectorswirler 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 signicant radial and circumferential components. A
single-cup test can only support a rst- or second-order mode, providing only one
or two resonant frequencies with which the fuel nozzleswirler combination can
interact. Second, the fuel system can potentially also play a role in affecting the
H. C. MONGIA ET AL. 54
Fig. 3.9 Tunable combustion acoustics (TCA) rig conceptual view.
phase relationship between a pressure disturbance and the resulting equivalence-
ratio uctuation that reinforces the pressure wave. Third, in general, the acoustic
inow and outow boundary conditions are not well represented by a single-cup
test. The choked outow boundary of a turbine nozzle can be simulated with a re-
stricted test combustor exit, but the inow boundary of the nal compressor stages
is more difcult 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 inow and outow boundaries, fuel-system dynamics, aerodynamic
pressureequivalence ratio interactions, ame 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 insuf-
ciently understood to permit rst-principles modeling. In particular, the interaction
of acoustic waves with the ame front, fuel ow, and airow rate is complex and
not well understood. Thus, the combustionacoustic 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 difculty, 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 nozzleswirler 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 nozzleswirler
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 rst
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
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 55
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 ow 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-dened 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 signicant circumferential component. Provided
that the primary acoustic coupling mechanism is through pressure-wave-induced
uctuations 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 uctuation, the mode shape will
not signicantly alter the response of the fuel nozzleswirler. However, the ame
dynamic response will also depend on the velocity uctuation, 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.
H. C. MONGIA ET AL. 56
minimumlength 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 gure. As is evident, the system response strongly depends
on the combined natural resonant frequency of the combustion chamber and the
response of the fuel nozzleswirler. Because of the relative simplicity of the phys-
ical experiment, computational modeling of the test vehicle can be accomplished
with condence. The combustoracoustic 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 shapecombustor
response independence described earlier, this transfer function is expected to be
applicable to the full annular model of the combustor as well.
Asignicant advantage of this approachis that it permits evaluationof 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 xed-
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 xed-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 combustionpressure wave interactions. Signicant progress has
been made in time-dependent modeling of the combustion process through large-
eddy simulation (LES) for modeling combustion-generated ow instability.
1921
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-
ed 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 inowand outowboundary conditions are approximations
to those in a gas turbine. Fuel-system interactions and feedback are either ignored
or reduced to simplied models that require their own assumptions of physical
coupling mechanisms.
Despite these difculties, 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
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 57
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 condence 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 ame 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 specied operating condition.
These gures 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 ame temperature is a critical
factor because of its inuence 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.
H. C. MONGIA ET AL. 58
a)
b)
c)
d)
Fig. 3.12 Effect of radial temperature nonuniformity on combustion dynamics.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 59
of the ame 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 rst-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 uniformtemperature in all three
domes. The measured spectra shown in these gures were based on dynamic pres-
sure measurements from pressure sensors installed just downstream of the ame
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 fromthe analysis, similar to those
observed from engine tests. These gures 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 ame 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 ow 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 ow 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 upstreamof 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 nite number of discontinuities at the
locations where they are installed. The incident and reected acoustic waves in the
diffuser cavity are signicantly 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
indexat the tuningfrequencyof the damper tubes is predictedbasedonthis analysis.
H. C. MONGIA ET AL. 60
Fig. 3.13 Schematic illustration of acoustic damper tube in combustor.
Fig. 3.14 Predicted impact of damper tubes on combustor acoustic response.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 61
Although analytical techniques have improved markedly during the past few
years, the difculty in representing the combustionpressure wave interaction will
likely remain a barrier to full predictive capability. Thus, it is likely that a combined
approach, including both experimental and high-delity unsteady CFD modeling,
will be required to develop models of the full annular combustor response that will
have utility in dening design direction.
VII. Conclusion
Combustion instability in conventional diffusion-ame gas-turbine combustors,
if encountered, can be eliminated, in general, with simple modications 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-certication phase. If encountered, simple analysis coupled with
empirical guidelines and a systematic testing and development process are used to
nd 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 ame stability and/or heat-release rates that
lead to strongly coupled nonlinear interaction between the ame exothermicity,
acoustic behavior of the system, and components. The approach for controlling
dynamics in the LM engines (bleed air, fuel staging, dome ame 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 signicant 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 combustionpressure 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. 222444.
2
Steele, R. C., Cowell, L. H., Cannon, S. M., and Smith, C. E., Passive Control of
Combustion Instability in Lean Premixed Combustors, American Society of Mechanical
Engineers, Paper 99-GT-52, 1999.
H. C. MONGIA ET AL. 62
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
Tr aptau, W., Development History and Field Experiences of the First F T8 Gas Turbine
with Dry Low NO
X
Combustion System, American Society of Mechanical Engineers,
Paper 99-GT-241, 1999.
9
Johnson, C. E., Neumeier, Y., Lubarsky, E., Lee, J. Y., Neumaier, M., and Zinn, B.
T., Suppression of Combustion Instabilities Using a Fast Adaptive Control Algorithm,
AIAA Paper 2000-16365, 2000.
10
Paschereit, C. O., Gutmark, E., and Weisenstein, W., Structure and Control of
Thermoacoustic Instabilities in a Gas-Turbine Combustor, Combustion Science and
Technology, Vol. 138, 1998, pp. 213232.
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. 17.
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, 1517 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
NO
X
, 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. 782794.
20
Stone, C., and Menon, S., Open-Loop Control of Combustion Instabilities in a Model
Combustor, Journal of Turbulence, Vol. 4, 2003, p. 20.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 63
21
Schl uter , J. U., Static Control of Combustion Oscillations by Coaxial Flows: An
LES Investigation, Journal of Propulsion and Power, Vol. 20, No. 3, 2004, pp. 460467.
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
=pressure uctuation
P3 =compressor discharge pressure
Re =real part of a complex number
s =spacing between two points of fuel injection
s
k
=strength of the kth fuel-injection source
T3 =compressor discharge temperature
T
i
=ame temperature of the i th combustion zone
u =instantaneous velocity
U =average velocity
u
=velocity uctuation
Copyright c 2005 by Rolls Royce Plc. Published by the American Institute of Aeronautics and
Astronautics, Inc., with permission.
i
=regression constants [Eq. (4.2)]
=ratio of heat capacities
=sound wavelength
=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 efciency of 43.1%.
Cross sections of the engine with a detail of the DLE combustor (as originally
congured 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
ow 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 ames out, the whole combustor ames 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 ame
stabilization. The result is a large turndown ratio in the achievable fuelair ratios
of the secondary and tertiary premixers.
Part of the secondary fuelair mixture is entrained into the primary zone to
mix with the primary premix stream. Once the ame temperature of the secondary
streamreaches a certain level, this entrainment results in a signicant 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 maximumtem-
perature for any of the three stages because of nitrous oxide (NOx) reasons, and a
minimumtemperature 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 denes an operating envelope, the axes of which are best dened 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, from1200 Kto 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-ame temperatures,
respectively, from which a reference temperature was subtracted. Any point inside
T. SCARINCI 68
-200
-150
-100
-50
0
50
100
150
200
250
-300 -200 -100 0 100 200 300 400
secondary premixer temperature, K
p
r
i
m
a
r
y
p
r
e
m
i
x
e
r
t
e
m
p
e
r
a
t
u
r
e
,
K
80% power
100% power
Fig. 4.2 Typical operating envelopes for the Trent 60 combustor (ISO Day).
the operating envelope denes 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 50100% 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 ame 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 signicant effort
to break away from the limit cycle, that is, there can be signicant 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.
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 69
The amplitude of these limit cycles is typically on the order of 12% of the
static pressure inside the combustion chamber, and typical frequencies seem to
range from approximately 100 Hz to roughly 12 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 systemof 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 signicantly. 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.
T. SCARINCI 70
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
0.00 0.25 0.50 0.75 1.00
L /
P
/
P
r
e
f
(
%
)
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 nd regions of high- and low-pressure-oscillation amplitude at any
power condition. In other words, the rms level of pressure uctuations 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 difcult 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 nd an optimumcondition
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-dened 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 uctuations is on the order of 0.10.2% of the reference pressure. When
oscillatory combustion sets in, it always appears at the frequency of one of the
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 71
0%
20%
40%
60%
80%
100%
-50 0 50 100 150 200 250
secondary temperature bias, K
n
o
i
s
e
a
m
p
l
i
t
u
d
e
(
%
o
f
m
a
x
i
m
u
m
)
primary bias = 0 K
primary bias = +75 K
3rd mode
2nd mode
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).
rst 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 fromthe 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 gure provides a good illustration of the exibility offered by the three-stage
design. At a xed-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
/ = 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
/ = 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.
T. SCARINCI 72
0%
20%
40%
60%
80%
100%
120%
140%
-100 -50 0 50 100 150 200 250
primary temperature bias, K
n
o
i
s
e
a
m
p
l
i
t
u
d
e
(
%
o
f
m
a
x
i
m
u
m
)
50% power
80% power
100% power
Fig. 4.6 Effect of primary-zone temperature on the pressure-oscillation amplitude
of the second unstable longitudinal mode (L
c
/ = 0.5).
0%
20%
40%
60%
80%
100%
120%
-100 -80 -60 -40 -20 0 20 40 60 80 100
Secondary minus Primary Zone Temperature (K)
n
o
i
s
e
a
m
p
l
i
t
u
d
e
(
%
o
f
m
a
x
i
m
u
m
)
Fig. 4.7 Amplitude characteristics of the third unstable mode (L
c
/ = 0.75). The
data in this gure are obtained at engine power levels ranging from 30% to 100%
power.
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 73
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 sufcient 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 rst 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 Janus
2
and Lieuwen and Zinn
3, 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 fromthe combustion zone lead
to airow modulations in the premixer that, in turn, lead to a fuelair ratio (FAR)
modulation. The resulting FARdisturbance is convected through the premixer into
the combustion zone, leading to an oscillation in heat release. If the fuel supply rate
is xed, the relationship between ow-velocity oscillations and FAR oscillations
is given approximately by
FAR
FAR
=
u
u
(4.1)
Lieuwen and Zinn
3
further noted that, because of the low Mach number of the
ow in the passage (typically, M 0.05), small pressure uctuations can result
T. SCARINCI 74
in signicant fuelair ratio oscillations. This can be seen by noting that
u
U
=
1
M
p
P
(4.2)
Thus, a 1% uctuation in static pressure inside the combustor can typically result
in a 15%FARuctuation fromthe 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 RichardsLieuwen
mechanism. This expression seems to be useful shorthand and it recognizes the
empirical explanation rst 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 fuelair ratio
oscillations,
5, 6
none seem as important, or at least as powerful, as the Richards
Lieuwen mechanism. At least, the RichardsLieuwen mechanism seems to be the
most powerful in DLE combustion systems as they apply to most industrial gas
turbines.
In Fig. 4.8, a simplied 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 fuelair 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 RichardsLieuwen mechanism). The
time-varying FARfromthe premixers will also lead to a time-varying density eld
inside the combustor, a time-varying ame-surface area, and a time-varying ame-
burning speed. All of these can lead to the generation of pressure waves either by
way of entropy noise or heat-release oscillations. Asignicant effort has been made
during the past fewyears, in particular, using active-control methodologies, to nd
ways of eliminating FARoscillations in DLEsystems. The approach taken at Rolls-
Royce has been, somewhat analogous to acoustic-damping devices, to damp the
presence of FARoscillations fromthe 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 aimof 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. Abrief description of our efforts to implement such acoustic
resonators is described in Sec. II.B.3.
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 75
Supply Plenum
fluctuations
(non-acoustic)
Fuel Flow
Oscillations
W
f
(t)
Premixer Velocity
Oscillations
u (t)
FAR Oscillations
FAR (t)
Flame Surface
Oscillations
A
f
)
Pressure Waves
in Premixer
Propagation of
Pressure Waves
in Combustor
Combustor Velocity
Oscillations
u (t)
Heat Release
Oscillations
Q (t)
Generation of
Pressure Waves
P (t)
Burning Speed
Oscillations
S
u
(t)
Fluid Density
Oscillations
(t)
Interaction with
Choked Nozzle
(Entropy Noise)
FAR-Wave
Damping
Pressure
Wave
Damping
Aerodynamic
Damping
FAR-Wave
Damping
Pressure
Wave
Damping
Aerodynamic
Damping
Fig. 4.8 Driving mechanisms for combustion instability and means for suppression
of instabilities. The thicker line refers to mixing-induced oscillation (R-Lmechanism).
Also noted in Fig. 4.8 are aerodynamic uctuations (i.e., broadband turbulent
uctuations) which disturb the ame and cause a general broadband level of in-
creased combustor pressure uctuations. Efforts to minimize these aerodynamic
sources are described briey in Sec. II.B.4.
1. Fuel Staging
The variable distribution of heat release along the length of the combustor
provides an effective mechanismto 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.
Indevelopingthis correlation, it was assumedthat the nondimensional amplitude
of the pressure oscillation is predominantly a function of the temperature ratio
T. SCARINCI 76
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-oscillationamplitude. It might be argued
that the overall heat release in the combustor (and the corresponding temperature
ratio fromcombustor 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:
p
P
= K
1
+ K
2
P
A
3
N
i =1
_
(T
i
T
3
)
k
i
_
i
(4.3)
where K
1
, K
2
, A, k
i
, and
i
are all constants to be tted 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 sufcient fuel-split
options to control pressure oscillations whilst meeting emissions requirements.
2. FuelAir Ratio Wave Damping
As previously suggested, the RichardsLieuwen mechanism is believed to be
the dominant source of instability in most modern DLE combustors. The existence
of a nite time delay between the generation of a pressure wave and the associated
FAR uctuation is at the heart of this mechanism. Once a perturbation in airow
(or fuel ow) occurs at the point of fuel injection, a nite 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 fuelair 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 fuelair 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 systemmay often be linearly proportional to the time delay introduced
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 77
x = 0 x = - d
fuel
Air
A
x = 0 x = - d
s
B
fuel-1
Air
fuel-2
x = 0 x = - d
air jets
L
m
D
fuel inlet
x = 0 x = - d
Air
L
m
C
fuel-1 fuel-N
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
sufciently long mixing time (ignoring autoignition concerns for a moment) then
no FAR uctuations 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 fuelair ratio (temporal) uctuations
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 congurations were examined experimen-
tally, analytically, and computationally. These congurations 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-ow conditions (u
and
p
/U) of 30%.
The mean airow rate and the total fuel ow rate were held constant for all test
congurations.
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 RichardsLieuwen mechanism are excited when the time for the mixture
to travel from the injection point(s) to the ame lies within some range of integer
multiple of the acoustic period. Consider N sources of fuel-injection locations,
T. SCARINCI 78
such as shown in conguration C in Fig. 4.9. Each location has a strength s
k
and is
located at a position d
k
. Neglecting turbulent-dispersion effects for the time being,
it can be shown
9
that the LieuwenRichards mechanism causes amplication of
acoustic waves when the following inequality is satised:
Re
_
N
k=1
s
k
e
i
(
d
k (
1
U
+
1
c
)
2
)
_
0 (4.4)
This equation assumes for simplicity that the reactive mixture is immediately
consumed at x = 0. Referring back to Eq. (4.1), we also dene a parameter as
the proportional change of FAR oscillations divided by the proportional change in
velocity uctuation, that is:
=
_
FAR
rms
FAR
__ _
u
rms
u
_
(4.5)
Parameter is a useful gure of merit for evaluating different mixer designs. A
good premixer is then one in which approaches zero, that is, velocity uctuations
in the premixer do not result in any FAR oscillations at the exit. It is expected that
0<<1, in general, and that will be frequency dependent.
Note that the convective wavelength of these FAR oscillations is given by
c
=
U/f . When 1 it means that all the airow 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, 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 (
c
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 rst 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 passagean 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 amplication in the range of 01000 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 01000 Hz.
Figure 4.10 thus shows the problem associated with a single-injector location;
it is very difcult 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.
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 79
Fig. 4.10 Regions of positive amplication of pressure waves as the parameter d/U
(premixer residence time) is varied.
Potentiallyina combustor withonlya single unstable mode couldsucha locationbe
identied (e.g., see Chap. 2); however, it is much more difcult in the Trent 60 DLE
in which one is dealing with three distinct modes (and frequencies). The only way
that such a conguration could be made sufciently 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 conguration (conguration A in Fig. 4.9).
Typical rawdata obtained for u
(t ) and FAR
which
would give = 1.0, that is, the level of FAR uctuation expected to occur from
the airow uctuation u
Sven Bethke,
Joachim Lepers,
Patrick Flohr,
Power Generation.
89
W. KREBS ET AL. 90
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 efciency, % 35.1 39.0 38.0 39.3
Performance data, combined cycle-single shaft
Net power output, MW 101 407 293 391
Net efciency, % 52.6 57.7 57.0 58.4
with a gross power output of 266 MWis the largest engine in the 60-Hz range. The
cross section of the engine is shown in Fig. 5.2. It is red by 16 Can-type combus-
tion systems with steam-cooled transitions. The SGT6-5000F is also red 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 dened 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 rst natural mode in an industrial gas-turbine
combustor is between 50 and 300 Hz, depending on the geometry and ring 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 dened 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 eld
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 050 Hz, the IFD
range of 501000 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 eld.
Note that different frequencies respond to changes in operating conditions dif-
ferently. Whereas one IFD mode may be damped by an increase in the fuelgas
fraction in one of the fuel stages, another modes amplitude may increase because
W. KREBS ET AL. 92
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 500 1000 1500 2000 2500 3000 3500
Frequency (Hz)
A
m
p
l
i
t
u
d
e
L
F
D
I
F
D
H
F
D
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 inuence 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
Combustor
Plenum
Air
Fig. 5.4 Thermoacoustically relevant inuence parameters of the feedback cycle.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 93
properties of the burner plenum have an impact on the impedance of the burner
exit, thus affecting the magnitude of the pressure-induced uctuations in volume
ow. The acoustic properties of the fuel lines determine the magnitude of fuel ow
uctuations at the fuel nozzles. The interaction between the fuel ow and airow
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 inuenced by the acoustic-boundary conditions
at the combustor exit. The source of the instability is the perturbed ame. In that
context it is important to understand and quantify the heat-release uctuations
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 ame 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 14 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 eld 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 models
1
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 airow is discharged. Ninety percent of
the air ows 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 orices are represented by the fuel-nozzle
impedance; in more elaborate models, whole fuel lines are explicitly included.
W. KREBS ET AL. 94
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
acoustic analysis
Transfer matrix network Evaluate acoustic properties
like impedances for
gas-turbine components
Three-dimensional
acoustic analysis
Finite element methods,
Sysnoise
Evaluate acoustic-pressure
distributions for
a) monitoring
b) developing resonators etc.
Flame response Time-lag models Evaluate impact of design
changes on ame response
Flame response Unsteady computational
uid dynamics
Evaluate impact of design
changes on Flame Response
Full-stability analysis Transfer-matrix approach
Galerkin method
Evaluate impact of design
changes on stability
Interaction of components
Component Testing
Tunable rig with
variable-exhaust-
passage impedance
Evaluate impact of design
changes on stability
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 Kr uger 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 eld is encountered. This makes the development and application of three-
dimensional codes necessary to investigate the acoustic-pressure distribution. At
Siemens AGthree-dimensional nite element codes are appliedthat solve the three-
dimensional acoustic equations in the presence of nonviscous ow. The governing
equations are described by Bethke et al.
5
Figure 5.6 gives an overview on the main tasks of the three-dimensional nite
element acoustic analysis. The acoustic mode shapes resulting from the nite
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.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 95
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 ame.
Determine Eigenfrequencies and related mode-
shapes of acoustic pressure and velocity at low
and intermediate frequencies.
Input for stability analysis
Determination of impedances (e.g. at burner exit)
Interaction of combustion chamber and plenum
Excitations for combustor mechanical design
Evaluation of IFD-resonators
Estimate Eigenfrequencies and
related mode-shapes of acoustic
pressure and velocity at high
frequencies.
Optimized positions of HFD-
resonators
Evaluation of HFD-resonators
Fig. 5.6 Finite element acoustic models for gas-turbine combustor designs.
W. KREBS ET AL. 96
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 rst
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 gure 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 exible approach has been de-
veloped that is outlined here. Three different types of congurations with varying
complexity have been selected.
1) Simplied 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 nal 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
Vane-simulation-section
Can-combustion-chamber
Plenum
Transition-piece
Diffuser-inlet
Simulated ambience
Exhaust passage
Burner
Fig. 5.7 Typical computational domain for nite element acoustic analyses ranging
from diffuser inletcompressor exit to exhaust.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 97
improve the accuracy of the prediction the computational domain can be extended
as shown in Fig. 5.7 (left-hand side). The computational domain now includes the
incoming ow 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 nite 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 ame 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 ame-response function expresses the heat-release uctua-
tions induced by acoustic-pressure waves. The instantaneous response of a ame
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:
F
1
(t ) =
q
(t )/ q
p
(t )/ p
(5.1)
F
2
(t ) =
q
(t )/ q
u
(t )|
burner exit face
/ u
burner exit face
(5.2)
q
describes the integral of heat-release uctuation over the ame surface, dened
as
q
(t ) =
( r, t ) dV[W]
p
and u
) 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 dened earlier, the amplitude of a pressure oscillation grows if
the imaginary part of the complex eigenfrequency becomes negative. This part
is called the damping coefcient. 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 fromthe interaction of several different
components, numerous designoptions are available toincrease the thermoacoustic-
stability range. As described in the preceding section, solution methods and design
tools are needed to identify promising design modications 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 ame response.
One approach is to reduce the interaction of the thermoacoustic source with the
pressure eld 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 amplied in a certain frequency range. Their ef-
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 100200 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 (airow, fuel ow, and fuel distri-
bution). Such methods are called active combustion control or automatic tuning
systems. Both the high-bandwidth and low-bandwidth control approaches require
reliable real-time measurement of the combustion-process oscillations; dynamic
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 101
Table 5.4 Design options for extending the safe operation range of
gas-turbine combustion systems
No. Design option Impact on
Burner
1 Change ame-front location by modiying Flame response
burner exit geomentry (cylindrical burner, outlet)
2 Changing of fuel-concentration proles 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
24-burner
configuration
Engine:
Fig. 5.10 Modulus of the dynamic-pressure distribution: rst azimuthal mode shape
in annular combustion chamber.
W. KREBS ET AL. 102
in-phase
alternating out-of-phase
Acoustic excitation
at burners
Phase
Phase
Can-Can
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 rst 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 cancan 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
conguration (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
modied 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 ame 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 ame front.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 103
Fig. 5.12 Cancan interaction in can-annular combustion chambers.
The temperature distributions for the different design congurations 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
ame stabilization over the whole gas-turbine operation envelope. In addition,
outer-recirculation zones are formed that also help to stabilize the ame. As typical
Modifications
NBO
CBO short
CBO long
NBO
CBO short
CBO long
Fig. 5.13 Modication of the burner outlet by cylinders of different lengths (NBO=
normal burner outlet; CBO =cylindrical burner outlet).
W. KREBS ET AL. 104
NBO CBO-Short
CBO-Long
Fig. 5.14 Reacting oweld for the different burner-exit congurations.
for swirl-stabilized burners, the location of heat release is mainly aerodynamically
determined by the size of the recirculation zones. This size can be modied by the
application of cylindrical burner outlets (CBOs), and, hence, the location of the
ame front, characterized by the steep temperature gradients, is shifted.
The change of the oweld 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
0.00
0.20
0.40
0.60
0.80
1.00
0.0000 0.1000 0.2000 0.3000 0.4000
t / t_0 [ - ]
D
i
s
t
r
i
b
u
t
i
o
n
[
-
]
NBO
Short CBO
Long CBO
Fig. 5.15 Normalized time-lag distribution for different burner-outlet modications.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 105
travel between the burner-outlet surface and the ame front has been statistically
evaluated. In this procedure the ame 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
nite extension of the ame. The time lag corresponding to the modication with
no CBO is the smallest and is characterized by the narrowest distribution. The
longer CBO leads to signicantly 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 veried in eld tests.
5
The concept works well because the reacting oweld is only shifted by the CBO
and the main parameter that has been changed is the time lag.
Other design options for affecting the ame responsetime-lag distribution are
changing the fuel-concentration prole 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 amevortex 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 amevortex interaction to
improve the design methodology in this aspect. Another design option widely used
is the implementation of fuel-staging concepts, which allows modication of the
ame 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 orice as shown in Fig. 5.16.
The eigenfrequency of a typical Helmholtz resonator conguration is given by
f
Helmholtz
=
c
2
S
neck
V(l +l)
(5.3)
The eigenfrequency is determined by the ratio of the area of the orice 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 0.85D
tube
. 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 congu-
rations of resonators, several resonator tubes are connected in parallel to one large
W. KREBS ET AL. 106
Helmholtz-
Volume V
Resonator Tube
Holes for
entering purge
air
Combustor
L
D
tube
Fig. 5.16 Typical Helmholtz Resonator Conguration.
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 proles 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 ame 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.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 107
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 rigis equippedwithanexit diffuser connectedtoa 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.
W. KREBS ET AL. 108
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.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 109
Fig. 5.21 Operation envelope of the annular combustor rig.
In Fig. 5.21 the rms value of the dynamic pressure is plotted vs airfuel ratio.
The air mass ow through the combustor is held constant at 8 kg/s and the pilot
mass ow is set to 48 g/s. For the baseline case without resonators the fuel mass
ow can increase toward an airow ratio (AFR) of 35. If the fuel mass ow is
further increased a sudden excitation of the 200-Hz mode occurrs. This mode can
only be damped if the fuel mass owis considerably reduced toward an AFRof 38.
Fig. 5.22 Modeling of resonators.
W. KREBS ET AL. 110
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 ows and pilot mass ows. 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 owis 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 signicantly wider
oscillation-free operation envelope.
To assess the damping achieved with this conguration 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
installedinthe test rig.
12
The transfer-matrixmodel of the Helmholtz resonators has
been qualied 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 sufcient damping can be achieved with
this conguration. 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 ame with the acoustic environ-
ment that is present in all types of combustion systems. The complexity of the
engineering treatment originates from the long wavelengths associated with the
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 111
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 modications affect the burner design and
hence the ame response. The motivation for burner-design modications is that
the ame 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 ame 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 ame 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 ame. Although time-lag models have been
used to characterize the dynamic-time constants of the ame, the impact of coher-
ent structures on the dynamic behavior of the ame has not been resolved. Unsteady
CFD is one of the best methods to give more insight in the dynamic interaction
between vortices and the ame 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. 124130.
2
Munjal, M. L., Acoustics of Ducts and Mufers 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.,
V64.3AGas Turbine Natural Gas Burner Development, American Society of Mechanical
Engineers, Paper GT-2002-30106, 2002.
4
Kr uger, U., H uren, 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, McGrawHill, New York, 1968.
7
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, R. T. Edwards,
Flourtown, PA, 2001.
8
B uchn er, 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. 116129, Jan. 2005.
W. KREBS ET AL. 112
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. 99128.
11
Hubbard, S., and Dowling, A. P., Acoustic Instabilities in Premix Burners, AIAA
Paper 982272, 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
e
(6.2)
If particles are too large, they become centrifuged inside turbulent or large-scale
structures so that their response to oweld 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 inuence on the
liquid droplets. Another reason could simply be the damping of the sprays 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 (350700 Hz). This is
important because it reects the ability of the droplets to respond to the exter-
nal excitation, thereby allowing for the possibility of a corresponding uctuating
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 (150350 Hz), the coherence remained at fairly
J. M. COHEN ET AL. 126
0.0
0.2
0.4
0.6
0.8
1.0
0 300 600 900
injector A
injector B
injector C
injector D
C
o
h
e
r
e
n
c
e
Forcing Frequency (Hz)
Fig. 6.10 Summary of coherence measurements for all injector congurations, 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 Cand D. Because
acoustic forcing had a negligible inuence on the sprays 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 signicant attenua-
tion of the instability amplitude to acceptable levels, and these modied injectors
were incorporated into the production engine design.
Further evaluation of the acoustic coupling was done through measurement of
the spray mass owacoustic 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 Acon-
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 uctuations 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-ltering
devices. The fuel injector acted as a bandpass amplier, only allowing signicant
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.
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 127
0.00
0.02
0.04
0.06
0.08
0.10
350 400 450 500 550 600 650
M
a
g
n
i
t
u
d
e
Frequency (Hz)
-180
-120
-60
0
60
120
180
350 400 450 500 550 600 650
P
h
a
s
e
(
d
e
g
.
)
Frequency (Hz)
Fig. 6.11 Magnitude and phase of the mass uctuation/ p
F, as shown
in Fig. 7.8. This gure shows combustor temperatures and pressure amplitude as
the engine is loaded, holding base load, and subsequently unloaded on operator
control.
The ashback thermocouple (T
fb
) 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.
J. B. SEWELL AND P. A. SOBIESKI 156
Upper line is peak hold
Lower line is live trace
Alarm
Threshold
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 ashback as engine is loaded.
COMBUSTION INSTABILITIES MONITORING: CALPINE EXPERIENCE 157
Fig. 7.9 Borescope pictures of combustor premixer showing owobstructions inswirl
vanes.
This temperature rise by itself would not have triggered an alarm, because typical
alarm limit set points for these engines are about 200
Stuart R. James,
Power Technology.
163
C. J. GOY ET AL. 164
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 nancial 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.
Acomplete on-line combustion-dynamics health-monitoring and screening sys-
tem includes the following:
1) Dynamic pressure transducers. Sensors record the uctuating 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 uctuat-
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
signicantly 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 modemconnec-
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
machines 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 condence when as-
sessing whether the gas turbine is operating normally. Analysis can be automated
when sufcient data characterizing the machines behavior have been acquired.
IV. Benets of Combustion-Dynamics Monitoring
On-line combustion-dynamics monitoring has been carried out for the past
three years and, already, signicant benets have been realized. The number of
combustion failures on monitored machines has signicantly declined. As a result,
the level of condence in the systems has greatly improved. Scheduled borescope
inspections can now be more focused because the specic 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. Aload restriction can often be applied and additional
C. J. GOY ET AL. 166
EDMS Engine Distress Monitoring System
Data exported by
Wide Area
Network
Power Station
Power Technology
Specialist Advice
Specialist PC
Evaluation
and Optimisation
Dynamic data
On Line
Combustion
On Line
Vibration
On Line
EDMS
On Line
Performance
On Line
Pyrometry
Plant data
Fig. 8.3 Integration of combustion dynamics with other monitoring systems.
alarms can be congured 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 specic combustion components. This information has been fed back into
the daily screening activity to focus on the most damaging modes and to ne-tune
alarm criteria.
V. Case Studies
Once a machines operating behavior is characterized, deviations from the nor-
mal dynamic response can quickly be identied. 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
MONITORING COMBUSTION INSTABILITIES 167
0.00
2.00
4.00
6.00
8.00
10.00
12.00
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
D
y
n
a
m
i
c
P
r
e
s
s
u
r
e
/
T
e
m
p
e
r
a
t
u
r
e
0
50
100
150
200
250
300
350
400
U
n
i
t
L
o
a
d
Low f level Mid f level Ambient T Unit Load
Mid frequency dynamics
dominate at lower loads
0.00
2.00
4.00
6.00
8.00
10.00
12.00
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
D
y
n
a
m
i
c
P
r
e
s
s
u
r
e
/
T
e
m
p
e
r
a
t
u
r
e
0
50
100
150
200
250
300
350
400
U
n
i
t
L
o
a
d
Low f level Mid f level Ambient T Unit Load
Mid frequency dynamics
dominate at lower loads
O
O
O
O
O
O
O
O
O O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
X
X
X
X
X
X
X
X
X
X
X
X
X X
X
X
X
X
X
X
X
X
X
X X
X
X
X
X X X
X X
X
X
X
X X
X
X X
X
X
X
X
X
X
X
X X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
O
O
O
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 rst 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 signicantly lower on the rst 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 ame 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 engines 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 exibility is required in todays 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 combustors
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 underring to counteract uctuations
in grid frequency. The response is achieved by continually varying the gas-turbine
C. J. GOY ET AL. 168
FS mode = 1 ACTIVE
FS mode = 0 INACTIVE
90
110
130
150
170
190
210
230
6-Oct AM 6-Oct PM 7-Oct AM 7-Oct PM 8-Oct AM 8-Oct PM
G
T
L
o
a
d
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
D
y
n
a
m
i
c
P
r
e
s
s
u
r
e
Load Max Dynamics FS Mode
90 90
110
130
150
170
190
210
230
6-Oct AM 6-Oct PM 7-Oct AM 7-Oct PM 8-Oct AM 8-Oct PM
G
T
L
o
a
d
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
D
y
n
a
m
i
c
P
r
e
s
s
u
r
e
Load Max Dynamics FS Mode
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 signicantly, 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 fuelair 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 airow 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
MONITORING COMBUSTION INSTABILITIES 169
Fig. 8.6 Damaged combustion liner (view looking downstream, burners removed).
0
1
2
3
4
5
6
7
1-Dec 2-Dec 3-Dec 4-Dec 5-Dec 6-Dec 7-Dec 8-Dec 9-Dec 10-Dec
D
y
n
a
m
i
c
P
r
e
s
s
u
r
e
0
20
40
60
80
100
120
140
N
O
x
Hot tone level NOx Dynamic pressure
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.)
C. J. GOY ET AL. 170
600
605
610
615
620
625
08 Dec
12:00
08 Dec
18:00
09 Dec
00:00
09 Dec
06:00
09 Dec
12:00
09 Dec
18:00
T
e
m
p
e
r
a
t
u
r
e
0
50
100
150
200
250
G
T
L
o
a
d
Can 8 Exhaust Temperature GT Load
Fig. 8.8 Combustor-relatedexhaust temperature during the combustionliner failure.
from about 20 mg/m
3
to 180 mg/m
3
. 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 ap of material that had become
partially detached (Fig. 8.6).
The ability to conrm, 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.
MONITORING COMBUSTION INSTABILITIES 171
0.00
0 50 100 150
Days
D
y
n
a
m
i
c
P
r
e
s
s
u
r
e
(
H
i
g
h
F
r
e
q
u
e
n
c
y
M
o
d
e
)
Failure
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
signicant 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
tocorrelate this dynamic mode tospecic hardware damage. Not showninFig. 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 ame to a diffusion ame. Because diffusion
ames tend to exhibit lower levels of combustion instability, the sudden transition
C. J. GOY ET AL. 172
0.00
125 135 145 155 165
Days
D
y
n
a
m
i
c
P
r
e
s
s
u
r
e
(
L
o
w
e
r
F
r
e
q
u
e
n
c
y
M
o
d
e
)
.
Step Decrease
In Dynamics At
Time Of Failure
Fig. 8.11 Dynamics data from affected combustor in the month before failure.
Fig. 8.12 New burner nozzle assembly.
MONITORING COMBUSTION INSTABILITIES 173
0
125 135 145 155 165
Days
N
O
x
0
E
G
T
S
p
r
e
a
d
NOx EGT Spread
Increase
In NOx
Levels
Increase
In EGT
Spread
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 ame 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 ame. Therefore, the high NOx emission arising from the failed combustor is
masked by the lowemissions fromthe 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 airow 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 ame 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 ame.
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.
C. J. GOY ET AL. 174
Fig. 8.14 Damage to a burner assembly caused by high-frequency dynamics.
In another instance, the continuous monitoring systemdetected abnormally high
levels of high-frequencydynamics ina combustor. Aborescope inspectionrevealed
that a st-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.
MONITORING COMBUSTION INSTABILITIES 175
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 signicant 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 exibility 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 specic 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 CombustorCombustion Noise Char-
acteristics, Journal of Engineering of Gas Turbines and Power, Vol. 122, April 2000,
pp. 280286.
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 LowEmissions 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. 11231130.
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
S ebastien Ducruix,
Thierry Schuller,
Daniel Durox,
Professor, ECP and Institut Universitaire de France, Laboratoire EM2C, Fellow AIAA.
179
S. DUCRUIX ET AL. 180
Upstream
dynamics
Feed line
dynamics
Impedance
conditions
Injection
Atomization/
vaporization/
mixing
Mixing
Stabilization
Organized
vortex
structures
Flame
wall
interactions
Flame/
vortex
interactions
Entropy
waves
Acoustics
Acoustics
Down-
stream
dynamics
Exhaust
impedance
conditions
Heat
release
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 systemcharacteristics, 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 Ch atelier.
1
An often-quoted
paper by Rayleigh
2
establishes a criterion that oscillations are sustained when
heat release and pressure uctuations 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 uctuations, nor the coupling path, which leads from acoustic
to heat-release uctuations, are known. They are both crucial to predict the ampli-
cation 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 difcult 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):
d
2
x
dt
2
+2
0
dx
dt
+
2
0
x(t ) = 0 (9.1)
Expanding Eq. (9.1) in a Taylor series to rst order yields
d
2
x
dt
2
+
0
(2
0
)
dx
dt
+
2
0
x(t ) = 0 (9.2)
The damping coefcient is negative if
0
> 2 . If the delay is long enough
with respect to the period T = 2,
0
, the amplitude of any perturbation will grow
exponentially. More generally, combustion instability occurs when the natural
COMBUSTION INSTABILITY MECHANISMS 181
resonant time of the ow conguration 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 ow 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.
38
Here, we will examine the aspects that typ-
ify what occurs in gas-turbine combustors. This chapter specically 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 conguration, and that the owis, 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. 915). 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 uctuations corresponding to ne-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. 1719).
Some of many possible interactions that need to be examined are especially
relevant because they directly cause uctuations in heat release or generate pres-
sure perturbations. According to Rayleighs 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) amevortex interactions;
2) acousticame coupling; 3) interactions of perturbed ames with boundaries;
4) mutual ame annihilations; 5) ame 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 rst case or leading fromacoustic 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
S. DUCRUIX ET AL. 182
(t)
Flame
F+O
P
Flame
F+O
Plane
acoustic
waves
Flame
Plane
acoustic
waves
Flame
Flame
Plane
acoustic
waves
Flame
Plate
F+O
P
P
F+O
P
F+O
P
Vortex
Equivalence
ratio
perturbation
a) b) c)
d)
e)
f)
g)
Fig. 9.2 Elementary processes: a) Unsteady strained diffusion ame, b) ame roll up
in a vortex, c) premixed ame/vortex interaction, d) equivalence ratio perturbation
interacting with a premixed ame, e) acoustically modulated conical ame, f) acous-
tically modulated V-ame, and g) perturbed ame interacting with a plate (adapted
from Ref. 7).
process is exemplied in Fig. 9.3 (from Ref. 20), which gives an illustration of
Rayleighs criterionina 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 ows (Sec. II). This
wave equation for the pressure uctuations 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 uctuations to
the acoustic variables (pressure, velocity). A classical representation based on the
(n ) 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 uctuations is crucial.
Fig. 9.3 Illustration of Rayleighs criterion (from Ref. 20).
COMBUSTION INSTABILITY MECHANISMS 183
Heat release is then considered as a pressure source in Sec. III, in which three
different situations are analyzed. The rst is that of vortex-driven uctuations, in
which vortices interact with a ame producing a heat-release pulse. This mech-
anism has been studied in many laminar and turbulent congurations, because it
constitutes a powerful driving process. In the second case studied in this section,
self-sustained oscillations of a ame impinging on a plate produce an intense radi-
ation of sound. This elementary process typies heat-release uctuations resulting
from amewall interactions. One may also infer from this example that similar
processes may take place as a result of mutual ame interactions in the core of
the ow. This third process effectively produces rapid changes of ame surface
area and correspondingly large uctuations 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 uctuations driven by waves or ow pertur-
bations. Three situations are envisaged. In the rst situation, a conical ame is
modulated by acoustic waves. It is shown that the response of the ame 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 ow impinge on a ame producing a uctuation
in heat release. The time delay between injection and combustion is the key pa-
rameter in the process and it denes conditions of oscillation. In the third situation,
heat-release perturbations result from a time-variable strain rate. A low-pass lter
behavior of the ame is found in this conguration. 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 eld and source terms associated with heat release and
turbulence. This equation is briey 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 exemplied in this book
or in the literature (see Ref. 22 for a review).
A. Role of Heat-Release Fluctuations
The following analysis provides a simplied framework for theoretical inves-
tigation of combustion oscillations; its intent is limited. A low-speed (low Mach
number) reactive ow 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 ows may be found in Ref. 23.
Our objective is to underline the role of heat-release uctuations and to demon-
strate that the rate of change of these uctuations acts as a source, driving pressure
waves in the system. Starting from the balance equations for a chemically reacting
S. DUCRUIX ET AL. 184
mixture of N species
24
and using various simplications, one can derive a wave
equation for the logarithm of the pressure
16
:
c
2
ln p
d
dt
_
1
d
dt
ln p
_
= (
1
)
d
dt
_
1
c
p
T
_
T ++
N
k=1
h
k
n
k
N
k=1
_
Y
k
c
pk
v
D
k
T
_
__
d
2
dt
2
(ln R) v : v (9.3)
where c designates the speed of sound; , p, T, Y
k
, v, and v
D
k
are the density,
pressure, temperature, species mass fractions, velocity, and diffusion velocity, re-
spectively. +, c
pk
, , , and R designate the viscous dissipation function, specic
heats, specic heat ratio, heat conductivity, and gas constant. h
k
and n
k
are, re-
spectively, the specic 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 Doak
25
in the
context of aerodynamic sound and by Kotake
26
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 uctuations.
26
Neglecting all other terms, one obtains
c
2
ln p
d
dt
_
1
d
dt
ln p
_
=
d
dt
_
1
c
p
T
N
k=1
h
k
n
k
_
(9.4)
Consideringlow-speedreactive ows, the convective terminthe material derivative
may be neglected d,dt ,t . Assuming, in addition, that the specic heat ratio
is constant, Eq. (9.4) becomes
c
2
ln p
2
t
2
ln p =
t
_
1
c
:
T
N
k=1
h
k
n
k
_
(9.5)
This equationis not linearized, andit canbe usedtodescribe nite 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
uctuating components: p = p
0
+ p
1
with p
1
,p
0
1. Then, ln p p
1
,p
0
and
Eq. (9.5) becomes
c
2
_
p
1
p
0
_
2
t
2
_
p
1
p
0
_
=
t
_
1
c
:
T
N
k=1
h
k
n
k
_
(9.6)
COMBUSTION INSTABILITY MECHANISMS 185
In practical continuous-combustion devices, the mean pressure does not change
by more than a few percent, the spatial derivatives of p
0
may be neglected, and,
hence, Eq. (9.6) may be written as
c
2
p
1
2
t
2
p
1
=
t
_
( 1)
N
k=1
h
k
n
k
_
(9.7)
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
v
1
t
=
1
0
p
1
(9.8)
where v
1
represents the velocity uctuations. 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
n
k
) and acoustic uctuations ( p
1
, v
1
) 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 ) 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 Lh
f
designates the change of formation enthalpy per unit mass
of the mixture, and if n represents the rate of reaction, the chemical source term
becomes (,t )( 1)(Lh
f
) n. 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 termassociated with chemical reaction may be written in the form:
( 1)
Q
1m
t
(9.9)
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 eld
is driven by the nonsteady release of heat. A coupled motion can take place if this
last quantity is inuenced 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 uctuations and
acoustic perturbations, it is instructive to consider the ow of a combustible mix-
ture through a long duct (an acoustic resonator), with a ame 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 ame thickness is small
compared with the acoustic wavelength, so that the region of heat release may be
S. DUCRUIX ET AL. 186
Fig. 9.4 Sketch of the model compact ame geometry. The ame 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 ame holder is denoted as region 1, with a fresh gas density
f
and sound
speed c
f
. Region 2 corresponds to the downstream side of the ame holder, with
a burnt-gas density
b
and sound speed c
b
. The acoustic velocity : 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 quantied by integrating the wave equation (9.7) over
a thin control volume containing the ame. This leads to (see Ref. 16 for details)
:
b
(a
+
, t ) :
f
(a
, t ) = ( 1)
Q
1a
f
c
2
f
(9.10)
where Q
1a
represents the instantaneous heat-release rate per unit area. When the
ame 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 ow variables and the dynamic response
of the ame. One approach
2729
uses a time-lag hypothesis to express Q
1a
in terms
of the time-delayed upstream velocity perturbation,
( 1)
Q
1a
f
c
2
f
= n:
f
(a
, t ) (9.11)
where n is an interaction index, and represents a time lag. The heat-release term
is modeled as a function of an acoustic-wave variable alone. This approach is
clearly a simplied representation of more complex processes involving the ow,
turbulence dynamics and large-scale motions, ame interactions with neighboring
ames and walls, heat transfer at the boundaries, etc. Some of these processes are
described subsequently.
COMBUSTION INSTABILITY MECHANISMS 187
The value of the time lag relative to the frequency often denes ranges of
instability.
30, 31
This value is recognized from the early work on rocket engine
instability
29
(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 mechanismthat feeds energy
into the oscillation. This mechanismis analyzed and illustrations are given. Section
III.B is devoted to the interaction of a ame with a wall. This mechanism may not
be of major importance for gas-turbine combustors, but it serves to show that
rapid changes in ame surface area can induce heat-release uctuations, which in
turn may feed energy into the pressure eld. This process, illustrated with ame
wall interactions, may also result from mutual interactions between neighboring
ame elements in the core of the ow, which may certainly arise in gas-turbine
combustors. Mutual interactions will be briey analyzed in the third subsection.
A. FlameVortex Interactions
Flamevortex interactions have been observed in many unstable combustion
systems. Two distinct mechanisms are usually involved. In the rst, the ame
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 ow
and the amplitude of the pressure pulse resulting from the vortex burnout. When
the ame is rolled up, the surface area increases rapidly. The growth is limited
by ame shortening, which results from interactions of neighboring elements, and
consumption of the reactants entrained by the vortex. Such rapid variations of
ame surface correspond to the rst mechanism. Flamevortex 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 ame
34
or an established strained diffusion ame,
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 ame 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 difcult to study experimentally
and is less well documented. Interactions between adjacent reactive vortices may
also take place, leading to formation of ne-grain turbulence.
S. DUCRUIX ET AL. 188
Many studies have also focused on the natural instabilities of wakes and jets.
3841
Indeed, the vortical structures that are involved in the amevortex 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 ow by using driver units or pistons and to analyze the response sensitivity of
the jet or wake to this forcing.
4246
The conversion of energy between vortices and
acoustics is also of great importance and has been studied in nonreactive
47, 48
and
reactive situations.
33
One example of self-sustained oscillations controlled by vortices is reported in
Ref. 36. Amultiple-inlet combustor is fed with a mixture of air and propane, and it
features a dumpplane (Fig. 9.5). The low-frequencyinstabilityobservedinthis case
Fig. 9.5 Geometry of the multiple-ame-holder dump combustor studied in Ref. 36.
Spark-schlieren photograph of the central jet for the 530 Hz unstable regime.
COMBUSTION INSTABILITY MECHANISMS 189
is acoustically coupled and occurs at one eigenfrequency of the system. The ame
visualization of Fig. 9.5 clearly shows that the largest-amplitude oscillations are
vortex driven. The following processes are involved: 1) Avortex is shed at the dump
plane when the velocity perturbation is maximum (v
1
1
in a driving path,
where
1
represents vorticityuctuations). 2) The vortexis convected, accelerated,
and entrains hot gases fromits surroundings. Acombustion pulse is produced when
two adjacent vortices interact, creating a large amount of small-scale turbulence
and ame 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
lagis ina suitable range withrespect tothe periodof the motionas showninFig. 9.6.
The self-sustained oscillations of a laminar V-ame 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 ame interactions with boundaries is detailed
in the next section, because it can constitute a source of heat-release uctuations
even in the absence of vortex shedding.
In the rst situation described in this section, the mechanism involves ow
perturbations producing vorticity, which results in rapid changes of ame 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:
O
1
Q
1
p
1
Many other studies
21, 37
have revealed the key role of vortex structures. Coming
backtothe theoretical expressions of Sec. II, one clearlysees that a model giving Q
1
as a function of O
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 ) 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 ames with solid walls constitute a source of heat-release uctu-
ations.
5254
Under certain conditions, such interactions can lead to self-sustained
oscillations, which are briey described in what follows. These experiments in-
dicate that heat-release uctuations of large amplitude can be induced by rapid
changes of ame area and that these uctuations generate an intense sound eld.
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.
COMBUSTION INSTABILITY MECHANISMS 191
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 uctuations produce nonsteady heat release, which
induces acoustic pressure radiation. It may be represented schematically by
A
1
Q
1
p
1
This sequence may be used to model the instability mechanism. It requires an
analytical description of the interactions between the ame and the wall and an
expression of the noise generated by the ame. Eventually, the determination of
the associated time lags denes regions of instability.
In experiments reported in Ref. 52, a laminar premixed ame, anchored on a
cylindrical burner, impinges on a horizontal plate, and a driver unit modulates the
upstream ow. This forcing generates perturbations in ame-surface area and heat
release (coupling path, typically ( p
1
, v
1
) Q
1
). The driving path (typically,
Q
1
( p
1
, v
1
)) is also easy to characterize, because the sound produced by the
S. DUCRUIX ET AL. 192
mixture of
gases
loudspeaker
M
thermocouple cooled plate
CH*
filter
M
PM
LDV
zone 1
zone 2
0
3
Fig. 9.8 Schematic view of the experimental setup used to study interactions of a
perturbed ame with a cooled wall. This conguration radiates an intense acoustic
eld (from Ref. 53).
system is 1020 dB higher than that emitted by a free ame submitted to the
same modulation, without the plate. The interaction of the ame with the plate
leads to rapid changes of the ame surface, which constitutes a major source of
sound in this situation. This well-controlled experiment typies more complicated
situations in which the ame spreads in a chamber (as in a gas-turbine combustor)
and can produce pressure oscillations when impinging on the walls or on adjacent
ame 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 ame 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 ame (a), when no sound emission is observed, and a complete cycle of
oscillation (be), when the instability is triggered. The ame 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 ame interaction with the plate and a representation of
the burner acoustics. The acoustic velocity :
1
and pressure p
1
at the burner exit
may be related by a second-order equation
53
:
M
d
2
:
1
dt
2
+ R
d:
1
dt
+k:
1
= S
1
dp
1
dt
(9.12)
COMBUSTION INSTABILITY MECHANISMS 193
Fig. 9.9 Different views of a ame interacting with a wall: a) Steady state; be)
Instantaneous images of the ame during an instability cycle (from Ref. 53).
