JEGTP 2003 Vol 125 N4 PDF

Transactions Journal of
of the ASME Engineering for Gas ®
Turbines and Power

Editor Published Quarterly by The American Society of Mechanical Engineers
LEE S. LANGSTON „2006…
Assistant to the Editor
LIZ LANGSTON VOLUME 125 • NUMBER 4 • OCTOBER 2003
Associate Editors
Fuels and Combustion Technologies
S. GOLLAHALLI „2004…
Internal Combustion Engines TECHNICAL PAPERS
D. ASSANIS „2005…
Nuclear Gas Turbines: Combustion and Fuels
R. DUFFY „2004… 861 Low NOx Combustion for Liquid Fuels: Atmospheric Pressure Experiments
International Gas Turbine Institute
IGTI Review Chair Using a Staged Prevaporizer-Premixer (2001-GT-0081)
R. NATOLE „2001… J. C. Y. Lee, P. C. Malte, and M. A. Benjamin
E. BENVENUTI „2002… 872 Investigation of the Response of an Air Blast Atomizer Combustion
H. R. SIMMONS „2003…
Chamber Configuration on Forced Modulation of Air Feed at Realistic
Combustion and Fuels
P. MALTE „2003… Operating Conditions (2000-GT-30059)
Structures and Dynamics C. Hassa, J. Heinze, and K. Stursberg
N. ARAKERE „2004… 879 Advanced Catalytic Pilot for Low NOx Industrial Gas Turbines
M. MIGNOLET „2005…
(2002-GT-30083)
BOARD ON COMMUNICATIONS H. Karim, K. Lyle, S. Etemad, L. L. Smith, W. C. Pfefferle, P. Dutta,
Chair and Vice-President and K. Smith
OZDEN OCHOA
885 Combustion System Development for the Ramgen Engine (2002-GT-30084)
OFFICERS OF THE ASME D. W. Kendrick, B. C. Chenevert, B. Trueblood, J. Tonouchi,
President, REGINALD VACHON
S. P. Lawlor, and R. Steele
Executive Director, V. R. CARTER
895 The Premixed Conditional Moment Closure Method Applied to Idealized
Treasurer, R. E. NICKELL
Lean Premixed Gas Turbine Combustors (2002-GT-30094)
PUBLISHING STAFF S. M. Martin, J. C. Kramlich, G. Kosály, and J. J. Riley
Managing Director, Engineering
THOMAS G. LOUGHLIN 901 Liquid Fuel Placement and Mixing of Genetic Aeroengine Premix Module
Director, Technical Publishing at Different Operating Conditions (2002-GT-30102)
PHILIP DI VIETRO J. Becker and C. Hassa
Managing Editor, Technical Publishing Gas Turbines: Controls, Diagnostics & Instrumentation
CYNTHIA B. CLARK 909 Fuzzy Fuel Flow Selection Logic for a Real Time Embedded Full Authority
Manager, Journals Digital Engine Control (2000-GT-0046)
JOAN MERANZE A. Martucci and A. J. Volponi
Production Coordinator
JUDITH SIERANT
917 The Use of Kalman Filter and Neural Network Methodologies in Gas
Turbine Performance Diagnostics: A Comparative Study (2000-GT-0547)
Production Assistant A. J. Volponi, H. DePold, R. Ganguli, and C. Daguang
MARISOL ANDINO
925 Active Combustion Instability Control With Spinning Valve Actuator
Transactions of the ASME, Journal of Engineering (2002-GT-30042)
for Gas Turbines and Power (ISSN 0742-4795) is published P. Barooah, T. J. Anderson, and J. M. Cohen
quarterly (Jan., April, July, Oct.) by The American
Society of Mechanical Engineers, Three Park Avenue, New Gas Turbines: Cycle Innovations
York, NY 10016. Periodicals postage paid at New
York, NY and additional mailing offices. POSTMASTER: 933 The Recuperative-Auto Thermal Reforming and the Recuperative-
Send address changes to Transactions of the ASME, Journal Reforming Gas Turbine Power Cycles With CO2 Removal—Part I:
of Engineering for Gas Turbines and Power, c/o THE
AMERICAN SOCIETY OF MECHANICAL ENGINEERS, 22 The Recuperative-Auto Thermal Reforming Cycle (2002-GT-30116)
Law Drive, Box 2300, Fairfield, NJ 07007-2300. D. Fiaschi, L. Lombardi, and L. Tapinassi
CHANGES OF ADDRESS must be received at Society
headquarters seven weeks before they are to be effective. 940 Thermodynamic Performance Analysis of New Gas Turbine Combined
Please send old label and new address.