S. DUCRUIX ET AL. 194
- 1
0
1
2
3
- 1
0
1
2
3
0 5 1 0 1 5 2 0
v
1
(
m
/
s
)
-
I
(
C
H
*
)
-
M
i
c
d
(
I
(
C
H
*
)
)
/
d
t
(
a
r
b
.
u
n
i
t
)
Time (ms)
LDV
CH*
d(I(CH*))/dt
Micro M
3
Fig. 9.10 Self-sustained oscillations of a ame interacting with a plate. Simultaneous
measurements of the velocity v
1
at the burner outlet, of the CH
(r, t ) =
4r
_
b
1
__
dQ
dt
_
t
a
(9.13)
In this equation,
,
f
,
b
are the densities in the far-eld air, the fresh gas, and
the burned gas, respectively;
a
is the time required by sound propagation over
a distance r from the sources to the detector. In gaseous premixed ames, the
far-eld radiated pressure p
(r, t ) =
4r
_
b
1
_
S
L
_
dA
dt
_
t
a
(9.14)
where S
L
is the laminar burning velocity. The fast rate of extinction of the ame
area at the cold boundary induces a signicant acoustic pressure radiation, which
shown in Fig. 9.10, where p
= 0.5
ms
1
(from Ref. 59).
(Fig. 9.11b). As the deformation travels along the ame 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 ame-surface area is maximum, and two
ame elements interact (Fig. 9.11d). The outcome of this mutual annihilation de-
pends on the spatial position of the rst 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 ame tori can be
produced.
During interaction of these ame elements, the shape of the reactive front under-
goes a strongalteration. As inthe ameplate situation, after the mutual interaction,
the ame quickly retrieves its initial shape at the beginning of the following cycle
(Fig. 9.11a). In this cycle, the short phase of ame-surface destruction produces a
faster rate of change of the ame-surface area thanthe longer phase of ame-surface
production by stretch. The same mechanism operates as in the ameplate
COMBUSTION INSTABILITY MECHANISMS 197
interaction, except that ame-surface destruction is produced by mutual anni-
hilation of neighboring front elements and not by thermal losses. The path is also
similar:
A
1
Q
1
p
1
The overall sound-pressure level is considerably enhanced. The pressure spectrum
is quite similar to that associated with the ameplate 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 ame annihilation is believed to control and limit ame-surface
area in turbulent combustion, the previous ndings suggest that this mechanism
could also be an important source of noise in turbulent combustors.
IV. Heat-Release Fluctuations Driven by Waves
Unsteady uctuations in pressure, temperature, strain rate, induced curvature,
and chemical composition directly inuence the rate of reaction in the ame. 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 McIntosh
61, 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 ame
elements or in the available ame-surface area are probably more relevant. These
unsteady changes are illustrated here by considering heat-release uctuations in-
ducedbyvarious perturbations; acoustic waves, equivalence ratioinhomogeneities,
and unsteady strain rates are successively discussed. These perturbations are not
the only possible sources of coupling (or feedback), which may drive heat-release
uctuations, but they are most signicant.
A. Modulated Conical Flames
The coupling may be represented schematically by
p
1
:
1
A
1
Q
1
This path may be investigated by modulating an initially stable ame 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 ame response to incident perturbations. Early investigations of this
type were proposed, for example by Markstein,
65
Blackshear,
66
and De Ste.
67
If the process remains in the linear regime, one may dene a transfer function
between the incident velocity uctuations and the nonsteady heat release, which
will depend on the burner geometry, operating parameters, and steady-state ame
conguration.
S. DUCRUIX ET AL. 198
Transfer functions proposed in Refs. 6871 indicated that the ame usually
behaves like a low-pass lter, providing a qualitative representation of the ame
response. Laminar conical ames are considered by Blackshear,
66
De Ste,
67
and,
more recently, Baillot et al.
72
Further theoretical efforts by Fleil et al.
73
and a
combination of theoretical analysis and detailed measurements by Ducruix et al.
74
have advancedthe status of this basic problem. More recent workbySchuller et al.
75
provides additional clues oname response inthe high-frequencyrange. Dowling
76
uses a similar approach to derive a model for the low-frequency nonlinear response
of a ducted V-ame in a geometry close to that considered in an earlier work by
Marble and Candel.
77
The modulation of a conical ame 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 ame is anchored on a cylindrical burner, and it is
submitted to acoustic waves generated by a loudspeaker placed at the bottomof the
burner. The ame response is driven by the acoustic velocity, and the aim is to nd
the transfer function between heat-release uctuations and velocity modulations:
F() =
Q
1
(),Q
0
:
1
(),:
0
(9.17)
where 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 uctuations. 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 uctuations as a function of velocity modulations, whereas its phase
characterizes the time lag existing between velocity and heat-release uctuations.
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 upstreamfromthe nozzle and containing various
grids and honeycombs to produce a laminar ow. The conical ame is stabilized on
a 22-mm-diam burner rim. A driver unit placed at the base of the burner generates
perturbations, which wrinkle the ame front. The shape of the perturbed ame
depends on the frequency and amplitude of modulation. The typical ame 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 intensied cameras, etc.] has provided new informa-
tion concerning the geometry of the ame front, the local and global heat-release
rates,
74
and the velocity eld at the burner exhaust and in the oweld.
80
This has
allowed direct measurements of the ame-transfer function dened by Eq. (9.17).
These measurements can be compared with theoretical and numerical predictions.
An analytical transfer function can be derived by decomposing the owin mean
and perturbed components. The geometry of the problem is sketched in Fig. 9.13.
A G equation is used to describe the ame position:
G
t
+v G = S
D
|G| (9.18)
COMBUSTION INSTABILITY MECHANISMS 199
x (mm)
y
(
m
m
)
-20 -10 0 10 20
0
10
20
30
40
x (mm)
y
(
m
m
)
-20 -10 0 10 20
0
10
20
30
40
Fig. 9.12 Methane air conical ame modulated by longitudinal acoustic perturba-
tions. f
e
= 150.5 Hz,
28, v = 1.44 ms
1
, v
o
o
(r)
(r,t)
n
LL
y
Fig. 9.13 Geometry of a) the conical ame in the steady situation and b) in the
perturbed case; from Ref. 74.
S. DUCRUIX ET AL. 200
where v = (u, :) is the velocity vector, and S
D
is the ame-displacement speed.
In what follows, S
D
is assumed to be a constant and equal to the laminar burning
velocity S
L
.
The G variable increases from the fresh mixture to the burnt gases, and one
contour G = G
0
represents the ame. In the simplest velocity-perturbation model,
the radial component u
1
is supposed to be negligible, compared with the vertical
component :
1
, which is assumed to be uniform and sinusoidal: :
1
= :
1
cos t .
This corresponds to a bulk motion of the fresh stream. Figure 9.13 shows that G
may be replaced by y, where designates the ame position. Substituting =
0
+
1
in Eq. (9.18), where
0
represents the steady ame shape, and considering
small perturbations
1
(Fig. 9.13b), one may expand the resulting equation to the
rst order and obtain
1
t
= S
L
cos
0
1
r
+:
1
(9.19)
where
0
denotes the half-angle of the steady ame cone.
The heat-release uctuations may be evaluated from the ame-surface varia-
tions:
A
1
= 2 cos
0
_
R
0
1
dr (9.20)
Heat-release uctuations Q
1
are directly related to the uctuations of ame-surface
area: Q
1
=
f
S
L
q A
1
, where
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 uctuations
74
:
Q
1
Q
0
=
:
1
:
0
2
[(1 cos
) cos (t ) +(
sin
) sin (t )] (9.21)
and the transfer function is easily deduced therefrom. The resulting expression
depends on a reduced frequency
= R,(S
L
cos
0
), 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 ame.
The analytical ame 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 PIV
80
show that this assumption may be
acceptable for weaklywrinkledames, witha small velocityradial component, that
is, in the low-frequency range (
*
Fig. 9.14 Comparisons between calculations (solid line), analytical results (dashed
and solid line) and measurements (symbols) for the transfer function of a conical ame
(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
ame 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 ame
shapes. In Fig. 9.12, the perturbation velocity is axial and uniformin 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 ,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 ames.
81
Schuller et al.
75
show
that this function depends on two parameters
and S
L
, :, and one obtains an
improved agreement with the experimental data. The phase of the transfer function
shifts from a purely convective behavior for elongated ames to a saturated value
for at ames.
S. DUCRUIX ET AL. 202
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 fuelair ratio.
8285
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 ow 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
1
, which is
then convected by the ow to the ame 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
p
1
1
convection Q
1
This mechanismis illustrated in Fig. 9.15. In a rst step, a pressure oscillation arises
in the system, which will modify the fuel ow 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 ame (from Ref. 82).
COMBUSTION INSTABILITY MECHANISMS 203
ratio. Three time delays dene the process. The rst
i
corresponds to a phase shift
between the pressure at the injector and fuel mass owrate m
F1
. Oscillations in this
owrate induce uctuations in the equivalence ratio
1
. An inhomogenous mixture
is then convected to the reaction zone with a delay
conv
. The response of the ame
to the impinging uctuations
1
comes after a combustion delay
chem
. Oscillations
will be sustained by this process if the pressure and heat-release uctuations are
in phase (Rayleighs criterion), which is the case if the total delay is such that
i
+
conv
+
chem
= (2n 1)
T
2
(9.22)
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
conv
(2n 1)T,2.
One fundamental aspect of this process is the response of the ame to incom-
ing equivalence ratio perturbations.
86
Another aspect that will also inuence this
mechanismis the level of mixing taking place between the injector and the ame.
87
If this mixing is efcient, the initial level of uctuations will be diminished to a
great extent by reducing the uctuation 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 eld can be induced by the resonant acoustic motion
acting on the ow. This eld may change the rate of heat release in two major
ways. The rst way consists of perturbations in the ame-surface area. To analyze
this rst possibility, let us consider a model equation for the ame-surface density:
dY
dt
= cY Y
2
(9.23)
Balance equations of this type are extensively used in turbulent combustion. In
Eq. (9.23), the rst termonthe right-handside represents productionof surface den-
sity by strain rate, and the second term describes mutual annihilation of ame sur-
face density(ame shortening). At equilibrium, dY
0
,dt = 0andc
0
Y
0
Y
2
0
= 0.
A sinusoidal perturbation of the strain rate c = c
0
+c
1
cos t is now assumed,
which produces a perturbation in surface density: Y = Y
0
+Y
1
. Injecting this
expression in the balance equation (9.23) and, retaining rst-order terms only, one
obtains
7
dY
1
dt
+c
0
Y
1
= (c
1
cos t )Y
0
(9.24)
The response in terms of Y
1
is that of a low-pass lter. The steady-state solution
takes the general form:
Y
1
Y
0
=
c
1
c
2
0
+
2
(c
0
cos t +sin t ) (9.25)
S. DUCRUIX ET AL. 204
In the low-frequency limit, c
0
, the relative perturbation of ame-surface den-
sity is in phase with the strain rate:
Y
1
Y
0
=
c
1
c
0
cos t (9.26)
In the high-frequency limit, c
0
, the relative perturbation of ame-surface
density is in quadrature with the strain rate, and it decreases with frequency:
Y
1
Y
0
=
c
1
sin t (9.27)
This mechanism applies equally well to premixed and nonpremixed ames. This
type of interaction modulates the ame-surface density and can be represented
schematically by
p
1
:
1
ow c
1
A
1
The second type of interaction involves a direct effect on the reaction rate per unit
ame surface and is represented by
p
1
:
1
ow c
1
1
This second effect is effective in the nonpremixed case, because the reaction rate
is directly related to the species gradients at the ame, which are xed by the strain
rate. In the premixed case, the consumption rate is weakly inuenced by the strain
rate, except near extinction conditions.
The ame response to strain rate has been extensively studied in turbulent com-
bustion.
88, 89
Other studies deal with the response of ames to external strain-rate
modulations. The problem is envisaged experimentally
90
and often treated by di-
rect calculations using time-dependent solutions of strained ames with complex
chemistry.
91, 92
Analytical expressions of the ame response have also been de-
termined by using asymptotics.
93
It was found that ames behave like low-pass
lters, when the perturbed strain-rate uctuations do not exceed the extinction
value. Considering nonpremixed ames and assuming the innitely fast chemistry
limit, the ame-transfer function is dened in the frequency domain as the ratio of
the relative reaction rate modulation to the relative strain-rate perturbation
7
:
F() =
_
m() m
0
m
0
___
c() c
0
c
0
_
(9.28)
This transfer function has the form of a low-pass lter:
F() =
1
2
1
1 +i (,2c
0
)
(9.29)
COMBUSTION INSTABILITY MECHANISMS 205
a)
c)
e) f )
d)
b)
Fig. 9.16 DNS computations of the mutual ame annihilation as a limitation mecha-
nismfor ame surface production. The different lines represent the peak consumption
rates of CH
4
, O
2
, H
2
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 ame time (from Ref. 96).
The effect of unsteadystrainonpremixedames cannot be describedinsuchsimple
terms. Numerical calculations by Im and Chen
94
indicate that the reponse of the
ame 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 ame-surface area is augmented when the strain rate acting on the reactive
elements is lower than the extinction value. Conversely, the ame area is limited by
a mechanismof mutual interactions of adjacent reactive elements. This mechanism
has been identied as a fundamental process reducing the ame-surface area (see,
for example, Echekki et al.
95
or Chen et al.
96
and Fig. 9.16). This mechanism may
also inuence the dynamics of turbulent ames. The mutual interaction of strained
ames is now well understood in cases in which the ame elements tend to propa-
gate away from each other. The case of strained elements approaching each other
and leading to a shortening of the ame 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 briey evoked in
Sec. III.
S. DUCRUIX ET AL. 206
Table 9.1 Summary of driving processes examined in this chapter
Effect Induced
Initial on ow eld
perturbation and ame Main result Consequences changes
Flamevortex
interaction
Vorticity
generation
Flame roll up
Flame
boundary
interaction
Flow
perturbation
Flame
wrinkling
Changes of
ame-surface
area
Heat-release
uctuations
Pressure-
wave
radiations
Mutual ame
interaction
Flame
wrinkling
Flame front
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 uctuations 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 ame surface generate an intense radiation of
sound. In practical situations, there are many possible mechanisms that may pro-
duce or destroy ame surface at a fast rate, such as amewall interactions and
collisions between adjacent ames or between neighboring ow 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 chapter
a
Effect
Initial on ow
perturbation and ame Main result Consequences
Flame response
to upstream
modulation
Flow
modulation
Flame surface
wrinkling
Flame response
to composition
inhomogeneities
Acoustic
wave motion
Injection
perturbation
Equivalence-
ratio
uctuations
Unsteady heat
release
Flame response
to strain rate
Flow
modulation
Fluctuations of
strain rate eld
a
When interacting with the proper phase lag, driving and coupling mechanisms can lead to combustion
instabilities.
COMBUSTION INSTABILITY MECHANISMS 207
they are properly phased with respect to the pressure. Fast changes in ame-surface
area constitute an important driving process of combustion instabilities.
The coupling (or feedback) mechanism between the pressure eld and the com-
bustion process may take many different forms. It is illustrated here with a set of
experiments with laminar ames, but previous experiments on turbulent ducted
congurations have indicated that premixed ames are quite susceptible to such
modulations. The ame is highly wrinkled by the external eld, giving rise to
surface and heat-release uctuations. In simple cases, it is possible to dene a
transfer function between the relative velocity and heat-release uctuations. Com-
parisons between analytical models, numerical simulations, and experiments are
reviewed. It is shown that simple lter models do not provide a suitable description
of the phase when the modulation frequency is high and that more rened methods
must be used to get a better description of this quantity. In practical systems, the
ame 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 uctuations when these perturbations convected downstream reach
the ame, which has been identied as a possible driving process for some types
of gas-turbine instabilities. Flame modulation may result from the eld of variable
strain rate, which can be induced by the nonsteady motion in the combustor. The
variable strain rate can produce or diminish the ame-surface area, and modify the
local rate of reaction per unit surface. If the uctuations 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 Ch atelier, H., Recherches Exp erimentales et Th eoriques sur la
Combustion de M elanges Gazeux Explosifs, Annales des Mines, Paris Series, Vol. 8,
1883, pp. 274377.
2
Lord Rayleigh, The Explanation of Certain Acoustic Phenomena, Nature, Vol. 18,
1878, pp. 319321.
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. 12771296.
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. 129.
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 Summereld, M. (eds.), Nonsteady Burning and Com-
bustion Stability of Solid Propellants, Vol. 143, Progress in Astronautics and Aeronautics,
AIAA, Reston, VA, 1992.
S. DUCRUIX ET AL. 208
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. 128.
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. 721829.
9
Poinsot, T. (ed.), Large Eddy Simulation of Reacting Flows, Flow Turbulence and
Combustion, Vol. 65, No. 2, 2000, pp. 111244.
10
Desjardins, P. E., and Frankel, S. H., Two-dimensional Large Eddy Simulation of Soot
Formation in the Near-eld of a Strongly Radiating Nonpremixed Acetylene-Air Turbulent
Jet Flame, Combustion and Flame, Vol. 119, No. 12, 1999, pp. 121133.
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. 16, 1999, pp. 2562.
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. 783791.
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. 20012008.
14
Huang, Y., Sung, H.-G., Hsieh, S.-Y., and Yang, V., Large-Eddy Simulation of Com-
bustion Dynamics of Lean-Premixed Swirl-Stabilized Combustor, Journal of Propulsion
and Power, Vol. 19, No. 5, 2003, pp. 782794.
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. 489505.
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. 722734.
18
Lieuwen, T., Modeling Premixed Combustion-Acoustic Wave Interactions: A Re-
view, Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 765781.
19
Dowling, A. P., andStow, S. R., Modal Analysis of Gas Turbine Combustor Acoustics,
Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 751764.
20
Yu, K. H., Trouv e, A., and Daily, J. W., Low-Frequency Pressure Oscillations in
a Model Ramjet Combustor, Journal of Fluid Mechanics, Vol. 232, Nov. 1991, pp. 4772.
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. 337.
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. 337347.
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. 527561.
COMBUSTION INSTABILITY MECHANISMS 209
26
Kotake, S., On Combustion Noise Related to Chemical Reactions, Journal of Sound
and Vibration, Vol. 42, No. 3, 1975, pp. 399410.
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. 256263.
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. 281289.
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. 11091115.
32
Hedge, U. G., Reuter, D., and Zinn, B. T., Sound Generation by Ducted Flames, AIAA
Journal, Vol. 26, No. 5, 1988, pp. 532537.
33
Renard, P. H., Th evenin, D., Rolon, J. C., and Candel, S., Dynamics of Flame-Vortex
Interactions, Progress in Energy and Combustion Science, Vol. 26, No. 3, 2000, pp. 225
282.
34
Sinibaldi, J. O., Mueller, C. J., and Driscoll, J. F., Local Flame Propagation Speeds
Along Wrinkled, Unsteady, Stretched Flames, Proceedings of the Combustion Institute,
Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 827832.
35
Renard, P. H., Rolon, J. C., Th evenin, 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. 189205.
36
Poinsot, T., Trouv e, A., Veynante, D., Candel, S., and Esposito, E., Vortex-Driven
Acoustically Coupled Combustion Instabilities, Journal of Fluid Mechanics, Vol. 177,
April 1987, pp. 265292.
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. 46, 1989, pp. 167186.
38
Crow, S. C., and Champagne, F. H., Orderly Structure in Jet Turbulence, Journal of
Fluid Mechanics, Vol. 48, No. 3, 1971, pp. 547591.
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. 119142.
40
Ho, C.-M., and Huerre, P., Perturbed Free Shear Layers, Annual Review of Fluid
Mechanics, Vol. 16, 1984, pp. 365424.
41
Kaiktsis, L., Karniadakis, G. E., and Orszag, S., Unsteadiness and Convective Instabil-
ities in Two-Dimensional Flowover a Backward-Facing Step, Journal of Fluid Mechanics,
Vol. 321, August 1996, pp. 157187.
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. 449491.
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. 493544.
44
Oster, D., and Wygnanski, I., The Forced Mixing Layer Between Parallel Streams,
Journal of Fluid Mechanics, Vol. 123, Oct. 1982, pp. 91130.
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. 2339.
S. DUCRUIX ET AL. 210
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. 706723.
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. 113142.
48
Bogey, C., Bailly, C., and Juv e, D., Numerical Simulation of Sound Generated by
Vortex Pairing in a Mixing Layer, AIAA Journal, Vol. 38, No. 12, 1999, pp. 22102218.
49
Durox, D., Schuller, T., and Candel, S., Combustion Dynamics of Inverted Conical
Flames, Proceedings of the CombustionInstitute, Vol. 30, The CombustionInst., Pittsburgh,
PA, 2004, pp. 17171724.
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. 3369.
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., andCandel, S., Dynamics of andNoise Radiatedbya Perturbed
Impinging Premixed Jet Flame, Combustion and Flame, Vol. 128, No. 12, 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. 6975.
54
Sch afer, 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. 1216.
56
Abugov, D. I., and Obrezkov, O. I., Acoustic Noise in Turbulent Flames, Combustion,
Explosions and Shock Waves, Vol. 14, 1978, pp. 606612.
57
Clavin, P., and Siggia, E., Turbulent Premixed Flames and Sound Generation, Com-
bustion Science and Technology, Vol. 78, No. 13, 1991, pp. 147155.
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. 343355.
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. 525538.
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., OnFlame Resonance inTubes, CombustionScience andTechnology,
Vol. 69, No. 46, 1990, pp. 147152.
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. 46,
1993, pp. 329346.
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. 13, 1996, pp. 373386.
COMBUSTION INSTABILITY MECHANISMS 211
64
Park, S., Annaswamy, A. M., and Ghoniem, A. F., Heat Release Dynamics Modeling
of Kinetically Controlled Burning, Combustion and Flame, Vol. 128, No. 3, 2002, pp. 217
231.
65
Markstein, G. H., Non Steady Flame Propagation, Pergamon Press, Elmsford, NY,
1964.
66
Blackshear, P. L., Driving Standing Waves by Heat Addition, Proceedings of the
Combustion Institute, Vol. 4, The Combustion Inst., Pittsburgh, PA, 1953, pp. 553566.
67
De Ste, G., Etude des Flammes Vibrantes. Application ` a la Combustion Turbulente,
Revue de lInstitut Francais du P etrole et Annales des Combustibles Liquides, Vol. 19, No. 6,
1964, pp. 766785.
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 G unther, R., The Transfer Function of Premixed Turbulent Jet Flames,
Proceedings of the Combustion Institute, Vol. 13, The Combustion Inst., Pittsburgh, PA,
1971, pp. 517526.
70
Baade, P. K., Design Criteria and Modes for Preventing Combustion Oscillations,
ASHRAE Transactions, Vol. 1, 1978, pp. 449465.
71
Matsui, Y., An Experimental Study on Pyro-acoustic Amplication of Premixed
Laminar Flames, Combustion and Flame, Vol. 43, No. 2, 1981, pp. 199209.
72
Baillot, F., Durox, D., and Prudhomme, R., Experimental and Theoretical Study of
a Premixed Vibrating Flame, Combustion and Flame, Vol. 88, No. 2, 1992, pp. 149
168.
73
Fleil, 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. 487510.
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. 765773.
75
Schuller, T., Durox, D., and Candel, S., A Unied Model for the Prediction of Flame
Transfer Functions: Comparison Between Conical and V-ame Dynamics, Combustion
and Flame, Vol. 134, No. 12, 2003, pp. 2134.
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. 761769.
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. 34,
1998, pp. 303318.
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. 99105.
80
Schuller, T., Ducruix, S., Durox, D., and Candel S., Modeling Tools for the Prediction
of Premixed Flame Transfer Functions, Proceedings of the Combustion Institute, Vol. 29,
The Combustion Inst., Pittsburgh, PA, 2002, pp. 107113.
81
Boyer, L., and Quinard, J., On the Dynamics of Anchored Flames, Combustion and
Flame, Vol. 82, No. 1, 1990, pp. 5165.
S. DUCRUIX ET AL. 212
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. 18091816.
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. 415421.
84
Hathout, J. P., Fleil, M., Annaswamy, A. M., and Ghoniem, A. F., Heat-Release Actu-
ation for Control of Mixture-Inhomogeneity-Driven Combustion Instability, Proceedings
of the Combustion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 721
730.
85
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., A Mechanism for Combustion
Instabilities inPremixedGas Turbine Combustors, Journal of Engineeringfor Gas Turbines
and Power, Vol. 123, No. 1, 2001, pp. 182190.
86
Marzouk, Y. M., Ghoniem, A. F., and Najm, H. N., Dynamic Response of Strained
Premixed Flames to Equivalence Ratio Gradients, Proceedings of the Combustion Institute,
Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 18591866.
87
Lieuwen, T., Neumeier, Y., and Zinn, B. T., The Role of Unmixedness and Chemical
Kinetics in Driving Combustion Instabilities in Lean Premixed Combustors, Combustion
Science and Technology, Vol. 135, No. 16, 1998, pp. 193211.
88
Law, C. K., Dynamics of Stretched Flames, Proceedings of the Combustion Institute,
Vol. 22, The Combustion Inst., Pittsburgh, PA, 1988, pp. 13811402.
89
Peters, N., Turbulent Combustion, Cambridge UniversityPress, 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 Counterow Propane Diffusion Flame, Proceedings of the Combus-
tion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 20212027.
91
Darabiha, N., Transient Behaviour of Laminar Counter Flow Hydrogen-Air Flames
with Complex Chemistry, Combustion Science and Technology, Vol. 86, No. 16, 1992,
pp. 163181.
92
Egolfopoulos, F. N., and Campbell, C. S., Unsteady Counterowing Strained Diffusion
Flames: Diffusion-Limited Frequency Response, Journal of Fluid Mechanics, Vol. 318,
July 1996, pp. 129.
93
Joulin, G., On the Response of Premixed Flames to Time-Dependent Stretch and
Curvature, Combustion Science and Technology, Vol. 97, No. 13, 1994, pp. 219229.
94
Im, H. G., and Chen, J. H., Effects of Flow Transients on the Burning Velocity of
Laminar Hydrogen-Air PremixedFlames, Proceedings of the CombustionInstitute, Vol. 28,
The Combustion Institute, Pittsburgh, 2000, pp. 18331840.
95
Echekki, T., Chen, J., and Gran, I., The Mechanism of Mutual Annihilation of
Stoichiometric Premixed Methane-Air Flames, Proceedings of the Combustion Institute,
Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 855863.
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. 12, 1999, pp. 1548.
Chapter 10
Flow and Flame Dynamics of Lean Premixed
Swirl Injectors
Ying Huang,
Shanwu Wang,
R
n
R
h
u wr
2
dr/
R
n
R
h
R
n
u
2
rdr (10.1)
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 217
0
x/H
S = 0.5
0 10
S = 0.3
x/H
5 10
5
Fig. 10.2 Streamlines based on mean axial and radial velocity components; swirl
numbers S =0.3 and 0.5.
where R
h
and R
n
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
S =
2
3
1 (R
h
/R
n
)
3
1 (R
h
/R
n
)
2
tan (10.2)
where 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 oweld 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 ow resulting from the
higher centrifugal force. A small separation bubble exists behind the centerbody,
and the ow rapidly merges along the centerline.
A simplied momentum equation indicates that a radial pressure gradient is
produced by the centrifugal force arising from the swirling effect:
p
r
=
U
2
r
(10.3)
The pressure tends to be minimized in regions where strong swirling motions
occur, that is, in the wake of the centerbody. As the owexpands and the azimuthal
velocity decays with the axial distance, the pressure is recovered in the downstream
Y. HUANG ET AL. 218
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. Apositive pressure gradient is consequently generated along the axial axis,
which may lead to the formation of a recirculation zone in a high swirling ow, a
phenomenoncommonlytermedvortexbreakdown. Inthe present congurationand
ow 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 oweld 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 owevolution 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 ow
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 nally disappears because of turbulent dif-
fusion and viscous dissipation at t = 41.4 ms. During this period, the coalesced
vortical bubble separates, and another newbubble appears in the upstreamregion 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 elds 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 KelvinHelmholtz instabilities in both
the axial and azimuthal directions. The ow evolution in the azimuthal direction
is presented on the two crosssections at x/H = 0.36 and 1.54. The shear layer
originating fromthe backward-facing step remains almost symmetric immediately
downstream of the dump plane (x/H = 0.36). It then rolls up and forms large
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 219
Fig. 10.4 Instantaneous axial velocity elds on xr 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) signicantly increases the strength of the shear layer, especially around
the boundary of the CTRZ. As a consequence of ow reversal, the effective ow
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 ow 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.
Y. HUANG ET AL. 220
Fig. 10.6 Instantaneous uctuating pressure eld on xr 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 ow, although the
whole structure follows the motion of the main ow. The swirl number plays an
important role in dictating the ow 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 ow evolves
downstream. A high swirl number apparently helps maintain ow coherence and
leads to strong ow reversal.
2. Vortico-Acoustic Interaction
The strong vortical motion in the chamber often produces acoustic waves prop-
agating throughout the entire eld. The shear layers, however, are susceptible to
acoustic excitations if suchdisturbances occur at appropriate locations andfrequen-
cies. A feedback loop can thus be established, depending on the mutual coupling
between the vortical and acoustic elds. To explore vortico-acoustic interaction,
the uctuating pressure eld is obtained by subtracting the longtime averaged
pressure from its instantaneous quantity.
Figure 10.6 shows snapshots of the uctuating pressure eld on several cross
sections for the high swirl-number case with S = 0.5. Considerable pressure uc-
tuations take place immediately downstream of the centerbody, where strong vor-
ticity is present. The entire eld 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 eld shown in Fig. 10.6 is spatially averaged in
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 221
Fig. 10.7 Time evolution of uctuating pressure eld 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 elds. The data are further reduced by spatial averaging
over each cross section along the axial axis. The averaged quasi-one-dimensional
elds 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 uctuations 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 uctuations are obtained to quantitatively
characterize the acoustic ow 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
Y. HUANG ET AL. 222
t = 35.6 ms
t = 35.9 ms
t = 36.2 ms
p
'
,
P
a
100
50
0
50
t = 35.0 ms
p
'
,
P
a
100
50
0
50
t = 34.7 ms
x/H
0 5 10 15 20
t = 36.5 ms
p
'
,
P
a
100
50
0
50
t = 34.4 ms
x/H
p
'
,
P
a
0 5 10 15 20
100
50
0
50
t = 35.3 ms
Fig. 10.8 Time evolution of uctuating pressure eld spatially averaged over r
cross section; swirl number S = 0.5.
frequency, Hz
0 2000 4000 6000
frequency, Hz
0 2000 4000 6000
frequency, Hz
p
'
,
P
a
0 2000 4000 6000
0
20
40
60
a) b) c)
Fig. 10.9 Frequency spectra of pressure uctuations at different axial locations near
the centerline: a) x/H = 0.38; b) x/H = 2.19; and c) x/H = 7.20, y/H = 0.06, and
z/H = 0.07. Swirl number S = 0.5.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 223
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 owproperties and chamber
geometry.
Asimple acoustic modal analysis, without accounting for the mean oweffects,
indicates that the eigenfrequencies of the rst 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 owdynamics need to be further explored. It
is well established that the shear layer originating from the centerbody is sensitive
to external forcing,
7880
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 dened as
St =
f
S
U
and
U =
1
2
(U
1
+U
2
) (10.4)
where is the initial momentum thickness of the shear layer, and U
1
and U
2
are the freestreamvelocities on the two sides of the shear layer. The mean averaged
axial velocity
U 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 ows.
78
The momentum thickness is roughly one-fourth of
the vorticity thickness,
79
which can be calculated from the axial velocity prole in
the radial direction. On the basis of Eq. (10.4), the frequency of the most unstable
mode of shear-layer instability, f
0
S
, is approximately on the order of 10
3
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 rst-tangential
mode of the acoustic oscillation in the chamber. At this response frequency, f
R
S
,
the shear layer rolls up into discrete vortices and reinforces the acoustic oscillation
in the chamber. A feedback loop between the acoustic uctuation and shear-layer
instability is thus established and leads to a large excursion of owmotions. When
the discrete vortices are convected downstream, they pair with the adjacent ones
to form larger structures with a characteristic frequency of f
R
S
/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 proles shown in Fig. 10.8.
The effects of swirl number on the acoustic eld were examined by considering a
lowswirl number of S = 0.3. The dominant frequencyover the entire eldbecomes
Y. HUANG ET AL. 224
3900 Hz, which corresponds to the mixed rst tangential (1T) and rst radial (1R)
acoustic mode in the chamber. A proper orthogonal decomposition (POD)
81
anal-
ysis of the uctuating pressure oweld also conrms 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
signicant inuence on the oscillatory ow characteristics. Their specic 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 ow
overshadows the shear layer originating from the corner region and dominates the
owdevelopment 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 lowswirl number of S = 0.3, no vortex breakdown occurs and
the importance of the oweld 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 indictatingthe acoustic owevolution, whose characteristic fre-
quency of 3900 Hz matches that of the mixed 1T/1Racoustic 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 ow 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
S
1
S
2
S
3
S
1
S
2
S
3
y
x
1st guide vane
2nd guide vane
fuel
nozzle
Fig. 10.10 Schematic of gas turbine swirl injector with radial entry.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 225
mixing duct includes a center cylindrical passage and two annular passages, which
are spaced radially outward from the axial axis. Three radial-entry swirlers
denoted S
1
, S
2
, and S
3
and counter-rotating with each otherare 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 rst 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 fuelair mixing is improved. The inclusion
of the second passage makes it easy to control the initial swirl number of the ow.
Third, a stronger CTRZ is generated, which increases the relight stability. Finally,
the relight stability and the total ow rate can be decoupled by shifting the airow
through the third passage.
82
The mixing duct in the injector, shown in Fig. 10.10, has a diameter of D
0
=
32 mm at the exit. Two different sets of swirl vanes are explored herein.
62, 63
The
lowswirl-number (LSN) case has swirl vane angles of S
1
= 30 deg, S
2
= 45 deg,
and S
3
= 50 deg; and the high swirl-number (HSN) case has S
1
= 45 deg, S
2
=
60 deg, and S
3
= 70 deg. The corresponding swirl numbers are 0.35 and 0.49,
respectively, based on the ow properties at the injector exit. The baseline ow
condition includes an ambient pressure of 1 atm, an inlet temperature of 293 K,
and a mass ow 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 10
5
.
1. Vortical Flow Evolution
Figure 10.11 shows snapshots of the vorticity magnitude elds on two cross
sections for both the high and low swirl numbers. The ow evolution exhibits
several distinct features, as follows. First, when the ow 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 ow 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 ow, a phenomenon commonly referred to as vortex breakdown or
vortex burst. The resultant ow 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
ow 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 ows 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 KelvinHelmholtz
instabilities. In comparison with the vortex-breakdown-induced central recirculat-
ing ow, the ow 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 owstructures in various parts of the
injector and their underlying mechanisms interact and compete with one another.
Y. HUANG ET AL. 226
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
owregion can be obtained fromthe isosurfaces of the azimuthal velocity shown in
Fig. 10.12. In the lowswirl-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 owelds 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
nally reaches the centerbody. The local owdevelopment depends on the relative
magnitudes of the downward momentum inertia of the incoming ow and the
outward ow motion arising from the centrifugal force. Although the downward
momentum inertia remains almost the same because of the xed inlet mass ow
rate employed, the weaker centrifugal force in the lowswirl-number case causes the
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 227
a)
b)
Fig. 10.12 Instantaneous isosurfaces of azimuthal velocities at u
= 10 and 50 m/s:
a) low swirl number and b) high swirl number.
Y. HUANG ET AL. 228
Fig. 10.13 Streamlines of mean owelds for swirl numbers of S = 0.35 and 0.49.
incoming owto penetrate all the way to the core region, as evidenced in Fig. 10.13.
The ensuing ow 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 ow.
83
The temporal evolution of the oweld 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 ow 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 ow 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. Flowelds spatially averaged in azimuthal direction. The
time interval between pictures is not constant.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 229
a)
b)
Fig. 10.15 Snapshots of azimuthal velocity elds 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 oweld, which is too complex to allowan
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 ow topology. The instantaneous
ow 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 ow
can penetrate deeply into the core region. The temporal variation in the vortical
structure affects the injector characteristics through its inuence on the effective
ow-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 ow 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 ows through the rst and second passages and
a center recirculating ow 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 ow. Large organized
structures then develop under the effect of the KelvinHelmholtz instability when
the incoming streams merge. The situation becomes more obvious for the high
Y. HUANG ET AL. 230
swirl-number case. The center recirculating ow even intersects the outer shear
layer, causing a complex oweld 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 conguration, the mean velocity,
U, is 50 m/s, and the mo-
mentum thickness of the shear layer, , is around 0.2 mm for both swirl numbers.
The frequency of the most unstable mode, f
n
, is estimated to be 1 10
4
Hz.
This value is comparable with the numerically calculated instability frequency
of 13,000 Hz, further demonstrating that the outer shear ow dynamics is dic-
tated by the KelvinHelmholtz 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 KelvinHelmholtz, helical, and centrifugal
instabilities in the azimuthal direction, the ow 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
ow also contributes to the eddy breakup and mixing processes.
c. Interaction and Competition of Instability Modes. As previously men-
tioned, three major ow mechanisms (i.e., vortex breakdown, KelvinHelmholtz
instability, and helical instability) exist and interact with one another within the
injector. The specic type of coupling depends on the swirl number and can be
classied in two categories. First, the outer shear layer may interact with the large
disorganized structures arising from evolution of the central recirculating ow
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
ow 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 fromthe centerbody, the high turbulence intensity in this
region can signicantly 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 ow 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 ow-
eld. In the high swirl-number case, the development of the billows is suppressed
and ow structures are severely distorted by the azimuthal ow instabilities.
Several other competing mechanisms may also exist in the oweld, such as the
one involving the KelvinHelmholtz and centrifugal instabilities. Swirling ows
usually result in an unstable radial stratication, thereby leading to centrifugal
instability,
83
which is enhanced by a higher azimuthal velocity gradient and which
further inuences the streamwise KelvinHelmholtz instability in the outer shear
layer.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 231
2. Spectral Characteristics
The injector dynamics involve an array of intricate ow 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 lowswirl-
number case, the dominant frequency of pressure oscillation along the main ow
passage is 13,000 Hz, corresponding to the most amplied mode of the shear-
layer instability downstream of the rst 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 conrmed by visual
inspection of the ow 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 oweld be-
comes very rich and is characterized by several different frequencies in various
regions. A low-frequency mode around 500 Hz dominates the ow oscillations
Fig. 10.16 Frequency spectra of pressure oscillations along main ow passage; high
swirl ow (S = 0.49).
Y. HUANG ET AL. 232
p: 4 3 2 1 0 1 2 3 4
Fig. 10.17 POD mode shape of pressure eld 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
ow displacement effect of the central recirculation zone. The occurrence of the
4000 Hz oscillation at the injector exit can be explained by considering the ow
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 ow is revealed. Figure 10.18 shows the characteris-
tic frequencies in different regions of the oweld. The prevalence of distinct
frequencies in different regions suggests that the ow instability mechanisms
vary in different regions, a phenomenon consistent with Martin and Meiburgs
expectation.
84
3. Injector Response to External Excitation
Most previous studies on gas-turbine combustion instabilities focused on
thermalacoustic interactions in the chamber. The dynamic behavior of an in-
jector was loosely modeled with an acoustic admittance function at the injector
exit; the specic value of this function was treated as an empirical coefcient.
Very limited effort was applied to investigate the injector internal ow 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 ow rate m are enforced at the injector entrance, similar
to the experiment conducted by Cohen and Hibshman,
85
m = m
0
[1 + sin(2 f
F
t )] (10.5)
where m
0
and f
F
denote the mean mass ow rate and the forcing frequency,
respectively. The amplitude of the oscillation, , is xed at 10%. The forcing
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 233
f
acoustic
~11k Hz
f
vortical
~1.7k Hz
f
passage
~1.2-1.8k Hz
f
PVC
~ 4.0k Hz
f
s
h
e
a
r
~
1
3
k
H
z
f
shear
~13k Hz
recirculation zone
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 ow dynamics. Only the higher swirl-number
case with S
1
= 45 deg, S
2
= 60 deg, and S
3
= 70 deg is considered herein.
The vortical and acoustic elds 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 ow 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, , is
2L/l
F
= 2 f
F
/f
a
(10.6)
where L is the length of the main ow passage and l
F
is the acoustic wavelength
at the forcing frequency.
a. Instantaneous Flow Structures. Figure 10.19 shows snapshots of the uc-
tuating vorticity magnitude elds, |
= 2 and 2 m/s in
azimuthal phase space ( = 0 to 360 deg) in shear layer originating from trailing edge
of rst guide vane: a) top view and b) bottom view.
Y. HUANG ET AL. 240
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 lm, which is then atomized to a spray of ne droplets by the local shear
ow near the rim of the second guide vane. The potential inuence of external
forcing on the breakup of the liquid lm appears in the two conicting areas.
On the one hand, the strong uctuation in the azimuthal direction promotes the
development of an instability wave on the fuel lming surface and the subsequent
atomization process.
88
On the other hand, as shown in Fig. 10.23, the external
forcing may signicantly modify the mean azimuthal velocity eld near the fuel
lming surface, especially when the forcing frequency approaches the shear-layer
characteristic frequency (i.e., 13 kHz). The ow near the downstream part of
the second guide vane even changes its direction from counter-rotating to co-
rotating with the ows in the main and the third (S
3
) passages. This qualitative
switch of ow pattern represents an undesired feature from the perspective of fuel
atomization.
23
In spite of the modication of the oweld between the rst 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 oweld. This may be attributed
to the weakness of the excitation as compared with the intrinsic high-intensity
ow 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 ow also discourages the modulation of the mean ow by a single-
harmonic excitation unless the forcing resonates with the local ow 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 ow motion in a combustion chamber.
90
The admittance function,
also the reciprocal of the impedance function, measures the velocity uctuation in
response to incident pressure uctuation. Following common practice, the acoustic
admittance function, A
d
, is dened as
A
d
( f ) =
u
a
/ a
p
a
/ p
(10.7)
where p and a 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 = R
0
is approached.
This outcome may be attributed to the relatively low pressure oscillation and high
velocity uctuations near the upper boundary. In this region, the pressure response
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 241
Magnitudeof A
d
r
/
R
0
0 5 10
0.0
0.5
1.0
500
900
1500
4000
f
F
(Hz)
Phase of A
d
, radian
r
/
R
0
1.0 0.5 0.0 0.5 1.0
0.0
0.5
1.0
500
900
1500
4000
f
F
(Hz)
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/R
0
< 1.0) becomes very small.
Since the liquid lm breaks up in the trailing edge of the second guide vane,
the ow 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 uctuation, which consequently exerts a strong inuence on spray
formation at that location.
The phase distribution of the admittance function indicates a lag between
the velocity and pressure uctuations of around 90 deg in the main ow passage
Y. HUANG ET AL. 242
(0.3 < r/R
0
< 0.8). The situation is consistent with the behavior of a simple trav-
eling acoustic wave without much inuence fromshear layers. The phase behavior
for the 4000 Hz case exhibits a trend distinct fromthat 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 ow. 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 ow
instabilities.
d. Mass Transfer Function. Another important measure of the injectors dy-
namic response is the transfer function of the total mass ow rate between the
injector entrance and exit, dened as
m
( f ) =
m
a
ex
m
a
i n
(10.8)
Here
m
a
is the Fourier component of the mass ow rate at the forcing frequency,
which is obtained by integrating the mass ux 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 uctuation of the mass ow rate
between the entrance and exit is clearly found. At rst glance, this observation
seems to violate the law of mass conservation for such an acoustically compact
M
a
g
n
i
t
u
d
e
o
f
m
0
1
2
Forcing frequency, Hz
P
h
a
s
e
o
f
m
,
r
a
d
i
a
n
0 5000 10000
2
0
Fig. 10.26 Effect of forcing frequency on transfer function of total mass ow rate.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 243
injector. The forcing frequency is much lower than the acoustic characteristic fre-
quency of the injector, f
a
. Under this condition, the oweld in the injector can
be treated as incompressible, and the instantaneous total mass ow 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 ow rates at the injector entrance and exit are
calculated. The result conrms the conservation of the overall mass ow rate for
all the forcing frequencies considered herein. The 1500 Hz forcing indeed excites
the oweld at the expense of suppressing uctuations at other frequencies. The
mass-owtransfer function for the 4000 Hz forcing is less than unity. In addition to
channeling mass ow among different Fourier components, ow 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 ux uctuation but also the temporal variation
of the overall mass ow 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, ,
in Eq. (10.6). The good agreement between the analytic estimation, Eq.(10.6),
and the numerical result further veries that the oscillation of the mass ow rate
propagates in the form of an acoustic wave.
C. Flow Dynamics of Axial-Entry Swirl Injector
This section deals with the ow dynamics of an axial-entry injector, the CFM56
swirl injector of GE Aircraft Engines, as shown in Fig. 10.27. It is a prelming
airblast injector, and it has been implemented in aero gas-turbine engines because
of its high combustion efciency, 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 are. The fuel nozzle is located at the
center of the primary-jet plane, where air swirlers give rise to a strong swirling
oweld 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 nely atomized spray is produced in this region. The fuel droplets carried by
the primary counterclockwise air stream mix with the counter-rotating secondary
ow to further promote rapid fuelair 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 lm, which is then atomized to a spray of ne droplets by
the local shear ow near the entrance of the secondary swirlers. In an effort to
optimize the combustor design, characterization of the oweld inside the swirl
injector becomes critical.
43, 75
The injector owwas studied in a 3
)
Fig. 10.33 Radial distributions of mean-ow axial, radial, and azimuthal velocity
components (symbol: co-rotation; line: counter-rotation).
Y. HUANG ET AL. 250
In addition to the central recirculation zone, another major difference between
the two congurations lies in the distributions of turbulent kinetic energy and shear
stress near the exit of the venturi. At that locationinthe counter-rotatingcase, 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 prelming surface. The co-
rotating conguration 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 ow evolution in the two congurations is identical.
The observation is further corroborated by the fact that the origin of PVCis located
within the venturi, in which the ow motion is mainly controlled by the swirling
ow 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 conguration produces a large recirculation zone, which
is more susceptible to ame 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 KelvinHelmholtz instability in the liquid
lm and subsequently the formation of ne droplets, as suggested by Chin and his
coworkers
23
in their experiments. Third, the counter-rotating ow 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
downstreamof the fuel nozzle, and facilitates the breakup of the liquid fuel. There-
fore, the counter-rotating design is expected to produce ner droplets and a much
more stable ame.
III. Flame Dynamics of Axial-Entry Swirl Injector
This section treats the ame 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 at vane swirler. Natural gas is injected radially from the centerbody through
10 holes immediately downstream of the swirler vanes. The fuelair mixture is
swirl
injector
combustion
chamber
38 mm
45 mm
choked
exit
235 mm
Fig. 10.34 Schematic of a model swirl-stabilized gas-turbine combustor (after
Broda et al., 1998, Ref. 14).
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 251
c
cross section view A A'
D
n
= 20.3 mm; D
h
= 9.53mm
D
n
D
h
A
A'
top view
Fig. 10.35 Schematic of top and cross-section views of a at 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. Achoked venturi is also installed at the inlet entrance to
acoustically isolate the test section fromthe air supply line. The baseline condition
includes an equivalence ratio of 0.57 and a chamber pressure of 0.46 MPa. The
mass ow rates of the natural gas and air are 1.7 and 50.7 g/s, respectively. The
inlet ow 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 T
i n
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 ame and an unstable ame with
an equivalence ratio of 0.6. As the inlet temperature increases and exceeds the
threshold value T
i n
, the ame structure transforms froma stable state to an unstable
one, and the amplitude of pressure oscillation increases and reaches a limit cycle.
This kind of bifurcation phenomenon in ame structure, as well as oscillatory
ame dynamics, is discussed systematically in this section.
The basis of the analysis is the LES technique developed specically for in-
vestigating lean premixed (LPM) combustion instabilities in swirl-stabilized com-
bustors.
55, 64, 65
The formulation employs the Favre-ltered conservation equations
in three dimensions and an appropriate subgrid-scale (SGS) model. A level-set
amelet library approach, which has been successfully applied to simulate pre-
mixed turbulent combustion,
55
is used. In this approach, the ltered ame surface
evolution is modeled using a level-set G-equation, where G is dened as a dis-
tance function outside the ame front. Thermophysical properties are obtained by
using a presumed probability density function (PDF) along with a laminar amelet
library.
A. Stable Flame Dynamics
Stable ame evolution was rst obtained for an inlet mixture temperature of
600 K (below the threshold value T
i n
for the onset of combustion oscillation). The
inlet swirl number dened by Eq. (10.2) is 0.76. The ame bifurcation phenomenon
Y. HUANG ET AL. 252
inlet air temperature, T
in
(K)
p'
rms
/P
c
15.0
7.5
p
' r
m
s
/
P
c
(
%
)
0.0
600 620 640 660 680 700
a)
Overall Equivalence Ratio,
Fig. 10.39 Stable ame evolution over one cycle of oscillation (3214 Hz): temperature
contours and streamlines for S = 0.76.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 255
inlet annulus exit. The phase angle 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 ame front,
which moves back and forth under the inuences of the vortical motion (indicated
by the concentrated streamlines) in the chamber. A new vortex begins to shed
from the center body at = 90 deg, accompanied by a higher local ow velocity.
As the vortex moves downstream ( = 180 deg 270 deg), it distorts the ame
front or even produces a separated ame pocket. At the same time, the higher-
speed mixture pushes the ame downstream. When the vortex moves away from
the ame ( = 360 deg) and dissipates into small-scale structures, the ame 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 ame dynamics in the system.
On the one hand, when the inlet temperature increases, for a xed mass ow rate,
the ow velocity also increases and pushes the ame downstream. On the other
hand, the increased inlet temperature leads to an increase in the ame speed and
consequently causes the ame to propagate upstream. In addition, ashback may
occur near the wall because of the small local ow velocity. The combined effects
of ow acceleration, ame-speed enhancement, and ashback determine the nal
form of the ame structure.
In the present study, as the inlet temperature increases from 600 to 660 K,
ame bifurcation takes place. The ame originally anchored in the center re-
circulation zone penetrates into the corner recirculation zone and ashes back.
Consequently, the ame is stabilized by both the corner-recirculating zone and the
center-recirculating ow and forms a compact enveloped conguration. The ame
aps dynamically and drives ow oscillations through its inuence 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 lling process, ame-
trapping process, and vortex-ashback 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 10.40a10.40c show the high-temperature-mixture lling process. As
the inlet mixture temperature increases, the ow speed increases because of
the decreased density for a xed mass ow rate. As a result, the original low-
temperature mixture is pushed downstream toward the ame. Although a ash-
back phenomenon is observed near the wall, the high-temperature mixture has not
reached the ame front near the wall and the ame speed remains unchanged at
this stage.
Figures 10.40d and 10.40e show the ame-trapping process. Once the high-
temperature mixture reaches the ame front, with the help of the increased ame
speed, the near-wall ashback overshadows the ow acceleration effects. Conse-
quently, the ame front penetrates into the corner recirculation zone and is trapped
by the local vortical motion.
Y. HUANG ET AL. 256
a)
b)
c)
d)
e)
f)
g)
h)
Fig. 10.40 Transition from stable to unstable ame with increased inlet temperature
from 600 to 660 K.
In the vortex-ashback process, as shown in Figs. 10.40f10.40h, the ame
propagates upstream under the inuence of the vortical motion. A counterclock-
wise rotating vortex originally shed from the edge of the backward-facing step
approaches the ame front in the corner recirculation zone and then pushes it
toward the dump plane. At the same time, a small ame pocket is produced
and separated from the main stream. After this vortex is convected downstream
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 257
and passes through the ame, another vortex approaches and interacts with the
ame. The process continues, and eventually the fresh reactants in the corner
recirculation zone are completely burnt. The ame is stabilized by both the corner-
recirculating ow and the center-recirculating ow, 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 ame becomes unstable (oscillatory) when the inlet ow temperature
exceeds the critical value T
i n
, it becomes difcult to reestablish stable operation
unless the inlet temperature is reduced to a level signicantly lower than T
i n
.
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 lled with high-temperature products,
and the chamber wall in this region is heated to reach the local ame tempera-
ture. To recover stable operation, the cold ow needs not only to extinguish the
ame stabilized by the corner recirculating ow through entrainment or ame
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 ash-
back. Consequently, a much lower inlet temperature is required to regain stable
operation.
In light of the preceding observations, we conclude that the ashback phe-
nomenon dictates the ame 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
rst involves ame propagation in the boundary layer along a solid wall, where
the local velocity diminishes toward the surface. The second mechanism is asso-
ciated with ow 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 ame speed S
L
increases with an
increase in the equivalence ratio . Thus, increases in the equivalence ratio and in-
let temperature exert similar effects on the ame 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 can
cause periodic extinction of the ame. As a result, the equivalence ratio oscil-
lation under lean conditions is prone to inducing ow oscillation
97
and subse-
quently increases turbulent velocity uctuation 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 ame bifurcation is largely determined by the ashback phenomenon in the
corner recirculation zone in the present case, one effective way to avoid its oc-
currence is to inject cold ow into that region. This procedure suppresses the
upstreampropagation of local ame and consequently leads to a much more stable
system.
Y. HUANG ET AL. 258
C. Oscillatory Flame Dynamics
When the inlet ow temperature exceeds the threshold value, unstable (oscilla-
tory) ames are clearly observed and are accompanied by large excursions of ow
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 ame dynamics. For S = 0.44, three dominant modes
exist at the frequencies of 1761, 10,367, and 17,618 Hz, corresponding to the rst
longitudinal (1L), rst 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 eld. 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 ow properties are obtained by taking the longtime average of the
instantaneous quantities. In spite of signicant ow motions in the azimuthal di-
rection, the mean oweld remains perfectly axisymmetric. Figure 10.41 shows
the streamline patterns and mean temperature elds on the xr plane. Three dis-
tinct recirculation zones are observed in the lowswirl-number case with S = 0.44,
includinga wake recirculationzone (WRZ) behindthe centerbody, a corner recircu-
lation zone attributable to the sudden enlargement of the combustor conguration,
and a central toroidal recirculation zone (CTRZ) resulting fromvortex breakdown.
The WRZ, however, disappears at the high swirl number of S = 1.10. The overall
ow 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 Chao
99
in his experimental study of the recirculation
structure in a coannular swirl combustor. The mean temperature elds clearly ex-
hibit enveloped ames anchored at the rim of the centerbody and the corner of the
backward-facing step. The ame is much more compact for the high swirl-number
case with S = 1.10, which is caused by the enhanced ame 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 ows 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 owpassage. The stronger the swirl strength is, the faster the
main ow 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 elds suggest that the ow 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
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 259
Fig. 10.41 Mean temperature elds and streamline patterns for S = 0.44, 0.76,
and 1.10.
the centerbody and the backward-facing step, where large velocity uctuations
are produced because of the strong turbulent mixing in the shear layers between
the incoming ow and the recirculation ows. The evolution of turbulent kinetic
energy is governed by the following equation:
D
u
2
i
/2
/Dt =
u
j
p/
0
u
2
i
u
j
/2 +2vu
i
e
i j
/x
j
u
i
u
j
U
i
/x
j
2ve
i j
e
i j
(10.9)
where e
i j
= (u
i
/x
j
+u
j
/x
i
)/2. The existence of the steep velocity gradi-
ent U
i
/x
j
because of the strong swirling and reverse ow in the high swirl-
number case, facilitates the production of turbulent kinetic energy, u
i
u
j
U
i
/x
j
.
Y. HUANG ET AL. 260
r
(
c
m
)
0 100
0
1
2
x
=
2
5
m
m
0 100
x
=
3
0
m
m
u
x
(m/s)
0 100
x
=
4
0
m
m
0 100
x
=
5
0
m
m
0 100
x
=
6
0
m
m
0 100
x
=
8
0
m
m
0 100
x
=
8
0
m
m
0 100
x
=
6
0
m
m
0 100
x
=
5
0
m
m
u
r
(m/s)
0 100
x
=
4
0
m
m
0 100
x
=
3
0
m
m
r
(
c
m
)
0 100
0
1
2
x
=
2
5
m
m
0 100
x
=
8
0
m
m
0 100
x
=
6
0
m
m
0 100
x
=
5
0
m
m
u
(m/s)
0 100
x
=
4
0
m
m
0 100
x
=
3
0
m
m
r
(
c
m
)
0 100
0
1
2
x
=
2
5
m
m
0 50
x
=
8
0
m
m
0 50
x
=
6
0
m
m
0 50
x
=
5
0
m
m
k (m/s)
0 50
x
=
4
0
m
m
0 50
x
=
3
0
m
m
r
(
c
m
)
0 50
0
1
2
x
=
2
5
m
m
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 Floweld
Vorticity is of concern because of its dominant inuence in determining the
ow entrainment in the reaction zone and the subsequent ame evolution. Fig-
ure 10.43 shows snapshots of the vorticity magnitude eld on an xr plane. For
the low swirl-number case with S = 0.44, large vortical structures, arising from
the shear layers downstream of the dump plane and centerbody, are convected
downstream and then dissipated into small-scale eddies. The same phenomenon is
observed for the high swirl-number case, in which well-organized vortices are shed
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 261
Fig. 10.43 Snapshots of vorticity magnitude eld on xr and x 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 ow. In both cases, the vortex shedding frequen-
cies are close to that of the rst tangential mode of the acoustic wave in the
chamber.
Figure 10.44 shows snapshots of the isovorticity surface at =75,000 1/s.
The oweld 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 KelvinHelmholtz 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-dened frequencies, as previously described.
Y. HUANG ET AL. 262
Fig. 10.44 Snapshots of isovorticity surface at = 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 conjecture
55
that the vortical motions in the shear layers resonate with
acoustic oscillations in the chamber. In the present conguration, two shear layers
exist downstream of the rear-facing step and the centerbody. The axial momen-
tum thickness,
0
, 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 ve, be-
cause of its stronger swirl strength. A linear stability analysis
100
has been carried
out to provide more insight into the shear-layer instability phenomena for annular
swirling ows 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-amplied frequencies are different
for the two swirl numbers considered herein, mainly because of the disparity of
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 263
the axial momentum thickness between the two cases. In addition, the predicted
values are much higher than the vortex-shedding frequency, which corresponds to
the frequency of the rst 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 ame dynamics and ow oscil-
lation, the total heat release and ame surface area were analyzed in the frequency
domain. The overall heat release in the chamber can be obtained from
Q =
u
h
0
f
S
T
A (10.10)
where
u
is the unburnt gas density, S
T
is the subgrid turbulent ame speed, h
0
f
is the heat of reaction, and A is the total ltered ame surface area.
Figure 10.45 shows the power spectral densities of the total ltered ame
surface-area and heat-release uctuations. At S = 0.44, a dominant mode exists
A
m
p
l
i
t
u
d
e
,
m
m
2
/
H
z
0 10 20
0
100
200
300
400
S = 0.44
1761
3320
flame surface area
A
m
p
l
i
t
u
d
e
,
m
m
2
/
H
z
0 10 20
0
100
200
300
400
S = 1.10
11,712
flame surface area
Frequency, kHz
k
J
/
s
/
H
z
0 10 20
0
2
4
6
8
S = 0.44
1761
3320
heat release
20,532
Frequency, kHz
k
J
/
s
/
H
z
0 10 20
0
2
4
6
8
S = 1.10
11,712
heat release
A
m
p
l
i
t
u
d
e
,
A
m
p
l
i
t
u
d
e
,
Fig. 10.45 Power spectral densities of total ame surface-area and heat-release uc-
tuations for S = 0.44 and 1.10.
Y. HUANG ET AL. 264
Fig. 10.46 Temporal evolution of temperature eld over one cycle of 1L mode of
oscillation for S = 0.44.
at 1761 Hz in the ame-surface oscillation, which corresponds to the 1L acoustic
mode of the combustor. Ahigher 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 ame 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 uctuations bears a close
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 265
resemblance to that of ame 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
uctuations in the subgrid turbulent ame speed S
T
.
55
In light of the preceding
observations, one can conclude that low-frequency acoustic perturbations exert a
strong inuence on the uctuations of the total ame surface area and heat release.
In contrast, high-frequency acoustic oscillations travel through the ame zone
without signicantly affecting the ame surface area and heat-release variations,
although they may impose a signicant impact on the local ame propagation. The
results qualitatively agree well with the prediction from a companion analytical
analysis of ame response.
101
The calculated mean ame surface area and the root
mean square of the uctuating 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 ame speed in the high
swirl-number case, the mean heat-release rate and the associated uctuation are
very close in these two cases.
4. Acoustic and Flame Interaction
Figure 10.46 presents the temporal evolution of the temperature eld 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 is referenced to the 1L acoustic pressure
at the chamber head end. The entire process is dictated by the cold-ow entrain-
ment into and mixing with hot gases in the vortical structures in the ame zone.
Figure 10.47 shows the time histories of the pressure immediately downstream of
the dump plane (top), the total ame surface area (middle), and the rate of heat
release (bottom). These signals involve a wide range of frequencies corresponding
to turbulent-ow and acoustic oscillations. The extracted 1L oscillations (denoted
by the thick black lines) of these quantities are also plotted for clarity. The ame
Fig. 10.47 Time histories of pressure immediately downstream of the dump plane
(top), ame surface area (middle), and heat release rate (bottom) for S = 0.44; the
thick black lines represent the extracted 1L oscillations.
Y. HUANG ET AL. 266
Time, ms
p
,
k
p
a
f
l
a
m
e
s
u
r
f
a
c
e
a
r
e
a
,
m
m
2
h
e
a
t
r
e
l
e
a
s
e
,
k
J
23 24 25
200
300
400
500
2000
3000
4000
100
150
200
250
total heat release
flame surface area
pressure
Fig. 10.48 Time histories of pressure immediately downstream of the edge of the
centerbody (top), ame 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 oweld. It lags behind the pressure oscillation by 76 deg. During the
period from = 166 deg (t = 24.09 ms) to 14 deg (t = 24.38 ms), a relatively
lower pressure eld exists near the dump plane, facilitating the delivery of the
fresh reactants into the chamber. Intensive heat release then occurs after a short
uid-mixing and chemical-induction time. The resultant ow expansion pushes
the ame outward and causes the ame 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 = 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 ame zone is thus reduced and becomes a lit-
tle more compact. The same process then repeats for another cycle of oscillation.
Figure 10.47alsoindicates that the heat-release andame-surface-area uctuations
are nearly in phase. The former only lags behind the ame 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 eld in the
upstream section of the chamber on the xr plane over one cycle of the 1T mode
of acoustic oscillation. As the swirl number increases, the ame anchored by
the center recirculating owmay propagate upstreamperiodically and cause ame
ashback. Two mechanisms, as previously stated, have been identied for the
occurrence of ame ashback. In this case, the ashback is closely linked to
the strong reverse ow in the center recirculation zone. The swirl strength is so
strong that it sometimes causes the center recirculating owto enter the inlet annu-
lus. As a consequence, the ame attached to the centerbody travels upstream and
ashback occurs.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 267
Fig. 10.49 Temporal evolution of temperature eld on an xr 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 eld at various times within one
cycle of the 1T acoustic oscillation for S = 0.76. The thick black line indicates the
ame front. Well-organized vortices are shed fromthe edge of the backward-facing
step. The process, however, becomes much more complex in the downstreamof the
Y. HUANG ET AL. 268
Fig. 10.50 Vortex and ame front interaction over one cycle of 1Tmode of oscillation
for the case S = 0.76.
centerbody because of the existence of a toroidal recirculating ow. New vortices
are produced at the tip of the backward-facing step at = 72 deg and bulge the
ame front. They continue to distort the ame or even produce separated ame
pockets when traveling downstream. Finally, these vortices move out of the ame
region and dissipate into small-scale structures. Another set of vortices appears at
= 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 ame surface possesses a characteristic frequency cor-
responding to the 1T acoustic wave. Because the vortex shedding affects the
shapes of the ame 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
uctuation.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 269
Fig. 10.51 Temporal evolution of isovorticity surface at =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 ow and ame dynamics of several different types of swirl-
injectors for contemporary gas turbine engines were studied. Both axial- and radial-
entry congurations were explored over a wide range of ow 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
Y. HUANG ET AL. 270
obtained from large eddy simulations of detailed ow structures under conditions
with and without external forcing. Where appropriate, experimental observations
are also given for the sake of completeness.
For cold-owcharacterization of injector dynamics, various fundamental mech-
anisms dictating the owevolution in swirl injectors were identied. These include
vortex breakdown; precession of vortex core; and KelvinHelmholtz, 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 ow uctuations, but
the trend is reversed for high-frequency excitations. The uctuation of the mass
owrate 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 ow. Results of this kind can be effectively used in analyzing combustion
instabilities in gas-turbine engines. The inuence of swirler orientation on injector
owdevelopment was also examined by considering both counter- and co-rotating
inow conditions. The counter-rotating conguration 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 ame dynamics of swirl injectors.
Detailed owstructures and ame 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. Aslight increase in the inlet airowtemperature across
the stability boundary leads to a sudden increase in the chamber owoscillation. Fi-
nally, the underlying mechanisms responsible for driving combustion oscillations
and the transition from a stable to an unstable ame were identied.
Acknowledgments
The work reported in this chapter was sponsored in part by the Ofce 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 Ofce of
Scientic 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. 617654.
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. 19103.
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. 441445.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 271
5
Wang, H., McDonell, V. G., Sowa, W. A., and Samuelsen, S., Experimental Study
of a Model Gas Turbine Combustor Swirl Cup, Part II: Droplet Dynamics, Journal of
Propulsion and Power, Vol. 10, No. 4, 1994, pp. 446452.
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., andMongia, H. C., Air Temperature Effects onNon-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. 554562.
10
Snyder, T. S., Rosfjord, T. J., McVey, J. B., and Chiappetta, L. M., Comparison of
Liquid/Air Mixing and NOx Emission for a Tangential Entry Nozzle, American Society
of Mechanical Engineers, Paper 94-GT-283, 1994.
11
Cohen, J. M., Rey, N. M., Jacobson, C. A., and Anderson, T. J., Active Control of
Combustion Instability in a Liquid-Fueled Low-NOx Combustor, Journal of Engineering
for Gas Turbines and Power, Vol. 121, 1999, pp. 281284.
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. 294302.
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. 776782.
14
Broda, J. C., Seo, S., Santoro, R. J., Shirhattikar, G., and Yang, V., An Experimental
Study of Combustion Dynamics of a Premixed Swirl Injector, Proceedings of the Com-
bustion Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 18491856.
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. 775782.
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. 909918.
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. 415421.
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. 735749.
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, JPGC2001/FACT-
19089, June 2001.
Y. HUANG ET AL. 272
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. 14811487.
21
Cohen, J. M., and Rosfjord, T. J., Inuence on the Sprays Formed by High-Shear
Fuel-Nozzle/Swirler Assemblies, Journal of Propulsion and Power, Vol. 9, No. 1, 1993,
pp. 1627.
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. 241247.
23
Chin, J., Rizk, N., and Razdan, M., Effect of Inner and Outer AirowCharacteristics on
High Liquid Pressure Prelming Airblast Atomization, Journal of Propulsion and Power,
Vol. 16, 2000, pp. 297301.
24
Hardalupas, Y., and Whitelaw, J. H., Interaction Between Sprays fromMultiple Coax-
ial Airblast Atomizers, Journal of Engineering for Gas Turbines and Power, Vol. 118,
1996, pp. 762771.
25
Hardalupas, Y., Pantelides, K., and Whitelaw, J. H., Particle Tracking Velocimetry in
Swirl-Stabilized Burners, Experiments in Fluids, [Suppl.], Vol. 29, 2000, pp. S220S226.
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. 2544.
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. 631638.
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. 18171824.
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. 12981304.
30
Acharya, S., Murugappan, S., ODonnel, M., and Gutmark, E. J., Characteristics and
Control of Combustion Instabilities in a Swirl-Stabilized Spray Combustor, Journal of
Propulsion and Power, Vol. 19, No. 3, 2003, pp. 484496.
31
Uhm, J. H., Acharya, S., Control of Combustion Instability with a High-Momentum
Air-Jet, Combustion and Flame, Vol. 139, 2004, pp. 106125.
32
Murugappan, S., Acharya, S., Allgood, D. C., Park, S., Anaswamy, A. M., andGhoniem,
A. F., Optimal Control of a Swirl-StabilizedSprayCombustionUsingSystemIdentication
Approach, Combustion Science and Technology, Vol. 175, 2003, pp. 5581.
33
Lee, J. G., Kim, K., Santavicca, D. A., Effect of Injection Location on the Effective-
ness of an Active Control System Using Secondary Fuel Injection, Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 739746.
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. 232240.
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. 795809.
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. 262268.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 273
37
Johnson, C. E., Neumeier, Y., Neumaier, M., Zinn, B. T., Darling, D. D., and Sattinger,
S. S., Demonstration of Active Control of Combustion Instabilities on a Full-Scale Gas
Turbine Combustor, ASME 2001-GT-519, 2001.
38
Conrad, T., Bibik, A., Shcherbik, D., Lubarsky, E., and Zinn, B. T., Slow Control of
Combustion Instabilities by Fuel Spray Modication 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 Modication 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. 353385.
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
Connement 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. 782794.
Y. HUANG ET AL. 274
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. 155160.
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. 7397.
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. 489505.
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. 17911798.
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. 383389.
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. 17751782.
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. 263304.
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. 251293.
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. 223244.
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. 13111333.
72
Wang, S., Yang, V. and Koo, J. Y., Large-Eddy Simulation of Spray-Field Dynamics
in Cross Flows, AIAA Paper 2005-0729, 2005.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 275
73
Menon, S., Acoustic-Vortex-Flame Interactions in Gas Turbines, Combustion Insta-
bilities in Gas Turbine Engines: Operational Experience, Fundamental Mechanisms, and
Modeling, edited by T. Lieuwen and V. Yang, Progress in Astronautics and Aeronautics,
AIAA, Reston, VA, 2005, pp. 277314.
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. 171195.
75
Wang, S., Yang, V., Hsiao G., Hsieh, S. Y., and Mongia, H., Large Eddy Simulation of
Gas-Turbine Swirl Injector FlowDynamics, 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. 2158.
79
Ho, C.-M., and Huerre, P., Perturbed Free Shear Layers, Annual Review of Fluid
Mechanics, Vol. 16, 1984, pp. 365424.
80
Panda, J., and McLaughlin, D. K., Experiments on the Instabilities of a Swirling Jet,
Physics of Fluids, Vol. 6, 1994, pp. 263276.
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. 539575.
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. 537566.
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. 19171928.
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. 241258.
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 Hopnger, E., Liquid Jet Instability and Atomization in a Coaxial
Gas Stream, Annual Review of Fluid Mechanics, Vol. 32, 2000, pp. 275308.
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. 131159.
90
Culick, F. E. C., and Yang, V., Overview of Combustion Instabilities in Liquid-
Propellant Rocket Engines, Chapter 1, Liquid Rocket Engine Combustion Instability, edited
by V. Yang and W. E. Anderson, Progress in Astronautics and Aeronautics, Vol. 169, AIAA,
Washington, DC, 1995, pp. 337.
Y. HUANG ET AL. 276
91
Gupta, A. K., Lilley, D. G., and Syred, N., Swirl Flows, Abacus Press, London, 1984.
92
Lucca-Negro, O., and ODoherty, T., Vortex Breakdown: A Review, Progress in
Energy and Combustion Science, Vol. 27, 2001, pp. 431481.
93
Syred, N., and Beer, J., The Damping of Precessing Vortex Cores by Combustion in
Swirl Generators, Astronautica Acta, Vol. 17, 1972, pp. 783801.
94
Novak, F., and Sarpkaya, T., Turbulent Vortex Breakdown at High Reynolds Number,
AIAA Journal, Vol. 38, No. 5, 2000, pp. 287296.
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. 182189.
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. 174.
99
Chao, Y. C., Leu, J. H., and Huang, Y.F., DownstreamBoundary Effects on the Spectral
Characteristics of a Swirling Floweld, Experiments in Fluids, Vol. 10, 1991, pp. 341348.
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. 56, 2005, pp. 11091150.
Chapter 11
Acoustic-Vortex-Flame Interactions in Gas Turbines
Suresh Menon
l/,
where u
D
Dt
(11.1)
Dv
Dt
=
p
(11.2)
De
Dt
=
p
( v) (11.3)
DY
k
Dt
=
k
, k = 1, . . . N (11.4)
Here, the substantial derivative is D/Dt = /t +v , where vis the velocity
vector, is the mixture density, p is the pressure, e is the internal energy per unit
mass dened as e =
k
e
k
Y
k
, where e
k
is the kth species internal energy (sum
of the translational, rotational, and heat of formation), Y
k
is the kth species mass
fraction, and
k
is the kth species production/destruction term. These equations
are usually closed by the equation of state for a perfect gas p = RT, where T
is the temperature and where R =
k
R
k
Y
k
is the mixture gas constant. Here, R
k
is the the kth species gas constant dened as R
k
= R/M
k
, where M
k
is the kth
species molecular weight.
It is instructive to rewrite these equations in a formthat 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 specic formulation. By using the denition of
internal energy for a multicomponent mixture in terms of entropy, we can rewrite
S. MENON 282
Eq. (11.3) as
39
(details are avoided for brevity)
DS
Dt
=
1
T
N
k=1
_
k
M
k
_
k
(11.5)
Here, S is the entropy and
k
is the chemical potential.
39
This equation shows
that entropy is convected because of uid 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 is
39
v =
1
p
Dp
Dt
1
k=1
k
k
(11.6)
where
k
= h
k
/(C
p
T) for an ideal gas mixture with constant average molecular
weight and where C
p
is the specic 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 pressure
31, 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, :
D
Dt
= ( )v ( v) +
p
2
(11.7)
For analysis purposes, Eq. (11.7) is often combined with Eq. (11.1) to obtain
D
Dt
_
_
=
_
_
v +
1
3
p (11.8)
Both forms of the vorticity equation show important features that are relevant
to the current discussion. The rst 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 ow. The quantity ( v) is the thermal expansion
term, which is nonzero only in compressible owand in reacting ow. In Eq. (11.8),
this term is absorbed into the convective term on the left-hand side. Therefore, the
left-hand side of Eq. (11.8) contains convection by both the hydrodynamic and
the acoustic velocity. As shown subsequently, this form can be used to identify
how acoustic and hydrodynamic disturbances can interact and contribute to vortex
motion.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 283
The last term, p/
3
, often called the baroclinic torque, is a consequence
of the relative orientation of the pressure and density gradients. This term can be
signicant in both compressible and reacting ows. For example, baroclinic torque
has been shown to affect ame wrinkling indirectly because of the changes induced
in the vorticity eld.
42
Pressure uctuations normal to the density gradient caused
by the ame also contribute to vorticity generation. In gas turbine combustors,
the ame (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 eld. Furthermore, this term can cause enhancement
or suppression of vorticity in the combustor depending on local conditions. This
behavior is discussed in the nal 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 neglectingtransport properties (e.g., dissipative effects) andexternal forces (and
these assumptions can be easily relaxed), these equations are exact, nonlinear, and
dene the oweld.
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
Kovaznay
3
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 ow 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 identied with hot spots).
As shown elsewhere
3
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 conned domain. It has been noted that vorticity and entropy modes can exist
even in the absence of pressure uctuations but cannot exist if there is no mean
ow; and to the rst approximation, weak vorticity uctuations do not generate
pressure or entropy uctuations of the same order.
Stability analysis
11, 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 growexponentially over time. In most of these studies, the linearized equations
of motion are solved subject to appropriate boundary conditions, jump conditions
across the ame, and by assuming that the initial perturbation is made up of the sum
of acoustic, entropy, and vorticity disturbances at a characteristic frequency, which
S. MENON 284
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 acousticvortex coupling. For example, in dump
combustors, because of the mean ow 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 fromthe
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 acousticvortex mode requires that the boundary conditions
reect 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 uctuation velocity u
, the boundary
condition for the vortex mode can be = u
= (x x
n
)/x has
been employed in the past.
11
Here, (x x
n
) is the Dirac delta function, and
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-
plications are needed to achieve a tractable formulation, since the complexity of
gas-turbine geometry, the swirl effects, the three-dimensional variation in mean
ow velocity and temperature, and the proper boundary conditions between mul-
tiple duct coupling and ow passages are difcult to implement in a simplied
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.
Adecomposition suggested by Hussain
43
could be used to obtain further insight
into AVF interactions, especially in highly turbulent ow containing large-scale
coherent structures (CSs). The ame structure can be substantially modied by
these CSs. Because of their deterministic nature and quasi-periodic motion, CSs
can contribute signicantly 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 ow variable is split as
(x, t ) =
(x) +
(x, t ; ) +
(x, t ) (11.10)
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 285
Here,
is the mean value,
is an unsteady coherent component (over a char-
acteristic time period ), and
is an unsteady incoherent contribution. This de-
composition is similar to the classical Reynolds decomposition, except that, in the
latter approach,
=
+
represents all the unsteady uctuations in the ow.
It has been shown
43
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 howthe 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 ow, coherent
motion, and incoherent motion. It would then be feasible to identify the terms
that control, for example, acoustic eld excitation by coherent structure transport,
and so on. Previous studies
44
have shown that the DNS database can be post-
processed to obtain CS information in shear ows. 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 eld can be characterized in two ways: 1) acoustics without any
owand 2) acoustics with ow. In the former case, the generation and the motion of
waves because of small perturbation on top of a stagnant or a steady mean owcan
be studied by using linearized analysis. In the latter case, especially for gas-turbine
combustors, the ow 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 eld 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 owin 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 fromnumerical
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):
t
( v) +v ( v) +v : v =
_
1
p
_
(11.11)
and combine with Eq. (11.1), to obtain
D
Dt
_
1
D
Dt
_
2
p = v : v
1
2
p (11.12)
If we consider the equation of state of the form = ( p, S), then
D
Dt
=
_
p
_
S
Dp
Dt
+
_
S
_
p
DS
Dt
(11.13)
S. MENON 286
On using Eq. (11.13) in Eq. (11.12) and noting that (/p)
S
= a
S
2
is the
isentropic (frozen) acoustic speed, we obtain
D
Dt
_
1
a
S
2
Dp
Dt
_
2
p = v : v
1
2
p
D
Dt
_
1
S
_
p
DS
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 rst 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 uctuation is in
phase with the pressure uctuation.
The term v : v is present even in nonreacting ow. It can be further inter-
preted (by using tensor notation with velocity q
i
) as
q
j
x
i
q
i
x
j
=
2
x
i
x
j
q
i
q
j
x
i
_
q
i
q
j
x
j
_
q
j
x
i
_
q
i
x
j
_
(11.15)
The rst term in the RHS of this equation is the classical Lighthills Reynolds
stress sound-generation term.
31
All other terms on the RHS are nonzero only for
reacting and compressible ow.
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 eld 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
Dt
_
_
=
_
D
Dt
_
0
_
+
_
D
Dt
_
p
_
_
+
_
D
Dt
_
_
1
0
__
(11.16)
The implication of this decomposition is that the rst term on the RHS is the
OrrSommereld operator (in the linearized limit). The second termcan be rewrit-
ten as
_
D
Dt
_
p
_
_
v
p
_
_
(11.17)
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 287
and represents the transport of vorticity by the acoustic disturbance; it acts as a
source for the OrrSommereld 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
(/), especially in the region of shear layers, should
be analyzed to understand the physics of acoustic-vortex interactions in turbulent
ows.
Sources and sinks for acoustics and the vorticity eld also exist in the ame zone.
Temperature rise because of combustioninthe ame regionincreases viscosity; and
viscosity, in turn, leads to dissipation of small-scale turbulent structures. However,
unsteady ame motion in an acoustic eld can result in turbulence enhancement or
suppression by the baroclinic torque term( p/
2
) in the vorticity equation.
Flame motion also can contribute to the pressure eld 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 fuelair mixture. Swirl also provides an
efcient mechanism to stabilize the ame in a compact region without requiring
a physical ame holder. In general, the extent of swirl is typically dened by the
swirl number S
i
, which is the ratio of the axial ux of angular momentum to the
product of the inlet radius and axial ux of axial momentum. Other parameters
such as inlet swirl-vane geometry, Reynolds number, connement geometry, and
inlet velocity (both mean and uctuation) prolescan all affect swirl effects.
With and without connement, the nature of the ow downstream changes
signicantly as a function of the swirl number S
i
. As long S
i
is below a critical
value, typically 0.6 for dump combustors,
46
the shear layer fromthe inlet separates
from the dump plane and rolls up into vortices that eventually coalesce into large-
scale coherent vortex structures. These structures propagate downstreamat a phase
velocity, which is of the order of the mean velocity, and maintains coherence for
some distance before breakingdownintomore irregular, three-dimensional vortical
structures.
Results from classical linear stability theory
47, 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 (St
= f
and p
> 0.
For the-high swirl case, Fig. 11.5b, the ame is very compact and the vortical
structures are no longer coherent rings, as in the low-swirl case. Flowis accelerated
around the VBB; and near the dump plane, heat release is no longer coupled to the
pressure uctuation. As a result, the Rayleigh parameter and pressure uctuation
are no longer in phase and combustion is more stable. An estimate for the pressure
root-mean-square uctuation intensity shows that p
rms
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, fuelair mixing and combustion in the combustor
depend in part on howthese 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 studies
22, 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 ows are carried out by using the previously noted
nite-volume code with a EulerianLagrangian approach. In this method, the gas
phase is simulated by using the usual nite-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 prole and perturbed by the rst 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 St
0
on particle dispersion. Here, St
0
is
dened as St
0
=
d
/
f
, the ratio of a particle response time
d
and a characteristic
ow time
f
. Here,
d
= (
d
D
2
)/18 is the particle response time,
d
is the
particle density, D is the particle diameter, and is the molecular viscosity of
the gas phase. The characteristic ow time is
f
= L/U
0
where L and U
0
are the
characteristic length and velocity scale for the ow.
As previously dened, St
0
1 implies that the particle can easily respond to
the changes in ow, and therefore the particle and the ow velocity will reach
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 297
an equilibrium. However, for St
0
1 particles will not have time to respond to
the ow velocity change and particle motion will not be signicantly affected.
For a given ow time, variation in St
0
is directly related to the particle diameter.
In the following, we compare the behavior of particles with St
0
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 (xed St
0
) and no vaporization is
included here.
Figure 11.6 shows the distribution of droplets at a nondimensional time of
T = 28. At very small St
0
, droplets are dragged along with the uid and can
a) b)
c) d)
Fig. 11.6 Droplet dispersion in a mixing layer as a function of Stokes number:
a) S
t
= 0.1, b) St
0
= 1.0, c) St
0
= 4.0, and d) St
0
= 100.
S. MENON 298
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 inuence of owon 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 St
0
= 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 ow, such as the braid regions between the larger
spanwise rollers. These results agree with past observations
6163
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
(N
rms
) over the whole eld. This number can be used to characterize the overall
accumulation tendency of the droplets in the oweld. The N
rms
is dened as
61
N
rms
=
_
N
c
i =1
N
2
i
N
C
(11.18)
where N
c
is the total number of computational cells and N
i
is the number of
droplets in the ith cell. N
rms
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 (St
0
= 14) have higher N
rms
than lighter
(St
0
1) and heavier droplets (St
0
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 ow. However, droplets with a Stokes number of the order
of unity (St
0
= 14) are preferentially dispersed by the ow, and hence their
distribution becomes nonhomogeneous even though their initial distribution is
homogeneous. The agreement with the spectral DNS data
61
is very good over the
whole range of St
0
.
Another way to quantify the dispersion of droplets with different St
0
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
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 299
a)
b)
Fig. 11.7 Droplet dispersion in mixing layer for St
0
= 4: a) T = 12 and b) T = 20.
S. MENON 300
a) b)
Fig. 11.8 Droplet distribution and spanwise vorticity in mixing layer for St
0
= 4:
a) T = 12 and b) T = 20.
dened as
61
D
y
(t ) =
_
_
1
N
p
N
p
i =1
[Y
i
(t ) Y
m
(t )]
2
_
(11.19)
where N
p
is the total number of droplets, Y
m
(t ) is the mean value of the droplet
displacement in the vertical direction at time t , and Y
i
(t ) is the displacement of an
i th droplet in the vertical direction at time t .
0.01 0.1 1 10 100 1000
Stokes number
1.5
2
R
M
S
o
f
p
a
r
t
i
c
l
e
n
u
m
b
e
r
s
p
e
r
c
e
l
l
T = 24 (Present DNS )
T = 28
T = 36
T = 24 (Spectral-DNS)
T = 28
T = 36
Fig. 11.9 Droplet concentration uctuation intensity per cell as a function of
Stokes number.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 301
0 10 20 30
Nondimensional time
0
1
2
3
4
D
y
(
t
)
St
O
= 0.1
St
O
= 1
St
O
= 4
St
O
= 10
St
O
= 100
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 St
0
. At very small times, droplets with St
0
= 0.1 showmore dispersion.
However, at later times, droplets with St
0
of order unity show more dispersion.
Droplets with very large Stokes numbers have much lower dispersion compared
with lighter droplets.
Vortexdroplet 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 conguration.
S. MENON 302
(GE-DACRS) combustor that was used in a recent study to understand droplet dis-
persion in swirling ow. 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 m
(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
ow. 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 (S
i
= 0.75) case and a high swirl
(S
i
= 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 ow 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 elsewhere
22
has shown that the
magnitude of the strain eld controls this breakdown.
With an increase in swirl, droplet dispersion increases signicantly. As observed
in temporal mixing layers, droplets tend to collect in the low-vorticity, high-strain
regions; and their dispersion is signicantly enhanced once the large-scale ring
structures have lost their coherence.
C. DropletVortexFlame Interactions
In spray combustion systems, additional complexity can be created by the in-
teraction between droplets and ame. Droplet vaporization is accompanied by
fuelair mixing and occurs in a swirling ow. 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 ame structure
can be quite complex in these regions. Both isolated ames 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 inthe combustor canvarysignicantlydependingonthe abilityof
the inlet swirl to mix and then redistribute the mixed uid. Very little experimental
data on operational gas turbines describe howthe ame structure actually correlates
with the vortical structures and with the acoustic uctuation as the operational point
is systematically changed.
However, some insight into the ame structure has been possible from sim-
ulation data. It appears that both spray and premixed combustion systems have
physical characteristics that are remarkably similar.
22, 65
These results also agree
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 303
a)
b)
Fig. 11.12 Spray dispersion and vortex structure in the GE-DACRS combustor:
a) low swirl, S
i
= 0.75 and b) high swirl, S
i
= 1.5.
S. MENON 304
Y
X
Z
Fig. 11.13 Schematic of the GE-1 combustor two-cup sector rig.
with classical DNS studies in simpler ows 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 amplied by the interaction between the strain rate and
the vorticity elds, thus increasing mixing and reaction in these regions.
As a nal 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 identied 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 signicant three-dimensional interaction between the
two cups occurs. Fuel (kerosene) is injected fromthe 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 ow 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
1015 droplets per group) are present when the simulation reaches the stationary
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 305
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 10
6
. 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, 6670
In
this approach, the scalar eldis simulatedwithineachLESbyusingthe LEMmodel
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 CO
2
mass fraction contours in the center
xy 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 quicklyvaporize. The combustionregions
fromthe two cups merge relatively quickly, since the fuel is injected frommultiple
locations; and fuelair mixing therefore occurs more uniformly.
Closer observation shows that the ame 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 fuelair
mixture can take place. It is observed that the ame is consistently lifted away
from the dump plane only for the lower cup. The shape of the combustor, which
pushes the owupward fromthe 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 CO
2
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 elds shows signicant three-dimensional variation in the
ame structure and in the burned regions.
Figures 11.15a and11.15bshowthe meanaxial velocitycontours andthe 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 VBBis
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 gure shows that the spray ame in this
combustor is anchored at multiple locations by both the BRB and the VBB in this
conguration.
Figure 11.15b shows the streamline pattern and temperature isocontours (1300
K) in the center xy plane. The streamlines fromboth the cups quickly interact, and
S. MENON 306
a)
b)
Fig. 11.14 Spray and reaction rate contours in the center xy plane: a) spray and
CO
2
mass fraction and b) reaction rate.
b)
a)
Fig. 11.15 Axial velocity contours and streamline pattern in the center xy plane:
a) axial velocity contours and b) streamlines and temperature contours.
S. MENON 308
Fig. 11.16 Instantaneous zcomponent of the baroclinic torque in the center xy
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 outow from this sector
is actually more uniform than in a single injector case.
To understand how vortexame interactions occur in this device and how this
process affects the acoustic uctuations, additional postprocessing of the results is
required. Analysis of the time-evolving ow 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 ow. As
noted, positive values indicate enhancement of vorticity, whereas negative values
indicate suppression. Overall, the pattern closely follows the ame 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 eld (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 signicant three-dimensional
structure anda periodic wavelike structure propagatingfromthe inlet. The unsteady
b)
a)
Fig. 11.17 Mean pressure and unsteady dilatation elds in the center xy plane:
a) mean pressure and b) unsteady dilatation.
S. MENON 310
pressure troughs (not shown) are closely related to the unsteady dilatation crests.
Time-series analysis indicates that these elds 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 crossowcaused by the geometrical convergence introduces
transverse disturbances that interact with the axially moving waves, so that further
downstream, both unsteady pressure and dilatation elds exhibit disturbances that
are more axially oriented.
Analysis of these elds shows that for the test conditions employed here, the
unsteady dilatation and the unsteady pressure elds are only weakly correlated with
the vortexame structures in the entire combustor. As a result, no enhancement
of the acoustic uctuations 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, fuelair
mixing (especially in liquid-fueled systems), ame stabilization, and combustor
geometry (multi-injector) are also important parameters in gas-turbine systems.
Understanding the sensitivity of the combustors 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 Ofce, General Electric Power Systems, and General Electric Aircraft
Engine Company. The simulations reported here were carried out by C. Stone, V.
Sankaran, and N. Patel.
References
1
Mongia, H. C., Held, T. J., Hsiao, G. C., and Pandalai, R. P., Challenges and Progress
in Controlling Dynamics in Gas Turbine Combustors, Proceedings of the Combustion
Institute, Vol. 19, The Combustion Inst., Pittsburgh, PA, 2003, pp. 822829.
2
Correa, S. M., Power Generation and Aeropropulsion Gas Turbines: FromCombustion
Science to Combustion Technology, Proceedings of the Combustion Institute, Vol. 27, The
Combustion Inst., Pittsburgh, PA, 1998, pp. 17931807.
3
Chu, B. T., and Kovasznay, L. S. G., Non-Linear Interactions in a Viscous Heat-
Conducting Compressible Gas, Journal of Fluid Mechanics, Vol. 3, 1958, pp. 494
514.
4
Culick, F. E. C., and Yang, V., Overview of Combustion Instabilities in Liquid-
Propellant Rocket Engines, LiquidRocket Engine CombustionInstability, editedbyV. Yang
and W. E. Anderson, Progress in Astronautics and Aeronautics, AIAA, Washington, DC
1995, pp. 337.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 311
5
Vuillot, F., Vortex-Shedding Phenomena in Solid Rocket Motors, Journal of Propul-
sion and Power, Vol. 11, No. 4, 1995, pp. 626639.
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. 12, 2002, pp. 110146.
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. 117132.
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. 5372.
9
Yu, K. H., Trouv e, A., and Daily, J. W., Low-Frequency Pressure Oscillations in a
Model Ramjet Combustor, Journal of Fluid Mechanics, Vol. 232, No. 11, 1991, pp. 4772.
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. 265292.
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. 535543.
12
Takahashi, F., Schmoll, W. J., Trump, D. D., and Goss, L. P., Vortex-Flame Interactions
and Extinction in Turbulent Jet Diffusion Flames, Proceedings of the Combustion Institute,
Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 145152.
13
Santoro, V. S., Kyritsis, D. C., and Gomez, A., An Experimental Study of Vortex-
Flame Interactions in CounterowSpray Diffusion Flames, Proceedings of the Combustion
Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 10231030.
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 FlamesADamaging Combination, Aeronautical
Journal, Vol. 104, No. 1033, 2000, pp. 105116.
16
Paschereit, O. C., Gutmark, E., and Weisenstein, W., Excitation of Thermoacoustic
Instabilities by Interaction of Acoustics and Unstable Swirling Flow, AIAAJournal, Vol. 38,
No. 6, 2000, pp. 10251034.
17
Lieuwen, T., and Zinn, B. T., The Role of Equivalence Ratio Oscillations in Driv-
ing Combustion Instabilities in Low NO Gas Turbines, Proceedings of the Combustion
Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 18091816.
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. 735750.
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. 8188.
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. 2562.
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. 119150.
22
Sankaran, V., and Menon, S., LES of Spray Combustion in Swirling Flows, Journal
of Turbulence, Vol. 3, No. 11, 2002.
S. MENON 312
23
Stone, C., and Menon, S., Open Loop Control of Combustion Instabilities in a Model
Gas Turbine Combustor, Journal of Turbulence, Vol. 4, No. 1, 2003.
24
Huang, Y., Sung, H.-G., Hsieh, S.-Y., and Yang, V., Large Eddy Simulation of Com-
bustion Dynamics of Lean Premixed Swirl-Stabilized Combustors, Journal of Propulsion
and Power, Vol. 19, No. 5, 2003, pp. 782794.
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. 6794.
27
Stow, S. R., Dowling, A. P., and Hynes, T. P., Reection of Circumferential Modes in
a Choked Nozzle, Journal of Fluid Mechanics, Vol. 467, 2002, pp. 215239.
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. 203209.
30
Sankaran, V., and Menon, S., Subgrid Combustion Modeling of 3-DPremixed Flames
in the Thin-Reaction-Zone Regime, Proceedings of the Combustion Institute, Vol. 30, The
Combustion Inst., Pittsburgh, PA, 2005, pp. 575582.
31
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, R. T. Edwards,
Philadelphia, 2001.
32
Lieuwen, T., Modeling Premixed Combustion-Acoustic Wave Interactions: A Re-
view, Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 765776.
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. 157197.
34
Ducruix, S., Durox, D., and Candel, S., Theoretical and Experimental Determination
of the Transfer Function of a Laminar Premixed Flame, Proceedings of the Combustion
Institute, Vol. 28,The Combustion Inst., Pittsburgh, PA, 2000, pp. 765772.
35
Borghi, R., Turbulent Combustion Modeling, Progress in Energy and Combustion
Science, Vol. 14, No. 4, 1988, pp. 245292.
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. 408409.
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. 297332.
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. 128.
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. 722734.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 313
42
Sinibaldi, J. O., Mueller, C. J., and Driscoll, J. F., Local Flame Propagation Speeds
Along Wrinkled Unsteady Stretched Premixed Flames, Proceedings of the Combustion
Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 827832.
43
Hussain, A. K. M. F., Coherent StructuresReality and Myth, Physics of Fluids,
Vol. 26, No. 10, 1983, pp. 28162850.
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. 110123.
45
Rivero, A., Ferre, J. A., and Giralt, F., Organized Motions in a Jet in Crossow,
Journal of Fluid Mechanics, Vol. 444, No. 10, 2001, pp. 117149.
46
Lilley, D. G., Swirl Flows in Combustion: A Review, AIAA Journal, Vol. 15, No. 8,
1977, pp. 10631078.
47
Michalke, A., On the Inviscid Instability of the Hyperbolic Tangent Prole, Journal
of Fluid Mechanics, Vol. 19, No. 4, 1964, pp. 543556.
48
Ho, C.-M., and Huerre, P., Perturbed Free Shear Layers, Annual Review of Fluid
Mechanics, Vol. 16, 1984, pp. 365424.
49
Gutmark, E., and Ho, C.-M., Preferred Modes and the Spreading Rate of Jets, Physics
of Fluids, Vol. 26, No. 10, 1983, pp. 29322938.
50
Stone, C., and Menon, S., Adaptive Swirl Control of Combustion Instability in Gas
Turbine Combustors, Proceedings of the Combustion Institute, Vol. 29, The Combustion
Inst., Pittsburgh, PA, 2002, pp. 155160.
51
Jeong, J., and Hussian, F., On the Identication of a Vortex, Journal of Fluid
Mechanics, Vol. 285, 1995, pp. 6994.
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. 263276.
55
Lucca-Negro, O. and Odoherty, T., Vortex Breakdown: AReview, Progress in Energy
and Combustion Science, Vol. 27, No. 4, 2001, pp. 431481.
56
Chen, R.-H., and Driscoll, J. F., The Role of the Recirculation Vortex in Improving
Fuel-Air Mixing within Swirling Flames, Proceedings of the Combustion Institute, Vol. 22,
The Combustion Inst., Pittsburgh, PA, 1988, pp. 531540.
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. 331378.
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. 10761086.
60
Syred, N., and Beer, J. M., Combustion in Swirling Flows: A Review, Combustion
and Flame, Vol. 23, No. 2, 1974, pp. 143201.
61
Ling, W., Troutt, J. N., and Crowe, C. T., Direct Numerical Simulation of a Three-
Dimensional Temporal MixingLayer withParticle Dispersion, Journal of FluidMechanics,
Vol. 358, No. 3, 1998, pp. 6185.
S. MENON 314
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. 143178.
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. 169209.
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. 577584.
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. 16, 1996, pp. 77106.
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., ALinear Eddy Mixing Model for Large
Eddy Simulation of Turbulent Combustion, LES of Complex Engineering and Geophysical
Flows, edited by B. Galperin and S. Orszag, Cambridge Univ. Press, Cambridge, U.K., 1993.
68
Menon, S., and Calhoon, W., Subgrid Mixing and Molecular Transport Modeling
for Large-Eddy Simulations of Turbulent Reacting Flows, Proceedings of the Combustion
Institute, Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 5966.
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. 148.
Chapter 12
Physics of Premixed Combustion-Acoustic
Wave Interactions
Timothy C. Lieuwen
W
k
=consumption rate of the kth species
x =axial coordinate
Y =mass fraction
Greek
=ratio of the ame length to radius, = L
f
/R
=normalized amplitude of velocity disturbance, = u/u
o
f
=disturbance amplitude for ashback dened in Eq. (12.33)
=equivalence ratio oscillation
=specic heats ratio
=stretch rate
=nondimensional disturbance convection velocity dened in Eq. (12.24)
=wavelength
=mean temperature ratio across the ame
=normalized mass burning rate response to acoustic pressure
perturbations
=momentum thickness
E
=dimensionless overall activation energy
=incident angle of acoustic waves
=density
=ame-brush thickness
=retarded time
M
=dimensionless timescale dened in Eq. (12.34)
=angular frequency
=volumetric reactant consumption rate
=axial ame position
Superscripts
( )
/
=uctuating quantities
( ) =mean quantities
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 317
Subscripts
1 =upstream side of the ame
2 =downstream side of the ame
a =acoustic disturbances
b =ame base, burned gas
c =conical ames
F =ame, fuel
in =inlet value
v =vortical disturbances
s =entropy disturbances
o =mean quantities
ox =oxidizer
w =wedge ames
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 uid mechanic disturbance with frequency f and phase
speed u
c
.
Several key questions this chapter addresses are as follows: 1) How does a
ame respond to an acoustic or vortical perturbation; in particular, what is the
subsequent heat-release uctuation? 2) How does this response scale with the
ame dimension L
F
, frequency f , disturbance-phase speed u
c
, method of ame
stabilization, or amplitude of perturbation ? 3) What are the differences between
acoustic interactions with laminar and turbulent ames? 4) What are the effects of
acoustic disturbances on inherent ame instabilities? 5) What are the relative roles
of chemical kinetic and large-scale adjustments of ame 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 ameacoustic 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
ames.
3
This chapter is organized in the following manner. The background section
briey describes the different regimes of premixed combustion (Sec. II.A), the
characteristics of the ow and thermodynamic perturbations that disturb the ame
(Sec. II.B), and the mechanism by which these disturbances are generated at the
ame (Sec. II.C). Sections III.A and III.B then focus on the effects of ow distur-
bances on premixed ames within the amelet and well-stirred reactor regimes,
respectively.
T. C. LIEUWEN 318
Acoustic disturbances
Fluid mechanic
disturbances
L
F
Fig. 12.1 Interaction of ow disturbances with a turbulent, premixed ame.
II. Background
A. Combustion Regimes
Acoustic waveame interactions involve unsteady kinetic, uid 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 temporalspatial scales. The
different regimes of interaction between acoustic waves, the combustion process,
and broadband turbulent uctuations can be readily visualized with the combustion
diagram in Fig. 12.2.