STATEMENT from By-Laws. The Society shall not be Cycles With No Emissions of Carbon Dioxide (2002-GT-30117)
responsible for statements or opinions advanced in papers R. Gabbrielli and R. Singh
or ... printed in its publications (B7.1, par. 3).
COPYRIGHT © 2003 by the American Society of Mechanical 947 Thermoeconomic Analysis of Gas Turbine Plants With Fuel
Engineers. For authorization to photocopy material for
internal or personal use under circumstances not falling
Decarbonization and Carbon Dioxide Sequestration (2002-GT-30120)
within the fair use provisions of the Copyright Act, M. Bozzolo, M. Brandani, A. Traverso, and A. F. Massardo
contact the Copyright Clearance Center (CCC), 222
Rosewood Drive, Danvers, MA 01923, Tel: 978-750-8400,
www.copyright.com.
INDEXED by Applied Mechanics Reviews and Engineering
Information, Inc. Canadian Goods & Services Tax „Contents continued on inside back cover…
Registration #126148048
Downloaded 03 Jun 2010 to 171.66.16.159. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
„Contents continued…
Journal of Engineering for Gas Turbines and Power OCTOBER 2003

Volume 125, Number 4
954 Comparative Analysis of Off-Design Performance Characteristics of Single and Two-Shaft Industrial Gas Turbines
(2002-GT-30132)
J. H. Kim, T. S. Kim, J. L. Sohn, and S. T. Ro
961 Steady and Dynamic Performance and Emissions of a Variable Geometry Combustor in a Gas Turbine Engine
(2002-GT-30135)
Y. G. Li and R. L. Hales
972 A New Generalized Carbon Exergy Tax: An Effective Rule to Control Global Warming (2002-GT-30139)
A. Traverso, A. F. Massardo, M. Santarelli, and M. Cali
979 A New Scaling Method for Component Maps of Gas Turbine Using System Identification (2002-GT-30150)
C. Kong, J. Ki, and M. Kang
Gas Turbines: Cycle Innovations, and Marine
986 Off-Design Performance Analysis of a Hybrid System Based on an Existing Molten Fuel Cell Stack (2002-GT-30115)
P. Bedont, O. Grillo, and A. F. Massardo
Gas Turbines: Heat Transfer
994 Experimental and Numerical Study of Heat Transfer in a Gas Turbine Combustor Liner (2002-GT-30183)
J. C. Bailey, J. Intile, T. F. Fric, A. K. Tolpadi, N. V. Nirmalan, and R. S. Bunker
Gas Turbines: Heat Transfer, and Turbomachinery
1003 Disintegration of Oil Jets Emerging From Axial Passages at the Face of a Rotating Cylinder (2001-GT-0201)
A. Glahn, M. F. Blair, K. L. Allard, S. Busam, O. Schäfer, and S. Wittig
1011 Disintegration of Oil Films Emerging From Radial Holes in a Rotating Cylinder (2001-GT-0202)
A. Glahn, M. F. Blair, K. L. Allard, S. Busam, O. Schäfer, and S. Wittig
Gas Turbines: Manufacturing, Materials & Metallurgy
1021 Mechanical Surface Treatment Technologies for Gas Turbine Engine Components
W. Zhuang and B. Wicks
Gas Turbines: Structures and Dynamics
1026 Imbalance Response of a Rotor Supported on Flexure Pivot Tilting Pad Journal Bearings in Series With Integral
Squeeze Film Dampers (2001-GT-0257)
L. San Andrés and O. De Santiago
1033 Contact Stresses in Dovetail Attachments: Alleviation via Precision Crowning (2001-GT-0550)
G. B. Sinclair and N. G. Cormier
Gas Turbines: Structures and Dynamics and Aircraft Engines
1042 Computation of the Optimal Normal Load of a Friction Damper Under Different Types of Excitation (99-GT-0413)
D. Cha and A. Sinha
Internal Combustion Engines
1050 Engine Torque Nonuniformity Evaluation Using Instantaneous Crankshaft Speed Signal
N. Cavina and F. Ponti
1059 Numerical Study of Lean-Direct Injection Combustor With Discrete-Jet Swirlers Using Reynolds Stress Model
S. L. Yang, Y. K. Siow, C. Y. Teo, R. R. Tacina, A. C. Iannetti, and P. F. Penko
1066 Development of an Exhaust Recirculation Distribution Prediction Method Using Three-Dimensional Flow
Analysis and Its Application
K. Yoshizawa, K. Mori, Y. Matayoshi, and S. Kimura
1075 Energy and Exergy Analyses of the Pulse Detonation Engine
T. E. Hutchins and M. Metghalchi