4
The regions denoted by wrinkled and corrugated amelets
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
/
/S
L
, where u
/
is the uctuating velocity and S
L
is the
laminar ame speed. Acoustic or vortical ow disturbances push these reaction
sheets around, causing additional wrinkling of the ame over well-dened 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 uctua-
tions in the wave.
In the distributed reaction zone, the smallest turbulent length scales are of the
same order as the ame thickness and thus alter the laminar ame structure. The
well-stirred reactor regime corresponds to the limit in which mixing occurs much
more rapidly than chemical kinetics, and reaction occurs homogeneously over a
distributed volume. Some debate exists about the characteristics of the combustion
process in the regions noted by well-stirred reactor and distributed reaction zone.
5
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 319
0.1
1
10
100
1 10 100 1000
Integral length scale/flame thickness ( /d)
R
M
S
v
e
l
o
c
i
t
y
/
f
l
a
m
e
s
p
e
e
d
(
u
'
/
S
L
)
Re =10
2
Corrugated
Flamelets
Distributed
Reaction Zone
Wrinkled
Flamelets
Well Stirred Reactor
Ka=1
Da=1
Re =10
4
Re =1
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
ame interactions. Note rst the following length scales: the thickness of a laminar
methaneair ame 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 ame and acoustic length
scales, the ame front essentially appears as a discontinuity to the acoustic wave.
As such, the uid dynamics of the ows up- and downstream of the ame can
often be treated separately from the dynamics of the ame structure. The situation
is quite different with respect to the relevant timescales. Forming a ame-response
timescale
M
fromthe ratio of the laminar ame thickness and ame speed leads to
values of between
M
0.002 and 0.07 s for methaneair ames. These values are
of magnitudes similar to acoustic perturbations with frequencies between 20 and
500 Hz. Thus, the interior ame structure and, consequently, quantities such as the
ame 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 ow and thermodynamic os-
cillations (e.g., p
/
, T
/
,
/
, u
/
, etc.). This is important because, as will be shown
subsequently, the response of a ame to an acoustic or vorticity velocity distur-
bance of a given magnitude is quite different.
It is useful to decompose an arbitrary disturbance eld into three canonical
types of disturbances
79
: vortical, entropy, and acoustic. In other words, each uc-
tuating quantity can be decomposed as: p
/
= p
/
a
p
/
v
p
/
s
,
/
=
/
a
/
v
/
s
,
T. C. LIEUWEN 320
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 uniformow, vorticity and entropy disturbances are convected
at the bulk ow velocity u
o
. Consequently, in low-Mach-number ows, these dis-
turbances have substantially different length scales. Acoustic properties vary over
an acoustic length scale, given by
a
= c/f , whereas entropy and vorticity modes
vary over a convective length scale, given by
c
= u
o
/f . Thus, the entropy and
vortical mode wavelength is shorter than the acoustic wavelength by a factor equal
to the mean-ow-Mach number
c
/
a
= u
o
/c = M. This can have important im-
plications on acousticame interactions. For example, a ame whose length L
f
is short relative to an acoustic wavelength, that is, L
f
_
a
, 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 ame front that results in heat-release disturbances generated at differ-
ent points of the ame that are out of phase with each other. Studies often nd
that a Strouhal number, dened as St = L
f
/u
o
, is a key parameter that affects
the ame response to perturbations. Note that St is proportional to the ratio of the
ame length and convective wavelength, St = 2L
f
/
c
. A ame 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 ow and
diffuse from regions of high to low concentration. In contrast, acoustic distur-
bances, being true waves, reect off boundaries, are refracted at property changes,
and diffract around obstacles. In general, the reection of acoustic waves from
multidimensional ame fronts results in a complex, multidimensional acoustic
eld in the vicinity of the ame. We make this point because analytical studies
often assume that the acoustic eld is one-dimensional. This is not the case in
reality, although it may be a reasonable approximation under certain conditions
or in specic regions of the ame. A planar-incident wave impinging on a ame
front not only generates planar (i.e., one-dimensional) reected and transmitted
waves, but also multidimensional disturbances that are often evanescent, or spa-
tially decaying. To illustrate, Fig. 12.3 reproduces a gure from Ref. 10 showing
computed velocity vectors in the vicinity of an axisymmetric conical ame per-
turbed by a plane sound wave at a frequency below the duct-cutoff mode. Note
that the acoustic eld has strong two-dimensional characteristics near the base of
the ame but it reverts to a one-dimensional structure up- and downstream of the
ame. Similar observations have been made from experiments.
11
Third, in a homogeneous, uniformow, 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 signicance of vortical mode interactions with
ame fronts.
12, 13
Figure 12.4 shows a simulated result of such an interaction,
14
where the ame 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 ow
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 321
Fig. 12.3 Instantaneous pressure contours (solidlines) andvelocityvectors inadump-
stabilized combustor geometry for which the ame was excited from upstream. Aver-
age ame location given by heavy inclined line. Adapted from Lee and Lieuwen.
10
instabilities. The phase velocity and growth rate of the owinstabilities is strongly
affected by the amplitude of forcing and the relationship between the acoustic-
forcingfrequencyandthe intrinsic owinstability. Acoustic excitationoftencauses
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 ow 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 ame disturbed by vortical structure.
14
Image courtesy of
S. Menon and C. Stone.
T. C. LIEUWEN 322
Fig. 12.5 Dependence of shear-wave convection velocity and growth rate in a jet ow
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 specic characteristics of the burner-
exit shear layer, such as coow velocity, and specically 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 ow 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 Michalkes
16
theoretical curves of the dependence of
the phase speed u
c
of shear-layer instability waves in a jet owon Strouhal number,
S
= f /u
o
, for several values of the momentum thickness , jet radius R, and
ratio R/. The gure shows that, for all R/ 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/ = 100, the phase velocity actually exceeds the maximum axial-ow velocity
in a certain S
and
velocity-oscillation amplitudes is linear at disturbance amplitudes u
/
/u
o
< 0.20.
T. C. LIEUWEN 324
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.00 0.05 0.10 0.15 0.20 0.25 0.30
u'/u
o
C
H
*
'
/
C
H
*
o
Fig. 12.6 Measured dependence of CH
1
S
2
1
= p
2
2
S
2
2
(12.4)
Tangential momentum: ( u
1
u
2
) f = 0 (12.5)
Energy:
1
S
1
_
h
1
u
1
u
1
2
_
=
2
S
2
_
h
2
u
2
u
2
2
_
(12.6)
where and h denote density and enthalpy, respectively. The dynamics of the
thermodynamic and ow variables up- and downstream of the ame 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 ames, only the spatially integrated heat release is impor-
tant. However, for high-frequency oscillations, where the ame is not acoustically
compact, its spatial distribution is also important.) The global heat-release rate of
the ame is given by
Q(t ) =
_
S
1
S
1
h
R
dA
FL
(12.7)
where the integral is performed over the ame surface A
FL
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 ame: uctu-
ations in density, ame speed, heat of reaction, or ame area. As noted by Clanet
et al.,
35
they can be classied based on either their modication of the local internal
structure of the ame (such as the local burning rate) or its global geometry (such
as its area).
Fluctuations in the mass ow rate of reactive mixtures into the ame, cor-
responding to
1
S
1
in Eq. (12.7), is the most basic mechanism of heat-release
oscillation
.
These density uctuations could be caused by both acoustic and en-
tropy uctuations. The ames burning rate S
1
is sensitive to the perturbations
in pressure, temperature, strain rate, or mixture composition that accompany the
acoustic wave. These pressure and temperature uctuations are usually generated
by acoustic perturbations, whereas the strain-rate uctuations are associated with
acoustic or vortical velocity uctuations.
Flame-area uctuations are associated with disturbances in the ames position
and orientation that, in turn, are generated by uctuations in either the local burning
rate or owvelocity. To illustrate the disturbance of a ame by an acoustic velocity
disturbance, Fig. 12.7 shows a photograph fromDucruix et al.
36
of a simple Bunsen
ame disturbed by acoustic-ow oscillations generated by a loudspeaker placed
upstream of the ame. The gure clearly shows the large distortion of the ame
front that is evidenced by the pronounced cusp in the center of the ame. This
ame disturbance is convected downstream by the mean ow, so that it varies
spatially over a convective wavelength.
37
T. C. LIEUWEN 326
Fig. 12.7 Photograph of ame disturbances generated by acoustic velocity
oscillations.
36
Photograph courtesy of S. Ducruix, D. Durox, and S. Candel, Centre
National de La Recherche Scientique and Ecole Centrale de Paris.
Finally, uctuations in heating value h
R
are driven by variations in reactive
mixture composition.
2. Effects of Flame on the Disturbance Field
Acousticame interactions are highly coupled because the ame has a strong
impact on the values of the acoustic oscillations that are disturbing it. Although
this point was briey noted in Sec. II.B, this section considers these effects in more
detail.
Consider a nominally at, vertically oriented ame front in a low-Mach-number
ow whose instantaneous position is described by the equation x = (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 ame thickness. This problem was rst
Incident Acoustic
Wave
Reflected Wave
Transmitted
Wave
Convected vortical
and entropy disturbances
Instantaneous
flame position
x= == =( ( ( (y,t)
Cold Reactants
Hot Products
Fig. 12.8 Illustration of planar ame disturbed by incident acoustic wave.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 327
reported by Manson,
38
who calculated the reection and transmission coefcients
of a planar ame modeled as a simple temperature discontinuity and generalized
by Chu,
38
who also considered sound amplication by perturbations in S
1
, h
r
,
, and entropy of the incoming mixture. This problem is treated by decomposing
each term in Eqs. (12.112.6) into the sum of a mean and uctuating term and
retaining only linear terms in uctuations; for example,
1
S
1
= (
1
1
/
)(
S
1
S
1
/
)
1
S
1
1
S
1
/
1
/
S
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 ame:
u
2
/
c
1
u
1
/
c
1
= (1)M
s
_
S
1
/
S
1
p
1
/
p
1
_
(12.9)
p
2
/
= p
1
/
(12.10)
where M
s
, , and refer to the laminar-ame-speed Mach number, M
s
=
S
1
/ c
1
,
specic heat ratio, and mean temperature ratio across the ame. This equation
neglects variations in and terms of O(M
2
s
). To this order, the unsteady pressure
is continuous across the ame. However, there is a jump in unsteady velocity
across the ame; that is, the ame 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 ames unsteady rate of heat release and result in the amplication
of acoustic waves. This jump is directly proportional to the temperature jump
(or, more fundamentally, the gas-expansion ratio) across the ame and the
ame-speed Mach number M
s
which typically has quite low values (0.001 for
a stoichiometric methaneair ame). Assuming a typical acoustic scaling, that
is, p
/
cu
/
, it can be seen that the second uctuating term on the right side
results in a velocity increment across the ame that is on the order of M
s
and,
thus, quite small. The relative magnitudes of the S
1
/
term on the right side of
Eq. (12.9) and the uctuating velocity-perturbation quantities on the left depend
on the specic processes causing the ame-speed perturbation. More detailed
analyses in Sec. III.A.4 suggest that ame-speed perturbations caused by pressure
and/or temperature uctuations are of similar magnitude S
1
/
/
S
1
O( p
/
/ p). Thus,
acoustic-wave amplication induced by the pressure or temperature sensitivity of
the ame 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 eld can be accurately calculated by ignoring them. As such, the leading-
order calculation of the acoustic eld is equivalent to replacing the ame front
with a passive-temperature discontinuity, as assumed by Mason.
38
Ignoring these
terms does not allow one to calculate the slight amplication or damping of sound
waves at the ame, but it does allow for an accurate calculation of the acoustic
eld that is disturbing the ame. We will proceed in a sequential fashion, rst
considering the leading-order problem, then considering the higher-order effects
needed to calculate the acoustic amplication and damping by the ame.
Consider rst the problem of an acoustic wave of pressure amplitude P
I
im-
pinging normally (i.e., = 0) on the ame. The acoustic pressure and velocity
T. C. LIEUWEN 328
are given by the following plane-wave equations:
p
1
/
(x, t ) = P
I
(e
i k
1
x
Re
i k
1
x
) p
2
/
(x, t ) = P
I
Te
i k
1
x
(12.11)
u
1
/
(x, t ) =
P
I
1
c
1
(e
i k
1
x
Re
i k
1
x
) u
2
/
(x, t ) =
P
I
T
1
c
1
e
i k
1
x
(12.12)
where R and T are the reection and transmission coefcients, denoting the am-
plitude of the acoustic waves reected from and transmitted through the ame.
Matching the pressure and axial velocity at the ame by using the leading-order
approximation of Eqs. (12.9) and (12.10), p
/
2
= p
/
1
and u
/
2
= u
/
1
, leads to the fol-
lowing solution for R and T:
R =
(
2
c
2
/
1
c
1
) 1
(
2
c
2
/
1
c
1
) 1
T =
2(
2
c
2
/
1
c
1
)
(
2
c
2
/
1
c
1
) 1
(12.13)
Neglecting variations in molecular weight and across the ame, note that
2
c
2
/
1
c
1
_
T
1
/
T
2
=
T
1
= 4, so
that R = 1/3 (the negative sign means that the reected wave is out of phase
with the incident wave). This implies that the acoustic pressure and velocity at
the ame are lower and higher, respectively, than that of the incident wave, that
is, p
/
ame
/P
I
= 2/3 and u
/
ame
/(P
I
/
1
c
1
) = 4/3. If the acoustic wave is incident
on the same ame from downstream, the pressure and velocity are higher and
lower, respectively, than those of the incident wave, that is, p
/
ame
/P
I
= 4/3 and
u
/
ame
/(P
I
/
2
c
2
) = 2/3. This implies that the presence of the ame alters the value
of the quantities that are disturbing it and causing its heat-release rate to oscillate.
Although not calculated here, the ame can exhibit an even more substantial impact
onthe local acoustic eldif the acoustic wave is obliquelyincident; 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 eld
by unsteady heat-release processes. In cases where the acoustic wave is obliquely
incident on the ame, vorticity is also produced via the baroclinic mechanism
because of the misaligned uctuating pressure and mean density gradients. The
energyinthese vorticityuctuations is derivedfromthe incident acoustic elditself
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 ux out of the ame I
a
is controlled by competing acoustic-energy production and dissipation processes.
Equation (12.14) is reproduced from Ref. 22 and illustrates these production (rst
two terms) and dissipation (third term) terms:
I
a
p
1
c
1
= (1)M
s
_
(2 )
_
p
/
p
_
2
_
p
/
p
__
S
/
1
S
1
_
_
_
p
/
p
__
u
/
2u
c
2
_
(12.14)
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 329
L
f
R
( , ) r t
( , ) r t
R
r
Fig. 12.9 Illustration of conical (left) and wedge-shaped (right) ame geometries.
Depending on the temperature ratio across the ame, the magnitude and phase of
the ame-burning velocity response, and the angle of incidence between the wave
and the ame, the acoustic disturbance can be damped or amplied. Plots showing
typical results can be found in Ref. 22.
Equation (12.14) only describes the acoustic-eld energy balance. Energy
is also added to the vortical and entropy elds. Although not shown here, Ref.
22 includes expressions for these convected waves. Although the vortical wave
couples with the acoustic eld at O(M
s
) (resulting in acoustic damping and
vorticity amplication), at this order entropy waves are forced disturbances and
do not have an impact on the acoustic or vortical elds. Full coupling between all
three disturbances occurs at O(M
2
s
).
Note that all energy amplication and damping processes in this equation are
relatively small, being of O(M
s
). Flame-area uctuations, 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 ame-surface area, which, as noted
in Eq. (12.7), constitutes a mechanism for heat-release uctuations. Flame-area
disturbances are generated by variations in ame-front orientation that, in turn,
are generated by disturbances either in the approach ow velocity or in the ame
speed. In this section, we restrict attention to area disturbances arising from ow-
velocity perturbations. Flame-speed perturbations, which introduce heat-release
disturbances through perturbations in both ame 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 ame stabilized on a tube, such as a Bunsen ame. On the right is an
axisymmetric wedge ame, stabilized on a bluff body. The ames have axial and
radial dimensions given by the ame length L
f
and radius R.
The instantaneous ame-sheet location at the radial location r is given by (r, t ),
assumed to be a single-valued function of r; thus, the ame-position surface f ( x, t )
[see Eqs. (12.1)(12.3)] is dened as f ( x, t ) = y (r, t ). This assumption nec-
essarily limits the range of amplitudes that can be treated with this formulation.
With Eq. (12.1), the ame dynamics are described by
t
= u v
r
S
1
_
_
r
_
2
1 (12.15)
T. C. LIEUWEN 330
where u and v are axial and radial velocity, respectively. Note that a differential
element of ame surface is related to the ame position through the relation
dA
FL
=
_
1
_
r
_
2
dr (12.16)
We assume that the mean velocity is uniform and purely axial (i.e., v = 0), that
v
/
= 0, and that the mean ame 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 ame areas linear dynamics. Nonlinear dynamics are
considered later. As shown in the equation below, the linear solution to the equation
for
/
/r, or the ame-surface area, can be decomposed into two canonical
components: the homogeneous solution (second term on the right side) containing
the inuence of boundary conditions and the particular solution caused by spatial
nonuniformities in owforcing (or ame speed). To simplify the equation, a ame
coordinate along the nominal ame surface position s is introduced:
(s, t )
s
=
1
u
o
s
_
0
s
_
u
/
_
x
/
, t
s x
/
u
o
__
dx
/
1
u
o
_
u
/
_
s = 0, t
s
u
o
_
u
/
base
_
(12.17)
where u
/
base
denotes the velocityof the endof the ame sheet at the attachment point.
Aspatially uniformvelocity disturbance u
/
/s = 0 excites only the homogeneous
solution. This disturbance can be understood by rst assuming that the ame edge
moves exactly in step with the particle velocity u
/
(s = 0) = u
/
base
. In this case,
the entire ame moves up and down in a bulk motion without change in ame
orientation or area. However, if a ame-anchoring boundary condition is imposed,
for example, u
/
base
= 0, such that the ame remains xed at a point, the ow
disturbance excites a ame-front disturbance that originates at the boundary and
propagates along the ame front.
If the disturbance oweld is spatially nonuniform, u
/
/s ,= 0, the particular
solution is excited. This results in waves originating at the spatial location(s) of
ow nonuniformity that also propagate along the ame at roughly the mean ow
velocity.
As will be shown subsequently, the ame area acts as a low-pass lter to ow
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 ame area to a
spatially uniform velocity disturbance (where only the homogeneous solution is
excited), (A
/
/A
o
)/(u
/
/u
o
) has a value of unity at zero frequency and decays with
frequency. In contrast, when the ame is perturbed by a spatially nonuniform dis-
turbance (so that both the homogeneous and particular solutions are excited), the
ame area consists of a superposition of the two solutions. As such, although
each solution decreases with frequency, their sum has oscillatory behavior in
cases where they constructively interfere and even causes the transfer function
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 331
(A
/
/A
o
)/(u
/
/u
o
) to exceed unity. This result was rst predicted and then experi-
mentally conrmed 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
/
/A
o
)/(u
/
/u
o
) identically equals
zero.
We nowconsider linear solutions to the nondimensionalized formof Eq. (12.15),
that is,
_
1
2
_
r
_
2
1
2
= u(, t ) (12.18)
for time harmonic velocity perturbations, given as
u(, t ) = u
o
u
/
cos [k
c
o
t ] (12.19)
where k
c
=
o
/u
c
and u
c
is the phase velocity of the disturbance The variables t ,
r, u, and 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 = (
o
L
F
)/u
o
, velocity perturbation, = u
/
/u
o
, and ratio of
the ame length to radius, = L
F
/R.
Assume that the ame remains anchored at the base, u
/
base
= 0:
(r = 1, t ) = 0 (12.20)
The effects of a nonstationary ame-anchoring point is addressed in Sec. III.A.7.
Consider the instantaneous ame-surface area, which, for a conical ame, is
given by
A
c
(t )
A
c, o
= 2
1
_
0
r
_
1
2
_
r
_
2
dr
_
1
2
(12.21)
where subscripts c and w are used to denote axisymmetric conical and wedge
ames, respectively. It is shown in Ref. 41 that the solution for the conical ame
area transfer function, G
c
= (A
/
c
/A
c, o
)/(u
/
/u
o
) is
G
c
(St
2
, ) = G
c, BC
G
c, Flow
= 2
_
exp (i St
2
) 1 i St
2
( 1) St
2
2
_
2
_
1 exp (i St
2
) i St
2
( 1) St
2
2
_
=
2
St
2
2
(1 )
_
1exp (i St
2
)
exp (i St
2
) 1
_
(12.22)
T. C. LIEUWEN 332
where
St
2
=
St (1
2
)
2
(12.23)
=
u
o
u
c
2
1
2
(12.24)
Note that the contribution of the ow nonuniformity and boundary conditions are
explicitly separated, following the discussion of Eq. (12.17).
The solution for the wedge-ame-transfer function, G
w
= (A
/
w
/A
w, o
)/(u
/
/u
o
),
is
G
w
(St
2
, ) = G
w, BC
G
w, Flow
= 2
_
1 (1 i St
2
) exp (i St
2
)
( 1) St
2
2
_
2
_
(1 i St
2
) exp (i St
2
) 1
( 1) St
2
2
_
=
_
2
( 1)St
2
2
_
[ 1i (i St
2
) exp (i )
(1 i ) exp (i )]
(12.25)
Thus, the linear ame-transfer functions for both the conical and wedge ames,
Eqs. (12.22) and (12.25), only depend on two parameters, St
2
and . The term
couples the effect of ame angle and phase speed of the disturbances. Alternatively,
the effects can be captured by dening another Strouhal number based on the
convective velocity u
c
of the ow disturbances St
c
. St
c
naturally arises in the two
transfer functions, Eqs. (12.22) and (12.25), and equals St
2
. These two Strouhal
numbers are related to the amount of time taken for a ow (St
c
) and ame-front
(St
2
) disturbance (which is ultimately created by a ow disturbance) to propagate
the ame length, normalized by the acoustic period.
Before looking at the total ame-transfer functions, it is useful to understand the
characteristics of its two contributing ow-forcing and boundary condition terms.
Their ratio is given by
G
c, Flow
G
c, BC
=
1 exp (i St
2
) i St
2
_
exp (i St
2
) 1 i St
2
_ (12.26)
G
w, Flow
G
w, BC
=
(1 i St
2
) exp (i St
2
) 1
_
1 (1 i St
2
) exp (i St
2
)
_ (12.27)
The magnitude of this ratio is identical for both wedge and conical ames (see
Fig. 12.10). The phase of this ratio is different for conical and wedge ames and
plotted in Fig. 12.11.
It is instructive to analyze the characteristics of this ratio for limiting values of the
parameters and St
2
. First, note that inthe 0limit (i.e., a spatiallyuniformdis-
turbance), the ame dynamics for both the wedge and conical ames is controlled
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 333
Fig. 12.10 Strouhal number dependence of the magnitude of the ratio of the transfer
functions causedbythe ow-forcingandboundary-conditionterms for different values
of .
Fig. 12.11 Strouhal number dependence of the phase of the ratio of the transfer
functions caused by the ow-forcing and boundary-condition terms for wedge ames
for different values of . Shaded regions indicate points where boundary-condition
and ow-forcing terms are in phase.
T. C. LIEUWEN 334
exclusively by the boundary-condition term, irrespective of Strouhal numbers:
Lim
0
_
G
c, Flow
G
c, BC
_
= Lim
0
_
G
w, Flow
G
w, BC
_
= 0 (12.28)
This result canbe anticipatedfromthe precedingdiscussionandreects the fact that
only the homogeneous solution is excited when the ow disturbance is uniform.
In the St
2
0 limit, the relative contribution of the two terms is determined by
the value of :
Lim
St
2
0
_
G
c, Flow
G
c, BC
_
= Lim
St
2
0
_
G
w, Flow
G
w, BC
_
= (12.29)
The boundary-condition and ow-forcing terms dominate when <1 and >1,
respectively. For long ames ( 1), this physically corresponds to situations
in which the disturbance-phase velocity is greater than and less than the mean
ow velocity, respectively. The two terms tend toward equal magnitudes when
= 1. These points can be clearly observed in Fig. 12.10. Note also that the ow-
disturbance and boundary-condition terms are 180 deg out of phase for low St
2
values (see Fig. 12.11).
In the St
2
1 limit, the contribution from both the boundary conditions and
ow-forcing term are equal, as shown in Fig. 12.10 and in Eq. (12.30):
Lim
St
2
_
G
c, Flow
G
c, BC
_
= 1
Lim
St
2
_
G
w, Flow
G
w, BC
_
= exp [i ( 1)St
2
] (12.30)
Equation (12.30) also shows that, in this limit, the relative magnitude contribu-
tion of these two terms is independent of (assuming that the St
2
product does
not simultaneously go to zero). Moreover, the two terms are always out of phase for
conical ames irrespective of the Strouhal number and ; typical relative phases
range between 140 and 220 deg. In contrast, for wedge ames the phase differ-
ence between the two contributions monotonically increases with St
2
, as shown in
Fig. 12.11 (the shaded bands in the gure indicate regions of constructive inter-
ference).
The dependence of the magnitude and phase of the conical ame-transfer func-
tion G
c
(St
2
, ) on St
2
is plotted in Figs. 12.12 and 12.13, respectively. Consider
the magnitude results rst. Note that the transfer-function gain is identical in the
cases in which = 0 or 1. Physically, this corresponds to cases in which the dis-
turbance velocity is uniform ( = 0) or its phase speed matches the ame-front
disturbance velocity ( = 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 St
2
, although some ripple occurs
at higher St
2
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 St
2
and initially increases monotonically with St
2
.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 335
Fig. 12.12 Axisymmetric conical linear transfer function G
c
(St
2
, ) amplitude de-
pendence on the reduced Strouhal number (St
2
) for different values of .
For wedge ames, the transfer-function gain G
w
(St
2
, ) is plotted in Fig. 12.14.
Note that all gain values tend toward values of unity at low St
2
. However, only
in the uniform velocity case, = 0, does the gain monotonically decrease with
increases in St
2
. 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
(St
2
, ) phase depen-
dence on the reduced Strouhal number (St
2
) for different values of .
T. C. LIEUWEN 336
Fig. 12.14 Axisymmetric wedge linear transfer function G
w
(St
2
, ) amplitude depen-
dence on the reduced Strouhal number (St
2
) for different values of .
Another striking feature is the resonance-like behavior at = 1, where the
wedge-ame response does not decrease with St
2
but tends toward a constant
value of two. This case corresponds to exact coincidence of ame-front and ow-
disturbance velocity. In reality, curvature effects on ame speed that are neglected
in this analysis, which increase with St
2
, cause the transfer function to decrease at
higher values of St
2
.
In general, the relationship between an unsteady heat-release rate and velocity
has a complex dynamic. However, for St _1 (convectively compact ame), the
A
/
(t ) relationship can be put in terms of a simple n model:
A
/
(t )/A
o
= n
u
u
/
(t ) (12.31)
where n
u
= 1/
S
1
,
conical
= [( 1)L
F
]/3u
o
, and
wedge
= [2( 1)L
F
]/3u
o
.
Eq. (12.31) indicates that the time response of the ame area to perturbations in
acoustic velocity is delayed by a retarded time . This retarded time equals the
time taken for the mean ow to convect some fractional distance of the ame
length, which is equivalent to replacing the distributed ame by a concentrated
source at this location; for example, for a conical ame this effective position of
concentrated heat release is L
eff
( 1)L
F
/3.
We next turn to the response of the ame 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(St
2
, ). For the general nonlinear case, however, the gain G also
depends on and ; that is, G = G(St
2
, , , ). Before considering specic
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 ame is perturbed into a corrugated front but
then allowed to relax back to its steady-state, planar position. Flame propagation
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 337
Fig. 12.15 Sketch of a ame that is initially wrinkled (top), showing the destruction
of ame 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 bottomsketch, opposed ame
branches merge to form a sharp cusp and propagate forward to destroy ame area.
The cusp-formation time
cusp
of a front with a corrugation of length scale
f
is
proportional to
f
/S
1
.
The rate of ame-area destruction depends nonlinearly on the amplitude of the
ame-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 ame cor-
rugations.
Consider the effects of these nonlinearities on the ame disturbances generated
at the boundaries and regions of ownonuniformity, as discussed after Eq. (12.17).
If only the homogeneous solution of Eq. (12.17) is excited, as in a spatially uniform
velocity-perturbation eld, nonlinear effects always cause the nonlinear transfer
function relating ame area (at the disturbance frequency) and velocity perturba-
tions, (A
/
/A
o
)/(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 eld 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
/
/A
o
)/(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 ame area for the boundary-condition termthat prop-
agates the entire ame length, as opposed to the ow nonuniformity terms that are
excited at each point along the ame. As will be shown subsequently, the result is
that the nonlinear transfer function can actually exceed its linear value.
T. C. LIEUWEN 338
Next, consider the dominant factors that affect ame-area nonlinearity. Note
that these nonlinearities arise from three sources. The rst factor is the nonlinear
ame dynamics through the term
_
1
2
(/r)
2
in Eq. (12.18). The second
factor is the static nonlinearity introduced through the dependence of the ame area
on ame-position gradient through a term with the same form,
_
1
2
(/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 ame-front
normal and ame-position gradient. The third nonlinearity is caused by the ow
forcing itself and the dependence of the disturbance velocity at the ame front on
the ame position u(, t ).
The fact that the rst two sources of nonlinearity are identical can be used to write
the nal expressions for the ame area, Eq. (12.21), in a revealing form. By substi-
tuting Eq. (12.18) into Eq. (12.21), note that the term
_
(1
2
(/r)
2
)/(1
2
)
which appears in both the area integrals can be written as
_
1
2
(/r)
2
1
2
= u(, t )
t
(12.32)
Thus, the explicit formof the nonlinearity disappears. Nonlinearities in ame-front
dynamics are included in the /t term, whereas those caused by the ow-forcing
nonlinearity noted previously are included in the u(, t ) and /t term. Based
on Eq. (12.32), the following observations can be made regarding the effects of
various parameters on nonlinearity in the ames response to ow perturbations.
a. Strouhal Number. At low Strouhal numbers, the unsteady term in
Eq. (12.32) is negligible. Moreover, the dependence of the velocity eld u(, t ),
is weak in the limit of low St
c
, at least for the velocity elds considered here. Thus,
the ame areas velocity response remains linear for low Strouhal numbers. This
point shows that the ames 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 ame base to convect the length of the
ame L
F
/u
o
to the ame-front cusp formation time
cusp
F
/S
1
. If this ratio is
small, the ame wrinkle will not have enough time to forma cusp, which is closely
associated with nonlinearity.
b. Flow Uniformity. Nonlinearities in the u(, t ) term are directly caused by
nonuniformity in ow disturbances. Thus, the contribution of this term to ame-
area nonlinearities is suppressed in the 0 limit.
c. Boundary Conditions. If the ame remains anchored at the attachment
point, as it is in this section, then /t is identically zero at this point for all
time. As such, the ame-area perturbations in the vicinity of the attachment point
[where 0 u u(0, t )] exhibit a linear dependence on velocity amplitude.
Nonlinearities only arise at points of the ame that are spatially removed from the
attachment point. As such, the axisymmetric conical ame exhibits a more linear
velocity response than the axisymmetric wedge ame for comparable values of ,
because most and very little, respectively, of the ame area is concentrated near
the attachment point.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 339
d. Flame Aspect Ratio. When 1, that is, when the ame is very long,
the ame dynamics are approximately described by the equation
t
r
= u(, t )
In this case, the ame dynamics are linear, although the ow-forcing termdoes not
need to be linear. Thus, is an important nonlinearity parameter for this problem;
that is, the ames area response can be anticipated to exhibit a linear dependence
on the perturbation velocity for much larger values at large values.
The rest of this section presents typical results comparing the linear and nonlin-
ear ame-transfer function. These results were obtained by numerically integrating
the governing equation (12.18).
41
The nonlinear ame-transfer function was deter-
mined by computing the ame 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 owvelocity is lower than the ame speed.
In this case, the ame ashes 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 ashback, given by the disturbance amplitude =
f
,
where
f
= 1
1
_
1
2
(12.33)
Figure 12.16 plots the St
2
dependence of the nonlinear transfer-function gain for
a wedge ame. The gain-transfer functions are normalized by their linear values,
Fig. 12.16 Strouhal number dependence of the ratio of the magnitude of the ame
areavelocity transfer function to its linear value for the axisymmetric wedge ame,
= 1, = 0.
T. C. LIEUWEN 340
Fig. 12.17 Strouhal number dependence of the ratio of the magnitude of the ame-
areavelocity-transfer function to its linear value for an axisymmetric wedge and
conical ame, = 1, /
f
= 0.99.
G/G
Lin
. Results are shown for a uniform-velocity eld = 0 and a value of = 1.
As predicted previously, the response tends to its linear value in all cases at low
St
2
. Note the substantial reduction in the ame 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 St
2
.
For the present case, the gain for the wedge ames decreases by about 75% at
=
f
. 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 =
f
.
Figure 12.17 shows the St
2
dependence of the gain and phase of the nonlinear
transfer function for a wedge and conical ame at a given velocity amplitude for a
range of values. The results are shown for a velocity amplitude of /
f
= 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 ame.
In the = 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 ow-forcing nonuniformity solutions
noted previously. To illustrate, Fig. 12.18 shows the ame response for a wedge
ame when = 2 (i.e., disturbances are traveling at approximately half the mean
owspeed). Note that the gain results are not normalized by their linear value here.
The gain result indicates that, in the 6 < St
2
< 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 St
2
values, the contributions attributable to the
boundary conditions and the ow-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 ow-forcing terms to decrease.
Because the individual gain decreases by different amounts, the total gain does not
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 341
Fig. 12.18 Strouhal number dependence of the magnitude of the ame-areavelocity-
transfer function for the axisymmetric wedge ame, = 2, = 2.
go to zero at the St
2
value at which the linear gain is zero but actually shifts to a
higher St
2
value in the = 0.2
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 ames, although less dramatically. In addition,
unlike the = 0 case, the phase exhibits a stronger amplitude dependence.
Althoughthese results have focusedontheoretical predictions, theyare 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
A
LC
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
/
/A
o
vs at the two conditions, St
2
= 2.5 and
St
2
= 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 ame-area-uctuation mechanisms may or
may not exhibit hysteresis and triggering, depending on the operating condition
and frequency.
T. C. LIEUWEN 342
A
/
A
o
Fig. 12.19 Dependence of ame-area uctuation, A
/A
0
on velocity amplitude for
wedge ames when = 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 denitive conclusions
about the ames saturation amplitude based on how strongly it responds to low-
amplitude uctuations. It is possible to nd 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 ame length is doubled
at a constant frequency. For example, such a trend could be associated with a
decrease in ame 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 ame-speed perturbations,
which, as noted in Eq. (12.7), constitute a mechanismfor heat-release uctuations.
It considers the effects of uctuating pressure and temperature, then strain rate,
and nally mixture composition.
Consider rst the response of the ame speed to the unsteady pressure and
temperature variations in an acoustic wave. Several analyses,
4246
have studied
the internal structure of a at ame 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
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 343
Fig. 12.20 Dependence of ame-area uctuation A
/A
0
on velocity amplitude for
wedge ames when = 0 for two different conditions.
the acoustic wave and ame preheat and reaction zone. Following McIntosh,
47
dene the following ratios of these length and timescales:
M
=
diffusion time
acoustic period
N
acoustic wavelength
diffusion length
(12.34)
These ratios are related by the Mach number of the ame-burning velocity:
M
s
=
S
L
c
u
=
1
M
N
(12.35)
Also, dene the dimensionless overall activation energy:
E
=
E
a
R
g
T
b
(12.36)
where E
a
is the overall activation energy, R
g
is the gas constant, and T
b
is the
burned-gas temperature. Four different regimes exist whose characteristics depend
on the relative magnitudes of these parameters:
1) N 1/M
s
(i.e.,
M
_1). Acoustic wavelength is much longer than the
ame thickness and the ame responds in a quasi-steady manner to acoustic
disturbances.
2) N O(1/M
s
) (i.e.,
M
O(1)). Acoustic wavelength is much larger than the
ame thickness, but acoustic- and ame-response times are commensurate.
T. C. LIEUWEN 344
0.1
1
10
0.01 0.1 1 10
Flame response time/Acoustic Period,
|
|
-60
-50
-40
-30
-20
-10
0
p
h
a
s
e
(
)
Fig. 12.21 Normalized mass burning-rate response to acoustic pressure perturba-
tions adapted from McIntosh.
42
3) N O(1/
2
E
M
s
) (i.e.,
M
O(
2
E
)). 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.,
M
O(1/M
s
)]. Pressure gradients occur over the same
length scale as ame 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 1 value in a
stoichiometric methaneair ame. For these cases, McIntosh derives the following
expression relating the mass burning rate and acoustic pressure perturbation
42
:
_
m
/
m
_
_
_
p
/
p
_
=
2
E
( 1)
(i
M
)(s 1 1/)
E
(1)
/
(x, t ) =
/
b
exp [i (t x/ u)] =
/
b
exp (i t ) exp [i St (1 r/R)]
(12.41)
/
b
in Eq. (12.41) denotes the perturbationin equivalence ratioat the ame base. The
ame-speed perturbation is related to the perturbation in mixture stoichiometry by
S
/
1
=
_
dS
1
d
_
/
(12.42)
From Eq. (12.7), the total heat-release perturbation is given as
Q
/
Q
=
_
/
1
d A
FL
_
1
d A
FL
_
S
1
/
d A
FL
_
S
1
d A
FL
_
h
/
R
d A
FL
_
h
R
d A
FL
A
/
FL
A
FL
(12.43)
Assume that the equivalence-ratio disturbance occurs at constant density; that is,
/
1
= 0. Dene the following ame-transfer functions to perturbations in equiva-
lence ratio, F
=
Q
/
/
Q
/
b
/
= F
H
F
S
= F
H
(F
S, dir
F
A
) (12.44)
where
F
H
=
d(h
R
/
h
R
)
d(/
)
2
St
2
{1 i St exp (i St )]
F
S, dir
=
d(S
1
/
S
1
)
d(/
2
St
2
{1 i St exp (i St )] (12.45)
F
A
=
d(S
1
/
S
1
)
d(/
2
St
2
{1 (1 i St ) exp (i St )]
Q
/
S
1
)/d(/
cannot be described, in
general, by an n model, even in the St _1 limit. This relationship is due
to the possible negative-phase dependence of F
is given by
Q
/
(t )/
Q = n
H
/
b
(t
H
) n
S
d
/
b
(t )
dt
(12.46)
where
n
H
=
d(h
R
/
h
R
)
d
,
H
=
L
F
3 u
, n
s
=
1
3
L
F
u
d(S
1
/
S
1
)
d
Q
/
1 ku
/
(t )/u
(12.47)
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 howmuch the preceding results,
derived for smooth, laminar ame fronts, can be generalized to highly corrugated,
turbulent ames. Although many turbulent ame effects have not been treated to
date, several key features have been worked out in a series of experimental and
theoretical papers.
We rst consider the effect of the ame on the acoustic eld. Analytical treat-
ments of this topic modeled the ame as a dynamically evolving discontinuity in
temperature with a pressure-sensitive ame speed.
59, 60
These studies prescribed
the ame position and, thus, did not consider the fully coupled amewave 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 ame is scattered as a monochromatic
disturbance of the same frequency. In the turbulent ame, 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
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 351
-20 -10 0 10 20
10
-6
10
-5
10
-4
10
-3
10
-2
f-f
i
P
S
D
(
a
r
b
.
u
n
i
t
s
)
f
i
=7.5 kHz
f
i
=15 kHz
Fig. 12.26 Measured spectra of scattered acoustic eld excited by 7.5- and 15-kHz
incident waves. Adapted from Lieuwen.
61
wave frequency f
i
. The broadened spectrum of the scattered eld is caused by the
randomly moving ame 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 ame. The
narrow band, coherent peak at the incident wave frequency, and distributed side
bands are clearly evident in the gure.
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 reecting
surface moving with a Mach number,
M, generates reected waves oscillating at
the Doppler-shifted frequency
f
re
= f
drive
(1
M n)
(1
M n)
(12.48)
where n denotes the unit normal direction of the incident wave. Noting that the
Mach number of ame-front motion is very small, this expression can be written
as
( f f
drive
)
2
)
1/2
= (f )
2
)
1/2
2 f
drive
(
M n)
2
)
1/2
(12.49)
This equation shows that the bandwidth of the scattered waves f grows with rms
ame-front velocity and incident-wave frequency.
If the ame does not add energy to the acoustic eld, the energy in the incoherent
sidebands is derived from the coherent wave. Thus, the wrinkled characteristics
of the ame act as a potential source of damping of coherent acoustic energy. In
cases in which the ame amplies sound waves, the overall energy balance of the
T. C. LIEUWEN 352
Acoustic
Disturbance
Flame
Fig. 12.27 Image of instantaneous pressure eld and ame front. Reproduced with
permission from A. Laverdant and D. Thevenin.
63
coherent eld 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 ame front because
of differences in distance the wave travels before impinging on the ame and
reecting. Phase mismatch between disturbances originating from different points
of the ame results in destructive interference between these different waves. This
distortion of the acoustic eld by a wrinkled ame front can be clearly seen in the
computational result in Fig. 12.27 which plots the spatial pressure-eld distribution
of an initially planar acoustic wave after impinging on a wrinkled ame.
63
In general, the characteristics of the scattered eld depend on the statistical
distribution of the ame front about its average position. In the limit where the
scales of ame wrinkling are much smaller than a wavelength, only the turbulent
ame-brush thickness is important and the coherent scattered eld is damped by the
amount 12(kcos
i
)
2
, where k = /c, , and
i
are the acoustic wavenumber,
ame-brush thickness, and relative angle between the incident wave and average
ame position. Besides the reduction in amplitude, the coherent eld has a phase
offset relative to its smooth-surface value if the ame position is asymmetrically
distributed about its mean position. These expressions predict, then, that turbulent
ame effects grow with increases in frequency or turbulent ame-brush thickness.
The increase of energy in the incoherent eld on frequency can be seen from
the data in Fig. 12.28. Because the total energy in the scattered eld is limited by
the energy in the incident wave plus any small amplication from the ame, the
energy in the incoherent eld saturates at high frequencies. The saturation at high
frequencies corresponds to situations in which no energy remains in the scattered
coherent eld, 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 ames. Apparently, the answer to this
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 353
0.1 0.2 0.3 0.4 0.5
10
-2
10
-1
10
0
/
I
n
c
o
h
e
r
e
n
t
S
c
a
t
t
e
r
e
d
P
o
w
e
r
Fig. 12.28 Dependence of scattered incoherent power on ratio of ame-brush thick-
ness and acoustic wavelength . Adapted from Lieuwen.
61
question lies in the value of the ratio of the disturbance wavelength, whether acous-
tic or convective, to the turbulent ame-brush thickness. For typical longitudinal
mode instabilities, the ratio of / is often very small, implying that wrinkled
ame effects provide only a small correction from laminar-ame analyses. At
these same frequencies, however, the convective wavelength could potentially be
of the same magnitude as . This potential similarity implies that the conclusions
of Sec. III.A.3 or III.A.4, where convective disturbances or ame wrinkles vary
over a convective scale, could be modied in the turbulent case. In the same way,
the response of ames to high-frequency acoustic waves, such as during screeching
instabilities, could also be substantially different than the response anticipated from
laminar-ame analyses.
6. Acoustic Field Interactions with Inherent Flame Instabilities
Even in the absence of acoustic oscillations, premixed ames are often un-
steady because of intrinsic instabilities. These instabilities are signicant because
their interaction with externally imposed acoustic oscillations results in qualita-
tive changes in the ames dynamics. We briey introduce these instabilities in
the following text; detailed discussions and analysis can be found in Williams
77
or Clavin.
49
We focus on three basic categories of intrinsic premixed-ame 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 notedinSec. III.B.1, andSaffman-type instabilities, whichappear
in ames propagating through thin channels.
The body-force instability is analogous to the classical buoyant mechanism in
which a heavy uid resting above a lighter one is destabilized by the action of
gravity. In the same way, ames propagating upward divide a higher- and lower-
density region and, thus, are unstable. Similar instabilities can be induced by
acceleration of the ame sheet, either through a variation of the burning velocity
T. C. LIEUWEN 354
or an externally imposed ow perturbation. As will be discussed further in the
next section, the latter mechanism plays an important role in certain acoustic
ame interaction phenomena, in which the acceleration is provided by the acoustic
velocity eld.
The hydrodynamic, or DarriusLandauinstability, has anunderlyingmechanism
that is purely uid 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 ame which causes the incident-ow
streamlines to diverge and/or converge in front of a ame disturbance that is convex
and/or concave to the unburned gas. The resulting ow divergence or convergence
causes the ow to locally decelerate or accelerate, respectively, causing the distur-
bance to grow further. The dependence of the local burning velocity on the radius
of ame curvature stabilizes short-wavelength perturbations. Longer-wavelength
perturbations are stabilized by gravity for downward-propagating ames.
The diffusive-thermal instability is caused by the effect of ame-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 ux that heats the incoming mixture. In the same way, it
results in focusing the diffusive ux of the decient reactant into the ame. 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
ame, also causing instability.
Two key interactions of acoustic waves with ame instabilities have been
noted.
35, 6469
These interactions are the stabilization of the DarriusLandau in-
stability by acoustic perturbations and a new parametric instability that occurs at
large-velocity oscillations.
Both instabilities can be observed in ames propagating down a pipe lled
with reactive mixture. Photographs obtained by R. C. Aldredge of the resulting
sequence of ame characteristics are illustrated in Fig. 12.29. As the ame prop-
agates down the tube, it develops a cellular shape caused by the DarrieusLandau
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 ame front, may also appear. These oscillations grow and can result in a re-
markable restabilization of the ame front, where the cellular structure disappears
and the ame reverts to a nearly planar front (see middle image in Fig. 12.29).
Measurements indicate that the ames 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
DarrieusLandau ame instability by the oscillatory acceleration imposed by the
acoustic eld.
66, 69
If the acoustic amplitude grows further, a violent secondary instability can occur
(see bottomimage in Fig. 12.29). The nearly planar ame develops small, pulsating
cellular structures whose amplitude increases rapidly. These cellular structures
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 355
Fig. 12.29 Sequence of ame-front characteristics as it propagates downward in a
tube.
68
Successive images show ame wrinkled by DarrieusLandau instability mech-
anism (top), planarization of ame 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 ame front by the unsteady
velocity eld, which separates two regions of differing densities. With increased
amplitudes, these organizedcellular structures breakdownintoa highlydisordered,
turbulent front. In the case in which the ambient oweld is highly turbulent, Vaezi
and Aldredge
70
found that the parametric instability still appeared. In addition, they
noted that for sufciently high-turbulence levels, the appearance of the parametric
instability did not result in additional acceleration of the ame front. This nding
is in contrast with the case in which the ambient oweld is quiescent, where the
parametric instability results in substantial ame acceleration.
Markstein rst 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 eld. The
ame-front dynamics can be described by a parametric oscillator equation of the
T. C. LIEUWEN 356
form
66
A
d
2
y(k, t )
dt
2
B
dy(k, t )
dt
[C
o
C
1
cos(wt )]y(k, t ) = 0 (12.50)
where A, B, and C are coefcients dened in Ref. 66, k is the wave number
of the perturbation, and is the frequency of imposed oscillation. The damping
coefcient, B, is always positive, whereas the coefcient C
o
is negative if the
planar ame front is nominally unstable. In the case in which C
o
is negative,
this equation has the properties that the solution is unstable in the absence of
imposed oscillations (i.e., C
1
= 0), is stabilized in the presence of small but nite
amplitude perturbations, and is destabilized in the presence of large-amplitude
parametric oscillations.
7. Flame Anchoring, Flashback, and Extinction
The dynamics of the ame-attachment point have a signicant impact on the
overall ame kinematics. This point can be appreciated by the discussion in
Sec. III.A.3, which noted that the solution for the ame area was controlled by the
superposition of the boundary condition (i.e., the ame-attachment condition) and
owame speed nonuniformity. The amplitude of ame 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 ame-attachment point in an oscillatory oweld is not
well understood. Most studies of laminar-ame dynamics have assumed that the
ame base remains motionless (i.e.,
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 ame that is not attached at a xed geometric point, such as in
a swirling ame that stands off from the burner and attaches at a ow-stagnation
point.
The ame-anchoring boundary condition is known to be amplitude and fre-
quency dependent at high forcing amplitudes. Baillot
52
found that laminar, conical
Bunsen ames subjected to high-amplitude, low-frequency velocity perturbations
exhibited a variety of transient ame-holding behavior, such as ashback, asym-
metric blowoff, unsteady lifting, and reanchoring of the ame. In addition, they
noted that its response is asymmetric and extremely disordered. However, at high
frequencies and forcing amplitudes, the ame remains rmly attached, but its over-
all shape dramatically changes. They found that the ame becomes collapsed with
a rounded-off tip region, and for sufciently high-forcing intensities (u
/
/ u > 1),
the ames mean shape becomes hemispherical.
54
Unsteady ashback phenomena, which occur when the instantaneous-ow ve-
locity falls belowthe ame speed, can be captured within the ame-sheet approach
considered here. For example, Dowling
71
modeled this with a nonlinear boundary
conditionwhichassumedthat the ame remainedanchored(i.e.,
attachment point
(t ) =
0) and propagated upstream when the total gas velocity exceeded and fell below,
respectively, the ame speed.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 357
Flow oscillations and the resulting oscillatory strain rates also cause local or
global ame extinction. Such extinction events in unsteady ows have been demon-
strated in several studies in fundamental geometries, such as counterowames.
72
In addition, the ame may extinguish at locations of sharp curvature, such as in
cusps. Such extinction phenomena are routinely observed in turbulent ames.
Unsteady extinction and reignition of local or global regions of the ame intro-
duces nonlinearity in the acousticheat-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 ame
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, Dowling
74
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 nite amplitude response of a ame 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 amelet regime, the well-stirred reactor (WSR) regime.