1081 Numerical Investigation of the Effects of Axial Cylinder Bore Profiles on Piston Ring Radial Dynamics
Y. Piao and S. D. Gulwadi
ERRATUM
1090 ‘‘Effect of Biodiesel Utilization of Wear of Vital Parts in Compression Ignition Engine’’
A. K. Agarwal, J. Bijwe, and L. M. Das
1091 Author Index
„Contents continued on facing page…
„Contents continued…
Journal of Engineering for Gas Turbines and Power OCTOBER 2003

Volume 125, Number 4
ANNOUNCEMENTS AND SPECIAL NOTES

1098 2004 Heat TransferÕFluids Engineering Summer Conference—Announcement
1100 Information for Authors
The ASME Journal of Engineering for Gas Turbines and Power is

abstracted and indexed in the following:
AESIS (Australia’s Geoscience, Minerals, & Petroleum Database), Applied Science &
Technology Index, AMR Abstracts Database, Aquatic Sciences and Fisheries Abstracts,
Civil Engineering Abstracts, Compendex (The electronic equivalent of Engineering
Index), Computer & Information Systems Abstracts, Corrosion Abstracts, Current
Contents, Engineered Materials Abstracts, Engineering Index, Enviroline (The electronic
equivalent of Environment Abstracts), Environment Abstracts, Environmental Science
and Pollution Management, Fluidex, INSPEC, Mechanical & Transportation Engineering
Abstracts, Mechanical Engineering Abstracts, METADEX (The electronic equivalent of
Metals Abstracts and Alloys Index), Pollution Abstracts, Referativnyi Zhurnal, Science
Citation Index, SciSearch (The electronic equivalent of Science Citation Index), Shock
and Vibration Digest
1104 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME
Low NOx Combustion for Liquid
Fuels: Atmospheric Pressure
Experiments Using a Staged
Prevaporizer-Premixer
J. C. Y. Lee1
Low emissions of NO x are obtained for a wide range of liquid fuels by using a staged
P. C. Malte prevaporizing-premixing injector. The injector relies on two stages of air temperature and
fires into a laboratory jet-stirred reactor operated at atmospheric pressure and nominal ␾
Department of Mechanical Engineering, of 0.6. The liquid fuels burned are methanol, normal alkanes from pentane to hexadecane,
University of Washington, benzene, toluene, two grades of light naphtha, and four grades of No. 2 diesel fuel.
Box 352600, Additionally, natural gas, ethane, and industrial propane are burned. For experiments
Seattle, WA 98195-2600 conducted for 1790 K combustion temperature and 2.3⫾0.1 ms combustion residence
time, the NO x (adjusted to 15% O 2 dry) varies from a low of 3.5 ppmv for methanol to a
M. A. Benjamin high of 11.5 ppmv for No. 2 diesel fuel. For the most part, the NO x and CO are positively
Gas Turbine Fuel System Division, correlated with the fuel carbon to hydrogen ratio (C/H). Chemical kinetic modeling sug-
Parker Hannifin Corporation, gests the increase in NO x with C/H ratio is caused in significant part by the increasing
9200 Tyler Boulevard, superequilibrium concentrations of O-atom created by the increasing levels of CO burn-
Mentor, OH 44060 ing in the jet-stirred reactor. Fuel bound nitrogen also contributes NO x for the burning of
the diesel fuel. This paper describes the staged prevaporizing-premixing injector, the
examination of the injector, and the NO x and CO measurements and their interpretation.
Optical measurements, using beams of He-Ne laser radiation passed across the outlet
stream of the injector, indicate complete vaporization and a small variation in the cross-
stream averaged fuel/air ratio. The latter is determined by measuring the standard devia-
tion and mean of the transmission of the laser beam passed through the stream. Additional
measurements and inspections indicate no pressure oscillations within the injector and no
gum and carbon deposition. Thus, the NO x and CO measurements are obtained for fully
vaporized, well premixed conditions devoid of preflame reactions within the injector.