It has beensuggestedthat ameacoustic interactions inthis regime canbe 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 uniform
77
:
dM
dt
= m
in
m (12.51)
dE
dt
= m
in
h
in
mh (12.52)
dM
k
dt
= m
in
Y
k,i n
mY
k
W
k
(12.53)
where m and
W
k
are the mass ow rate and consumption rate of the kth species,
respectively. M, E, and M
k
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 ow rate and reactor volume,
res
= m/V. The reactor volume or
residence time cannot, in general, be specied by simplied analysis, because it
T. C. LIEUWEN 358
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 simplied models.
78
Note that the preceding equations assume that the perturbations are spatially
uniform. If the ame zone is acoustically compact, such an approximation may
be adequate to describe acoustic perturbations. Entropy and vorticity uctuations
could potentially be of much shorter length scales than the reactor size, how-
ever, indicating that the perturbation variables and ow-strain eld are spatially
distributed in the reactor.
The total heat release from the reactor is given by the volume integral
Q(t ) =
_
v
h
r
dV (12.54)
where 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
1
S
1
and d A are replaced by and dV,
respectively. Note that heat-release uctuations are generated by reaction rate,
reactor volume, and heat of reaction uctuations.
2. WSR Response to Flow Perturbations
Consider rst the factors affecting the reaction rate . Some understanding of
the sensitivity of to ow and mixture perturbations can be gained from the
following generic global reaction-rate expression:
= AY
a
f
Y
b
ox
p
c
e
E
u
/RT
(12.55)
where A is a pre-exponential factor, Y
F
and Y
ox
denote fuel and oxidizer mass
fractions, and a, b, and c denote sensitivity coefcients. Whereas sensitivity of
to disturbances in any of the quantities in Eq. (12.55) can be determined from the
value of the exponential coefcients, it must be emphasized that any disturbance
does not occur in isolation. For example, uctuations in temperature have an impact
on the residence time or reactor fuel and oxidizer concentration. The sensitivities
of to perturbations in pressure, mass owrate, 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
E
u
/RT
to temperature varia-
tions grows with temperature, the effect of temperature perturbations in the inlet-
reactant stream on 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 oscillations grows with decreas-
ing equivalence ratio. This point is analogous to the dependence of the ame speed
on the equivalence ratio discussed in Sec. III.A.4 and can be understood by noting
that reaches a maximum near = 1 (i.e., it has no sensitivity to perturba-
tions) and decreases as becomes leaner. Both of these sensitivities were deter-
mined assuming that the reactor residence time was xed; that is, oscillations in
reactive-mixture composition or temperature did not affect the residence time.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 359
Reactor Temperature
V
o
l
u
m
e
t
r
i
c
H
e
a
t
R
e
l
e
a
s
e
R
a
t
e
Decreasing
Residence Time
Blowout
(minimum residence
time)
Maximum
Reaction Rate
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 showthat 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
gure, respectively. Energy convection rate curves are drawn at several residence
times. Note that maximumreaction 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 gure, 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 maximumreaction-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 h
r
, the reactor volume as well (i.e., higher heat content reactant may
result in faster kinetics, reducing the reactor volume).
T. C. LIEUWEN 360
To extend the reactor approach to situations where the ame was convectively
noncompact and, thus, owdisturbances varied substantially over the ame region,
Lieuwen et al.
84
treated the combustion zone as a distribution of innitesimal, inde-
pendent reactors whose input conditions were given by that of the local ow 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 amelet 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 nite amplitude effects into unsteady, well-stirred reactor calcula-
tions is straightforward, although it may require numerics for time stepping through
Eqs (12.5112.53).
79, 81
The WSR equations are often used for model problems in
nonlinear dynamics studies because of the complex, even chaotic dynamics
85
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 ames
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 ame have ameletlike 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 difcult 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 ow rates,
but at different pressures, p
1
and p
2
. 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 ) = p
1
( p
2
p
1
)sin t. 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 uctuations, such as temperature. Only uctuations
in the heat content of the inlet fuel stream will cause a quasi-steady uctuation
in reactor heat release. So what is happening? Clearly, the changes in pressure
or temperature inuence the reaction rate . 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 uctuations in overall heat release.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 361
Thus, any uctuation 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 sufciently large frequency . 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 amelet studies (see Sec. III.A.4), in
which ame speed and area uctuations are coupled; for example, quasi-steady
uctuations in ame speed do not cause the global heat release to oscillate be-
cause of the accompanying oscillations in ame area. The difference is that these
coupled dynamics can be reasonably modeled in the amelet case from rst 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 ame response in realistic com-
bustors on geometric and fuel composition parameters. Reasonable, quantitative
predictions of ameacoustic interaction phenomena have been demonstrated
for a few simple congurations, such as the laminar Bunsen burner of Ducruix
et al.
36
or Baillot et al.,
53
or the nominally at ame of Searby and Rochwerger.
66
These successes illustrate the rapid progress being made in modeling kinematic
processes in ameacoustic interactions. In addition, progress is being made in
developing hybrid models that use computational simulations to determine various
components of the combustor systemame interactions.
86
The development of
accurate, predictive combustion-response models for realistic (i.e., turbulent)
congurations 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 signicant part of the relevant liter-
ature consists of essentially decoupled theoretical models or experimental studies,
even in rather fundamental congurations. For example, a substantial number of
fundamental studies have theoretically investigated the response of at, laminar
ames to pressure perturbations.
4247
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 frompractical ames,
they are prerequisite building blocks toward modeling realistic systems.
Second, work is needed to develop simplied models of vortexame interac-
tions. The existing theoretical work on this subject is largely numeric. Analytical
methodologies for modeling unstable, reacting shear ows have been developed
87
and need to be extended to incorporate the unsteady ow effects on the ame.
Third, predicting the response of ames to nite 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.
T. C. LIEUWEN 362
In addition, effects such as the stabilization or parametric destabilization of ames
discussed in Sec. III.A.6 may cause nite amplitude acoustic oscillations to sub-
stantially change the mean characteristics of the turbulent ame with which it
is interacting. These effects merit further investigation.
Fourth, ameacoustic wave interactions in realistic environments occur in a
very noisy atmosphere where the ame 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-ame fronts to velocity,
equivalence ratio, or vortical disturbances needs to be generalized to include the
fact that the average ame is highly unsteady. For example, the successful work
performed to date on laminar, Bunsen ames should be extended to turbulent-ow
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 ames with thickened amelets, 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
signicant conceptual problems. Progress in this area will require clarication 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 Summereld, M., Nonsteady Burning and Combustion Sta-
bility of Solid Propellants, Progress in Aeronautics and Astronautics, Vol. 143, AIAA,
Washington, DC, 1992.
2
Sirignano, W., Fluid Dynamics and Transport of Droplets and Sprays, Cambridge Univ.
Press, Cambridge, England, U.K., 1999.
3
Harrje, D., Reardon, F. (eds.) Liquid Propellant Rocket Instability, SP-194, NASA, 1972.
4
Peters, N., Laminar Flamelet Concepts in Turbulent Combustion, Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 21, 1986, pp. 1231120.
5
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, Edwards,
Philadelphia, PA, 2001.
6
Turns, S., An Introduction to Combustion, McGrawHill, 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. 494514.
8
Kovaszay, L. S. G., Turbulence in Supersonic Flow, Journal of the Aeronautical Sci-
ences, Vol. 20, No. 10, 1953, pp. 657674.
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. 535543.
10
Lee, D. H., and Lieuwen, T., Acoustic Neareld Characteristics of a Conical, Premixed
Flame, Journal of the Acoustical Society of America, Vol. 113, No. 1, 2003, pp. 167177.
11
Ferguson, D., Richards, G., Woodruff, S., Bernal, S., and Gautam, M., Proceedings of
the 2nd Joint Meeting U.S. Sections of the Combustion Institute, Pittsburgh, PA, 2001.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 363
12
Schadow, K., and Gutmark, E., Combustion Instability Related to Vortex Shedding
in Dump Combustors and Their Passive Control, Progress in Energy and Combustion
Science, Vol. 18, 1992, pp. 117132.
13
Coats, C., Coherent Structures in Combustion, Progress in Energy and Combustion
Science, Vol. 22, 1996, pp. 427509.
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. 119142.
16
Michalke, A., Zeitschrift f ur Flugwissenschaften (in German), Vol. 19, 1971.
17
Baillot, F., Durox, D., and Prudhomme, R., Experimental and Theoretical Study of a
Premixed Vibrating Flame, Combustion and Flame, Vol. 88, 1992, pp. 149168.
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. 135138.
21
Marble, F., and Candel, S., Acoustic Disturbance fromGas Non-uniformity Convected
Through a Nozzle, Journal of Sound and Vibration, Vol. 55, 1977, pp. 225243.
22
Lieuwen, T., Theoretical Investigation of Unsteady FlowInteractions With a Premixed
Planar Flame, Journal of Fluid Mechanics, Vol. 435, 2001, pp. 289303.
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. 761769.
26
Subbaiah, M. V., Nonsteady Flame Spreading in Two Dimensional Ducts, AIAA
Journal, Vol., 21, No. 11, 1983, pp. 15571564.
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. 125.
29
Boyer, L., and Quinard, J., On the Dynamics of Anchored Flames, Combustion and
Flame, Vol. 82, 1990, pp. 5165.
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. 487510.
31
Ashurst, W., and Sivashinsky, G., On Flame Propagation through Periodic Flow
Fields, Combustion Science and Technology, Vol. 80, 1991, pp. 159164.
32
Joulin, G., Sivashinsky, G., Pockets in Premixed Flames and Combustion Rate, Com-
bustion Science and Technology, Vol. 77, 1991, pp. 329335.
33
Aldredge, R., The Propagation of Wrinkled, Premixed Flames in Spatially Periodic
Shear Flow, Combustion and Flame, Vol. 90, No. 2, 1992, pp. 121133.
34
Matalon, M., and Matkowsky, B., Flames as Gas Dynamic Discontinuities, Journal
of Fluid Mechanics, Vol. 124, 1982, pp. 239260.
T. C. LIEUWEN 364
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. 157197.
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. 553566.
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. 187199.
39
Chu, B. T., On the Generation of Pressure Waves at a Plane Flame Front, Proceedings
of the Combustion Institute, Pittsburgh, PA, Vol. 4, 1953, pp. 603612.
40
Schuller, T., Durox, D., and Candel, S., A Unied Model for the Prediction of Lam-
inar Flame Transfer Functions: Comparisons between Conical and V-Flame Dynamics,
Combustion and Flame, Vol. 134, 2003, pp. 2134.
41
Preetham, and Lieuwen, T., Nonlinear Flame-Flow Transfer Function Calculations:
Flow Disturbance Celerity Effects, AIAA Paper 20044035.
42
McIntosh, A. C., Pressure Disturbances of Different Length Scales Interacting with
Conventional Flames, Combustion Science and Technology, Vol. 75, 1991, pp. 287309.
43
Peters, N., and Ludford, G. S. S., The Effect of Pressure Variations on Premixed
Flames, Combustion Science and Technology, Vol. 34, 1983, pp. 331344.
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. 982995.
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. 141159.
46
Ledder, G., and Kapila, A. K., The Response of Premixed Flames to Pressure Pertur-
bations, Combustion Science and Technology, Vol. 76, 1991, pp. 2144.
47
McIntosh, A. C., Deagration Fronts and Compressibility, Philosophical Trans-
actions of the Royal Society of London, Series A: Mathematical and Physical Sciences,
Vol. 357, 1999, pp. 35233538.
48
Huang, Z., Bechtold, J., and Matalon, M., Weakly Stretched Premixed Flames in
Oscillating Flows, Combustion Theory Modeling, Vol. 2, 1998, pp. 115133.
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. 219229.
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. 18331840.
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. 303318.
53
Baillot, F., Bourehla, A., and Durox, D., The Characteristic Method and Cusped Flame
Fronts, Combustion Science and Technology, Vol. 112, 1996, pp. 327350.
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. 295310.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 365
55
Sankaran, R., and Im, H., Dynamic Flammability Limits of Methane/air Premixed
Flames with Mixture Composition Fluctuations, Proceedings of the Combustion Institute,
Pittsburgh, PA, Vol. 29, 2002, pp. 7784.
56
Cho, J. H., and Lieuwen, T., Laminar Premixed Flame Response to Equivalence Ratio
Oscillations, Combustion and Flame, Vol. 140, No. 12, pp. 116129.
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. 12, 2001, pp. 14891505.
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 Amplication/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. 1119.
64
Searby, G., Acoustic Instability in Premixed Flames, Combustion Science and Tech-
nology, Vol. 81, 1992, pp. 221231.
65
Markstein, G., Flames as Ampliers of Fluid Mechanical Disturbances, Proceedings
of the Sixth National Congress on Applied Mechanics, 1970, pp. 1133.
66
Searby, G., and Rochwerger, D., A Parametric Acoustic Instability in Premixed
Flames, Journal of Fluid Mechanics, Vol. 231, 1991, pp. 529543.
67
Pelce, P., and Rochwerger, D., Vibratory Instability of Cellular Flames Propagating
in Tubes, Journal of Fluid Mechanics, Vol. 239, 1992, pp. 293307.
68
Vaezi, V., and Aldredge, R., Laminar Flame Instabilities in a Taylor-Couette Com-
bustor, Combustion and Flame, Vol. 121, 2000, pp. 356366.
69
Bychkov, V., Analytical Scalings for Flame Interaction with Sound Waves, Physics
of Fluids, Vol. 11, No. 10, 1999, pp. 31683173.
70
Vaezi, V., and Aldredge, R. C., Inuences of Acoustic Instabilities on Turbulent-Flame
Propagation, Experimental Thermal and Fluid Science, Vol. 20, 2000, pp. 162169.
71
Dowling, A. P., A Kinematic Model of a Ducted Flame, Journal of Fluid Mechanics,
Vol. 394, 1999, pp. 5172.
72
Sung, C. J., and Law, C. K., Combustion and Flame, Vol. 123, 2000, pp. 375388.
73
Poinsot, T., Veynante, D., Bourienne, F., Candel, S., Esposito, E., and Surget, J., Pro-
ceedings of the Combustion Institute, Pittsburgh, PA, Vol. 22, 1988.
74
Dowling, A. P., Nonlinear Self-Excited Oscillations of a Ducted Flame, Journal of
Fluid Mechanics, Vol. 346, 1997, pp. 271290.
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. 214230.
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. 4561.
T. C. LIEUWEN 366
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. 7585.
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. 193211.
82
Park, S., Annaswamy, A., and Ghoniem, A., Heat Release Dynamics Modeling of
Kinetically Controlled Burning, Combustion and Flame, Vol. 128, 2002, pp. 217231.
83
Park, S., Annaswamy, A., and Ghoniem, A., Dynamic Characteristics of Kinetically
Controlled Combustion and their Impact on Thermoacoustic Instability, Proceedings of
ICDERS 2001.
84
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., A Mechanism for Combustion
Instabilities inPremixedGas Turbine Combustors, Journal of Engineeringfor Gas Turbines
and Power, Vol. 123, No. 1, 2001, pp. 182190.
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. 483510.
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
Du
Dt
= p +
i, j
x
j
e
i
(13.1b)
where p is the pressure, is the density, u is the velocity, and
i, j
is the viscous
stress tensor. Here D/Dt is the material derivative /t +u and e
i
represents
the unit vector in the direction of coordinate i . For a perfect gas, we have the gas
law p = R
gas
T, where T is the temperature, R
gas
= c
p
c
v
is the gas constant,
and c
p
and c
v
are the specic heats at constant pressure and volume, respectively.
The internal energy per unit mass e is equal to c
v
T, and the enthalpy h is c
p
T =
e + p/. Conservation of energy gives the energy equation,
D
Dt
_
e +
1
2
u
2
_
= ( pu) +q + (kT) +
x
j
(
i, j
u
i
) (13.2)
where k is the conductivity and q is the rate of heat added to the uid per unit
volume. By using Eq. (13.1b), this can be written as
Dh
Dt
=
Dp
Dt
+q + (kT) +
i, j
u
i
x
j
(13.3)
We dene entropy S by the thermodynamic relation Dh = TDS +(1/) Dp.
Hence, Eq. (13.3) gives that
T
DS
Dt
= q + (kT) +
i, j
u
i
x
j
(13.4)
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, = u,
D
Dt
_
_
=
_
_
u +
1
3
p +
1
_
1
i, j
x
j
e
i
_
(13.5)
The rst term on the right-hand side describes how the stretching of vortex lines
intensies the local vorticity, and the last term clearly represents generation of
A. P. DOWLING AND S. R. STOW 372
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 ame
front (density gradient), so that, for instance, circumferential waves will generate
vorticity at combustion zones.
We will nowassume inviscid ow(
i, j
0). We will also assume the uid is an
ideal gas (i.e., in addition to being a perfect gas, it does not conduct heat), and we
take c
p
and c
v
to be constant. Fromthe preceding denition of entropy, we nd that
S = c
v
log( p/
(x, t ) (13.6)
and similarly for the other ow variables. From Eqs. (13.1), (13.4), and (13.5), the
linearized equations for these perturbations are
Dt
+ u
= 0 (13.7a)
Du
Dt
+
1
p
= 0 (13.7b)
T
DS
Dt
= q
(13.7c)
D
Dt
= 0 (13.7d)
where
D/Dt = /t + u and we have used
= 0. Combining Eqs. (13.7a
13.7c) and using S
= c
v
p
/ p c
p
D
2
p
Dt
2
2
p
=
1
c
2
Dq
Dt
(13.8)
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 disturbances
11
: 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
= 0; hence,
= p
/ c
2
. Since q
,
_
1
c
2
D
2
Dt
2
2
_
p
= 0 (13.9)
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 373
and the corresponding u
= 0 and u
= 0 and
u
= 0, and Eq. (13.7d) shows that this type of disturbance (a vorticity wave)
is also convected with the mean ow. If the mean ow is zero, only acoustic
disturbances propagate.
A. Conditions Across a Flame Zone
We now consider the effect of a thin ame 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 ow parameters across the ame; we denote conditions at x = 0
and x = 0
+
by subscripts 1 and 2, respectively. From Eqs. (13.1) and (13.3), we nd that
2
u
2
=
1
u
1
(13.10a)
p
2
+
2
u
2
2
= p
1
+
1
u
2
1
(13.10b)
2
u
2
H
2
=
1
u
1
H
1
+ Q
A
(13.10c)
where H = h +
1
2
u
2
is the stagnation enthalpy. To calculate the mean ow, we
assume that
Q
A
is known (fromknowledge of the fuel type, equivalence ratio, etc.).
A ame model is used to describe the dependence of Q
A
on the ow 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
u
+
p
= 0 (13.11)
A. P. DOWLING AND S. R. STOW 374
Stowet 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, Stowet al.
13
considered the interaction of the shock
position and the ow perturbations (see also Yang and Culick
14
and Culick and
Rogers
15
), ndingthat, for one-dimensional disturbances, the perturbations inmass
ux and energy ux 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 ow
M
1
just downstreamof the choking plane. This gives the inlet boundary
conditions:
+
u
u
=
p
+( 1)
M
1
u
u
= w
= 0 (13.12)
For a weak shock, one would expect that there is negligible entropy production.
However, the equations implythat the (usuallyignored) entropyperturbationdown-
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 signicant
vorticity perturbation is also produced.
Other analytical inlet and outlet boundary conditions, such as acoustically closed
ends (u
t
=
1
p
x
=
1
c
t
f (t x/ c)
1
c
t
g(t + x/ c) (13.14a)
that is,
u
(x, t ) =
1
c
( f (t x/ c) g(t + x/ c)) (13.14b)
For perturbations of frequency , it is convenient to write f (t ) = Re(
f e
it
),
where the circumex denotes a complex amplitude. With this notation
p(x) =
f e
ix/ c
+ ge
ix/ c
(13.15a)
u(x) =
1
c
_
f e
ix/ c
ge
ix/ c
_
(13.15b)
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
p(0) = u(l) = 0 (13.16)
Equation (13.15a) then leads to g =
2
p
t
2
2
p
x
2
=
1
c
2
q
t
(13.19)
The term on the right-hand side describes how the unsteady addition of heat gen-
erates pressure disturbances. For a specied rate of heat release q
(x, t ) =
2
1
p
(x, t ) (13.20)
The formof the pressure perturbation can be determined by substituting for q
(x, t )
in Eq. (13.19) and seeking a separable solution, p
2
+2ie
i
2
n
= 0 (13.21)
where
n
is dened in Eq. (13.17).
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 377
When = 0, the roots of Eq. (13.21) are the undamped resonant organ-pipe
frequencies
n
.
When = 0, = 0, the quadratic equation (13.21) for can be readily solved
to give
= i
_
2
n
2
_
1/2
(13.22)
is nowcomplex. Because the time dependence is e
it
, Im() is the growth rate
of the disturbances. Here e
it
= exp[t i (
2
n
2
)
1/2
t ], showing that the oscil-
lations grow exponentially in time if is positive. We have recovered Rayleighs
criterion
2
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 , that is, heat release in antiphase
with the pressure, the oscillations are damped.
When = 0, = 0, Eq. (13.21) would, in general, need a numerical solution,
and some results are shownin Fig. 13.2. InFig. 13.2 andsubsequently, a normalized
frequency f
N
= Re()/
1
and a normalized growth rate g
N
= Im()/
1
are
used. However, the general characteristics of the solution can be investigated by
considering small , and determining the roots iteratively. We have already noted
1
2
n
o
r
m
a
l
i
z
e
d
f
r
e
q
u
e
n
c
y
,
f
N
0 0.2 0 4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0.97
0.98
0.99
1
1.01
1.02
1.03
/
( )
.
n
o
r
m
a
l
i
z
e
d
g
r
o
w
t
h
r
a
t
e
,
g
N
0 0.2 0 4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0.03
0.02
0.01
0
0.01
0.02
1
2
/
( )
.
Fig. 13.2 Variation with of the root of Eq. (13.21) near
1
: , /
1
= 0.01;
, /
1
=0.02; and , /
1
=0.01. a) Frequency. b) Growth rate.
A. P. DOWLING AND S. R. STOW 378
that, for = 0, a root of Eq. (13.21) is at =
n
. For small , this root moves to
=
n
+, where is small; substitution into Eq. (13.21) shows that
= ie
i
n
= i cos(
n
) sin(
n
) (13.23)
From this we see that any cos(
n
) > 0 leads to a positive growth rate, that is,
any unsteady heat release with /2 < phase( q/ p) < /2 is destabilizing. It is
also clear that the resonant frequency is shifted whenever sin(
n
) = 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 are conrmed by
the numerical results shown in Fig. 13.2. For > 0, the growth rate is increased
for cos(
1
) > 0, that is, (2n
1
2
) <
1
< (2n +
1
2
) and decreased when
cos(
1
) is negative. Also the frequency is decreased for sin(
1
) > 0 and
increased for sin(
1
) < 0, and the behaviors are reversed for negative . For
nonzero , Eq. (13.21) becomes transcendental and has additional solutions that
are primarily related to
1
rather than the downstream geometry. For example,
for small || these are at Im() and Re() 2m/ for negative and
(2m +1)/ for positive , where m is an integer. These are the even and odd
harmonics for the convection time . The choice of even and odd comes from
a balance of the right-hand side and the rst term on the left-hand side in Eq.
(13.19); the rst 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 signi-
cant of which is usually a saturation in the heat-release response,
16
lead to a nite
amplitude limit-cycle oscillation.
However, this rst example is an oversimplication of what occurs in practice.
In LPP gas turbines, it is not the unsteady pressure that has the greatest inuence on
the rate of heat release. Rather, the rate of heat release is related to the instantaneous
fuelair 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 inuence 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 ,
q
(x, t ) = Q
(t )(x b) (13.24a)
Q
(t ) =
c
2
1
u
1
(t ) (13.24b)
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 379
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 can be expected to range from 0 to about 10 and that
in a LPP system 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
ame. However, for consistency, with being the fuel-convection time, the ame
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=b
+
x=b
= 0 (13.25a)
_
p
x
_
x=b
+
x=b
=
1
c
2
dQ
dt
(13.25b)
Equation (13.25b) is equivalent to
_
u
_
x=b
+
x=b
=
1
c
2
Q
(t ) (13.25c)
relating the volumetric expansion to the instantaneous rate of heat release. After
substitution for the particular Q
(b
+
, t ) = u
(b
, t ) u
(b
, t ) (13.26)
We will consider solutions with time dependence e
it
and want to nd the resonant
frequencies and the mode shapes.
In x < b, the solution of the homogeneous wave equation that satises the inlet
boundary condition p(0) = 0 is
p(x) = A sin(kx) (13.27a)
u(x) =
i
c
A cos(kx) (13.27b)
where k is the wave number / c and the complex constant A has yet to be deter-
mined. Similarly, in x > b, the boundary condition u(l) = 0 leads to
p(x) = B cos(k(l x)) (13.28a)
u(x) =
i
c
B sin(k(l x)) (13.28b)
A. P. DOWLING AND S. R. STOW 380
n
o
r
m
a
l
i
z
e
d
f
r
e
q
u
e
n
c
y
,
f
N
0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1
0
0 2
0 4
0 6
0 8
1
Fig. 13.3 Variation of frequency with for the root of Eq. (13.30) near
1
, taking =
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 e
i
(13.30)
The resonant frequencies follow from a numerical solution of Eq. (13.30). Their
dependence on and is shown in Figs. 13.3 and 13.4.
For = 0, the roots are at =
n
. As varies, for = 0, the rate of heat release
is in quadrature with the pressure perturbation [note the 90-deg phase difference
between p
and u
1
2
/
( )
.
Fig. 13.4 Variation of growth rate with for the root of Eq. (13.30) near
1
, taking
b=l/10: , =0.2; , =0.4; , =0.6; and , =0.8.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 381
= 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
< Re() phase( 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
p(x) =
_
C sin(kx)/ sin(kb) for 0 x b
C cos(k(l x))/ cos(k(l b)) for b x l
(13.32)
where the constant C is arbitrary.
C. Galerkin Series
Another wayof solvingthe inhomogeneous wave equationEq. (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 ) =
m=1
m
(t )
m
(x) (13.33)
where the functions
m
(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
m
.
Substitution for the pressure perturbation from Eq. (13.33) into Eq. (13.19) then
leads to
m=1
_
d
2
m
dt
2
+
2
m
m
_
m
(x) = ( 1)
q
t
(13.34)
After multiplication by
n
(x) and integration with respect to x, the orthogonality
of
n
(x) shows that Eq. (13.34) becomes
d
2
n
dt
2
+
2
n
n
=
1
E
n
_
l
0
q
t
n
(x) dx n = 1, . . . , (13.35)
where E
n
=
_
l
0
2
n
dx. Equation (13.35) is a complicated system of equations,
because q
(x, t ) is related to the local ow and so involves all the unknown coef-
cients
m
(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
A. P. DOWLING AND S. R. STOW 382
Culick and Yang.
17
When q
/t , replacing
the pressure and velocity perturbations by
n
(t )
n
(x) and (
n
(t )/
2
n
) d
n
/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,
n
(t )
2
n
n
(t ). The errors introduced by these approximations can be checked
by applying the method to nd the lowest frequency of oscillation in example 2.
Example 2 by Galerkin Series
After applying Culicks rules, the rate of heat release in Eq. (13.24) leads to
q
t
(x, t ) =
c
2
1
1
(t )
d
1
dx
(b)(x b) (13.36)
and substitution into Eq. (13.35) gives
d
2
1
dt
2
+
2
1
1
=
c
2
E
1
1
(t )
d
1
dx
(b)
1
(b) (13.37)
The solutions
n
of the homogeneous wave equation are given in Eq. (13.18) and
1
(x) = sin(x/2l), leading to E
1
=
1
2
l. Equation (13.37), therefore, simplies
to
d
2
1
dt
2
+
2
1
1
=
c
2
l
2
1
(t ) cos
_
b
2l
_
sin
_
b
2l
_
(13.38)
The frequency of oscillation can be found by substituting
1
(t ) = Ce
it
into
Eq. (13.38) to give
2
=
2
1
c
2
2l
2
e
i
sin
_
b
l
_
(13.39)
The root of this equation is shown as a dashed line in Fig. 13.3 for the particular
case = 0. Comparison with the exact solution given in Eq. (13.30) shows that
the one-termGalerkin expansion gives the correct frequency and gradient at = 0
but that it rapidly diverges from the exact solution as 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 articial; they have had an unsteady
heat release q
(x, t ) and yet the mean temperature has been uniform. In practice,
of course, heat release is associated with temperature gradients and the mean
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 383
temperature and density are functions of position. We will introduce these effects
through discussion of the zero-mean-ow case. Then the momentum equation
(13.1b) ensures that the mean pressure is uniform and for linearized perturbations,
u
t
= p
(13.40)
in an inviscid ow. We show in the Appendix that the mass conservation equation
(13.1a) and the entropy equation (13.4) can be combined to give
1
c
2
p
t
= u
+
1
c
2
q
(13.41)
when heat conduction and viscous effects are neglected. Eliminating u
from Eqs.
(13.40) and (13.41), we obtain
1
c
2
2
p
t
2
_
1
p
_
=
1
c
2
q
t
(13.42)
In this equation, and c vary with position, but c
2
= p is uniform if the small
dependence of on temperature is neglected. We can illustrate the inuence of
temperature variation by extending example 2 to the case in which the mean
temperature rises from
T
1
to
T
2
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 in the system
illustratedinFig. 13.5. Just as inexample 2, we againapplythe boundaryconditions
(13.16) and the ame model (13.24).
2
T
2
1
T
1
c
1
c
2
l
x = b
Fig. 13.5 System for example 3.
A. P. DOWLING AND S. R. STOW 384
Outside the ame 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 ow variables are used. Hence, using
Eq. (13.27) and Eq. (13.28), we can write in x < b
p(x) = A sin(k
1
x) (13.43a)
u(x) =
i
1
c
1
A cos(k
1
x) (13.43b)
and in x > b
p(x) = B cos(k
2
(l x)) (13.43c)
u(x) =
i
2
c
2
B sin(k
2
(l x)) (13.43d)
where k
1
= / c
1
and k
2
= / c
2
.
Integration of Eqs. (13.40) and (13.42) across the region x = b with q
(x, t ) =
Q
(t )(x b) leads to
_
p
_
x=b
+
x=b
= 0 (13.44a)
_
1
x
_
x=b
+
x=b
=
1
1
c
2
1
dQ
dt
(13.44b)
Equation (13.44b) is equivalent to
_
u
_
x=b
+
x=b
=
1
1
c
2
1
Q
(t ) (13.44c)
After substituting for the particular Q
T
1
700 K,
T
2
2000 K).
So far, we have assumed that the ame is compact, that is, axially short com-
pared with the wavelengths of the perturbations. If the ame is not compact, we
may approximate the axial heat-release distribution by discretizing into a series of
compact ames, each having the formdescribed earlier. Between these series, there
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 385
T
2
T
1
n
o
r
m
a
l
i
z
e
d
f
r
e
q
u
e
n
c
y
,
f
N
1 1.5 2 2.5 3 3.5 4 4.5 5
1
1.2
1.4
1.6
1.8
2
/
Fig. 13.6 Variation of frequency with mean temperature ratio for solution near
1
taking = = 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 dasheddotted lines in Fig. 13.6 show results for applying this
approximation when, instead of a compact ame 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)
2025
and also for certain area
variations.
26, 27
E. Mean Flow
Most combustion systems involve a mean ow that brings fresh reactants into
the combustion zone. The Mach number of the oncoming owis 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 ow has two main consequences. Trivially, it affects the speed of prop-
agation of the acoustic waves, with one-dimensional disturbances then traveling
downstreamwith speed c + u and upstreamat c u. In addition, the mean owad-
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 ow.
For deniteness we again apply an open-end inlet boundary condition p
(0) = 0.
At the downstream end, we assume an area restriction in which the ow becomes
choked, and so Eq. (13.11) is the appropriate boundary condition. Note that the
hard-end boundary condition u
(t ) with Q
(x, t ) = Ae
it
_
e
ix/( c
1
(1+
M
1
))
e
ix/( c
1
(1
M
1
))
_
(13.46)
for disturbances of frequency and
M
1
= u
1
/ c
1
. For this isentropic ow
=
p
/ c
2
1
and for a perfect gas c
p
T
= p
(x, t ) = Ae
it
_
e
ix/( c
1
(1+
M
1
))
+e
ix/( c
1
(1
M
1
))
_
(13.47)
The uxes of mass, momentum, and stagnation enthalpy into the combustion zones
[dened 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 ) = e
it
_
Ce
ix/( c
2
(1+
M
2
))
+ De
ix/( c
2
(1
M
2
))
_
(13.48a)
2
c
2
u
(x, t ) = e
it
_
Ce
ix/( c
2
(1+
M
2
))
De
ix/( c
2
(1
M
2
))
_
(13.48b)
(x, t ) =
p
(x, t )
c
2
2
S
2
c
p
e
i(t x/ u
2
)
(13.48c)
c
p
T
(x, t ) =
p
(x, t )
2
+
S c
2
2
( 1)c
p
e
i(t x/ u
2
)
(13.48d)
for b x l and
M
2
= u
2
/ c
2
. 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. (13.10a13.10c).
Care needs to be taken to recover the jump conditions for zero mean ow from
Eq. (13.10). In the limit u
1
, u
2
0, Eq. (13.10b) clearly simplies to p
2
= p
1
,
the zero-mean-ow jump condition [Eqs. (13.25a) and (13.44a)]. At rst sight
one might assume that Eq. (13.10a) gives
2
u
2
=
1
u
1
as u
1
, u
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 u
2
S. In the limit u
2
0, S tends to innity,
in such a way as to keep the product u
2
S and, hence, u
2
2
and u
2
T
2
, nite. For
low-Mach-number mean ows, very large entropy uctuations occur downstream
of the ame zone. To see these uctuations mathematically it is convenient to rst
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 387
use Eq. (13.10a) to recast Eq. (13.10c) into the form
2
u
2
(c
p
T
2
+ u
2
u
2
) = Q
+
1
u
1
(c
p
T
1
+ u
1
u
1
)
(
H
2
H
1
)(
1
u
1
+
1
u
1
)
(13.49)
for linear perturbations. After using Eq. (13.48d) to expand c
p
T
2
and taking the
limit u 0, this equation simplies to
2
u
2
c
2
2
c
p
( 1)
Se
i(t b/ u
2
)
= Q
c
p
(
T
2
T
1
)
1
u
1
(13.50)
Physically, Eq. (13.50) shows that entropy is generated unsteadily at the combus-
tion zone whenever Q
= c
p
(
T
2
T
1
)
1
u
1
, that is, whenever there is unsteadiness
in the rate of heat addition per unit mass. In particular, the preceding assertion that
uS remains nite for small u is conrmed. Equation (13.48c) clearly shows that,
in this limit, the left-hand side of Eq. (13.50) is equal to u
2
c
2
2
2
/( 1), and
hence, the equation can be rearranged to give
u
2
2
=
1
c
2
2
Q
+(
1
2
)u
1
(13.51)
where we have used the perfect gas relationships to rewrite c
p
( 1) (
T
2
T
1
)
1
/ c
2
2
as
1
2
. Finally, substitution for u
2
2
in the equation of mass conserva-
tion leads to
2
u
1
=
2
u
1
c
2
2
Q
(13.52)
thereby recovering the zero-mean-ow jump condition Eq. (13.44c).
Once Q
/ c
2
in the downstreamboundary condition (13.11). We would expect the entropy
uctuations to be of importance only for the lowest-frequency acoustic mode, if
at all. Figure 13.7 shows the effects of a mean ow 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 ow is that it admits a different mode of
oscillation, one with a much lower resonance frequency (typically, 40150 Hz for
aeroengines), where the period of oscillation is set by the convection time of the
entropy uctuations from the ame 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 rst 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
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 389
(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: rst, 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 , we are interested in a straight
cylindrical duct 0 r b. Because we are assumingthat the meanowis uniform,
we must have v = w = 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 rst consider a pressure disturbance. We seek a separable solution by sub-
stituting p
F
+2 u
F
FX
+ u
2
X
X
c
2
_
X
X
+
(r B
r B
+r
2
_
= 0 (13.53)
where the prime denotes a derivative with respect to the argument. We see that
solutions take the form F(t ) = e
it
, X(x) = e
ikx
, and () = e
in
, with
(r B
+(
2
n
2
r
2
)r B = 0 (13.54)
where
2
= ( + uk)
2
/ c
2
k
2
. For continuity in the circumferential wave num-
ber n must be an integer. The axial wave number k and complex frequency may
take any complex value, but they are dependent. The general solution of Eq. (13.54)
is B(r) = c
1
J
n
(r) +c
2
Y
n
(r), where J
n
and Y
n
are the Bessel functions of the
rst and second kind, respectively. Since Y
n
is singular at r = 0, we must have
c
2
= 0, and the rigid wall boundary condition v(b) = 0 implies
dJ
n
dr
(b) = 0 (13.55)
For a given n, this gives an innite number of discrete solutions for . The solutions
are all real,
29
and without loss of generality we may take 0. We dene
n, m
to
be the (m +1)th solution. The full solution can be expressed as an acoustic wave
A. P. DOWLING AND S. R. STOW 390
of the form
30
p
= A
e
it +in+ik
x
B
n, m
(r) (13.56a)
=
1
c
2
A
e
it +in+ik
x
B
n, m
(r) (13.56b)
u
=
k
e
it +in+ik
x
B
n, m
(r) (13.56c)
v
=
i
e
it +in+ik
x
dB
n, m
dr
(r) (13.56d)
w
=
n
r
e
it +in+ik
x
B
n, m
(r) (13.56e)
with B
n, m
(r) = J
n
(
n, 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
= + uk
,
k
M (
2
2
c
)
1/2
c(1
M
2
)
(13.57)
and
M is the mean Mach number (which is assumed to be less than unity). Also,
c
= c
n, m
(1
M
2
)
1/2
is the complex cutoff frequency of the duct, and A
, which
may be complex, are the wave amplitudes. For real >
c
, A
+
represents a
downstream-propagating wave and A
=
1
c
2
A
E
e
it +in+ik
0
x
E(r) (13.58)
with p
= u
= v
= w
= 0, where k
0
= / u 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 rst type has the form
u
=
n
c
A
V
e
it +in+ik
0
x
V(r) (13.59a)
w
=
k
0
r
c
A
V
e
it +in+ik
0
x
V(r) (13.59b)
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 391
with p
= v
=
1
cr
A
W
e
it +in+ik
0
x
dW
dr
(r) (13.60a)
v
=
ik
0
cr
A
W
e
it +in+ik
0
x
W(r) (13.60b)
with p
= w
(t, x, ) = p(x)e
it +in
. For circumferential waves, n = 0,
for integer m, is given by
_
2
m
+
2
c
_
1/2
(13.63d)
where
c
= n c/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()/(2)] and growth rates
[= Im()] 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 reection 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 u/(2l) Hz apart, that is, Re() u/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 gure. 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.
0
0.5
1
1
0
1
1
0
1
1
0
1
x l
y R
z
R
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
spin direction
0
0.5
1
1
0
1
1
0
1
1
0
1
x l
y R
z
R
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
spin direction
0
0.5
1
1
0
1
1
0
1
1
0
1
x l
y R
z
R
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
spin direction
0
0.5
1
1
0
1
1
0
1
1
0
1
x l
y R
z
R
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
spin direction
a)
b)
/
/
/
/
/
/
/
/
/
/
/
/
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.
A. P. DOWLING AND S. R. STOW 394
plenum
combustor
fuel injection premix duct and swirler unit
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. Atypical geometry is illustrated in Fig. 13.10.
The high-speed ow 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 ow 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 sufciently 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 owis nearly choked, which leads
to inlet boundary conditions for the linear waves of frequency (as discussed in
Sec. II.B). The relative wave strengths at AA
2
u
2
= A
1
1
u
1
(13.67a)
H
2
= H
1
(13.67b)
A
2
p
2
+ A
2
2
u
2
2
= A
2
p
1
+ A
1
1
u
2
1
(13.67c)
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 397
where subscripts 1 and 2 denote the ow parameters and areas before and after
the area change, respectively. Here the pressure on the abrupt expansion has been
taken to be p
1
; however, some pressure recovery could be included through the use
of a loss coefcient. To nd 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 ow.
An area decrease can be assumed to be isentropic, hence,
p
2
/
2
= p
1
/
1
(13.68)
and conservation of mass and energy give Eqs. (13.67a) and (13.67b) as before.
For no mean ow, the jump conditions at any area change simplify to
[ p]
2
1
= [Au]
2
1
= 0 (13.69)
The ame is also treated as compact, and so Eq. (13.10) applies across it.
However, this approach needs a ame model relating the instantaneous rate of heat
release to the oncoming ow. Flame models are discussed in Chapter 12, but here
we note that they can be determined either by analytical descriptions of the ame
dynamics
9, 39
or through numerical
36
or experimental investigations
4042
of the un-
steady combustion response to inlet ow disturbances. Measurements carried out
at lowand high pressure have remarkably similar forms
42
but different amplitudes,
supporting the idea that the ame-transfer function can be investigated by suitably
scaledexperiments or throughlocal computational uiddynamics (CFD) solutions.
C. Combustor
Once the uxes 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 ow at exit. For a general
value of frequency , this will not satisfy the downstream boundary condition.
The resonant frequencies are the values of at which the downstream boundary
condition is satised.
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 ame model,
Q
= k
m
i
m
i
e
i
(13.70)
is used at the start of the combustor, where m
i
is the air-mass ow 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 sufcient blockage that it acts approximately like a hard
A. P. DOWLING AND S. R. STOW 398
Table 13.1 Geometry and ow conditions for simple
combustor (based on an atmospheric test rig)
Description Value
Choked inlet, mass ow rate 0.05 kg s
1
Choked inlet, temperature 300 K
Plenum, cross-sectional area 0.0129 m
2
Plenum, length 1.7 m
Premixer, cross-sectional area 0.00142 m
2
Fuel-injection point, fuel-convection time 0.006 s
Premixer, length 0.0345 m
Combustor, cross-sectional area 0.00385 m
2
Flame zone, temperature after combustion 2000 K
Combustor, length 1.0 m
Open outlet, pressure 101,000 Pa
frequency (Hz)
g
r
o
w
t
h
r
a
t
e
,
I
m
(
s
1
)
0 50 100 150 200 250 300 350 400 450 500 550 600
450
400
350
300
250
200
150
100
50
0
50
100
150
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.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 399
end (u
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
5 1 0. 1.5 2 2.5 3
0
0.25
0.5
0.75
1
x (m)
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
x (m)
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
x (m)
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
x (m)
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
x (m)
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
x (m)
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
x (m)
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
x (m)
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
p
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
x (m)
0 0.5 1 1.5 2 2.5 3
0
0.25
0.5
0.75
1
a) b)
c) d)
e) f)
g) h)
i) j)
Fig. 13.14 Mode shapes for simple combustor, k=1. a) 30-Hz mode. b) 104-Hz mode.
c) 168-Hz mode. d) 203-Hz mode. e) 300-Hz mode. f) 312-Hz mode. g) 396-Hz mode.
h) 415-Hz mode. i) 495-Hz mode. j) 514-Hz mode.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 401
be found as before. At the inlet to the combustor, the ring of phase-shifted one-
dimensional disturbances creates a circumferential wave of an order identical with
that in the plenum. The resonant modes for circumferential waves of this selected
order can then be found by investigating the propagation of this circumferential
mode through the combustor and determining the resonant frequencies at which
the downstream boundary condition is satised.
VI. Modal Coupling
In uniformcylindrical and annular ducts, the solutions inEq. (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 ow, only acoustic waves are present, and
so from Eq. (13.56) for a given n, we may write for x < 0
p
= e
it +in
m=1
_
A
+(1)
n, m
e
ik
+
n, m
x
+ A
(1)
n, m
e
ik
n, m
x
_
B
n, m
(r) (13.71a)
and for x > 0
p
= e
it +in
m=1
_
A
+(2)
n, m
e
ik
+
n, m
x
+ A
(2)
n, m
e
ik
n, m
x
_
B
n, m
(r) (13.71b)
with similar expressions for the other owvariables. Miles
43
and Alfredson
44
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
, v
, and w
followfromcontinuity of p
),
and on the rigid wall b
(1)
r b
(2)
we require u
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.2 0 4 0.6 0.8 1
0
0.25
0.5
0.75
1
1.25
u
(
a
r
b
i
t
r
a
r
y
s
c
a
l
e
)
0 0.2 0 4 0.6 0.8 1
0
0.25
0.5
0.75
1
1.25 a) b)
/
( )
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 Polifke
46
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 narrowannular 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 nding linear
resonances in problems with modal coupling. We write p
(t, x, ) = p(x, )e
it
with p(x, ) =
n=
p
n
(x)e
in
, and similarly for the other variables. For |n|
large (for example, |n| > N), the mode will be highly cut off; hence, in a way sim-
ilar to the radial modes earlier, we approximate circumferential modes by taking
p(x, ) =
N
n=N
p
n
(x)e
in
. At the inlet of the geometry, there are boundary con-
ditions that apply to each mode independently. These dene the perturbations for
each circumferential mode n except for an unknown parameter
n
. For instance, if
it is an open end, p = 0 for all , implying that p
n
= 0 for all n, and so we may set
A
+
n
= A
n
=
n
(with no entropy or vorticity waves). Here, = [
N
, . . . ,
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 dene
n
to be the error in this
boundary condition for circumferential mode n; for example, for an open end we
may take
n
= p
n
. Given and , all the circumferential components at the inlet
are known. We can step through the geometry calculating all the circumferential
modes at each sec. before continuing to the next. For the solution thus found, each
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 403
mode will have an error at the outlet
n
. We must nd and to satisfy
n
= 0,
thus, giving a resonance of the geometry. For a given , we dene the matrix M
to be such that M
n, m
is the value of
n
for the solution with
i
=
i, m
. For a gen-
eral ,
n
= M
n, m
m
because the perturbations are linear. Hence, for the correct
values of and , M = 0. For a solution to exist, = 0, and so this implies that
det M = 0. Thus the procedure to nd a complex resonant frequency is to rst
guess the value of and calculate the matrix M and then iterate the value of to
achieve det M = 0. For this value of , a will exist giving M = 0. Finally, this
correct is calculated by using an inverse iteration method (M
new
=
old
). The
mode shape for the resonance can then be calculated using this . As before, the
resonant frequency and growth rate are given by .
VII. Acoustic Absorbers
Passive control of combustion instability can signicantly 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 massspring 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 uctuating pressure at the neck entrance
frequency (Hz)
s
o
u
n
d
p
r
e
s
s
u
r
e
l
e
v
e
l
(
d
B
)
0 100 200 300 400 500 600 700 800 900 1000
110
120
130
140
150
160
170
Fig. 13.16 Power spectra of experimental results for anadjustable-volume Helmholtz
resonator: , resonator volume is minimum, hence, damping is negligible; and
, resonator volume to suppress oscillations.
A. P. DOWLING AND S. R. STOW 404
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
signicant kinetic energy is dissipated in the unsteady jets that form. Specically,
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 ow through the combustor
and hence across the neck of the Helmholtz resonator, and the requirement to cool
the resonator may lead to an additional ow through the neck; these ows 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 conguration 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 sufciently strong cooling ow 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 upstreamof the resonator and the point just down-
stream. However, one needs to account for the mass ow perturbation m entering
the duct through the neck of the resonator. In no mean ow, as stated previously,
nonlinear effects are important and so m 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 crossow and/or neckow, the system remains linear and the calculation
techniques described previously can be applied directly. In this case, writing p
1
and p
2
for the pressure perturbations in the combustor and inside the resonator,
respectively, we may dene the Rayleigh conductivity of the neck by
=
i m
p
2
p
1
(13.72)
The rate of decrease of mass inside the resonator must equal m and so, assuming
isentropic conditions there, we have m = iV p
2
/ c
2
, where V and c are the
volume of the resonator and the speed of sound inside, respectively. If the mean
ow through the neck is much larger than the crossow, 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 ow is across the neck,
the result by Howe
53
is applicable. The case in which the cross- and throughows
are comparable has been considered by Dup` ere and Dowling.
54
Details on the
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 405
modeling of Helmholtz resonators with and without mean ows and a discussion of
practical issues for their use in real combustors are given by Dup` ere and Dowling.
55
The placement of Helmholtz resonators can be an important consideration. For
maximumeffect, 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 gure.]
Liners with bias ow 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. Eldredge
58
extended their
analysis to higher-order modes. The conguration 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 ow
passes into the duct. This ow 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 ow. 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 owthrough 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 ow; 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;
A. P. DOWLING AND S. R. STOW 406
frequency (Hz)
g
r
o
w
t
h
r
a
t
e
,
I
m
(
s
1
)
0 50 100 150 200 250 300 350 400 450 500 550 600
150
100
50
0
50
100
150
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 outow
of 10 ms
1
. (The crossowin 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
m
3
. 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 nd 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 signicant 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
ow 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
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 407
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 uctuating 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
uctuations 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 Dowling
16
). The main effect determining the limit-
cycle amplitude is likely to be a saturation of the heat-release oscillation from the
ame. Consider a ame being forced by a time-varying input (such as the air-mass
ow at the fuel-injection point m
i
) at a single frequency. The heat release from
the ame 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 ame
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 ame can be characterized by a nonlinear ame-
transfer function relating the heat-release component at the forcing frequency to
the owperturbation as a function of both frequency and amplitude of forcing. For
example,
Q
= T(, A)
m
i
m
i
(13.73)
where T is the nonlinear transfer function and A = | m
i
|/ m
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 signicant. We have
already seen in example 5 that decreasing the gain of the ame-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 in the linear problem, the solution now
amounts to nding real and A such that the exit boundary condition is satised.
This solution gives the limit-cycle frequency, amplitude, and dimensional mode
shape.
A. P. DOWLING AND S. R. STOW 408
We nowgive a simple illustrative example of a nonlinear ame-transfer function.