关DOI: 10.1115/1.1584476兴
Introduction stages of air temperature, including a relatively low temperature

air to accomplish as much of the vaporization and mixing as pos-
Lean-premixed combustion of natural gas is widely practiced in
sible and a high temperature air to bring the injector outlet stream
gas turbine engines applied to electrical generation and mechani-
cal drive. Most of the engines emit NOx less than 25 ppmv 共at to the desired combustor inlet temperature without causing pre-
15% O2 dry兲, though for some engines in the field, the NOx emis- flame reaction and autoignition. The inside of the staged
sion is less than 10 ppmv 共at 15% O2 dry兲. In general, lean pre- prevaporizer-premixer 共SPP兲 is monitored and inspected to con-
mixed 共LP兲 combustors are sensitive to fuel composition. For ex- firm the absence of reaction within the SPP. The outlet stream of
ample, natural gas high in non-methane hydrocarbon content the SPP is optically examined for completeness of vaporization
共NMHC兲, when premixed with the high temperature compressor and degree of mixedness between the fuel vapor and air. The
discharge air, may experience autoignition and flashback. Further, objective of the study is to show DLN combustion for a wide
for many dual fuel engines, the combustor is switched to diffusion range of fuels burned in a high-intensity combustor fed by the
flame mode for liquid fuel firing and uses water or steam injection reactants from the SPP injector. Most significantly, the study pro-
for NOx control. A preferred alternative is dry low NOx 共DLN兲 vides a database of NOx as a function of fuel type for lean, fully
combustion for all fuels, since elimination of the water/steam in- vaporized, well premixed combustion. The combustor is an atmo-
jection generally improves the engine economics and efficiency. spheric pressure jet-stirred reactor 共JSR兲 operated at combustor
The prevaporizing-premixing injector used should be designed to inlet temperatures 共up to 773 K兲 and flame temperature 共normally
minimize preflame fuel oxidation 共e.g., coking and autoignition兲 1790 K兲 typical of LP combustion turbines. The gaseous fuels
and should not be prone to flashback. Further, it is important to burned are natural gas, ethane, and industrial propane. The liquid
understand the NOx behavior of the combustor fitted with the fuels burned are methanol, normal alkanes of C5 to C16, benzene,
prevaporizing-premixing injector. toluene, two grades of light naphtha, and four grades of No. 2
The present study involves the application of the staged diesel fuel 共DF2兲. The DF2s contains small amounts of fuel bound
prevaporizing-premixing injector, 关1兴, for lean premixed combus- nitrogen 共FBN兲, about 50 to 100 ppm by weight. One of the pure
tion of liquid fuels. The staging is accomplished by using two liquid fuels 共n-dodecane兲 was previously burned, 关2兴, with and
without a FBN dopant 共n-ethylethylenediamine兲. The experi-
1
Currently with Solar Turbines Incorporated. ments, conducted under conditions very close to those of the
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN present study, indicated complete conversion of small levels of
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- FBN to NOx.
tional Gas Turbine and Aeroengine Congress and Exhibition, New Orleans, LA, June The paper contains the following sections:
4 –7, 2001; Paper 2001-GT-0081. Manuscript received by IGTI, Dec. 2000, final
revision, Mar. 2001. Associate Editor: R. Natole. • description of the SPP injector and JSR combustor,
Journal of Engineering for Gas Turbines and Power OCTOBER 2003, Vol. 125 Õ 861
Copyright © 2003 by ASME
• description of the optical method used to evaluate the mixed- Gaseous fuel is injected through a single port into the first stage
ness of the SPP outlet stream, air. As noted in Fig. 1, this port is located slightly upstream of the
• data and analysis on the mixedness of the SPP outlet stream, film atomizer. Liquid fuel is atomized by a Nukiyama-Tanasawa
• experimental data of NOx and CO as a function of fuel type, type, plain-jet, air-blast atomizer custom built for this application
• analysis of the NOx and CO data by chemical kinetic model- and located on the centerline of the first stage. The Nukiyama-
ing, and Tanasawa type atomizer is described by Lefebvre 关3兴. The previ-
• conclusions and recommendations. ous embodiment of this atomizer for research at the University of
Washington is discussed by Lee et al. 关4兴. Within the Nukiyama-
Stage Prevaporizer-Premixer „SPP… Injector and Diag- Tanasawa type nozzle, pressure atomization occurs first, followed
by air-assist atomization and spray cone angle adjustment.