Written in the time domain, the linear ame model in Eq. (13.70) becomes
Q
L
(t ) = k
m
i
(t )
m
i
Q (13.74)
The subscript L denotes that this a linearized result. One could consider nonlinear
effects on the equivalence ratio, convection time, and ame response (see Stow
and Dowling
60
); however, here we will assume a simple saturation in Q(t ):
Q
(t ) =
_
Q
L
(t ) for |Q
(t )|
Q
Q sign
_
Q
L
(t )
_
for |Q
(t )| >
Q
(13.75)
This nonlinear ame 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(t ) m
i
, the transfer
function is found by calculating the component of Q
(t ) at frequency ,
T(, A) =
A
Q
_
2/
0
Q
(t )e
it
dt (13.76)
In this model, for A /|k| no saturation occurs and so the transfer function
is the same as the linearized version, that is, T(, A) = T
L
() = ke
i
. For
A > /|k|, it can be shown that
T(, A) =
2e
i
A
__
/
/
e
it
dt +
_
()/
/
|k| A cos(t )e
it
dt
_
(13.77)
with = cos
1
(1/), where = |k| A/ is a scaled amplitude parameter. Evalu-
ating these integrals and combining with the low-amplitude result gives
T(, A)/T
L
() =
_
_
_
1 for 1
1
2 cos
1
(1/)
+
2(1 1/
2
)
1/2
for > 1
(13.78)
Figure 13.18 shows the variation of T/T
L
with . We see that for A > /|k|
the ame 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 ame model. The describing-function approach is not
strictly applicable unless there is a single dominant frequency in the limit cycle,
so it is questionable to use the technique for this case. However, if instead we take
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 409
T
T
L
0 1 2 3 4 5 6 7 8 9 10
0
0 2
0 4
0 6
0 8
1
1 2
Fig. 13.18 Variation of nonlinear ame-transfer function (normalized by linear
value) with amplitude parameter =|k| A/.
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 ame model
in Eq. (13.78) with set to be 0.1, we nd 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 ow 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 ow effects are not signicant for Mach numbers less than about 0.2;
x (m)
p
(
P
a
)
0 0 25 0 5 0 75 1 1 25 1 5 1 75 2 2 25 2 5 2 75
0
1000
2000
3000
4000
Fig. 13.19 Limit-cycle mode shape for simple combustor.
A. P. DOWLING AND S. R. STOW 410
however, a mean ow 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 sufcient 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 congurations 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)
simplies to
T
DS
Dt
= q (13.A1)
Replacing S by the perfect gas form S = c
v
log p c
p
log , we obtain
Tc
v
p
Dp
Dt
= c
p
T
D
Dt
+q (13.A2)
After substitution for D/Dt from the equation of mass conservation, we obtain
c
v
R
gas
Dp
Dt
= c
p
T u +q (13.A3)
which is equivalent to
Dp
Dt
= c
2
u +( 1)q (13.A4)
since c
2
= R
gas
T and R
gas
/c
v
= 1. Equation (13.41) is the linearized form
of Eq. (13.A4).
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 411
References
1
Dowling, A. P., andStow, S. R., Acoustic Analysis of Gas Turbine Combustors, Journal
of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 751764.
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 Summereld, M. (eds.), Nonsteady Burning and Com-
bustion Stability of Solid Propellants, Vol. 143, Progress in Astronautics and Aeronautics,
AIAA, New York, 1992.
5
Keller, J. J., Thermoacoustic Oscillations in Combustion Chambers of Gas Turbines,
AIAA Journal, Vol. 33, No. 12, 1995, pp. 22802287.
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., andShah, S. N., Modelingof CombustionChamber 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. 317332.
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. 225243.
13
Stow, S. R., Dowling, A. P., and Hynes, T. P., Reection of Circumferential Modes in
a Choked Nozzle, Journal of Fluid Mechanics, Vol. 467, Sept. 2002, pp. 215239.
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. 222228.
15
Culick, F. E. C., and Rogers, T., The Response of Normal Shocks in Diffusers, AIAA
Journal, Vol. 21, No. 10, 1983, pp. 13821390.
16
Dowling, A. P., A Kinematic Model of a Ducted Flame, Journal of Fluid Mechanics,
Vol. 394, Sept. 1999, pp. 5172.
17
Culick, F. E. C., and Yang, V., Overview of Combustion Instabilities in Liquid-
Propellant Rocket Engines, Liquid Rocket Engine Combustion Instability, edited by
V. Yang, and W. E. Anderson, Vol. 169, Progress in Astronautics and Aeronautics, AIAA,
New York, 1995, pp. 337.
18
Dowling, A. P., The Calculation of Thermoacoustic Oscillations, Journal of Sound
and Vibration, Vol. 180, No. 4, 1995, pp. 557581.
19
Annaswamy, A. M., Fleil, M., Hathout, J. P., and Ghoniem, A. F., Impact of Linear
Coupling on Design of Active Controllers for Thermoacoustic Instability, Combustion
Science and Technology, Vol. 128, No. 16, 1997, pp. 131160.
A. P. DOWLING AND S. R. STOW 412
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. 5567.
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. 459473.
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. 389402.
23
Kumar, B, M., and Sujith, R. I., Exact Solution for One-dimensional Acoustic Fields
in Ducts with Polynomial Mean Temperature Proles, Journal of Vibration and Acoustics,
Vol. 120, No. 4, 1998, pp. 965969.
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. 3843.
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. 169175.
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. 705715.
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. 779786.
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. 101163.
31
Tyler, J. M., and Sofrin, T. G., Axial Compressor Noise Studies, SAE Transactions,
Vol. 70, No. 31, 1962, pp. 309332.
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. 239247.
33
Ohtsuka, M., Yoshida, S., Inage, S., and Kobayashi, N., Combustion Oscillation Anal-
ysis of PremixedFlames at ElevatedPressures, AmericanSocietyof Mechanical Engineers,
Paper 98-GT-581, June 1998.
34
Hobson, D. E., Fackrell, J. E., and Hewitt, G., Combustion Instabilities in Industrial
Gas TurbinesMeasurements on Operating Plant and Thermoacoustic Modeling, Journal
of Engineering for Gas Turbines and Power, Vol. 122, No. 3, 2000, pp. 420428.
35
Dowling, A. P., Thermoacoustic Instability, 6th International Congress on Sound and
Vibration, July 1999, pp. 32773292, http://icsv6.dat.dtu.dk.
36
Kr uger, U., H urens, J., Hoffmann, S., Krebs, W., and Bohn, D., Prediction of Ther-
moacoustic Instabilities with Focus on the Dynamic Flame Behavior for the 3A-Series Gas
Turbine of Siemens KWU, American Society of Mechanical Engineers, Paper 99-GT-111,
June 1999.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 413
37
Kr uger, U., H urens, 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. 785808.
41
Krebs, W., Hoffmann, S., Prade, B., Lohrman, M., and B uchner, 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 Reection 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. 1419.
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. 433442.
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., andPolifke, 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. 269274.
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. 271275.
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 Orices, AIAA Journal,
Vol. 22, No. 6, 1984, pp. 786792.
52
Howe, M. S., Scott, M. I., and Sipcic, S. R., The Inuence of Tangential Mean Flow
on the Rayleigh Conductivity of an Aperture, Proceedings of the Royal Society of London
Series A, Vol. 452, 1996, pp. 23032317.
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. 205223.
A. P. DOWLING AND S. R. STOW 414
54
Dup` ere, I. D. J., andDowling, A. P., The Absorptionof SoundbyHelmholtz Resonators
with and Without Flow, AIAA Paper 2002-2590, June 2002.
55
Dup` ere, 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. 268275.
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. 307335.
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. 303331.
59
Culick, F. E. C., Some Recent Results for Nonlinear Acoustics in Combustion Cham-
bers, AIAA Journal, Vol. 32, No. 11, Jan. 1994, pp. 146169.
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 Modelingof a Gas Turbine Combustor EquippedwithAcoustic 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
Professor. Member AIAA. Present address: Beijing University of Aeronautics and Astronautics,
Beijing, China.
415
D. YOU AND V. YANG 416
R
h
= hub radius of annular chamber
R
p
R
u
= combustion response coefcient, 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
x
o
= x coordinate at cell interface
Y
m
= mth-order Bessel Function of the second kind
Greek Symbols
= axial wave number
a
= reection coefcient, Eq. (14.66)
s
= reection coefcient, Eq. (14.68)
= density of two-phase mixture
= Kronecker delta
= normal mode function
= circumferential coordinate
= series coefcient of Fourier-type expansion, Eq. (14.35)
= specic heat ratio for mixture
= frequency
= rate of entropy generation
Overscripts and Superscripts
= mean quantity
= perturbation quantity
= uctuation 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 ow oscillations in combustion chambers, has been
a serious concern in the development of gas-turbine engines, as discussed in Chap-
ters 19 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 ow and ame
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 existinganalytical models for treatingcombustioninstabilities are basedon
a wave equation of some kind that characterizes the oscillatory oweld in a cham-
ber. The equation, along with its boundary conditions, is then solved by using either
Greens function
1, 2
or Galerkin
35
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 ow property variations and Mach number are
assumed to be small. Under such conditions, the frequencies and spatial variations
of unsteady motions deviate only slightly fromthe classical acoustic eld obtained
for the same geometry but lack any source terms. The acoustic eld 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 ow 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 ow 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 congurations and nonuniform ow 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
D. YOU AND V. YANG 418
compressor
acoustic damper
primary air
cooling air
swirler
fuel
nozzle
diffuser shroud
unsteady
combustion
flow
oscillation
s
turbine
vane
diffuser
Fig. 14.1 Schematic of a gas turbine combustor.
for the effects of geometric and ow 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 ow 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 owelds 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-dened 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 ow 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, ow 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:
t
+u = W (14.1)
u
t
+u u = p +F (14.2)
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 419
p
t
+u p = p u +P (14.3)
where u and p are the velocity and pressure of the gas phase, respectively but
where and 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 uctuating quantity:
u(r, t ) = u(r) +u
(r, t )
(r, t ) = (r) +
(r, t ) (14.4)
p(r, t ) = p(r) + p
(r, t )
The densityuctuationcontains twocomponents: anisentropic 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 ow. Thus, from the equation of
state, we have
d =
_
p
_
s
dp +
_
s
_
p
ds (14.5)
After some straightforward manipulations to rst-order approximation, the density
uctuation can be written as
=
1
a
2
p
C
p
s
(14.6)
where C
p
is the constant-pressure specic heat of the mixture. Similarly, the tem-
perature uctuation can be written as
T
=
1
C
p
p
+
a
2
( 1) C
2
p
s
(14.7)
Substituting the decomposed variables into Eqs. (14.2) and (14.3) and linearizing
the result yields
p
=
u
t
( u )u
(u
) u
( u ) u +F
(14.8)
p
t
+ u p
+u
p = ( p u
+ p
u) +P
(14.9)
We differentiate Eq. (14.9) with respect to time and substitute Eq. (14.8) for u
_
t
to nd the wave equation governing the oscillatory eld in a two-phase mixture:
2
p
1
a
2
2
p
t
2
= h (14.10)
D. YOU AND V. YANG 420
where the inhomogeneous term h has the following form:
h = h
I
+h
I I
+h
I I I
(14.11)
And where
h
I
=
_
u
t
_
[ ( u )u
] [ (u
) u]
_
p
a
2
( u ) u
_
+
1
a
2
t
( u p
) +
1
a
2
t
(u
p) (14.11a)
+
a
2
t
( p u
) +
1
a
2
t
( p
u)
h
I I
=
_
s
C
p
( u ) u
_
(14.11b)
h
I I I
=
1
a
2
t
P
+F
(14.11c)
The rst term, h
I
, represents the linear gasdynamic effect. The second term, h
I I
,
arises from entropy uctuations. 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 eld 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
= n
_
u
t
( u )u
(u
) u
( u ) u +F
_
(14.12)
It can be treated conveniently by using an acoustic admittance function, A
d
, which
relates the local velocity uctuation to its pressure counterpart:
A
d
= n
u
/ a
p
/ p
(14.13)
If all perturbations are absentwith both functions h and f vanishedthe wave
equation for classical acoustics in a closed volume having a rigid wall is recovered.
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 421
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
ow 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-ow properties can be taken as uniform within each cell. Furthermore,
because the mean owMach number and variations of owproperties 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 eld obtained for the same cell geometry, but without
any source terms. The acoustic eld in each cell can be synthesized as a Fourier-
type series in terms of the eigenfunctions for the transverse plane , along with
temporal and axial variations. In the cylindrical coordinates, the expansion can be
expressed as
p
(r, t ) =
n=0
m=
[
mn
(, r)
mn
(x, t )] (14.14)
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
coefcients,
mn
(x, t ).
The eigenfunction, also called the normal mode, satises the Helmholtz equation
in the transverse plane,
2
T
mn
+k
2
mn
mn
= 0 (14.15)
subject to the following boundary condition for a rigid surface along the combustor
wall.
n
T
mn
= 0 (14.16)
D. YOU AND V. YANG 422
where k
mn
is the wave number. The transverse Laplacian operator
2
T
in the cylin-
drical coordinates is dened as
2
T
=
1
r
r
_
r
r
_
+
1
r
2
2
(14.17)
The eigenfunction
mn
can be constructed to be orthonormal:
__
mn
n
ds =
mm
nn
(14.18)
where the integral is performed throughout the entire cross section and where
denotes the Kronecker delta function.
For an annular duct with the inner and outer radii of R
t
and R
h
, respectively,
the eigenfunction can be expressed as
mn
(, r) =
1
2
exp(i m)[C
Jmn
J
m
(k
mn
r) +C
Ymn
Y
m
(k
mn
r)] (14.19)
where
C
Jmn
=
_
R
2
t
2
_
1
m
2
k
2
mn
R
2
t
_
B
2
m, t
R
2
h
2
_
1
m
2
k
2
mn
R
h
_
B
2
m, h
_
1/2
(14.20)
and where
C
Ymn
= C
Jmn
_
dJ
m
(k
mn
r)
_
dr
dY
m
(k
mn
r)
_
dr
_
r=R
t
(14.21)
The coefcients B
m, t
and B
m, h
are
B
m, t
= J
m
(k
mn
R
t
) +Y
m
(k
mn
R
t
)
_
dJ
m
(k
mn
r)
_
dr
dY
m
(k
mn
r)
_
dr
_
r=R
t
B
m, h
= J
m
(k
mn
R
h
) +Y
m
(k
mn
R
h
)
_
dJ
m
(k
mn
r)
_
dr
dY
m
(k
mn
r)
_
dr
_
r=R
t
(14.22)
The eigenvalues k
mn
can be determined by applying the boundary condition,
Eq. (14.16), at both the inner and outer walls:
_
dJ
m
(k
mn
r)
dr
r=R
t
dY
m
(k
mn
r)
dr
r=R
h
_
_
dJ
m
(k
mn
r)
dr
r=R
h
dY
m
(k
mn
r)
dr
r=R
t
_
= 0 (14.23)
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 423
For a circular duct with radius R
c
, the transverse eigenvalue takes the form
mn
(, r) =
1
2
exp(i m)C
Jmn
J
m
(k
mn
r) (14.24)
where
C
Jmn
=
_
R
2
c
2
_
1
m
2
k
2
mn
R
2
c
_
J
2
m
(k
mn
R
c
)
_
1/2
(14.25)
The eigenvalue k
mn
can be obtained from the following wall condition:
d
dr
J
m
(k
mn
r) = 0 at r = R
c
(14.26)
After obtaining the eigenfunction
mn
, a spatial-averaging technique equivalent
to the Galerkin method is applied to the transverse plane to solve for the series co-
efcient
mn
. By multiplying Eq. (14.15) by p
2
T
mn
+ p
k
2
mn
mn
mn
2
p
+
mn
1
a
2
2
p
t
2
_
ds =
__
mn
hds
(14.27)
Applying Greens theoremand substituting boundary conditions (12) and (16) into
Eq. (14.27) yields
__ _
p
k
2
mn
mn
mn
2
p
x
2
+
mn
1
a
2
2
p
t
2
_
ds =
__
mn
hds
_
mn
f
T
dl
(14.28)
where f
T
= n
T
p
mn
2
mn
x
2
+
1
a
2
mn
t
2
_ __
2
mn
ds =
__
mn
hds
_
mn
f
T
dl
(14.29)
For linear stability analysis, each uctuating quantity can be decomposed into a
spatial and a time-harmonic temporal part, i.e.,
mn
(x, t ) =
mn
(x) exp(i t ), h (r, t ) =
h (r) exp(i t ), etc. (14.30)
D. YOU AND V. YANG 424
The overhat
2
a
2
k
2
mn
_
mn
=
___
mn
hds +
_
mn
f
T
dl
_
(14.32)
The source terms
h and
f
T
are functions of both the mean and oscillatory ow
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 specic 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
mn
and coefcients C
h, mn
and C
f, mn
, which incorporate all the distributed and surface effects at a given cross
section through spatial averaging:
__
mn
hds = C
h, mn
mn
(x)
_
mn
f
T
dl = C
f, mn
mn
(x) (14.33)
Equation (14.32) then reduces to a second-order ordinary differential equation
with constant coefcients, whose solution
mn
(x) takes the form
mn
(x) = p
+
mn
exp
_
i
+
mn
x
_
+ p
mn
exp
_
i
mn
x
_
(14.34)
where p
mn
and p
+
mn
are the complex amplitudes of the upstream and downstream
traveling waves, respectively. The axial wave number
mn
is related to the fre-
quency , eigenvalue k
mn
, and source-term coefcients:
_
mn
_
2
=
2
a
2
k
2
mn
C
h, mn
_
mn
,
_
C
f, mn
_
mn
,
_
(14.35)
So far, by means of normal-mode expansion, the solution of the wave equation
has been given in the formof 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
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 425
each cell can thus be explicitly expressed by combining these results:
p
(r, t ) = e
i t
n=0
m=
_
mn
(, r)
_
p
+
mn
e
i
+
mn
x
+ p
mn
e
i
mn
x
__
(14.36)
The axial velocity uctuation u
(r, t ) =
1
n=0
m=
_
e
i t
mn
(, r)
_
+
mn
p
+
mn
e
i
+
mn
x
+ u
+
mn
+
mn
p
mn
e
i
mn
x
+ u
mn
__
(14.37)
Similarly, the radial and circumferential velocity uctuations are derived by sub-
stituting Eq. (14.36) into the linearized momentum equations in the radial and
circumferential directions, respectively:
v
(r, t ) =
e
i t
n=0
m=
_
i
mn
(, r)
r
_
p
+
mn
e
i
+
mn
x
+ u
+
mn
+
p
mn
e
i
mn
x
+ u
mn
__
(14.38)
w
(r, t ) =
e
i t
r
n=0
m=
_
m
mn
(, r)
_
p
+
mn
e
i
+
mn
x
+ u
+
mn
+
p
mn
e
i
mn
x
+ u
mn
__
(14.39)
In addition to the acoustic eld, entropy uctuation arising in the ame zone
must be treated with care. The following transport equation is used:
Ds
Dt
=
q
T
+ (14.40)
The rst termon the right-hand side represents the rate of entropy change caused by
the heat ux, q. The second termstands for irreversibilities, which can be modeled
as follows if we only consider the entropy generated from heat release,
Q:
=
Q
T
(14.41)
Following the procedure described in Ref. 7, we obtain the equation governing the
transport of entropy oscillation:
s
= exp(i t ) exp(i x/ u)
_
m=
n=0
__
s
mn
+C
s, mn
_
mn
(r, )
_
_
(14.42)
D. YOU AND V. YANG 426
where s
mn
is the amplitude of the entropy disturbance and where C
s, mn
is a co-
efcient associated with the unsteady heat release represented by pressure and
velocity perturbations
C
s, mn
(x, r, ) =
Q
T
x
u
__
R
p
p
+
mn
R
u
a(+ u
+
mn
)
_
e
i
+
mn
x
p
mn
+
_
R
p
p
mn
R
u
a(+ u
mn
)
_
e
i
mn
x
p
mn
_
(14.43)
The interaction between entropy uctuation and mean-ow 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
h and
f
T
in Eq. (14.32) and subsequently the coefcients 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 specic 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 uctuating quantities. The results are then
substituted into Eq. (14.33) to determine the coefcients, C
f, mn
and C
h, mn
, so as to
facilitate model closure. As examples, the source terms resulting from mean ow,
combustion heat release, and boundary conditions are formulated below.
1. Effect of Mean Flow
If a uniform mean ow in the axial direction is the only one considered, the
inhomogeneous term in the wave equation, Eq. (14.10), becomes
h =
2 u
a
2
2
p
xt
+
u
2
a
2
2
p
x
2
(14.44)
Substituting of Eq. (14.44) into Eq. (14.33) leads to
C
h, mn
=
__
2
mn
_
2
M
a
mn
M
2
_
mn
_
2
_
ds (14.45)
2. Effect of Combustion Heat Release
The second case is concerned with combustion heat addition, expressed as
h = ( 1) i
Q
/ a
2
(14.46)
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 427
Following common practice, the oscillatory heat-release rate can be conveniently
related to local pressure and velocity uctuations, as follows:
Q
= R
p
p
p
+ R
u
u
a
+ R
v
v
a
+ R
w
w
a
(14.47)
where R
p
and R
u
, R
v
, and R
w
are complex variables commonly referred to as the
pressure- and velocity-coupled response functions, respectively. Substituting the
oscillatory ow properties, Eqs. (14.36)(14.39), into Eq. (14.47) and applying
Eq. (14.33) yields
C
h, mn
=
__
mn
G
h, mn
ds (14.48)
where
G
hmn
=
i ( 1)
Q
a
2
_
R
p
p
mn
R
u
mn
mn
a(+ u
mn
)
+
iR
v
a(+ u
mn
)
mn
r
mR
w
mn
ar(+ u
mn
)
_
(14.49)
The formulation is thus closed after those combustion response functions are estab-
lished. Several empirical and analytical models, including the time-lag
9
and ame
response
10, 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 ow inuence
is ignored, Eq. (14.12) reduces to
f
T
=
v
t
= (i ) v (14.50)
where v
j
s
j
j-1
1 j
p
j
p
+
j
s
j
p
R
c,j
R
c,j-1
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 coefcient
C
f, mn
:
C
f, mn
=
_
_
2
mn
i A
d
/ a
_
dl (14.52)
E. Matching Conditions
The oscillatory eld 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 uctuating quantities on both sides of the interface at x
0
between
two neighboring cells having different cross-sectional areas. The matching condi-
tions require continuities of mass, momentum, and energy uxes over the region
0 r R
c, j 1
. It is assumed that R
c, j 1
R
c, j
without loss of generality.
Mass ux:
(u)|
j 1, x
0
= (u)|
j, x
+
0
(14.53)
Momentum ux:
_
p +u
2
_
j 1, x
0
=
_
p +u
2
_
j, x
+
0
(14.54)
Energy ux:
_
C
p
T +
1
2
u
2
_
j 1, x
0
=
_
C
p
T +
1
2
u
2
_
j, x
+
0
(14.55)
Because no mass owis allowed to pass through the solid region R
c, j 1
r R
c, j
at the interface,
(u)|
j, x
+
0
= 0, R
c, j 1
r R
c, j
(14.56)
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 429
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 p
+
mn
, p
mn
, and s
mn
and corresponding coefcients. The deriva-
tion is achieved by 1) decomposing the variables in Eqs. (14.53)(14.56) into the
mean and uctuating parts; 2) linearizing the results; 3) substituting the uctuating
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:
_
=0
_
A
+
m
p
+
m
+ A
_
+ A
sm
n
s
m
_
j 1
+
_
p=
q=0
_
A
+
pq
p
+
pq
+ A
pq
p
pq
_
+ A
spq
s
pq
_
j
= 0 (14.57)
The explicit expressions of the coefcients in the preceding equation can be found
in Ref. 7.
Equation (14.57) indicates that the interfacial matching conditions contain in-
nite summations of unknown acoustic and entropy wave amplitudes, p
+
mn
, p
mn
,
and s
mn
. In practice, a nite number of modes is sufcient 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
mn, j 1
and integrating the result over the cross-sectional area
S
j 1
. This operation yields
__
S
j 1
_
M1
=
N1
=0
_
A
+
m
p
+
m
+ A
_
+ A
sm
n
s
m
_
j 1
mn, j 1
ds
+
__
S
j 1
_
M1
p=
N1
q=0
_
A
+
pq
p
+
pq
+ A
pq
p
pq
_
+ A
spq
s
pq
_
j
mn, j 1
ds =0 (14.58)
As a result of the orthonormal property of the transverse eigenfunctions,
Eq. (14.18), the double summation of m
and n
mn
p
mn
_
+ B
smn
s
mn
_
j 1
+
__
S
j 1
_
M1
p=
N1
q=0
_
A
+
pq
p
+
pq
+ A
pq
p
pq
_
+ A
spq
s
pq
_
j
mn, j 1
ds = 0
m = 0, 1, . . . , M 1
n = 0, 1, . . . , N 1
(14.59)
Equation (14.59) can be rearranged in the following form:
__
B
+
mn
p
+
mn
+ B
mn
p
mn
_
+ B
smn
s
mn
_
j 1
+
_
M1
p=
N1
q=0
_
B
+
pq, mn
p
+
pq
+ B
pq, mn
p
pq
_
+ B
s, pq, mn
s
pq
_
j
= 0
m = 0, 1, . . . , M 1
n = 0, 1, . . . , N 1
(14.60)
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
mn, j
and integrating the result over the cross-sectional area S
j
, yielding
_
M1
=
N1
n=0
_
B
+
m
, mn
p
+
m
+ B
, mn
p
_
+ B
s, m
, mn
s
m
_
j 1
+
__
B
+
mn
p
+
mn
+ B
mn
p
mn
_
+ B
smn
s
mn
_
j
= 0,
m = 0, 1, . . . , M 1
n = 0, 1, . . . , N 1
(14.61)
Consequently, 3MN matching conditions are obtained fromthe 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 A
d
, which quanties the magnitude and direction
of the energy ow across the boundary, as evidenced in the expression for the
acoustic intensity, I :
I = p
= A
d
p
2
/( a) (14.62)
It is apparent that energy is delivered into the system if the pressure and velocity
uctuations are in phase, thereby exerting a destabilizing inuence. A more rig-
orous theory of linear combustion stability clearly shows that the growth rate and
frequency of oscillation depend on the real and imaginary parts of the admittance
function, respectively.
4
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 431
By substituting the expressions of pressure and velocity uctuations, Eqs.
(14.36) and (14.37), into Eq. (14.13) and manipulating the result, we obtain the
condition at the combustor entrance:
_
A
d
+ a
+
mn
+ u
+
mn
_
p
+
mn
+
_
A
d
+ a
mn
+ u
mn
_
p
mn
= 0 (14.63)
The entropy uctuation at the combustor inlet is assumed to be zero,
s
mn
= 0 (14.64)
At the combustor exit, the boundary conditions can be more conveniently spec-
ied by the reection coefcients. The acoustic wave reected from the boundary
consists of contributions fromthe incident acoustic andentropydisturbances. Thus,
the pressure of the reected wave takes the form
p
= p
a
+ p
s
=
a
p
+
+
s
s ( p)
_
C
p
(14.65)
where the acoustic reection coefcient is dened as
a
= p
a
/p
+
(14.66)
It can be related to the admittance function as follows:
a
=
_
1 +
a
+
mn
A
d
(+
+
mn
u)
___
1 +
a
mn
A
d
(+
mn
u)
_
(14.67)
The entropy reection coefcient
s
is dened as
s
=
_
p
s
_
p
___
s/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 reected acoustic
waves, as well as the entropy uctuation:
a
exp(i
+
mn
dx) p
+
mn
exp(i
mn
dx) p
mn
+
s
p
C
p
s
mn
exp (i dx/ u) = 0
(14.69)
where dx is the axial length of the boundary cell.
The acoustic admittance function, or the reection coefcient, 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 uctuation var-
nishes, and the admittance function becomes innity. If the boundary is rigid, the
vanished velocity uctuation 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 uctuating quantities in
the axial direction at the nozzle entrance (or combustor exit) satisfy the following
D. YOU AND V. YANG 432
relation.
8
2u
u
+
p
= 0 (14.70)
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 reection
coefcient at the nozzle entrance under a choked condition.
a
=
1 ( 1)
M/2
1 +( 1)
M/2
and
s
=
M/2
1 +( 1)
M/2
(14.71)
In the limit of zero Mach number,
a
and
s
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 p
+
mn
, p
mn
, and s
mn
(n = 0, 1, , N 1 and m = 0, 1, , M 1)
for the oscillatory oweld 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:
_
_
_
_
_
_
_
_
_
_
_
C
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
C
j 1
C
j
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
C
NT
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
D
1
.
.
.
D
j 1
D
j
.
.
.
D
NT
_
_
_
_
_
_
_
_
_
_
_
= 0 (14.72)
where C
j
is a matrix consisting of the coefcients 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 p
+
mn
, p
mn
, and s
mn
. To nd 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).
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 433
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) Dene the domain of concern.
2) Obtain the mean owproperties fromnumerical simulations or experimental
measurements.
3) Determine the acoustic-boundaryconditions at the combustor inlet andoutlet.
4) Discretize the combustor into a number of cells along the axial direction
according to the mean ow distribution and chamber geometry.
5) Calculate the bulk ow 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 elds 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 elds 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 oweld in a step duct with uniform mean
temperature, as shown in Fig. 14.4. The ratio of the cross-sectional area is four,
L
1
L
2
R
2
R
1
Fig. 14.4 Schematic of a chamber with sudden expansion.
D. YOU AND V. YANG 434
Table 14.1 Acoustic oscillations frequencies
of step duct (R
2
/R
1
= 2, L
1
= L
2
= R
1
)
Mode 1T 1T/1R
Frequency ANSYS 1.95 4.38
R
2
/ a 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 rst longitudinal (1L), the rst tangential (1T), and the
mixed rst tangential/rst radial (1T/1R) mode. If all source terms are absent and
if the mean-ow Mach number is ignored, the frequency of oscillation depends
only on the sound speed and chamber conguration. In other words, the acoustic
characteristics in each cell can be determined by the following Helmholtz equation,
subject to appropriate boundary conditions:
d
2
mn
dx
2
+
_
2
a
2
k
2
mn
_
mn
= 0 (14.73)
If the duct is acousticallyclosed, the frequencyof the 1Lmode canbe analytically
determined from the following equation:
R
2
1
tan (L
1
/ a) + R
2
2
tan (L
2
/ a) = 0 (14.74)
where L
1
, L
2
, R
1
, and R
2
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 nite
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 elds 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 a/R
2
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
mn
for a duct with an area variation can-
not be zero. For the 1T mode, the acoustic eld 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 R
2
(i.e., 1.84). This frequency yields a nontrivial complex axial
wave number in the small cell to satisfy the boundary condition, thereby leading to
an axially attenuated 1T wave. In other words, the 1T mode is cut off in the small
cell, in which the oscillation has a much lower amplitude, as shown in Fig. 14.5a.
Similarly, for the 1T/1R mode in the entire chamber, the prevalence of the wave
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 435
c) d)
a) b)
1.00
0.60
0.20
0.20
0.60
1.00
Fig. 14.5 Distributions of acoustic pressures in a step duct (R
2
/R
1
= 2, L
1
= L
2
=
R
1
, axial coordinate extended for better resolution): a) and b) rst tangential mode
from different views (R
2
/ a = 1.96); c) and d) rst tangential/rst radial mode from
different views (R
2
/ a = 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 eld
on a longitudinal plane along the centerline shown in Fig. 14.6. The transition
is inuenced 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 = T
1
for x L
1
and T = T
2
for
x > L
1
, as shown in Fig. 14.7. Other conditions remain identical to those in
the rst case. The calculation only involves two cells corresponding to the low-
and high-temperature regions. The frequency of the 1L mode can be analytically
D. YOU AND V. YANG 436
b)
1.00
0.60
0.20
0.20
0.60
1.00
a)
Fig. 14.6 Distributions of acoustic pressures in step duct (R
2
/R
1
= 2, L
1
= L
2
=
R
1
) showing mode transition: a) rst tangential mode (R
2
/ a = 1.96) and b) rst
tangential/rst radial mode (R
2
/ a = 4.40).
determined from the following equation:
a
1
/ tan (L
1
/ a
1
) + a
2
/ tan (L
2
/ a
2
) = 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 a
2
in a straight duct. Thus, the normalized frequency
/ ( a
2
k
mn
) becomes unity. As L
1
increases, the effect of the lower sound speed
a
1
becomes stronger. The frequency decreases and nally reaches its minimum
of a
1
/ a
2
as L
2
approaches zero. Similar to the step-duct case, mode transition
phenomena are observed in the acoustic pressure eld 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 upstreamwith
T
1
T
2
R
L
1
L
2
Fig. 14.7 Schematic of a straight chamber with temperature jump.
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 437
L
1
/(L
1
+ L
2
)
/
(
a
2
k
m
n
)
0 0.25 0.5 0.75 1
0.5
0.6
0.7
0.8
0.9
1
a
1
/ a
2
= 0.5
a
1
/ a
2
= 0.7
a
1
/ a
2
= 0.9
Fig. 14.8 Normalized frequency of rst tangential mode in a straight duct with tem-
perature jump.
1
0.2
0
1
0.2
X Y
Z
Fig. 14.9 Acoustic pressure eld of rst tangential mode in a straight duct with tem-
perature jump( a
1
/ a
2
= 0.5, L
1
= L
2
= R); cross-sections at x = 0(z > 0is blanked),
x = L, and slice z = 0.
swirl
injecto
r
combustion
chamber
38 mm
45 mm
choked
exit
235 mm
Fig. 14.10 Schematic of a swirl-stabilized combustor.
D. YOU AND V. YANG 438
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 specication 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 ow properties
can be acquired from the numerical simulation of the conservation equations by
using either the Reynolds-averaged NavierStokes (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 conguration. In case 1, the ame spreads
from the corner of the centerbody to the chamber wall. In case 2, the ame is
anchored by both the corner-recirculating ow and the center-recirculating ow
and forms a compact enveloped shape, which is in sharp contrast with the shape of
case 1.
The combustion responses of these two ames 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,
ame speed, and ame-surface area. Briey, the uctuation of the heat of the re-
action is attributed to changes in the mixture-equivalence ratio resulting from ow
disturbances. The density uctuation, mainly arising from pressure perturbation,
has a negligible effect on unsteady heat release, as compared with the other
224 mm
20 mm
2
0
.
3
m
m
9.53 mm
4
5
m
m
Fig. 14.11 Physical domain of a model combustor.
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 439
a)
b)
Fig. 14.12 Mean temperature contours and streamlines in swirl-stabilized com-
bustor
15,16
: a) simple ame (T
i n
= 600K,S = 0.76, = 0.57, p = 0.463 MPa); b) en-
veloped ame (T
i n
= 660K,S = 0.76, = 0.57, p = 0.463 MPa).
Axial coordinate (m)
N
o
r
m
a
l
i
z
e
d
a
m
p
l
i
t
u
d
e
0 0.1 0.2
0.0
0.5
1.0
1.5
2.0
2.5
pressure (experiment)
pressure
velocity
, Valter Bellucci
Chaired Professor.
Group Leader.
Senior Scientist.
P
e
2
P
e
1
(15.5)
By calculating the cross spectra between the microphone signals and a signal of a
third reference microphone, only the randomnoise will be averaged out. In this way
p
e, 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 =
P
e, s
2
H p
e, s
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:
_
f
e
d A
f
e
d B
g
e
u A
g
e
uB
_
=
_
SC
11
SC
12
SC
21
SC
22
__
f
e
u A
f
e
uB
g
e
d A
g
e
d B
_
(15.7)
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, . 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):
_
f
s
g
s
_
=
_
f
e, s
d
g
e, s
d
_
_
SC
11
SC
12
SC
21
SC
22
_
_
f
e, s
u
g
e, s
d
_
(15.8)
This equation is also valid if there is no forcing from the speakers (e = 0).
C. O. PASCHEREIT ET AL. 452
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:
_
_
f
e, s
d A
f
e, s
d B
f
e, s
dC
g
e, s
u A
g
e, s
uB
g
e, s
uC
_
_
=
_
SC
11
SC
12
SC
21
SC
22
_
_
_
f
e, s
u A
f
e, s
uB
f
e, s
uC
g
e, s
d A
g
e, s
d B
g
e, s
dC
_
_
+
_
f
s
g
s
_
(111)
(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
scatteringmatrixof a swirl-stabilizedpremixburner at certainoperatingconditions.
Arst 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
reection coefcient, duct with ow, measured scattering matrix of the burner and
ame, measured source term, duct with ow, and a measured reection coefcient.
The reection coefcients 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 tothe 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 nd the spectrum of the uctuations 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) nonreecting boundary conditions, and
2) reecting boundary conditions. The nonreecting boundary conditions in the
experiment were achieved by an orice at the exit of the test rig.
11
The simulation
of the nonreecting 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 reection coefcients; and it shows that modeling acoustic
networks with measured transfer matrices and source terms yields valid results.
A real test for checking the predictive capabilities of the method is to change
one of the boundary conditions of the test rig and then compare measured pressure
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 453
1.0
0.8
0.6
0.4
0.2
N
o
r
m
a
l
i
z
e
d
a
m
p
l
i
t
u
d
e
1.0 0.8 0.6 0.4 0.2
Normalized frequency
a)
4
3
2
1
0
N
o
r
m
a
l
i
z
e
d
a
m
p
l
i
t
u
d
e
1.0 0.8 0.6 0.4 0.2
Normalized frequency
b)
Fig. 15.4 The predicted spectrum of the pressure uctuation (dashed lines) and the
measured uctuations (dotted lines): a) in the combustion test rig with a nonreecting
exit; b) in the combustion test rig with a reecting exit.
spectra with computational results obtained with the adjusted reection coef-
cient and the previously determined burner transfer matrix and source terms. The
boundary at the exit has been changed to an almost fully reecting 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 reecting end and a nonreecting end, not only are higher overall amplitudes
observed but a shift in the resonance frequencies can also be seen. This outcome
occurs because the reection coefcient is a complex quantity; by changing the
geometry of the end, the absolute values as well as the argument of the reection
coefcient differ. This change in phase of the reection coefcient 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
C. O. PASCHEREIT ET AL. 454
1.0
0.8
0.6
0.4
0.2
0.0
N
o
r
m
a
l
i
z
e
d
a
b
s
(
T
2
2
)
2.5 2.0 1.5 1.0 0.5
Strouhal number
3
2
1
0
1
2
3
a
r
g
(
T
2
2
)
abs(T
22
)
arg(T
22
)
Fig. 15.5 Comparison of T
22
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, specically 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, modications 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 T
22
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 ow
is signicantly accelerated, fuel is injected and is homogeneously mixed with
the passing airstream. Additional swirl is often imparted on the burner ow to
increase the mixing efciency. The fuelair mixture then enters the combustor,
and a ame can stabilize at the recirculation zones that form at the ow expansion;
inner recirculation zones associated with strongly swirling ows may additionally
act as ame holders. Inhomogeneities in the fuelair mixture are convectively
transported from the fuel injection points (i ) into the ame front. Each of the fuel
injectors can be associated with a specic 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.
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 455
1 2
i
plenum fuel injector
and mixer
flame / combustor
0
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 ame front.
A. Lossy Flow Through Burner Nozzle
The ow in the burner element is described by the unsteady, incompressible
Bernoulli equation. In this model, the effect of unsteady uctuations is associated
with inertia work; losses caused by the complex three-dimensional ow inside the
burner are taken into account by using an integral loss coefcient:
_
p
1
u
1
_
=
_
_
_
1
0
c
0
_
M
0
_
1
_
A
0
A
1
_
2
_
i
c
0
L
red
_
0
A
0
A
1
_
_
_
p
0
u
0
_
(15.10)
where
L
red
=
_
1
0
A
0
A(s)
ds =
_
1
0
u(s)
u
0
ds (15.11)
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 ame does not have a steady position in the combustor, but varies in
position. This uctuation is taken into account in the model. The ame is fed with
a premixed fuelair stream, where the fuel injection takes place at the location i
inside the burner, before the ame front with upstream and downstream states
1 and 2; see Fig. 15.6. The common approach to model the acoustic behavior
of such a ame is based on the assumption that the acoustic and heat-release
uctuations at the ame front are coupled with uctuations in the fuelair mixture
that are attributable to acoustic disturbances at the fuel injectors. This implies
the existence of a characteristic time lag , after which the fuel particles reach the
C. O. PASCHEREIT ET AL. 456
ame; thus, for the fuelair mixture and its uctuation in time
(t ),
1
(t )
1
=
i
(t )
i
(15.12)
The heat release Q in the ame can be written as
Q =
1
1
S
f
h
fuel
(15.13)
The turbulent ame speed is not assumed to be constant but is assumed to be
affected linearly by the fuel supply, S
f
, which is reasonable for lean ames
( - 1). This result implies for the linearized perturbations from Eq. (15.13),
Q
Q
= 2
1
+
1
(15.14)
By using the RankineHugoniot jump conditions across the ame and the pre-
ceding equations, we can formulate the ame model as
_
p
2
u
2
_
=
_
_
1
1
c
1
_
T
2
T
1
1
_
M
1
_
1 2e
i
_
0 1
_
T
2
T
1
1
_
2e
i
_
_
_
p
1
u
1
_
(15.15)
In particular, the amplitude of the velocity uctuations across the ame has been
changed, which is important because no free parameter (such as the interaction
parameter n in the n model) exists to adjust this condition.
The model is still based on a time-delay . It is an oversimplication for a
realistic ame, in which fuel injectors are spread over an axial distance inside the
burner, the ame is nonplanar, or both. As will be shown subsequently, the effect of
the time-delay spread can in fact signicantly inuence the stability characteristics
of a burner system. This effect can be incorporated in the ame 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 ame front itself). The fuel from each of
the inlet points reaches the ame after a certain time delay
j
, and the following
relation for the velocity uctuations across the ame can be derived:
u
2
=
_
1
p
j =1
a
j
2
p
_
T
2
T
1
1
_
e
i j
_
u
1
(15.16)
Here, only a burner conguration 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
j
c [
max
L;
max
] is considered. In this case, the ame speed
model reads
u
2
=
_
1 2
_
T
2
T
1
1
__
1
i L
__
_
e
i (
max
L)
e
i
max
_
u
1
(15.17)
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 457
4
3
2
1
0
1.0 0.8 0.6 0.4 0.2
Normalized frequency
3
2
1
0
1
2
3
a
r
g
(
T
2
2
)
5
4
3
2
1
0
a
b
s
(
T
2
2
)
1.0 0.8 0.6 0.4 0.2
Normalized frequency
3
2
1
0
1
2
3
a
r
g
(
T
2
2
)
a
b
s
(
T
2
2
)
Fig. 15.7 T
22
element of the ame transfer function, with measured (solid) and mod-
eled (dashed) values: a) best t of the constant time-delay model; b) best t 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 t experi-
mental results.
C. Computational Fluid Dynamics Analysis of Time Delays
Computational uid dynamics (CFD) of the burner ow is now used to deter-
mine the model parameters
j
in the transfer function [Eq. (15.16)], or rather its
distribution f
ref
/
ref
D
B
f
Fig. 15.10 Top: normalized distribution of time delays, as obtained from particle
trajectories. Bottom: For conguration B, the effect of turbulent diffusion on redis-
tributing particle trajectories is shown (D).
location). The distributions AC (top part of Fig. 15.10) explicitly study the effect
of variations infuel penetration. The distributionBdenotes 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
j
because that different streamlines inside the burner
will be fed by the fuel if the penetration changes, and as a consequence, the ame
shape changes, as well.
All three variants show the same characteristic behavior, namely, a distributed
time delay with two distinct peaks, one at ,
ref
1, and one at ,
ref
2 to 2.5.
Only the peak at large times is affected by the modication in fuel penetration,
and this effect can in fact be correlated with changes in the shape of the ame,
which are not shown here. Unfortunately, when incorporating these distributions
into the ame 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 is primarily affected. It is
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 461
1.0 0.8 0.6 0.4 0.2
Normalized frequency
4
3
2
1
a
b
s
(
T
2
2
)
0
3
2
1
0
1
2
3
a
r
g
(
T
2
2
)
absolute
phase
Fig. 15.11 The T
22
ame transfer matrix element, derived from the time-delay dis-
tributions in Fig. 15.10; measured (solid) and modeled (dashed) values.
much atter if turbulent diffusion is included. The peak at ,
ref
= 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 ame that local deviations from the particle path have no signicant effect.
The ame 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 elds. The k c turbulence model is
known to be too diffusive to fully capture peak values of the axial burner ow,
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 ow; see Flohr et al.
15
In that paper it was concluded that
peak velocities inside the burner nozzle ahead of the ame front were not captured
correctly by the k c 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 newapproach, a methodology
is developed that includes geometries of any complexity, and the resulting systems
can be analyzed in a time-efcient, straightforward manner.
First, we demonstrate how a statespace representation of geometries, without
combustion, can be obtained. As an example, the statespace representation of an
C. O. PASCHEREIT ET AL. 462
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 nite 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 nite 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 rst one is
a typical frequency-domain approach. The second method yields a statespace
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
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 463
p
1
u
1
p
2
u
2
p
3
u
3
p
4
u
4
P B F C E
Plenum Burner Flame
Combustion
chamber
Exit
u
s
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 transformof 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() that contains the transfer matrices and of a vector
P() that contains the
unknown pressures and velocities. The right-hand side of this equation contains
2
p
2
p
t
2
= 0 (15.21)
n p
= f (15.22)
A solution can be obtained by making use of Greens 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
n
, speed of sound c, and the eigen-
vectors :
p(x) =
n=0
c
2
n
(x)
A(
2
2
n
)
_
s
(x
s
)
f (x
s
)dS (15.23)
where A =
_
2
dV. The eigenfunctions and eigenfrequencies
n
can be ob-
tained analytically for simple geometries or numerically (e.g., nite element
method) for more-complex systems. If the source function is a source of acoustic
velocity on the boundary, it can be written as:
f (, x
s
) = i u
s
(, x
s
)
An acoustic transfer function can be dened 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 A
s
centered on the boundary at x
s
. If the extent of the area A
s
is
small compared with the wavelength, then Eq. (15.23) can be rewritten to obtain
the transfer function H() between p(x) and u
s
(x
0
):
H() =
p(x)
u
s
(x
s
)
= i A
s
c
2
n=0
n
(x)
n
(x
s
)
A(
2
2
n
)
(15.24)
Equation (15.24). relates the acoustic pressure at one location to the acoustic
velocityat 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 x
i n
are then related to the J pressures at x
out
by
a J K transfer matrix H: p(x
out
) = H u(x
in
), in which the elements of H are
given by
H
j k
= i A
k
c
2
(n=0)
n
(x
j
)
n
(x
k
)
A(
2
2
n
)
(15.25)
1. State-Space Representation
Because all elements of the transfer matrix Hhave the same eigenvalues (
n
) the
transfer matrix can be expressed more conveniently by a state-space representation.
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 469
One mode of the SISO system of Eq. (15.24) can be represented as
x(t ) = A
n
x(t ) +B
n
u(t ) (15.26)
p
n
(t )
c
= C
n
x(t ) +D
n
u(t )
A
n
=
_
n
n
n
n
_
, B
nj
=
_
0
n
(x
j
)
_
C
nk
=
_
0
cA
k
A
n
(x
k
)
_
, D = [0]
Note that , the modal damping, has been introduced here. The value of is
assumed to be small compared with
n
. 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 x
n
(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:
_
_
_
x
1
.
.
.
x
2N
_
_
=
_
_
_
A
1
.
.
.
A
N
_
_
_
_
_
x
1
.
.
.
x
2N
_
_
+
_
_
_
B
11
. . . B
1J
.
.
.
.
.
.
.
.
.
B
N1
. . . B
N J
_
_
_
_
_
u
1
.
.
.
u
j
_
_
1
c
_
_
_
p
1
.
.
.
p
K
_
_
=
_
_
_
C
11
. . . C
1N
.
.
.
.
.
.
.
.
.
C
K1
. . . C
K N
_
_
_
_
_
x
1
.
.
.
x
2N
_
_
+
_
_
_
0 . . . 0
.
.
.
.
.
.
.
.
.
0 . . . 0
_
_
_
_
_
u
1
.
.
.
u
J
_
_
(15.27)
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 ow loss coefcient of the burner.
Acoustic losses purely associated with wave propagation through the combustion
chamber [taken into account by the parameter 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
n
and the values of the eigenvectors at the interface locations
n
(x)
C. O. PASCHEREIT ET AL. 470
need to be known. For practical (often very complicated) systems, the eigen-
frequencies and vectors can be obtained from a nite element analysis. In nite
element analysis, only a modal analysis, which is very computationally efcient, 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 inputoutput 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
n, m
=
_
_
2cm
D
_
2
+
_
cn
L
_
2
(15.28)
n, m
= cos
_
nx
L
_
_
cos(m)
sin(m)
A
n, m
=
_
_
LD
h
(2
kron
(n))(2
kron
(m))
LD
h
2(1 +
kron
(n))
kron
(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:
A
n, m
=
_
_
_
n, m
n, m
n, m
n, m
_
_
(15.29)
B
n, m
=
_
_
_
_
0 . . . 0
cos(m
1
) . . . cos(m
j
)
0 . . . 0
sin(m
1
) . . . sin(m
j
)
_
_
C
n, m
=
(2
kron
(n))(2
kron
(m))
J L
B
T
n, m
(15.30)
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
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 471
0 2 4 6 8 10 12 14 16 18
0
0.1
0.2
0.3
0.4
/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 nite 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 difculties.
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:
x = Ax +B
I
u
1
+B
r
u
r
(15.31)
p
I
c
= C
1
x +D
ll
u
1
+D
1r
u
r
p
r
c
= C
r
x +D
r1
u
1
+D
rr
u
r
in whichl andr 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 B
1
and C
1
contain the values
of the eigenvector on the left side of the duct and are identical to the matrices Band
C in Eq. (15.29). The matrices B
r
and C
r
contain the values of the eigenvectors
at x = L. Because cos (nx,L) = (1)
n
if x = L, the following expressions are
obtained for B
r
and C
r
in annular ducts:
B
rn, m
= (1)
n
B
1n, m
C
rn, m
= (1)
n
C
In, m
C. O. PASCHEREIT ET AL. 472
Although the D matrices are empty again, they are shown here to be consistent
with a more general notation of partitioned state-space systems:
H =
_
H
11
H
11
H
21
H
22
_
=
_
A B
l
B
r
C
l
D
ll
D
lr
C
r
D
rl
D
rr
_
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-eld, but of which the generation
itself is not inuenced by the acoustic eld.
2) External excitation with fuel ow actuators, necessary for active control.
The frequency spectra of the combustion source terms have been determined
experimentally by Schuermans.
3
A transfer function H
source
was then tted to the
magnitude of the frequency spectra. A time-domain source signal can then be
obtained by ltering a white-noise signal n(t ) with the transfer function H
source
.
In a multiburner conguration, the source terms of the individual burners are, by
denition, linearly independent. Thus, different white-noise sequences have to be
generated for each burner.
So far, the entire systemis considered to be linear. It is very likely that the actual
systemis not linear, especially when the linearized systemis 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 rst example, the one-dimensional thermoacoustic system described in
Lang et al
4
is analyzed. This system consists of a straight duct, closed on one side,
openonthe other side, witha ame stabilizedinthe middle of the duct. The pressure
drop across the ame sheet is assumed to be negligible. The acoustic velocity jump
is modeled by the so-called n model: u
2
(t ) = u
1
(t ) +nu
1
(t ), in which is
a delay time and n is referred to as the interaction coefcient. The impedance of the
open end is simply Z
3
= 0. The n model contains a delay and is thus of innite
order. To avoid systems of innite order, the time delay is approximated by a Pad e
approximation, a technique commonly used in control theory. The upstreamduct 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 rst resonant mode have been calculated for several
values of . According to Lang et al.,
4
the eigenvalues of the system are shown to
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 473
R
e
a
l
(
)
/
0
I
m
(
)
/
0
1.04
1.00
0.98
0.96
1.02
0.04
0.00
0.02
0.04
0.02
0 3 4 2
0 3 4 2
Fig. 15.17 Eigenfrequencies of the one-dimensional combustion system as a funtion
of normalized time delay, , left: real part of frequency; right: imaginary part. Solid
line: analytic solution, dotted line: modal expansion.
be the roots of
cos
_
2
L
c
_
sin
2
_
L
c
_
ne
i
= 0 (15.32)
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
0
is the
resonance frequency in the case n = 0. For large values of
0
both curves deviate
because the order of the Pad e 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 rst 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
fromthe temperature in the rst 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 downstreamhot, 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 upstreamgeometry 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 PB is the upstream system
and C is the downstream system, the interconnected system can be represented as:
PB - C. Note that once the eigenfrequencies
n, ref
are obtained for the geometry
at a specic temperature, the eigenfrequencies at different temperatures are
C. O. PASCHEREIT ET AL. 474
0 5 10 15 20 25 30 35
0
1
2
3
4
5
6
Eigenvalue number
/
c
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:
n
= (c,c
ref
)
n, ref
. The scaled eigenvectors: ,
A remain
unchanged. Although the system interconnected in this way approximates the
numerical solution, a perfect match is not obtainedeven for a systemof very high
order because the interconnectionat eachinterface positionis 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 eld. 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
rst-order expansion is sufcient. The value of the length correction has been set
to (A
s
,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 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 uctuating in the burner nozzle, the
parameter represents the effect of dissipation of acoustic energy to the mean
ow. The block diagram of the L model is given in Fig. 15.19, in which,
L (s) = [(L,c)s + M(1 (A
1
,A
2
)
2
]
1
. The values of L and are ob-
tained from a t to measured transfer functions as described in Schuermans et al.
7
For one burner, this element (denoted by B) has two inputs ( p
1
and p
2
) and two
outputs (u
1
and u
2
). In a multiburner conguration with J burners, a block diago-
nal matrix has to be formed: B = diag[B
1
, B
2
, B
J
]. If geometrically different
burners are used, the elements B
j
will be different.
The ame module, in which the interaction of the combustion process with
the acoustic eld is modeled, is again represented as an n model. However,
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 475
p
2
p
3
2 2
3 3
c
c
p
2
L(s)
2
1
A
A
p
1
u
2
u
1
( )
2 2
2
1
2
3
1
e e
i
T
T
-
u
2
u
3
-
Fig. 15.19 Block diagrams of the burner model (left), and the ame 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 ame is then given by u
2
(s) = (1 n
_
0
() e
s
dr) u
1
(s), in
which is the convective time delay between fuel injection and consumption
and () is the probability density distribution function of time delays. The in-
teraction coefcient is given by n = 1 T
2
,T
1
. The distribution of time delays
can either be obtained numerically
13, 26
or from experimental ts.
27
If a Gaussian
distribution of time delays is assumed with mean value and standard deviation
, then, after carrying out the integration, the ame model can be written as
u
2
(s) = (1 ne
s2
1,2
e
s
) u
1
(s). In this case, the values of and
have been
generated from ts to experimentally obtained frequency responses of the ame
transfer function. The interconnection of the ame block is given in Fig. 15.19.
The ame subsystem will be denoted by F; the diagonal system containing the J
ame 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 ALSTOMGT11N2
(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 ow 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 nite element method (FEM) modal expansion.
For the L representation of burners, the end correction is obtained by FEM
applied to a combustor-burners-hood model, and the loss coefcient is obtained
from impedance tube acoustic measurements. The ame 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 ame-transfer function
is tted 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 justied by the large area jump between air-supply
C. O. PASCHEREIT ET AL. 476
Fig. 15.20 GT11N2 ALSTOM gas turbine.
1
-1
0
Fig. 15.21 Hood and combustor modes.
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 477
Burner
Flame
Burner
Flame
Resonator
Combustor
j j
u p
,
Hood
Fig. 15.22 Thermoacoustic network of the GT11N2 gas turbine.
channels and hood and by the large-ow Mach number at the combustor exit. For
acoustic wavelengths that are much larger than burner and resonator dimensions,
the acoustic pressure p
j
and normal acoustic velocity u
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
Tijdeman
28
:
Z
1
=
i
I
_
e
kI
e
kI
_
+ Z
2
_
e
kI
+e
kI
_
_
e
kI
+e
kI
_
+ Z
2
_
e
kI
e
kI
_ (15.33)
where the I factor is given by the Kirchhoff solution,
I = i +
i +1
2Sh
_
1 +
1
Pr
_
(15.34)
and where Sh and P
r
are the shear number and Prandtl number, respectively. The
area jump between neck and resonator volume is modeled by forcing the continuity
of p and u 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
uid mass inside the combustor that is involved in uctuations by the air uctuating
inside the neck. The Helmholtz dampers were also tested in the impedance lab in
atmospheric conditions and showed excellent agreement with theory.
29
C. O. PASCHEREIT ET AL. 478
p Normalized
Normalized frequency Normalized frequency
p Normalized
Resonators for two unstable modes Additional resonators to suppress
the third unstable mode
simulation
engine data
simulation
engine data
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
dampingpower was thus limitedas well. The designtasktherefore hadthe extended
goal of suppressing the instability by the most efcient 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 ames 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 ame acoustics is validated by determining ex-
perimentally the transfer functions at several acoustic pressure levels. The ame
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 ame transfer functions. The in-
uence of changing boundary conditions was then predicted by using this model.
The predicted and measured spectra corresponded very well.
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 479
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 inuence of the thermal power of the combustion process on the transfer
function is investigated. According to the classical n model, the transportation
time of a fuel particle between fuel injection in the burner and fuel consumption
in the ame will result in a phase shift between velocity uctuations at the burner
and acoustic heat release in the ame. Since the velocity of the fuelair 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 ame model that
considers the three-dimensional properties of the ame is shown to be consistent
with measured transfer functions.
References
1
Munjal, M. L., Acoustics of Ducts and Mufers, 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. 281289.
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. 115.
9
Bod en, H. and
Abom, M., Modelling of Fluid Machines as Sources of Sound in Duct
and Pipe Systems, Acta Acustica, 1995, pp. 549560.
10
Lavrentjev, J., and
Abom, M., Characterization of Fluid Machines as Acoustic Mul-
tiport Sources, Journal of Sound and Vibration, Vol. 197, 1996, pp. 116.
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. 10251034.
C. O. PASCHEREIT ET AL. 480
12
Polifke, W., Poncet, A., Paschereit, C. O., andD obbeling, K., Reconstructionof Acous-
tic Transfer Matrices by Instationary Computational Fluid Dynamics, Journal of Sound
and Vibration, Vol. 245, 2001, pp. 483510.
13
Flohr, P., Paschereit, C. O., and van Roon, B., Using CFD for Time-Delay Modeling
of Premix Flames, ASME Turbo Expo 01, New Orleans, LA, June 2001.
14
Polifke, W., Flohr, P., and Brandt, M., Modeling of Inhomogeneously Premixed Com-
bustion with an Extended TFC Model, ASME Turbo Expo 00, Munich, Germany, May
2000.
15
Flohr, P., and Paschereit, C. O., Mixing Prediction in Premix Burners Using Industrial
LES Tools, Symposium on Computational Modeling of Industrial Combustion Systems,
ASME International Mechanical Engineering Congress and Exposition, New Orleans, LA,
Nov. 2002.
16
Dowling, A. P., The Calculation of Thermoacoustic Oscillations, Journal of Sound
and Vibration, Vol. 180, 1995, pp. 557581.
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., Fleil, 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, ASMETurbo Expo 02, No. GT-2002-30064,
Amsterdam, The Netherlands, June 2002.
24
Rienstra, S., and Hirschberg, A., An Introduction to Acoustics, Report IWDE9902,
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. 133.
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
, OH
, andCO
2
(the asteriskindicates anexcitedspecies), whereas
in rich hydrocarbon ames 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
110mm dia x 375mm
fused silica
combustor section
100mm dia x 500mm
stainless steel
combustor section
19mm dia
exit
choked
inlet
swirl vanes
Fig. 16.5 Schematic drawing of optically accessible lean premixed combustor.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 487
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 fromCH
(309 nm)
occur at distinctly different and relatively narrow-wavelength intervals, whereas
the CO
2
chemiluminescence lies over a broad-wavelength interval (350600 nm)
and overlaps the CH
and OH
,
CH
, and CO
2
chemiluminescence. First, the measured CO
2
chemiluminescence
signal strength can be signicantly increased over that of OH
and CH
chemilu-
minescence by using a very broad lter, for example, = 100200 nm. Second,
to detect OH
2
chemiluminescence from an
OH
or CH
2
chemiluminescence background. The in-
dependent measurement of the CO
2
chemiluminescence background has not been
done in most OH
and CH
2
chemiluminescence.
Measurements of the chemiluminescence emission from lean premixed ames
have been used in numerous studies to indicate the location of the reaction zone and
to infer local and overall heat-release rates.
2440
The rationale for such measure-
ments is usually based on the experimental observation that, for a xed equivalence
ratio, the intensity of chemiluminescence emission fromthe entire ame, hereafter
T
inlet
= 650 K
3 4 5 6 7 8 9
0.0
0.5
1.0
1.5
2.0
= 0.45
0.50
0.55
0.60
0.65
0.70
Fuel Flow Rate (scfm)
O
v
e
r
a
l
l
C
O
2
C
h
e
m
i
l
u
m
i
n
e
s
c
e
n
c
e
I
n
t
e
n
s
i
t
y
(
a
.
u
.
)
Fig. 16.6 Overall CO
2
chemiluminescence emission vs fuel ow rate from a lean
premixed combustor operating on natural gas at 100 kPa with an inlet temperature
of 650 K.
J. G. LEE AND D. A. SANTAVICCA 488
0.4 0.6 0.8
1
Equivalence Ratio
0
2
4
6
8
I
n
t
e
n
s
i
t
y
/
F
u
e
l
F
l
o
w
R
a
t
e
(
a
.
u
.
)
V
inlet
= 67 m/s T
inlet
= 673 K
Fig. 16.7 Overall CO
2
chemiluminescence emissiondividedby fuel owrate vs equiv-
alence ratio.
referred to as the overall chemiluminescence emission, increases linearly with the
fuel ow rate, where the slope increases with increasing equivalence ratio.
2428
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 ow rate for a range of xed 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
ame onto a photomultiplier tube through a glass lter (BG-40) that transmits over
the wavelength interval from325 to 650 nm. The results shown in Fig. 16.6 indicate
that the overall chemiluminescence intensity is a function of both the fuel owrate,
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 ow 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
41
The fact that it is the ame 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
was detected with a bandpass lter centered at 430 nm and a FWHM of 10 nm.
In addition to the effects of fuel ow 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 ow 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 detailedchemical kinetic calculations of leanpremixed
laminar methaneair ames have been conducted to investigate the relationship
between the local rate of heat release, that is, the rate of heat release per unit ame
area, and the local chemiluminescence emission, that is, the rate of chemilumi-
nescence emission per unit ame area.
4144
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
ame curvature, for example, at cusps, where the calculations show that the local
chemiluminescence emission can effectively go to zero without local extinction of
the ame. 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 (HR
local
) and the local chemiluminescence emission
(I
local
) are affected by unsteady strain and ame curvature and that they increase
J. G. LEE AND D. A. SANTAVICCA 490
exponentially with temperature, leading to a power-law relationship between the
local chemiluminescence emission and the local rate of heat release, that is,
I
local
(HR
local
)
(16.1)
where the exponent is a positive number and depends on the ame temperature
(as determined by the equivalence ratio, unburned gas temperature, dilution, and
radiation losses) and the effects of unsteady strain and ame curvature.
41
To determine the relationship between the overall chemiluminescence emission
(I
overall
) and the overall rate of heat release (HR
overall
), one must integrate the local
values over the ame area, that is,
I
overall
=
A
I
local
dA
ame
and HR
overall
=
A
HR
local
dA
ame
(16.2)
If the ame temperature, that is, the equivalence ratio, unburned gas temperature,
dilution, and radiation losses, is constant and the effects of strain and ame cur-
vature are negligible or constant, then I
local
, HR
local
, and are constant over the
ame. These constant values result in the proportionality of the overall chemilu-
minescence emission and the overall rate of heat release, that is,
I
overall
= C HR
overall
(16.3)
where the constant C depends on the ame 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 xed equivalence ratio and inlet temperature the
overall chemiluminescence emission increases linearly with fuel ow 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 ow 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 ame 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 ame surface. For example, a more compli-
cated relationship would occur in a partially premixed turbulent ame in which the
equivalence ratio is not constant over the ame surface. In this case, the exponent
, in the equation relating the local chemiluminescence emission to the local rate
of heat release, varies with location on the ame 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 xed overall equivalence ratio vs a parameter that is referred to as
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 491
V
inlet
= 84 m/s
T
inlet
= 650 K
0 20 40 60 80 100
0.0
0.5
1.0
1.5
2.0
O
v
e
r
a
l
l
C
O
2
C
h
e
m
i
l
u
m
i
n
e
s
c
e
n
c
e
I
n
t
e
n
s
i
t
y
(
a
.
u
.
)
= 0.70
= 0.64
= 0.58
= 0.51
= 0.45
% Premixed
Fig. 16.9 The effect of incomplete fuelair 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 ow, 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 ame. Any factors causing the ame area to
change, for example, amevortex interaction, will result in a change in the overall
chemiluminescence emission and the overall rate of heat release. Changes in the
ame area will not alter the relationship between the overall chemiliuminescence
emission and the overall rate of heat release, as long as the ame temperature and
the effects of strain and curvature are constant. In other words, as the ame 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 ame surface as the ame area
changes, as might be expected during amevortex interaction, then the relation-
ship between the overall chemiluminescence emission and the overall rate of heat
release is likely to change as the ame 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
J. G. LEE AND D. A. SANTAVICCA 492
release in lean premixed ames, 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 uorescence.
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
uctuations in both the overall heat release
12, 20, 30, 31, 35, 3840
and the spatial distri-
bution of the local heat release.
7, 12, 20, 3032, 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 ow rate and changes in the equivalence ratio indepen-
dently affect the overall chemiluminescence emission, whereas only changes in
the fuel owrate 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 uctuating but for different reasons. In the rst case, the equivalence-
ratio uctuations are the result of uctuations in the airow rate, whereas the fuel
ow rate is constant. Under these conditions, uctuations 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 ame speed, changes with the local equivalence ratio. This discrepancy
is compensated for by changes in the ame area such that the overall rate of heat
release remains constant.) In the second case, the equivalence ratio uctuations are
the result of uctuations in the fuel ow rate, whereas the airow rate is constant.
Under these conditions, uctuations will occur in the overall chemiluminescence
emission, that, in part, will be caused by equivalence ratio uctuations and, in
part, by fuel ow rate uctuations. In this situation, the chemiluminescence uc-
tuations overestimate the uctuations in the overall rate of heat release. The only
situation where the uctuation in the overall chemiluminescence can be attributed
solely to uctuations 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, uctuations in the equivalence ratio. Under
such conditions, measurements of the overall chemiluminescence-emission uc-
tuations without simultaneous measurements of the equivalence ratio uctuations
can potentially give misleading information about both the amplitude and phase
of the overall heat-release uctuations.
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 signicantly reduced by using an optical arrangement with a very
short depth of eld; however, this reduction is at the expense of signicantly re-
duced signal strength.
37
Another approach, if the ame is axisymmetric, is to
use a deconvolution technique to reconstruct the two-dimensional emission eld
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 ame thickness.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 493
Because of the line-of-sight nature of the chemiluminescence measurement, the
two-dimensional chemiluminescence results effectively integrate over the local
three-dimensional ame structure. In other words, the intensity of the chemilumi-
nescence emission depicted in the two-dimensional chemiluminescence images
(I
2D
) represents the product of the local chemiluminescence emission and the
local ame area, that is,
I
2D
= I
local
A
local
(16.4)
where
A
local
is the ame area within a volume dened by the resolution of the
line-of-sight measurement.
Similarly, one can dene a two-dimensional rate of heat release (HR
2D
), which
represents the product of the local rate of heat release and the local ame area,
that is, HR
2D
= HR
local
A
local
. 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 I
2D
= (I
local
/HR
local
) HR
2D
. Using
the power-lawrelationship between the local chemiluminescence emission and the
local rate of heat release, discussed previously, gives the following relationship:
I
2D
= (HR
local
)
1
HR
2D
= C
2D
HR
2D
(16.5)
where C
2D
depends on the local ame temperature and any factors that affect
the ame 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 ame temperature and
any factors that affect the ame temperature. As with overall chemiluminescence
measurements, the most likely concern would be in a partially premixed ame 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 uctuations 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 uctuation is shown in Fig. 16.10b. The
overall CO
2
chemiluminescence emission plotted in Fig. 16.10a was measured
by imaging the entire ame onto a photomultiplier tube through an appropriate
bandpass lter as described previously. Care must be taken when making such
measurements to collect the chemiluminescence emission from the entire ame
so as to obtain an accurate indication of the total heat-release rate and to avoid
erroneous uctuations caused by the ame moving in and out of the eld 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 uctuations that is related to the overall system
damping and gain characteristics.
3
Such measurements also provide information
on how the ames heat release responds to pressure uctuations. For example,
Fig. 16.11 shows a plot of the rms overall heat-release uctuation normalized
by the mean overall heat release (measured by using CO
2
chemiluminescence)
vs the rms combustor pressure uctuation during unstable combustion. These
J. G. LEE AND D. A. SANTAVICCA 494
Time (sec)
0.000 0.025 0.050 0.075
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
O
v
e
r
a
l
l
H
e
a
t
R
e
l
e
a
s
e
(
a
.
u
.
)
Time (sec)
0.000 0.025 0.050 0.075
-2
-1
0
1
2
p
(
p
s
i
)
a)
b)
Fig. 16.10 Simultaneous measurement of a) the overall heat-release uctuations and
b) the combustor pressure uctuations 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 uctuation
increases linearly as the pressure uctuation increases until it becomes saturated at
high-pressure uctuations, indicating that there is a nonlinear relationship between
the pressure and heat-release uctuations during unstable combustion.
45, 46
Amore
comprehensive assessment of the nonlinear response of lean premixed ames to
pressure uctuations can be obtained from forced-response studies in which the
amplitude and relative phase of heat-release uctuations resulting from imposed
pressure uctuations over a range of frequencies and amplitudes are measured.
40
Chemiluminescence emission can also be recorded by using an intensied
charge-coupled device (CCD) camera to obtain an image of the ame structure
during unstable combustion that represents the spatial distribution of the ames
heat release. An example of such a measurement is shown in Fig. 16.12a, which
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 495
0.0 0.1 0.2 0.3 0.4 0.5 0.6
0
5
10
15
20
25
30
35
P
rms
(psi)
q
r
m
s
/
q
r
m
s
(
%
)
Fig. 16.11 The rms heat-release uctuation normalized by the average heat release vs
rms pressure uctuation at various unstable operating conditions in a lean premixed
combustor.
presents a CO
2
chemiluminescence image of the ame 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 ow 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.
J. G. LEE AND D. A. SANTAVICCA 496
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 ame. If the ame is assumed to be axisymmetric, one
can use a deconvolution procedure to reconstruct the two-dimensional ame struc-
ture, including onion-peeling, Abel transformation, and ltered 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 ame 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,
Time (msec)
0.0 1.0 2.0 3.0 4.0
P
r
e
s
s
u
r
e
(
p
s
i
)
-1.0
-0.5
0.0
0.5
1.0
Image Number
1 3 5 7 9 11 13
#1 #4
#7
#10
#2
#5
#8
#11
#3
#6
#9 #12
Low High
Fig. 16.13 Flame-structure evolution during one period of unstable combustion with
a frequency of 235 Hz.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 497
as it is in Fig. 16.5; hence, a marked change occurred in the frequency of the insta-
bility compared with the previously presented results. Note that only the upper half
of the ame 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
ame-structure image sequence reveals the temporal evolution of the ame struc-
ture during the instability and provides insight regarding the phenomenology of
the instability. For example, the images in Fig. 16.13 showa ame 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 ame shape remains very nearly the same
during the instability but there is a noticeable change in the overall intensity of the
ames 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
ame 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 ame-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 operatingconditions (T
inlet
= 623 K, V
inlet
= 59 m/s, p = 100kPa, and
= 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 showevidence of amevortex
interaction, but the details of the interaction are noticeably different in the two
cases. The ame in Fig. 16.14a appears to be wrapped around the vortex, which
results in stretching and contraction of the ame zone, whereas the ame in Fig.
16.14b appears to be contained within the vortex, exhibiting periodic extinction
and reignition of the entire reaction zone.
Two-dimensional ame-structure images, suchas those showninFigs. 16.13and
16.14, reveal the location and intensity of the ames heat release and its temporal
evolution during one period of the instability. Combining this information with the
measured pressure uctuation, one can calculate the Rayleigh index distribution,
R(x, y), which is given by the following equation:
R(x, y) =
1
T
(x, y, t ) q
(x, y, t ) dt (16.6)
where q
t
o
+T
t
o
p
secondary
dt (16.7)
where T is the period of the secondary fuel ow modulation, t
o
is the time delay
between the pressure signal zero crossing and the secondary fuel valve trigger
signal, p
secondary
is the ame-response
function. An example of the ame-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 ame-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 ame-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 ame-response Rayleigh index as a function of
the time delay t
o
, is plotted in Fig. 16.16c for the ame-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 oweld param-
eters such as gas concentration, temperature, pressure, and velocity.
12, 20, 4961
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 specic atoms or molecules. The absorption process is
described by BeerLamberts Law, that is,
I
I
o
= 10
l
o
cdx
(16.8)
where I
o
is the intensity of incident light, I is the intensity of transmitted light,
is the decadic molar absorption coefcient (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 coefcient. The
major limitation of this technique is the fact that it is a line-of-sight measurement,
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 501
q
(
C
O
2
*
c
h
e
m
i
l
u
m
i
n
e
s
c
e
n
c
e
)
Time (msec)
0 1 2 3 4 5 6 7 8 9 10 11
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
Valve opening pulse
time delay (t
o
)
Period (T)
0 2 4 6 8 10 12
-0.02
-0.01
0
0.01
0.02
0.03
Time (msec)
H
e
a
t
r
e
l
e
a
s
e
(
a
.
u
.
)
q
secondary
time delay (msec)
0 1 2 3 4
G
l
o
b
a
l
R
e
s
p
o
n
s
e
R
a
y
l
e
i
g
h
I
n
d
e
x
-0.020
-0.015
-0.010
-0.005
0.000
0.005
0.010
a)
b)
c)
Fig. 16.16 a) Flame-response function, b) ame-response function superimposed
on the unstable pressure trace, and c) ame-response Rayleigh index vs time-delay
prediction.
J. G. LEE AND D. A. SANTAVICCA 502
250 300 350 400 450 500 550 600 650 700
I
/
I
o
0.00
0.10
0.20
0.30
0.40
0.50
Temperature (K)
Pressure (kPa)
100 200 300 400 500 600
0.001
0.01
0.1
1
T = 295 K
T = 473 K
T = 573 K
T = 673 K
I
/
I
o
a)
b)
Fig. 16.17 a) Temperature and b) pressure dependence of the normalized transmit-
tance (I/I
o
) of a 3.39-m He-Ne laser beam through a homogeneous methaneair
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 oweld 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-m 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 coefcient. An example of this empirical determination for
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 503
methane is shown in Fig. 16.17a and 16.17b, in which plots of the normalized
transmittance (I /I
o
) vs temperature at a xed pressure and vs pressure for values
of constant temperature, respectively, are shown.
12
These measurements were
made in a owcell with a homogeneous methaneair mixture at known conditions.
The 3.39-m 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 coefcient can be obtained from BeerLamberts Law, that is,
=
1
cl
log
10
I
I
o
(16.9)
By using the data shown in Fig. 16.17, the following expression for the decadic
molar absorption coefcient for methane as a function of pressure and temperature
is obtained
12
:
= 84737
P
o
P
T
293 K
C
1
+C
2
P
P
o
1
(16.10)
with
C
1
= 0.1131 +1.1875 (293 K/T)
and
C
2
= 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 uctuations 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-m laser beam
passes through the annular mixing section in the injector, just upstream of the en-
trance to the combustor. Beampassage through the annular mixing section required
a modication 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 xed 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 coefcient 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,
J. G. LEE AND D. A. SANTAVICCA 504
Data Acquisition
System
InAs Detector
HeNe Laser
(3.39 m)
InAs Detector
Data
Acquisition
System
Center bluff-body
Adapters
Fuel injection
Location
Preheated Air
7 mm
Diffuser
Measurement Location
Swirl vanes
a)
b)
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 Kand an inlet velocity of
100 m/s. It is assumed that the temperature of the mixture in the nozzle is constant
and, therefore, only uctuations 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 signicant role in this instability, resulting in
peak-to-peak uctuations in the equivalence ratio of approximately 0.05 about a
mean of 0.65. The frequency spectrumcorresponding to these uctuations is shown
in Fig. 16.19d. The equivalence-ratio uctuations 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 uctuations are clearly the same, that is, 465 Hz,
although there are signicant differences in the relative magnitude of the second
harmonic oscillations. Most noticeable is the 10-dBdifference between the rst and
second harmonics of the equivalence-ratio uctuation vs 23-dBdifference between
the rst and second harmonics of the heat-release uctuations. This suggests that
the ame, which is spatially distributed, acts to average out the higher-frequency
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 505
0.55
0.60
0.65
0.70
0.75
E
q
u
i
v
a
l
e
n
c
e
R
a
t
i
o
-1
0
1
P
c
(
p
s
i
)
t/T
0.0 0.2 0.4 0.6 0.8 1.0
q
0.15
0.20
0.25
0.30
0.35
0.40
-90
-70
-50
-30
-10
d
B
V
r
m
s
(
P
c
)
0 400 800 1200 1600
-70
-50
-30
-10
Frequency (Hz)
d
B
V
r
m
s
(
q
)
-60
-50
-40
-30
-20
-10
d
B
V
r
m
s
(
)
a) d)
b)
c)
e)
f)
Fig. 16.19 ac) Time traces and df) power spectra of equivalence ratio , combustor
pressure P
c
, and heat release q during one period T of unstable combustion.
uctuations in the equivalence ratio. Also note that the relative magnitudes of the
pressure, equivalence-ratio, and heat-release uctuations for this instability are
3%, 5%, and 23% of the mean, respectively.
Simultaneous pressure, heat-release, and equivalence-ratio uctuation measure-
ments canbe usedtodetermine the phase delayor time lagbetweenthese processes.
Of particular interest is the time lag between the equivalence-ratio uctuation and
the heat-release uctuation, because it is important in assessing the role of feed-
system coupling. To estimate whether the equivalence-ratio uctuation produced
by feed-system coupling arrives at the ame front in-phase with the heat-release
uctuation, one must estimate the convection time between the fuel-injection loca-
tion and the ame front. The most difcult part of the convection time to estimate
is the time required for the fuel to travel from the entrance of the combustor to
the ame front where the fuel burns. The phase delay or the time lag between the
equivalence ratio and the heat-release uctuation shown in Fig. 16.19a and 16.19c
is a direct measurement of that quantity.
The 3.39-m helium-neon laser-absorption technique has also been imple-
mented in a ber-optic probe
52, 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
J. G. LEE AND D. A. SANTAVICCA 506
uctuations 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 ow is steady; they are impractical if the ow is unsteady.
Laser-induced uorescence, 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 uorescence 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 uorescence or as laser-induced uorescence, and its intensity can be
related to the number density of the species of interest. To quantify the relation-
ship between the uorescence 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 uorescence 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 uorescence intensity vs number density of the uorescing species as a func-
tion of temperature and pressure. When calibrating a uorescence 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 uorescence 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 uorescence signal, S
f
(joules per square
centimeter), can be related to the mole fraction of the uorescence species,
abs
,
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 507
by the following equation:
S
f
(E/h)
abs
P
T
(, T) (, P, T) (16.11)
where E is the laser irradiance (J/cm
2
), h is Plancks constant, is the laser fre-
quency,
abs
is the molar concentration of the uorescence species, (, T) is the
molecular absorption cross section of the uorescence species, and (, P, T) is
the uorescence quantum yield. In this equation, the effect of gas composition on
the uorescence yield is assumed to be constant. For a xed excitation wavelength
and under isothermal and isobaric conditions, the uorescence signal is only pro-
portional to the mole fraction of the uorescence seed. However, in ows 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 uorescence quantum yield, and therefore the uorescence
signal, must be accounted for, as discussed before.
Laser-induced uorescence can be used to make point measurements with sub-
millimeter spatial resolution or to make two-dimensional measurements, also re-
ferredtoas planar laser-induceduorescence (PLIF) measurements.
67, 68
Inmaking
point measurements, the laser beam is focused with a spherical lens to a small-
diameter beam waist and the uorescence 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 uorescence signal
is detected by imaging a portion of the laser sheet onto an intensied CCDcamera.
In both cases, an appropriate interference lter is used to isolate and selectively
detect the desired uorescence wavelength. The strength of the uorescence sig-
nal for gas-phase uorescence measurements is typically very low, requiring the
use of high-power pulsed lasers, where depending on the uorescence species and
its concentration, single-pulse measurements are often possible. Unfortunately,
high-power pulsed lasers operate at relatively low pulse rates, that is, typically
1020 Hz; therefore, this technique does not provide a continuous measure of the
uorescence 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 uorescence in the study of combustion
dynamics is in the characterization of fuelair mixing. Measurement of fuelair
mixing is important because both the temporal and spatial fuel distribution can have
a signicant effect on the stability characteristics of the combustor.
32
Many fuels
of interest, however, are not well suited for uorescence measurements, and so a
common approach is to seed the fuel with a small amount of a uorescence seed, in
which the concentration of the uorescence seed, as measured by the uorescence
technique, is assumed to be an indicator of the fuel concentration. An advantage
of using a uorescence seed is that its concentration can be controlled. In addition,
a uorescence seed can be selected that has optimum physical and spectroscopic
properties, making quantitative equivalence ratio measurements possible. Various
J. G. LEE AND D. A. SANTAVICCA 508
species have been used as uorescence seeds for fuelair-mixing studies and de-
tailedinformationontheir uorescence characteristics canbe foundinRefs. 6974.
Several factors must be considered when selecting a uorescence seed to charac-
terize mixing, including boiling point, autoignition temperature, absorption and
uorescence characteristics, mass diffusion coefcient, cost, and toxicity.
For gaseous fuels, acetone is commonly used as a uorescence seed because
of its low-boiling point (50
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 (225320 nm) with maximum absorption between 270 and 280 nm.
The uorescence emission is broadband in the blue (350550 nm) and short lived
( 4 ns) with a uorescence efciency of 0.2%.
7174
Because acetone is a stable species, the effects of pressure and temperature
on the absorption cross section and the uorescence yield are best accounted
for empirically. Figures 16.20a and 16.20b show the effect of temperature and
pressure, respectively, on the acetone uorescence signal with 266-nm excita-
tion, that is, the fourth harmonic output of a Nd:YAG laser. Each panel shows
two curves. The lled circles are the actual measurements, and the lled squares
have been corrected to constant number density. The measurements were made in
a ow cell with the volume fraction of acetone xed at 1%; therefore, changes
in pressure and temperature also affect the uorescence signal as a result of
changes in the density. The actual measurements show that the uorescence sig-
nal decreases with increasing temperature and increases with increasing pressure.
After correcting for changes in density, however, the uorescence signal is in-
dependent of pressure, but it decreases with increasing temperature. Knowledge
of the effect of pressure and temperature on the uorescence signal, indepen-
dent of their effect on density, is critical to the proper interpretation of uores-
cence measurements. The temperature and pressure dependence of acetone-laser-
induced uorescence with different excitation wavelengths can be found in the
Refs. 7174.
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-uorescence
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
Fig. 16.20 Acetone LIF signal with 266-nm excitation: a) temperature dependence
and b) pressure dependence.
exit of the annular mixing section. The uorescence signal is recorded by using
an intensied CCD camera positioned downstream of the combustor and perpen-
dicular to the laser sheet. Subtraction of background noise and a uniform eld
correction are applied to each of the acetone PLIF images. In addition, the images
are corrected for pulse-to-pulse uctuations 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
J. G. LEE AND D. A. SANTAVICCA 510
Equivalence Ratio Scale
PM 50% CB/50% DS CB DS
0.25
0.35
0.45
0.55
0.65
0.75
0.85
0.95
1.05
0.00 0.04 0.08 0.12 0.16 0.20 0.24 0.28 0.32 0.36
Radial Distance (in)
E
q
u
i
v
a
l
e
n
c
e
R
a
t
i
o
V
inlet
= 50 m/s, T = 373K,
avg
= 0.70
CB
50CB/50DS
DS
PM
(2)
(1)
(3)
Center bluff-body
Swirl vanes
(1) Premixed injection (PM)
(2) Center body injection (CB)
(3) Downstream injection (DS)
Fused Silica Combustor
Laser sheet
ICCD Camera
Air
a)
b)
c)
Fig. 16.21 a) Side view of the optically accessible axial dump combustor and
schematic diagram of PLIF setup; b) processed acetone PLIF images; and
c) equivalence-ratio distribution over the annular mixing section at the inlet of the
combustor.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 511
combustor-inlet velocity (50 m/s), inlet temperature (373 K), and pressure (100
kPa). In the case labeled PM, the fuel and air are premixed well upstream of the
combustor. In this case the fuel distribution is expected to be perfectly uniform.
In the case labeled CB, all the fuel is injected through holes in the centerbody
at a location approximately 25 mm upstream of the dump plane, that is, location
(2). In this case the fuel penetrates to the outer wall of the mixing section, which
results in fuel-rich conditions along the outer wall and fuel-lean conditions along
the centerbody. In the case labeled DS, all the fuel is injected through holes in
the outer wall of the mixing section at a location approximately 25 mm upstream
of the dump plane, that is, location (3). In this case the fuel penetrates to the
centerbody, which results in fuel-rich conditions along the centerbody and fuel-
lean conditions along the outer wall of the mixing section. Note also the evidence
of the effect of the six-vane swirler on the fuel distribution in the DS image
shown in Fig. 16.21b. Last, in the case labeled 50%CB/50%DS, half of the fuel
is injected through the holes located in the centerbody, location (2), and half is
injected through the holes in the outer wall of the mixing section, location (3). The
resulting fuel distribution is very uniform. These results are further quantied by
calculating an average radial fuel distribution for each case. This calculation is an
average of the radial fuel distribution over 12 radial proles 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 ame-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
ame 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 ame 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 ame 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
DS CB PM
Fig. 16.22 The effect of inlet fuel distribution on ame structure for the PM, CB, and
DS fuel distributions shown in Fig. 16.21.
J. G. LEE AND D. A. SANTAVICCA 512
distribution, it can be argued that differences in the susceptibility of the different
ame 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 ber-optic probe for making point mea-
surements of the equivalence ratio and thereby determine the fuel distribution.
A ber-optic laser uorescence equivalence ratio probe has been developed and
successfully used for this purpose.
76
The overall diameter of the probe is 16 mm(
5
8
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 bers, one for transmitting the laser beam and the other for
transmitting the collected uorescence 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 fuelair 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 uorescence 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 uorescence intensity. This scattering necessitates the use of a band-
pass lter with very good stray-light rejection, which transmits the uorescence
while rejecting the Mie scattering. Another issue related to the presence of liq-
uid droplets is the fact that the uorescence signal strength, which is proportional
to the density of the uorescence 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,
betweenregions that primarilycontainfuel liquidandregions that primarilycontain
fuel vapor. One can also make simultaneous two-dimensional Mie scattering (see
Sec. VI) and uorescence measurements, by using a split-image lter or two cam-
eras and using the Mie image to identify the location of liquid drops in the uo-
rescence image. With either of these approaches, however, measurements made in
regions containing a mixture of vapor and liquid are very difcult to quantify. (Note
that the exciplex uorescence technique was specically developed to distinguish
betweenthe liquidandvapor uorescence, but this technique is signicantlylimited
because it cannot be usedinthe presence of oxygenbecause of quenchingeffects.
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 uorescence characteristics of the individ-
ual fuel components can make quantitative fuel-concentration measurements ex-
tremely difcult. One way to address this problem is to use a simulated fuel that
is composed of nonuorescing fuel components, typically alkanes, which have
vaporization characteristics similar to the components in the actual fuel, and u-
orescence seeds, typically ketones, with matching boiling points for tagging the
individual fuel components.
78
An example of such a systemfor 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 uorescence tracers, any of
the individual fuel components can be followed.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 513
Table 16.1 Fuel component simulants and matching uorescence tracers
Fuel component T
BP
,
C Fluorescence tracer T
BP
,
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 uorescence is the use of HCO uores-
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 HCOuorescence
technique, as compared with the chemiluminescence technique, are that the HCO
mole fraction has been shown to be an accurate indicator of the ames heat release,
even in the presence of unsteady strain and ame curvature, and that the uores-
cence measurement is spatiallyresolved.
43, 44
The HCOuorescence 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 uorescence
signals are very weak, it is unlikely that two-dimensional uorescence 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 uorescence technique is the use
of OHplanar laser-induced uorescence to obtain detailed two-dimensional ame-
structure measurements during unstable combustion.
7981
The reaction zones in
lean premixed combustors can be expected to be predominantly in the so-called
wrinkled laminar-ame regime,
8284
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-ame regime, the leading
edge of the reaction zone, that is, the ame front, is characterized by a steep gradient
in OHconcentration. However, because OHis relatively long lived, it persists well
into the high-temperature products downstream of the ame front.
65
The location
of the steep gradient in OHconcentration can be used as an indicator of the location
of the reaction zone or ame front. Because OHPLIF signal strengths are typically
strong, single-shot measurements that provide a detailed space- and time-resolved
map of the two-dimensional ame 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 ame is
anchored on the centerbody. It also shows a well-dened and highly wrinkled
ame front, which indicates that combustion is occurring in the so-called wrinkled
J. G. LEE AND D. A. SANTAVICCA 514
a)
b)
Fig. 16.23 Flame-surface area calculation procedure: a) Normalized OHPLIFimage
and b) thresholded ame surface.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 515
laminar-ame regime.
82
Last, there is clear evidence of the interaction between the
ame front and the vortex that is shed from the dump-plane shoulder. The effect of
this interaction on the area of the ame, because the ame area is directly related
to the ames rate of heat release, is of particular interest.
The rst step in calculating the ame area is to determine the location of the
ame 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 ame
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
uorescence 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 ame
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 ame front determined by this threshold procedure from the OH
PLIF image shown in Fig. 16.23a is shown in Fig. 16.23b. Once the ame front is
determined, the total ame area is calculated by revolving the ame front around
the centerline of the combustor. This calculation assumes that the ame front
determined from the OH PLIF image is representative of the ame front at other
cross sections of the ame. In addition, because the OHPLIF image is a single-shot
image, one must calculate the area for many such images from which an average
ame area can be determined.
To understand the role of ame-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 ame 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 ame front and the vortex and
the resulting changes in the ame length. Five single-shot OH PLIF images were
acquired at each phase angle, the ame area was then calculated for each image
by using the procedure described earlier, and an average ame area at each phase
angle was calculated. The resulting ame area vs phase angle over one period of
the instability is plotted in Fig. 16.25, along with the measured overall heat-release
uctuation, that is, the overall CO
2
chemiluminescence-intensity uctuation. In
this case the area and heat-release uctuations are very nearly in phase, indicat-
ing that ame-area changes caused by ame-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 ame-structure measurements can also be used
to calculate the local ame-surface density, which is a measure of the local reac-
tion rate in turbulent ames. Such measurements have been made in a lean pre-
mixed combustor under stable and unstable operating conditions and have been in
J. G. LEE AND D. A. SANTAVICCA 516
t = 0ms t = 0.22ms t = 0.44ms t = 0.66ms
t = 2.42ms t = 2.20ms t = 1.98ms
t = 1.76ms
t = 0.88ms t = 1.10ms t = 1.32ms t = 1.54ms
Fig. 16.24 Sequence of digitized ame surfaces over one period of unstable combus-
tion.
0 0.5 1.0 1.5 2.0 2.5
2
3
4
5
6
7
H
e
a
t
r
e
l
e
a
s
e
(
a
.
u
.
)
Heat release
Area
8
10
12
14
16
18
20
22
24
F
l
a
m
e
a
r
e
a
(
s
q
.
i
n
c
h
e
s
)
A
r
e
a
A
r
e
a
H
e
a
t
r
e
l
e
a
s
e
H
e
a
t
r
e
l
e
a
s
e
Time (msec)
Fig. 16.25 Variation of ame area and heat release with time during one period of
unstable combustion.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 517
good agreement with OH
2
Chemilumines-
cence in Premixed Flames, Combustion Science and Technology, Vol. 109, No. 1, 1995,
pp. 183203.
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. 131150.
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. 312332.
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. 369403.
45
Dowling, A. P., Nonlinear Self-Excited Oscillations of a Ducted Flame, Journal of
Fluid Mechanics, Vol. 346, 1997, pp. 271290.
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. 11461152.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 527
48
Kim, K., Lee, J. G., and Santavicca, D. A., Optimization of the Spatial and Tem-
poral Fuel Distribution for Active Control of Combustion Dynamics in Lean Premixed
Combustors, AIAA Paper 2002-4024, July 2002.
49
Mallard, W. G., and Gardiner, W. C. Jr., Absorption of the 3.39 m He-Ne Laser
Line by Methane from 300 to 2400K, Journal of Quantitative Spectroscopy and Radiative
Transfer, Vol. 20, No. 2, 1978, pp. 135149.
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. 60906103.
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 m
He-Ne Laser, JSAE Review, Vol. 17, No. 4, 1996, pp. 339345.
52
Mongia, R. K., Tomita, E., Hsu, F. K., Talbot, L., and Dibble, R. W., Use of an Optical
Probe for Time-Resolved In situ Measurement of Local Air-to-Fuel Ratio and Extent of Fuel
Mixing with Applications to Low NOx Emissions in Premixed Gas Turbines, Proceedings
of the Combustion Institute, Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 2749
2755.
53
Koenig, M., and Hall, M. J., Measurements of Local In-Cylinder Fuel Concentration
Fluctuations in a Firing SI Engine, Society of Automotive Engineers, Paper 971644, 1997.
54
Wehe, S. D., Baer, D. S., and Hanson, R. K., Tunable Diode-Laser Absorption Mea-
surements of Temperature, Velocity and H
2
O in Hypersonic Flows, AIAA Paper 97-3267,
July 1997.
55
Mihalcea, R. M., Baer, D. S., andHanson, R. K., Diode-Laser Sensor for Measurements
of CO, CO2 and CH4 in Combustion Flows, Applied Optics, Vol. 36, No. 33, 1997,
pp. 87458752.
56
Seitzman, J. M., Tamma, R., and Vijayan, R., Infrared Absorption Based Sensor
Approaches for High Pressure Combustion, AIAA Paper 970318, 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 CO
2
, H
2
O and Gas Temperature, Pro-
ceedings of the Combustion Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998,
pp. 95101.
59
Mongia, R., Torres, J., Dibble, R., Lee, D., Anderson, T., and Sowa, W., Fast Re-
sponse Extraction Probe for Measurement of Air-Fuel Ratio Fluctuations in Lean Premixed
Combustors, American Society of Mechanical Engineers, Paper 99-GT-277, 1999.
60
Ebert, V., Fernholz, T., Giesemann, C., Pitz, H., Teichert, H., and Wolfrum, J.,
Simultaneous Diode-Laser-Based In-Situ-Detection of Multiple Species and Tempera-
ture in a Gas-Fired Power-Plant, Proceedings of the Combustion Institute, Vol. 28, The
Combustion Inst., Pittsburgh, PA, 2000, pp. 423430.
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. 407413.
62
Drallmeier, J., Hydrocarbon-Vapor Measurements in Pulsed Fuel Sprays, Applied
Optics, Vol. 33, No. 33, Nov. 1994, pp. 77817788.
J. G. LEE AND D. A. SANTAVICCA 528
63
Edwards, J. L., Gouldin, F. C., and MacDonald, M. A., High Speed Absorption
Tomography with Advanced Reconstruction Algorithms, AIAA Paper 2003-1013, Jan.
2003.
64
Eckbreth, A. C., Laser Diagnostics for Combustion Temperature and Species, 2nd ed.
Gordon and Breach, New York, 1996.
65
Nguyen, Q.-V., and Paul, P. H., The Time Evolution of a Vortex-Flame Interaction
Observed via Planar Imaging of CH and OH, Proceedings of the Combustion Institute,
Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 357364.
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. 439441.
67
Hanson, R. K., Combustion Diagnostics: Planar Floweld 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. 343362.
69
Pringsheim, P., Fluorescence andPhosphorescence, Interscience Publishers, NewYork,
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. 369376.
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. 93101.
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. 229240.
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. 384387.
77
Melton, L. A., Exciplex-Based Vapor/Liquid Visualization Systems Appropriate for
Automotive Gasolines, Applied Spectroscopy, Vol. 47, No. 6, 1993, pp. 782786.
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 Mechanismin Lean Premixed
Dump Combustors, Ph.D. Thesis, The Pennsylvania State University, University Park, PA,
2000.
80
Lee, S.-Y., Seo, S., Broda, J. C., Pal, S., and Santoro, R. J., An Experimental Estimation
of Mean Reaction Rate and Flame Structure During Combustion Instability in a Lean
Premixed Gas Turbine Combustor, Proceedings of the Combustion Institute, Vol. 28, The
Combustion Inst., Pittsburgh, PA, 2000, pp. 775782.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 529
81
Santhanam, V., Knopf, F. C., Acharya, S., and Gutmark, E., Fluorescence and Tem-
perature Measurements in an Actively Forced Swirl-Stabilized Spray Combustor, Journal
of Propulsion and Power, Vol. 18, No. 4, 2002, pp. 855865.
82
Peters, N., Laminar Flamelet Concepts in Turbulent Combustion, Proceedings of the
Combustion Institute, Vol. 21, The Combustion Inst., Pittsburgh, PA, 1986, pp. 12321250.
83
Buschmann, A., Dinkelacker, F., Sch afer, T., and Wolfrum, J., Measurement of the
Instantaneous Detailed Flame Structure in Turbulent Premixed Combustion, Proceedings
of the Combustion Institute, Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 437
445.
84
Dunkelacker, F., Soika, A., Most, D., Hofman, D., Leipertz, A., Polifke, W., and
D obbeling, K., Structure of Locally Quenched Highly Turbulent Lean Premixed Flames,
Proceedings of the Combustion Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA,
1998, pp. 857865.
85
Grehan, G., Gouesbet, G., and Rabasse, C., Monotonic Relationships Between Scat-
teredPowers andDiameters inLorenz-Mie Theoryfor Simultaneous VelocimetryandSizing
of Particles, Applied Optics, Vol. 20, No. 5, March 1981, pp. 796799.
86
Stenzler, J. N, Lee, J. G., and Santavicca, D. A., Penetration and Dispersion of Liquid
Jets in a Heated Crossow, AIAA Paper 2003-1327, 2003.
87
Lin, K. C., Kennedy, P. J., and Jackson, T. A., Penetration Heights of Liquid Jets in
High-Speed Crossows, 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 962760, 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 841199, 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. 441452.
V. Combustion Instability Control
Chapter 17
Passive Control of Combustion Instabilities in
Stationary Gas Turbines
Geo A. Richards
Energy Systems Dynamics Focus Area Leader, National Energy Technology Laboratory.
Mechanical Engineer.
Scientist.
533
G. A. RICHARDS ET AL. 534
= ratio of specic heats
= the ratio of the speed of sound to area, c/s (m s)
1
= bulk gas velocity, m/s
= acoustic mass velocity, kg/s
= gas density, kg/m
3
= the transfer function relating acoustic pressure to velocity source, Pa
s/kg
= bulk time lag, s
= circular frequency, 2 f , rad/s
I. Introduction
C
OMBUSTION dynamics has become a signicant 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 signicant 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 ame-transfer functions has been
condensed signicantly. 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 modications for
solving dynamics problems is discussed. Section III also describes the similarities
betweenthe control concepts introducedinSec. II andtime-lagmodels. Techniques
to enhance stability, such as using multiple time lags or adding a pilot ame are
discussed in Sec. III, as well. In Sec. IV, a reviewof 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
interacts with the ame 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
Acoustic
Processes
H
Combustion
Processes
G
p'
q'
-
+
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
creation of sound by unsteady heat release is a complex process,
5
but, in simple
terms, q
Flame
Generated Reflected
Fig. 17.2 Schematic illustration of an unstable premixed combustion process.
produced in the ame 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
ame-surface area,
11
changes in equivalence ratio,
1, 1217
vortex shedding,
1823
changes in the bulk ow,
24
and changes in ame anchoring.
2527
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 fewpercent
of the operating pressure. Therefore, modest perturbations in the combustor pres-
sure will create signicant variations in premixer airow, and subsequent variations
in fuelair ratio in the premixer. These variations in fuelair mixture are transported
to the ame after a convection time lag , creating a heat-release perturbation that
may add to perturbations produced by other mechanisms, such as a variable ame
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 sufcient detail. An operational transfer
function is simply a mathematical model that relates the output froman 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(t ) enters at the left, and a resulting signal B cos(t +) exits at
the right. Considering a range of frequencies , the ratio B/A is the gain of the
transfer function, and 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
j t
and B e
j (t +
, respectively. By using this notation, the constant
B can be redened as a complex quantity including the phase angle (B
= Be
j
).
In this manner, the time dependence is not needed in a feedback-loop analysis
because all the blocks have the same time dependence e
j t
. Thus, the transfer
function is the complex ratio of the output to the input, or B
/A.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 537
Transfer
function
A cos(t) B cos(t+)
a) Transfer function nomenclature
B'/A D'/C
connect
B' D'
A C
equivalent to
b) Series connection of transfer functions
B'/A
D'/C
+
+
B'/A +D'/C
equivalent to
c) Parallel connection of transfer functions
jt
A e
jt+
B e
jt
= B' e
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
/C
process, the net transfer function is (B
/A)
(D
/A) +(D
/
/
0.0
0.1
0.2
0.3
0 100 200 300 400
Frequency (Hz)
M
a
g
n
i
t
u
d
e
-200
-100
0
100
200
0 100 200 300 400
Frequency (Hz)
P
h
a
s
e
(
d
e
g
)
H
Q q
P p
=
/
/
0.0
0.1
0.2
0.3
0 100 200 300 400
Frequency (Hz)
M
a
g
n
i
t
u
d
e
-200
-100
0
100
200
0 100 200 300 400
Frequency (Hz)
P
h
a
s
e
(
d
e
g
)
-200
-100
0
100
200
0 100 200 300 400
Frequency (Hz)
P
h
a
s
e
(
d
e
g
)
0.0
2.0
4.0
6.0
0 100 200 300 400
Frequency (Hz)
M
a
g
n
i
t
u
d
e
e G
P p
Q q
6
/
/
= =
-200
-100
0
100
200
0 100 200 300 400
Frequency (Hz)
P
h
a
s
e
(
d
e
g
)
0.0
2.0
4.0
6.0
0 100 200 300 400
Frequency (Hz)
M
a
g
n
i
t
u
d
e
b)
j j
e G
P p
Q q
6
/
/
= =
0.0
0.5
1.0
1.5
2.0
0 100 200 300 400
Frequency (Hz)
M
a
g
n
i
t
u
d
e
-200
-100
0
100
200
0 100 200 300 400
Frequency (Hz)
P
h
a
s
e
(
d
e
g
)
GH
0.0
0.5
1.0
1.5
2.0
0 100 200 300 400
Frequency (Hz)
M
a
g
n
i
t
u
d
e
-200
-100
0
100
200
0 100 200 300 400
Frequency (Hz)
P
h
a
s
e
(
d
e
g
)
GH c)
Fig. 17.5 Frequency response (Bode plots) of transfer functions used in the rst
example problem with a single acoustic mode (Fig. 17.4, top).
normalized pressure response ( p
/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 ame at 140 Hz. At this frequency, there
is no pressure response at the ame 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
G. A. RICHARDS ET AL. 540
to the acoustic losses or damping in the system. In this example, acoustic losses
arise from the step expansion, mean ow, 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 fuelair ratio, variable ame
area). For the purposes of this example, a simple ame-transfer function will be
considered. Again referring to the normalized perturbations ( p
/P and q
/Q), the
ame will be treated as having a constant gain of magnitude 6.0, but with a time
delay = 2 ms relative to the acoustic pressure at the ame. This transfer function
is 6.0e
j 2 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 = 2 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
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 541
Nyquist Plot
-2
-1
0
1
2
-2 -1 0 1 2
Real
I
m
a
g
i
n
a
r
y
258 Hz
a)
Nyquist Plot
-2
-1
0
1
2
-2 -1 0 1 2
Real
I
m
a
g
i
n
a
r
y
251 Hz
263 Hz
241 Hz
= 2 ms
Nyquist Plot
-2
-1
0
1
2
-2 -1 0 1 2
Real
I
m
a
g
i
n
a
r
y
241 Hz = 2.6 ms
c)
= 1.6 ms
Nyquist Plot
-2
-1
0
1
2
-2 -1 0 1 2
Real
I
m
a
g
i
n
a
r
y
258 Hz
Nyquist Plot
-2
-1
0
1
2
-2 -1 0 1 2
Real
I
m
a
g
i
n
a
r
y
251 Hz
263 Hz
241 Hz
= 2 ms
b)
Nyquist Plot
-2
-1
0
1
2
-2 -1 0 1 2
Real
I
m
a
g
i
n
a
r
y
241 Hz = 2.6 ms
= 1.6 ms
Fig. 17.6 Nyquist plots for the single acoustic mode example (Fig. 17.4, top) at three
different time lags ( = 1.6, 2.0, and 2.6 ms).
G. A. RICHARDS ET AL. 542
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 denition 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 reection of the Bode plot into the
negative-frequency axis, and the open-loop systemmust 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 systemwill 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 benet of the Nyquist analysis becomes very apparent when assessing how
different time lags affect systemstability. Figure 17.6 shows the open-loop system,
GH, at three different values of the time lag: = 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 signicant
(230270 Hz), changes to the phase angle = 2 f are dominated by changes
in . Changes in 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 that almost produce
encirclement of the 1 point are stability boundaries for the system. Each of these
stabilityboundaries, or values of , has a correspondingfrequencyat whichthe lobe
crosses the negative real axis. In this example, the system would be unstable for
frequencies of 258 and 241 Hz at = 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 systemin this example were actually operated in the closed-loop mode, the
limit-cycle frequencies could be estimated from the so-called describing-function
theory.
30
Under the assumption of real-valued describing functions, the limit-cycle
frequency would correspond to the frequency at which the Nyquist plot crosses
the negative real axis. Considering the sequence of time lags in Fig. 17.6, the
implication is that as the time lag increases from 1.6 to 2.6 ms, the frequency
would change from 258 to 241 Hz. This frequency shift as a function of time lag
is a general feature of the Nyquist analysis, and the range of frequencies depends
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 543
on the specic 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
j 2 f
. The resulting Bode plot
with = 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 difcult 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 = 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
0.0
1.0
2.0
3.0
4.0
5.0
0 100 200 300 400 500
Frequency (Hz)
M
a
g
n
i
t
u
d
e
-200
-150
-100
-50
0
50
100
150
200
0 100 200 300 400 500
Frequency (Hz)
P
h
a
s
e
(
d
e
g
r
e
e
s
)
Fig. 17.7 The open-loop frequency response (Bode plot) for the example with two
resonant acoustic modes (Fig. 17.4, bottom), = 1.5 ms.
G. A. RICHARDS ET AL. 544
Nyquist Plot
-3
-2
-1
0
1
2
3
-3 -2 -1 0 1 2 3
Real
I
m
a
g
i
n
a
r
y
184
Nyquist Plot
-3
-2
-1
0
1
2
3
-3 -2 -1 0 1 2 3
Real
I
m
a
g
i
n
a
r
y
408 Hz
186 Hz
= 1.5 ms
= 2.0 ms
Nyquist Plot
-3
-2
-1
0
1
2
3
-3 -2 -1 0 1 2 3
Real
I
m
a
g
i
n
a
r
y
411 Hz
= 0.9 ms
a)
b)
c)
Nyquist Plot
-3
-2
-1
0
1
2
3
-3 -2 -1 0 1 2 3
Real
I
m
a
g
i
n
a
r
y
184
Nyquist Plot
-3
-2
-1
0
1
2
3
-3 -2 -1 0 1 2 3
Real
I
m
a
g
i
n
a
r
y
408 Hz
186 Hz
= 1.5 ms
= 2.0 ms
Nyquist Plot
-3
-2
-1
0
1
2
3
-3 -2 -1 0 1 2 3
Real
I
m
a
g
i
n
a
r
y
411 Hz
= 0.9 ms
Fig. 17.8 Nyquist plots for the example with two resonant acoustic modes (Fig. 17.4,
bottom) at three different time lags ( = 0.9, 1.5, and 2.0 ms).
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 545
the time lag to = 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 = 2.0 ms causes the
instability to shift frequency modes and become unstable at 184 Hz.
The rotation of these lobes underscores the fundamental problemwith 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 ame response is a constant magnitude.
In real applications, the transfer-function magnitude and phase are governed by
the ame response to acoustic perturbations. As noted earlier, this response can
involve multiple physical processes that are often difcult to differentiate. The
next section reviews the physical processes associated with the ame response and
various approaches used to describe a combustion-transfer function.
D. Physical Processes Contributing to the Combustion-Transfer Function
The heat release froma premixed ame is the product of the reactants consumed
by the ame and the heat of reaction. The heat release can be written:
Q = Y
f
SA
f
H (17.1)
where is the density of the reactants, Y
f
is the mass fraction of fuel in the
premixed gases, S is the ame speed, A
f
is the area of the ame, 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, ame speed,
and ame 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 signicant modulation in ame area and may not be neglected. Likewise, changes
in the owof either fuel and/or air will change the fuel mass fraction and the ame
speed.
In short, numerous mechanisms can generate perturbations in the heat release
at the ame. In most practical applications, it is difcult 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 ames with constant fuelair ratio will be discussed.
Following this discussion, the combined problem of fuelair variation and ame-
area response will be considered.
1. Simple Premixed Flames
The transfer function of fully premixed ames has been investigated by various
authors. Blackshear
31
proposed one of the earliest models for the response of a
G. A. RICHARDS ET AL. 546
premixed burner to acoustic perturbations. This work showed that variations in the
ame area were responsible for the driving, or damping, acoustic waves imposed on
the burner. Companion experiments demonstrated that the ame response depends
signicantly on the mean ow velocity and the fuelair ratio. Merk
32
presented an
improved analysis of a premixed burner ame and was able to derive an explicit
expression for the ame-transfer function [see Eq. (17.2)].
A
=
1
1 + j
1
u
(17.2)
Here, the perturbation quantities are normalized by their corresponding steady-
state values. This transfer function represents a rst-order response between the
dimensionless ame area A
. The analysis
identies a characteristic time
1
that represents the average time for gas exiting
the burner to be consumed by the ame cone. For clarity, the notation
1
is used
to make a distinction with the convective time lag identied in Fig. 17.2. Note
that the convective time lag includes the premixing process.
Note that for large values of
1
, Eq. (17.2) predicts that the ame-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
j
described in the example problem,
which will have arbitrarily large phase angles for large . This distinction will
be noted in a comparison to experiments discussed later.
Further investigations of the ame-transfer functionhave beencarriedout, which
include both analytical work and experimental data.
11, 3337
Mugridge
35
reports
preliminary attempts to measure the ame-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.
Matsui
36
investigated experimental data from three multiport premixed burner
congurations and again identied a characteristic time lag in the ame-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 Matsuis transfer function includes a multiplier
with a pure time-lag term (e
j
). As noted previously, this pure time-lag term
allows the transfer function to reach large phase angles for large .
More recently, Fliel et al.
11
developedananalytical model for the ame-transfer
function that describes the distortion of the ame-surface area in response to im-
posed velocity perturbations at the base of the ame. Unlike earlier work,
31, 32, 36
this model accounted for the distortion of the ame surface by tracking the kine-
matics of the ame movement in the oscillating ow. The variation in heat release
again results from the variable ame-surface area. The predicted transfer function
is qualitatively similar to earlier models and can be approximated as a rst-order
system. However, no experimental data were used to verify the predictions.
Using a theoretical model very similar to Fliel et al.,
11
Ducruix et al.
37
com-
pared the measured and predicted ame-transfer function of a bunsen ame 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 congurations and at several operating conditions. In contrast, the phase
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 547
of the transfer function was poorly predicted for frequencies beyond 30 Hz. For
frequencies higher than 30 Hz, the phase of the response depended signicantly
on the mean ow velocity and the burner dimensions. Additional experiments (not
reported) showed that the spatial velocity distribution at the base of the ame must
be accounted for to improve model predictions.
Although the preceding discussion is limited to relatively simple premixed Bun-
sen or jet ames, the difculty of accurately predicting a ame-transfer function
is clear. In bluff or step-stabilized ames (e.g., afterburners or dump combustors),
ame-area variations may originate from oscillations in the shear layer and from
vortex merging.
1824, 39
In swirl-stabilized ames, 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 ow dynamics. Thus, for
swirl-stabilized ames, there is no general approach to estimate the contribution
of the ame-area variation to the ame response. Some recent attempts to measure
or predict transfer functions for swirl ames are discussed below.
2. Practical Swirl-Stabilized Flames
In addition to aerodynamic processes, perturbations in the fuelair ratio that
occur at the fuel injector (Fig. 17.2) will result in variations in the heat release
q
after a convective time lag . Although the supply-system dynamics are not
formally a part of a ame-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, 4044
which describe the importance of the supply system dynamics. Several recent
papers
45, 46
suggest that it is also necessary to account for the dispersion of fuelair
perturbations in the premixer when describing the supply-system dynamics.
For some combustor congurations, ame-area variations occur simultaneously
with the feed-system dynamics. For example, Peracchio and Proscia
6
have pro-
posed a model that includes both the variation in ame area and the simultaneous
variation in fuelair 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 ame-area variations is very signicant and probably larger than expected from
a simple ame model. These observations were made based on visual observations
of the ame and unpublished computations of uid dynamics. In contrast, other
experiments
47
show only modest structural change in practical turbine ames dur-
ing oscillations. This result is not general, however, because the same combustor
exhibited signicant ame-area variation when tested at atmospheric pressure.
47
In addition to the ame-area effects, the ame-anchoring method also plays an
important role in the combustion-transfer function. Kendrick et al.
26
compared the
dynamics of two different fuelair premixers and showed that ame-anchoring
methods will signicantly affect the ame 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 ame anchoring, allowing the ame-reaction zone to oscillate
axially in the ow throughout the pressure cycle. This axial movement may add
another complexity to the ame-transfer function. Schuermans and Polifke
25
also
G. A. RICHARDS ET AL. 548
Frequency (Hz)
M
a
g
n
i
t
u
d
e
(
d
B
)
= 0.55
= 0.60
= 0.65
Frequency (Hz)
P
h
a
s
e
(
d
e
g
r
e
e
s
)
= 0.55
= 0.60
= 0.65
10 100 1000
0
-200
-400
-600
-800
-1000
100 1000 10
-15
-25
-35
-45
M
a
g
n
i
t
u
d
e
(
d
B
)
P
h
a
s
e
(
d
e
g
r
e
s
)
Frequency (Hz)
Frequency (Hz)
Frequency (Hz)
M
a
g
n
i
t
u
d
e
(
d
B
)
= 0.55
= 0.60
= 0.65
Frequency (Hz)
P
h
a
s
e
(
d
e
g
r
e
e
s
)
= 0.55
= 0.60
= 0.65
10 100 1000
0
-200
-400
-600
-800
-1000
100 1000 10
-15
-25
-35
-45
M
a
g
n
i
t
u
d
e
(
d
B
)
P
h
a
s
e
(
d
e
g
r
e
s
)
Frequency (Hz)
Frequency (Hz)
Fig. 17.9 Measured ame response to perturbations in acoustic velocity equivalence
ratios () from Khanna.
51
noted that it is necessary to include ame translation in their analytical model of
the ame response.
Several recent papers have attempted to measure the dynamic response of
premixed, swirl-stabilized ames that are characteristic of low-emission gas
turbines.
4852
As an example, Khanna
51
and Khanna et al.
52
measured the ame-
transfer function of a swirl-stabilized ame with fuelair 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-eld acoustics that are not accounted for in a
simple feedback model between the local acoustic pressure p
andq
and q
perturbations be precisely
in phase to meet the Rayleigh criterion. In principle, oscillations can occur for a
range of 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
L
L
Volume, V
Radius, a
Area, S
Fuel
Air
Helmholtz
Resonator
Quarter
wave
resonator
f
0
c
2
S
L 1.7 a + ( ) V
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 onthe resonator geometryandthe
operating conditions. Because the acoustic dissipation occurs at the entrance/exit
ownear the mouth, the geometry of this region is very important. Laudien et al.
72
show that rounded versus square corners at the resonatorcombustor connection
produce a signicant 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 specic
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 nalized 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 specic 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
specic situations. Some examples are described in the next section.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 563
B. Application of Acoustic Dampers
Gyslinget al.
74
examinedthe use of Helmholtz resonators ona sector rigcombus-
tor. Their analysis identied many of the important design variables for installing
a resonator. For example, the resonator-to-combustor volume ratio, the resonator
frequency, and the loss coefcient at the resonator mouth are all important consid-
erations in a practical acoustic damper design. The nal 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 tting the theoretical parameters to
the resulting data. Note that this t was made at each operating condition of inter-
est. As noted earlier, the gas temperature and owconditions in the combustor will
affect the resonator natural frequency and damping, which needs to be accounted
for in resonator performance.
Five resonator congurations 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 systemmust 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 nal example of the use of resonators, Pandalai and Mongia
3
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 fuelair 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 nowin 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
signicant role in reducing pressure oscillations in stationary engines.
G. A. RICHARDS ET AL. 564
A few words of clarication 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 sufcient 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. Mitchell
76
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 specic example to demonstrate both the method of
design and the potential benets of acoustic dampers. However, this analysis is
theoretical, and as previously mentioned, experimental rig tests should be used to
nalize 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 . 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 specic 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(410Hz) of the combustor showninFig. 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 rst 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 conguration. 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.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 565
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 ow 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 ame 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 difculty in resonator design. Because
the resonator contents may include a mix of purge gas and combustion products,
it is difcult 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 ame 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 dened as the length of the cylindrical combustor needed to achieve a plug
ow residence time of 30 ms, which is typical of the time needed for complete CO
oxidation in premix combustors. Note that the specic geometry of the resonator
volume does not need to be precisely dened. 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:
f
0
=
c
2
_
S
(L +1.7a)V
(17.5)
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 signicant 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.
G. A. RICHARDS ET AL. 566
10 cm
Flame
Resonator
D
C
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 dened, the resonator location can be assessed
by using the system model described in the example problem. Experimental ver-
ication 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:
Z
r
=
2 f
2
c
+
4
S
_
+
L
2a
_
_
vis
f
+ j
2 f (L +1.7a)
S
_
1
f
2
0
f
2
_
(17.6)
Inthis expression, f is the frequencyand f
0
is the natural frequencyof the resonator
calculated from Eq. (17.5). The speed of sound c and the dynamic viscosity
vis
are evaluated at the (assumed) gas conditions in the resonator. The dimensionless
term is a resistance factor that depends on the quality of the orice and nonlinear
processes at the mouthof the resonator. This termis difcult toevaluate analytically
and depends on the magnitude of the oscillating pressure. The resistance factor
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 dene the range of possible behavior before
conducting experimental testing. For this example, investigation of the effect of
(not reported) showed only a modest effect on resonator attenuation.
Given the geometry and parameter selections described in preceding text,
Fig. 17.20 presents the open-loop frequency response of the example combus-
tor without the resonator. Note that the baseline case without the resonator is
shown in the top of Fig. 17.20. Recall that the baseline corresponds to the spec-
tra of Fig. 17.7. With the resonator tuned to 410 Hz, the amplitude of the higher
combustor mode is replaced by two much smaller adjacent modes. Gysling et al.
74
noted that the addition of a resonator produced two new eigenvalues, representing
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 567
0.0
1.0
2.0
3.0
4.0
5.0
0 100 200 300 400 500
Frequency (Hz)
M
a
g
n
i
t
u
d
e
0.0
1.0
2.0
3.0
4.0
5.0
0 100 200 300 400 500
Frequency (Hz)
M
a
g
n
i
t
u
d
e
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 newfrequencies
G. A. RICHARDS ET AL. 568
Nyquist Plot
-3
-2
-1
0
1
2
3
-3 -2 -1 0 1 2 3
Real
I
m
a
g
i
n
a
r
y
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 ame, 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 systemresponse 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
0.0
0.2
0.4
0.6
0.8
1.0
300 400 500
Frequency (Hz)
M
a
g
n
i
t
u
d
e
430
390
410
Fig. 17.22 Effect of resonator tuning on the Bode plot for the higher-frequency
range. Resonators with natural frequencies 390 and 430 Hz are compared with
the baseline (410 Hz).
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 569
0.0
0.2
0.4
0.6
0.8
1.0
300 400 500
Frequency (Hz)
M
a
g
n
i
t
u
d
e
A
B
C
D
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 ame-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 ame response have been
reviewed, and some examples of the complex details of practical ame-transfer
functions were provided. In Sec. III, applications of time-lag modications 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 eld applications that conrmed the benecial effect of increas-
ing the time-lag distributions. Improvements to stability from ame-anchoring
modications, and pilot ames 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 elded-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 gasication 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 efciency. At the same time, it is desirable to reduce
the cooling and dilution ows 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
G. A. RICHARDS ET AL. 570
ows 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.
17.517.8. The acoustic computations are described rst, followed by a deriva-
tion of the response of the system to a velocity source located at the ame and
incorporation of a ame 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
identied by using subscripts A
i, 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 Aand B, with air entering at the upstreamside of region Aand
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 fuelair mixture enters
the combustion chamber at region D where it continues to ow 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 Eand G. Computation stations, numbered 0 to 9, are also shown in
Station #'s
Air
Fuel
Flame
Combustion
products
A
0 1 2 3 4 5 6 9
B D E
G
Fig. 17.A1 Generic combustor geometry (not to scale).
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 571
Table 17.A1 Geometry and gas conditions used in the two-mode combustor model
Geometric region
Dual-mode combustor
simulation parameters A+B D E G
Diameter, m 0.06858 0.13335 0.13335 0.17780
Length, m 0.06350 0.01270 1.39700 0.63500
Area, m
2
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/m
3
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
p
in
p
4
= p
3
in
0
= 0
p
out
p
3
= p
4
4
=
3
3
+
s
out
9
= 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 specic 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,
_
p
1
1
_
= C
_
p
2
2
_
(17.A1)
_
p
0
0
_
= AB
_
p
1
1
_
(17.A2)
and
_
p
2
2
_
= D
_
p
3
3
_
(17.A3)
The acoustic pressures p and acoustic velocities are represented by complex
phasors. The 2 2 matrices AB, C, and D are acoustic transfer matrices for their
G. A. RICHARDS ET AL. 572
respective regions. The matrices C and F do not correspond to specic 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. Asimilar linear relationship does not exist
between stations 3 and 4 because of the presence of the ame, which acts as an
acoustic source.
The 2 2 acoustic transfer matrices for various geometries are dened in
Munjal.
28
For simple cylindrical elements:
R(s, L, , f, c) =
_
_
_
cos
_
2 f L
c
_
j c
s
sin
_
2 f L
c
_
j s
c
sin
_
2 f L
c
_
cos
_
2 f L
c
_
_
_
_
exp
_
j
c
2 f
c
L
_
(17.A4)
For step expansions such as the region between stations 1 and 2, the transfer matrix
is
S(
a
,
b
, M
a
, M
b
, k, ) :=
_
1 M
b
b
M
b
b
1
_
1
_
_
1
kM
2
a
1M
2
a
kM
a
a
1M
2
a
( 1)kM
3
a
(1M
2
a
)
a
1
( 1)kM
2
a
1M
2
a
_
_
_
1 M
a
a
M
a
a
1
_
(17.A5)
where k, the loss coefcient for a step expansion, is
k =
_
s
b
s
a
1
_
2
(17.A6)
The acoustic impedances at each station in the model are calculated starting from
known boundary conditions. At station 0,
0
= 0 such that
Z
0
= p
0
/
0
= (17.A7)
At station 1,
0
= 0 = AB
1, 0
p
1
+AB
1, 1
1
, which implies
Z
1
= p
1
/
1
= AB
1, 1
/AB
1, 0
(17.A8)
At station 2,
Z
1
=
p
1
1
=
C
0, 0
p
2
+C
0, 1
2
C
1, 0
p
2
+C
1, 1
2
1
2
=
C
0, 0
Z
2
+C
0, 1
C
1, 0
Z
2
+C
1, 1
(17.A9)
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 573
therefore
Z
2
=
(Z
1
C
1, 1
C
0, 1
)
(Z
1
C
1, 0
C
0, 0
)
(17.A10)
And, similarly at station 3,
Z
3
=
(Z
2
D
1, 1
D
0, 1
)
(Z
2
D
1, 0
D
0, 0
)
(17.A11)
At station 9,
9
= 0 such that:
Z
9
= p
9
/
9
= (17.A12)
At station 6,
Z
6
=
p
6
6
=
G
0, 0
p
9
+G
0, 1
9
G
1, 0
p
9
+G
1, 1
9
=
G
0, 0
G
1, 0
(17.A13)
At station 5,
Z
5
=
p
5
5
=
F
0, 0
p
6
+F
0, 1
6
F
1, 0
p
6
+F
1, 1
6
=
F
0, 0
Z
6
+F
0, 1
F
1, 0
Z
6
+F
1, 1
(17.A14)
And at station 4,
Z
4
=
E
0, 0
Z
5
+E
0, 1
E
1, 0
Z
5
+E
1, 1
(17.A15)
The ame is modeled as an acoustic velocity source located in the plane be-
tween stations 3 and 4. The boundary conditions at this interface are p
3
= p
4
and
4
=
3
(
4
/
3
) +
s
. The transfer function relating the acoustic pressure to the
velocity source, that is,
=
p
3
s
(17.A16)
is derived in terms of system impedances, starting with the denition of the
impedance at station 4:
Z
4
=
p
4
4
=
p
3
3
4
3
+
s
=
p
3
3
4
3
+
s
1
p
3
1
p
3
=
1
1
Z
3
3
+
1
(17.A17)
Therefore,
=
Z
3
Z
4
Z
3
3
Z
4
(17.A18)
G. A. RICHARDS ET AL. 574
The transfer function relating acoustic pressure to variations in heat release is
derived by relating heat release and an acoustic velocity source.
25
s
=
p
3
=
_
4
_
T
4
T
3
1
_
3
s
3
_
q
Q
+
_
4
_
T
4
T
3
1
_
3
s
3
_
p
3
P
3
(17.A19)
Equation 17.A19 provides the last link needed to derive H, the relative pressure
response to oscillations in heat release. Derivation of the Htransfer function started
with equation A19 by separating terms involving p
3
from
3
and dividing through
by P
3
to yield
p
3
P
3
_
1
1
P
3
_
4
_
T
4
T
3
1
_
3
s
3
__
=
q
Q
_
4
_
T
4
T
3
1
_
3
s
3
P
3
_
(17.A20)
Therefore,
H =
p
3
/P
3
q
/Q
=
{
4
[(T
4
/T
3
) 1] (
3
s
3
/P
3
)}
((1/) (1/P
3
) {
4
[(T
4
/T
3
) 1]
3
s
3
})
(17.A21)
A simple form for transfer function G, the relative heat-release response to pres-
sure uctuations, 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:
G =
q
/Q
p
3
/P
3
=
1
2
e
1 j w
(17.A22)
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, GH, 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 handled
28
by adding one more acoustic matrix T that
includes the impedance of the resonator mouth Z
r
:
T =
_
1 0
1/Z
r
1
_
(17.A23)
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 dene 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 dening
sec. E.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 575
References
1
Hobson, D. E., Fackrell, J. E., and Hewitt, G., Combustion Instabilities in Indus-
trial Gas Turbines - Measurements on Operating Plant and Thermoacoustic Modeling,
Journal of Engineering for Gas Turbines and Power, Vol. 122, No. 3, 2000, pp. 420
428.
2
Sholz, M. H., and Depietro, S. M., Field Experience on DLN Typhoon Industrial Gas
Turbines, American Society of Mechanical Engineers, Paper 97-GT-61, 1997.
3
Pandalai, R. P., and Mongia, H. C., Combustion Instability Characteristics of Industrial
Engine Dry Low Emission Combustion Systems, AIAA Paper 98-3379, 1998.
4
Richards, G. A., Straub, D. L., and Robey, E. H., Passive Control of Combustion
Dynamics in Stationary Gas Turbines, Journal of Propulsion and Power, Vol. 19, No. 5,
2003, pp. 795810.
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. 603612.
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. 415421.
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., Hofnan, S., Krebs, W., and Bohn. D., Prediction of Thermoa-
coustic Instabilities with Focus on the Dynamic Flame Behavior for the 3A-Series Gas
Turbine of Siemens KWU, American Society of Mechanical Engineers, Paper 99-GT-11,
1999.
9
Harrje, D. T., and Reardon, F. H., Liquid Propellant Rocket Combustion Instability,
NASA SP-194, 1971.
10
Arana, C. A., Sekar, B., and Mawid, M. A., Determination of Thermoacoustic Re-
sponse in a Demonstrator Gas Turbine Engine, Journal of Engineering for Gas Turbines
and Power, Vol. 124, No. 1, 2000, pp. 4657.
11
Fleil, 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. 487510.
12
Steele, R. C., Cowell, L. H., Cannon, S. M., and Smith, C. E., Passive Control of
Combustion Instability in Lean Premixed Combustors, Journal of Engineering for Gas
Turbines and Power, Vol. 122, No. 3, 2000, pp. 412419.
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. 294302.
14
Keller, J. J., Thermoacoustic Oscillations in Combustion Chambers of Gas Turbines,
AIAA Journal, Vol. 33, No. 12, 1995, pp. 22802287.
15
Lieuwen, T., and Zinn, B. T., The Role of Equivalence Ratio Oscillations in Driv-
ing Combustion Instabilities in Low NOx Gas Turbines, Proceedings of the Combustion
Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 18091816.
16
Lee, J. G., Kim, K., and Santavicca, D. A., Measurement of Equivalence Ratio Fluc-
tuation and Its Effect on Heat-release During Unstable Combustion, Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 415421.
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
G. A. RICHARDS ET AL. 576
Mixing with Applications to Low NOx Emission in Premixed Gas Turbines, Proceedings
of the Combustion Institute, Pittsburgh, PA, Vol. 26, 1996, pp. 27492755.
18
Schadow, K. C., and Gutmark, E., Combustion Instabilities Related to Vortex Shedding
in Dump Combustors and Their Passive Control, Progress in Energy and Combustion
Science, Vol. 18, 1992, pp. 117132.
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. 161166.
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. 265292.
23
Langhorne, P. J., Reheat Buzz: An Acoustically Coupled Combustion Instability,
Part 1, Experiment, Journal of Fluid Mechanics, Vol. 193, 1988, pp. 417443.
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. 4772.
25
Schuermans, B. H., and Polifke, W., Modeling Transfer Matrices of Premixed Flames
and Comparison with Experimental Results, American Society of Mechanical Engineers,
Paper 99-GT-132, 1999.
26
Kendrick, D. W., Anderson, T. J., Sowa, W. A., and Snyder, T. S., Acoustic Sensitivities
of Lean-Premixed Fuel Injectors in a Single Nozzle Rig, Journal of Engineering for Gas
Turbines and Power, Vol. 121, No. 3, 1999, pp. 429436.
27
Pascheriet, C. O., Flohr, P., Knopfel, H., Geng, W., Steinbach, C., Stuber, P., Bengtsson,
K., and Gutmark, E., Combustion Control by Extended EVBurner Fuel Lance, American
Society of Mechanical Engineers, Paper GT-2002-30462, 2002.
28
Munjal, M. L., Acoustics of Ducts and Mufers, 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., PrenticeHall,
Upper Saddle River, NJ, 2000.
31
Blackshear, P. L. Driving Standing Waves by Heat Addition, Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 4, 1953, pp. 553556.
32
Merk, H. J. An Analysis of Unstable Combustion of Premixed Gases, Proceedings
of the Combustion Institute, Vol. 6, 1956, pp. 501512.
33
Becker, R., and Gunther, R., The Transfer Function of Premixed Turbulent Jet Flames,
Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 13, 1971, pp. 517526.
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. 437452.
36
Matsui, Y., An Experimental Study on Pyro-Acoustic Amplication of Premixed
Laminar Flames, Combustion and Flame, Vol. 43, 1981, pp. 199209.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 577
37
Ducruix, S., Durox, D., and Candel, S., Theoretical and Experimental Determination
of the Transfer Function of a Laminar Premixed Flame, Proceedings of the Combustion
Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 765773.
38
Hadvig, S., Combustion Instability: SystemAnalysis, Journal of the Institute of Fuel,
Vol. 44, 1971, pp. 550558.
39
Hegde, U. G., Reuter, D., Daniel, B. R., and Zinn, B. T., Flame Driving of Longitudinal
Instabilities in Dump Type Ramjet Combustors, Combustion Science and Technology,
Vol. 55, 1987, pp. 125138.
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. 13531361.
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 Inuence on Pressure Wave Amplication, American Society of
Mechanical Engineers, Paper 2000-GT-0106, 2000.
46
Sattlelmayer, T., Inuence of the Combustor Aerodynamics on Combustion Instabil-
ities From Equivalence Ratio Fluctuations, American Society of Mechanical Engineers,
Paper 2000-GT-0082, 2000.
47
Anderson, T. J., Sowa, W. A., and Morford, S. A., Dynamic Flame Structure in a Low
NOx Premix Combustor, American Society of Mechanical Engineers, Paper 98-GT-568,
1998.
48
Paschereit, C. O., Polifke, W., Schuermns, B., and Mattson, O., Measurement of
Transfer Matrices and Source Terms of Premixed Flames, Journal of Engineering for Gas
Turbines and Power, Vol. 124, No. 2, 2002, pp. 239247.
49
Lawn, C. J., Thermo-Acoustic Frequency Selection by Swirled Premixed Flames,
Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 823830.
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. 713720.
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.
G. A. RICHARDS ET AL. 578
55
Kruger, U., Hoffman, S., Krebs, W., Judith, H., Bohn, D., and Matouschek, G., Inu-
ence of Turbulence on the Dynamic Behavior of Premixed Flames, American Society of
Mechanical Engineers, Paper 98-GT-323, 1998.
56
Smith, C. E., and Leonard, A. D., CFD Modeling of Combustion Instability in Pre-
mixed Axisymmetric Combustors, American Society of Mechanical Engineers, Paper 97-
GT-305, 1997.
57
Zhu, M., Dowling, A. P., and Bray, K. N. C., Flame Transfer Calculations for Com-
bustion Oscillations, American Society of Mechanical Engineers, Paper 2001-GT-0374,
2001.
58
Walz, G., Krebs, W., Hoffmann, S., and Judith, H., Detailed Analysis of the Acous-
tic Mode Shapes of an Annular Combustion Chamber, American Society of Mechanical
Engineers, Paper 99-GT-113, 1999.
59
Cronemyr, P. J. M., Hulme, C. J., and Troger, C., Coupled Acoustic-Structure Analysis
of an Annular DLE Combustor, American Society of Mechanical Engineers, Paper 98-GT-
502, 1998.
60
Pankiewitz, C., and Sattelmayer, T., Time Domain Simulation of Combustion Instabil-
ities in Annular Combustor, American Society of Mechanical Engineers, Paper GT-2002-
30063, 2002.
61
Evesque, S., and Polifke, W., Low-Order Acoustic Modeling for Annular Combustors:
Validation and Inclusion of Modal Coupling, American Society of Mechanical Engineers,
Paper GT-2002-30064, 2002.
62
Krebs, W., Walz, G., Flohr, P., and Hoffmann, S., Modal Analysis of Annular Com-
bustors: Effect of Burner Impedance, American Society of Mechanical Engineers, Paper
2001-GT-0042, 2001.
63
Straub, D. L., and Richards, G. A., Effect of Fuel Nozzle Conguration 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.,
V64.3AGas Turbine Natural Gas Burner Development, American Society of Mechanical
Engineers, Paper GT-2002-30106, 2002.
65
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., A Mechanism of Combustion
Instability in Lean Premixed Gas Turbine Combustors, Journal of Engineering for Gas
Turbines and Power, Vol. 123, No. 1, 2001, pp. 182189.
66
Lovett, J. A., and Uznanski, K. T., Prediction of Combustion Dynamics in a Staged
Premixed Combustor, American Society of Mechanical Engineers, Paper GT-2002-30646,
2002.
67
Berenbrink, P., and Hoffmann, S. Suppression of Dynamics Combustion Instabilities
by Passive and Active Means, American Society of Mechanical Engineers, Paper 2000-
GT-0079, 2000.
68
Richards, G. A., Straub, D. L., and Robey, E. H. Control of Combustion Dynamics
Using Fuel System Impedance, American Society of Mechanical Engineers, Paper GT-
2003-38521, 2003.
69
Lee, J. G., Kim, K., and Santavicca, D. A., A Study of the Role of Equivalence Ratio
Fluctuations During Unstable Combustion in a Lean Premixed Combustor, AIAA Paper
2002-4015, 2002.
70
Lovett, J. A. , Bluffbody Flameholders for Low-Emission Gas Turbine Combustors,
U.S. Patent No. 5,471,840, 1995.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 579
71
Straub, D. L., Richards, G. A., Baumann, W. T., and Saunders, W. R., Measurement of
Dynamics Flame Response in a Lean Premixed Single-Can Combustor, American Society
of Mechanical Engineers, Paper 2001-GT-0038, 2001.
72
Laudien, E., Pongratz, R., Pierro, R., and Preclik, D., Experimental Procedures Aiding
the Design of Acoustic Cavities, Liquid Rocket Engine Combustion Instability, Progress
in Astronautics and Aeronautics, AIAA, Washington, DC, Vol. 169, 1995, pp. 377399.
73
Lewis, G. D., and Garrison, G. D., The Role of Acoustic Absorbers in Preventing
Combustion Instability, AIAA Paper 71-699, 1971.
74
Gysling, D. L., Copeland, G. S., McCormick, D. C., and Proscia, W. M., Combus-
tion System Damping Augmentation with Helmholtz Resonators, American Society of
Mechanical Engineers, Paper 98-GT-268, 1998.
75
Bellucci, V., Paschereit, C. O., Flohr, P., and Magni, F., On the Use of Helmholtz
Resonators for Damping Acoustic Pulsations in Industrial Gas Turbines, American Society
of Mechanical Engineers, Paper 2001-0039, 2001.
76
Mitchell, C. E., Analytical Models for Combustion Instability in Liquid Rocket En-
gine Combustion Instability, Liquid Rocket Engine Combustion Instability, Progress in
Astronautics and Aeronautics, AIAA, Washington, DC, Vol. 169, 1995, pp. 403430.
77
DeBiasi, V., Gasication on Track to Turn Problem Fuels into Electricity and Power,
Gas Turbine World, NovDec, 1999, pp. 1220.
Chapter 18
Factors Affecting the Control of Unstable Combustors
Jeffrey M. Cohen
1
= performance bandwidth
2
= control bandwidth
ii
= power-spectral density function of the input disturbance
pp
= power-spectral density function of the combustor pressure
= required attenuation level for sensitivity function over
performance bandwidth
= fuelair equivalence ratio
= mean fuelair ratio
= frequency
b
= lower-boundary control bandwidth
c
= higher-boundary control bandwidth
= standard deviation of valve command
r
= real part of unstable pole
= delay
Copyright c 2005 by United Technologies Corporation. Published by the American Institute of
Aeronautics and Astronautics, Inc., with permission.
(18.1)
where is the standard deviation of the Gaussian process at the input of the valve.
One can shown that, given the noise-input PSD
i i
( j ), can be found from the
Gaussian process balance equation:
=
_
1
2
_
G
0
( j )G
c
( j )
1+
(2/)(b/) G
0
( j )G
c
( j )
i i
( j ) d (18.2)
which can be solved numerically. Once the value of is known, the pressure PSD
pp
( j ) can be obtained from the formula
pp
( j ) = G
0
( j )/1 + N()G
0
( j )G
c
( j )
i i
( j ) (18.3)
The closed-loop transfer from noise to pressure is G
0
( j )/[1 + G
0
( j )G
C
( j )N()], indicating that for |1 + G
0
( j )G
C
( j )N()| < 1, pressure oscil-
lations are amplied by the controller, and for |1 + G
0
( j )G
C
( j )N()| > 1,
pressure oscillations are attenuated by the controller.
Figure 18.18 shows the Nyquist plot of G
0
( j )G
C
( j )N() in the complex
plane for the value of b corresponding to three valves. [For this plot G
0
( j )G
C
( j )
and were obtainedfrommodel simulationrather thancalculation, mainlybecause
Fig. 18.18 Nyquist diagram for single-nozzle, closed-loop control near the optimum-
control phase, showing that the controller excited secondary peaks (B and C) and
attenuated the primary peak (A).
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 601
the controller G
C
( j ) usedinthe experiments andinthe simulationwas a nonlinear
phase-shifting controller based on a frequency-tracking extended kalman lter,
which does not have a simple linear-transfer function, even though for a xed
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 amplies 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
( j )G
C
( j )N().
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 controllers effectiveness.
Note, however, that increasing the number of actuated nozzles does not correspond
to a proportional gain increase, because the standard deviation 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 colleagues
21, 22
showed (by using the approach of Freudenberg and
Iooze
23
) 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
( j ) can be obtained from the
formula
pp
( j ) = |G
0
( j )S( j )|
2
i i
( j ) (18.4)
where
S( j ) :=
1
1 + G
0
( j )G
c
( j )
(18.5)
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 amplied)
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
r
of
J. M. COHEN AND A. BANASZUK 602
the combustor. This requirement can be stated as
|S( j )| < for
1
(18.6)
where
1
is the so-called performance bandwidth, that is, the interval containing
the resonant frequency
r
over which reduction of pressure oscillations by the fac-
tor of relative to uncontrolled level is enforced. The fundamental limitations
23, 24
yield controller-independent lower bounds on the maximumof the sensitivity func-
tion. Assume that the combustor response-transfer function G
0
( j ) has, at most,
one unstable complex conjugate pole pair with the real part denoted by
r
. If
the combustor model is stable, we dene
r
= 0. An example of fundamental
limitations is the Bode integral formula for the sensitivity function:
_
0
ln |S( j )|d = 2
r
(18.7)
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 (amplication of pressure oscillations) in some
other band. If the control bandwidth is innite, the positive area may be distributed
over a wide frequency range so that amplication at any given frequency may be
designed to be arbitrarily small. However, if the control bandwidth is nite be-
cause of factors such as actuator bandwidth [so that G
0
( j )G
C
( j ) 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.
0 300 400 500 600 700 800 9001000
Frequency (Hz)
0
0
0 200 0 100
0
0.5
C
o
m
b
u
s
t
o
r
P
r
e
s
s
u
r
e
(
p
s
i
)
1
1.5
2
2.5
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.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 603
0 100 200 300 400 500 600 700 800 900 1000
-8
-6
-4
-2
0
2
4
6
8
+
+
-10
dB
-
0 100
+
+
Frequency (Hz)
-
Fig. 18.20 A typical sensitivity function, showing the sensitivity tradeoffs caused by
nite-controller performance bandwidth.
Assume that the combustor-transfer function G
0
( j ) is a relative degree of
at least two (i.e., it has at least two more poles than zeros). This assumption is
typically satised if actuator and sensor dynamics are included in the combustor
response-transfer function. To model the effect due to nite control bandwidth, we
require the open-loop gain to satisfy the inequality:
|G
0
( j )G
c
( j )|
_
_
1+k
for >
c
(18.8)
Here, it is assumed that < 1/2 and k > 0 (relative degree of at least two). We
impose a similar constraint on the loop gain:
|G
0
( j )G
c
( j )|
_
b
_
1+k
for <
b
(18.9)
Let us dene the control bandwidth
2
:=
c
b
. Figure 18.21 illustrates
the nite-bandwidth performance specication 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 specication |S( j )| < for
1
, a control bandwidth specication
must also be met.
We now compute the performance limitations as peaking in the sensitivity-
function magnitude. Let S
:= Sup
|S(j
1
Frequency
2
(Control bandwidth)
(
Performance bandwidth)
Loop gain rolls off
at high frequencies
Loop gain small at
low frequencies
|S(j)|
1
Frequency
2
(Control bandwidth)
(
Performance bandwidth)
Loop gain rolls off
at high frequencies
Loop gain small at
low frequencies
b
Fig. 18.21 Illustration of performance and control bandwidth.
(as in Freudenberg and Iooze
23
) to show that
log S
1
_
2
r
+
1
log
1
b
log
1
1
3
c
2k
_
1
1
(1 +/
c
)
__
(18.10)
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
1
log(1/).
2) The limitation on the actuator bandwidth relative to required performance
bandwidth, which is represented by the amplifying term 1/(
2
1
).
3) The real part of the unstable combustor pole. This inuence is represented
by the term 2
r
. The larger the growth rate of the pole, the larger the peak of the
sensitivity function.
2 3 4 5 6 7 8 9 10
1
1.5
2
2.5
3
3.5
4
4.5
5
2
/
1
= 40 z
= 0.1 (20 dB)
r
= 0
20 dB
15 dB
10 dB
||S||
7
2
3
5
7
2
3
2 1
1
= 40 z
= 0.1 (20 dB)
r
= 0
20 dB
15 dB
10 dB
||S||
8
Fig. 18.22 Lower bounds on sensitivity-function norm as function of control band-
width for three values of .
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 605
2 3 4 5 6 7 8 9 10
1
= 40z
= .3 (10 dB)
= 0.1 (20 dB)
r
= 0
= 43 ms
= 4.3 ms
= 0.6 ms
= 0 (no delay)
2
/
||S||
8
2 3 4 5 6 7 8 9 10
0.6
0.8
1
1.2
1.4
1.6
1.8
2
1
= 40z
= .3 (10 dB)
= 0.1 (20 dB)
r
= 0
= 43 ms
= 4.3 ms
= 0.6 ms
= 0 (no delay)
2
/
1
Fig. 18.23 Lower bounds on sensitivity-function norm as function of control band-
width for four values of .
4) The combustion response delay . One can verify that the lower bound on the
sensitivity peak is an increasing function of the delay (assuming other parameters
are xed).
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 , an increase in the performance requirements (lower ), or
an increase in the real part of the unstable pole of the open-loop plant
r
.
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
( j ) replaced with G
C
( j )N(), where N() is the random
2 3 4 5 6 7 8 9 10
= 4 %
= 2%
= 0.4%
stablel
1
= 40z
= .3 (10 dB)
= 0.1 (20 dB)
= 4.3ms
2
/
1
||S||
8
2 3 4 5 6 7 8 9 10
1
2
3
4
5
6
7
= 4 %
= 2%
= 0.4%
1
= 40z
= .3 (10 dB)
= 0.1 (20 dB)
= 4.3ms
Fig. 18.24 Lower bounds on sensitivity function norm as function of control band-
width for four values of combustor pole-damping ratio.
J. M. COHEN AND A. BANASZUK 606
input-describing function of the on/off valve and 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 identied and discussed. This list is not complete, by any measure.
Many effects can be systemdependent; they are critical to one systembut 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 ow with the remainder of the premixed reactants. In effect, the actuated fuel
ow 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 ow 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 mechanismthrough which pressure
oscillations outside this bandwidth could be amplied. This nding 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 arbitrarilydecrease the level of pressure oscillations usinglinear controllers.
This limit is strongly inuenced 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 signicant 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 Ofce of Scientic Research and by United
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 607
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 Stufe-
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
Cohen, J. M., Rey, N. M., Jacobson, C. A., and Anderson, T. J., Active Control of
Combustion Instability in a Liquid-Fueled Low-NO
x
Combustor, Journal of Engineering
for Gas Turbines and Power, Vol. 121, No. 2, 1999, pp. 281284.
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. 721726.
4
Hoffmann, S., Weber, G., Judith, H., Hermann, J., and Orthmann, A., Application
of Active Combustion Instability Control to Siemens Heavy Duty Gas Turbines, Sym-
posium of the AVT Panel on Gas Turbine Engine Combustion, Emissions and Alternative
Fuels, NATO Research and Technology Organization, Neuilly-Sur-Seine Cedex, France,
Oct. 1998.
5
Paschereit, C. O., Gutmark, E., and Weisenstein, W., Control of Combustion Driven
Oscillations by Equivalence Ratio Modulations, American Society of Mechanical Engi-
neers, Paper 99-GT-118, June 1999.
6
Richards, G. A., Yip, M. J., Robey, E., Cowell, L., and Rawlins, D., Combustion
Oscillation Control by Cyclic Fuel Injection, American Society of Mechanical Engineers,
Paper 95-GT-224, June 1995.
7
Anson, B., Critchley, I., Schumacher, J., and Scott, M., Active Control of Combustion
Dynamics For Lean Premixed Gas Fired Systems, American Society of Mechanical Engi-
neers, Paper GT-2002-30068, June 2002.
8
Sattinger, S. S., Neumeier, Y., Nabi, A., Zinn, B. T., Amos, D. J., and Darling, D. D.,
Sub-scale Demonstration of the Active Feedback Control of Gas-Turbine Combustion
Instabilities, Journal of Engineering for Gas Turbines and Power, Vol. 122, April 2000,
pp. 262270.
9
Bloxsidge, G. J., Dowling, A. P., Hooper, N., and Langhorne, P. J., Active Control of
Reheat Buzz, AIAA Journal, Vol. 26, 1988, pp. 783790.
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. 324333,
1981.
11
Fleil, 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. 227260.
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.
J. M. COHEN AND A. BANASZUK 608
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. 125.
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., Fleil, M., and Ghoniem, A. F., AModel-Based Ac-
tive Control Design for Thermoacoustic Instability, Combustion Science and Technology,
Vol. 132, 1998, pp. 99105.
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. 19651972.
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. 99104.
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. 595601.
27
Krstic, M., Performance Improvement andLimitations inExtremumSeekingControl,
Systems and Control Letters, Vol. 39, April 2000, pp. 313326.
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., Stufebeam, 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. 537542.
30
Stufebeam, J. H., Kendrick, D. W., Sowa, W. A., and Snyder, T. S., Quantifying
Fuel/Air Unmixedness in Premixing Nozzles Using an Acetone Fluorescence Technique,
American Society of Mechanical Engineers, Paper 99-GT-399, June 1999.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 609
31
Peracchio, A. A., and Proscia, W., Nonlinear Heat-Release/Acoustic Model for Ther-
moacoustic Instability in Lean Premixed Combustors, Journal of Engineering for Gas
Turbines and Power, Vol. 121, No. 3, 1999, pp. 415421.
32
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B., A Mechanism for Combustion
Instabilities inPremixedGas Turbine Combustors, Journal of Engineeringfor Gas Turbines
and Power, Vol. 123, No. 1, 2001, pp. 182190.
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. 6167.
35
Mezic, I., and Banaszuk, A., Comparison of Systems with Complex Behavior: Spectral
Methods, Physica D, Nonlinear Phenomena, Vol. 197, No. 12, 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