nostics The Nukiyama-Tanasawa type nozzle provides good atomiza-
The experimental setup of the SPP injector with the JSR com- tion and spray quality for the low fuel flow rates associated with
bustor is schematically depicted in Fig. 1. the atmospheric pressure firing of this study. The nominal flow
rate of the fuel is 0.2 kg/hr and that of the unheated atomizing air
SPP Injector. The SPP injector is shown in the lower three-
is 0.4 kg/hr. The atomizer nozzle flow is choked and the calculated
fourths of Fig. 1. First stage air, of relatively low temperature,
SMD of the spray is about 10 ␮m. However, for high pressure
enters at the bottom, i.e., upstream end of the SPP. The first stage
industrial applications, with significantly greater flow rates of fuel,
air is accelerated through jets located at the upstream end of the
a commercial air blast, pressure swirl or macrolaminated, 关5兴, at-
film atomizer that is coaxial with the fuel nozzle. Two streams of
omizer will be used for the SPP.
the jetted first stage air are obtained, separated by the thin annular
For the present study, the first stage temperature is varied from
splitter plate constituting the film atomizer. Second stage air enters
through a manifold near the top, i.e., downstream end of the SPP. 293 to 623 K and that of the second stage, which is also the total
This air is mixed into the internal flow of the SPP through several temperature of the outlet stream, is varied from 293 to 773 K
staggered jets. These jets enter the SPP at an angle of 45 deg to 共though the data shown are limited to 723 K兲. The outlet nozzle of
the axial direction and do not have a tangential velocity compo- the SPP, which couples the SPP to the jet stirred reactor, is 4 mm
nent. The total air flow rate of the SPP as used for atmospheric in diameter. Nominal pressure within the SPP, downstream of the
pressure firing is 4.7 kg/hr. In the first stage of the SPP, the light film atomizer, is 1.15 atm. The nominal nozzle exit velocity 共from
ends and middle boiling compounds of a blended fuel are vapor- 293 to 773 K兲 is 85 to 230 m/s. The ratio of the first stage to
ized, while in the second stage, the vaporization process is com- second stage air mass flow rates is set at 0.5, 1.0, and 2.0. Some
pleted. As the temperature of the second stage air is increased, the variation in the performance of the SPP is obtained with respect to
residence time of the gases in the second stage can be decreased to this air split, however, as shown in the results section, the varia-
small values to prevent reaction within the SPP. tion is relatively small. For use in gas turbine engines, a low ratio
of first stage to second stage air is preferable, since this would
require the least amount of modification to the engine cycle for
cooling the compressor discharge air to the desired first step tem-
perature of the SPP.
In the present work, the nominal residence time of the first
stage of the SPP is 12 ms and that of the second stage is either 5
or 12 ms, depending on whether the short or long version of the
second stage is installed. These nominal values correspond to an
airflow ratio of 1.0 and a temperature of 623 K in each stage.
Absence of preflame reactions in the liquid fueled SPP is con-
firmed by the absence of C1 and C2 hydrocarbons, by the absence
of wall deposits of gum and carbon and by the absence of pressure
oscillations. The absence of the C1 and C2 hydrocarbons is deter-
mined by probe sampling of the internal flow of the SPP, followed
by GC analysis of the gas samples. The absence of pressure os-
cillations is determined by continuous monitoring of the SPP via a
pressure probe and transducer. In earlier work 关4兴, with a single-
stage prevaporizer-premixer of about 100 ms residence time and
2–3 atm pressure, several preflame reaction indicators 共i.e., gum
deposits and pressure oscillations兲 were observed when the tem-
perature was approximately 100°C above the normal boiling point
of the fuel.
The supporting components of the SPP system are the
following:
• electrical air preheaters,
• cascade-type temperature controller for setting and maintain-
ing air preheating and SPP temperature levels, and
• air and fuel control: The mass flow rates of the air and gas-
eous fuels are controlled and measured by electronic mass
flow controllers. The mass flow rates of the liquid fuels are
controlled by placing the liquid fuels under nitrogen pressure
and pushing the fuels through a control valve. The liquid flow
rates are measured with a calibrated rotameter.
Laser Diagnostics. Two laser-based methods are used for ex-
Fig. 1 Schematic drawing of the atmospheric pressure JSR, of amining the outlet stream of the SPP. 共1兲 Completeness of vapor-
15.8 cm3 volume, with the staged prevaporizing-premixing in- ization is confirmed by observing the absence of droplet and aero-
jector sol scattering from a 5 mW, 0.63 ␮m He-Ne laser beam passed
Fig. 2 Schematic diagram of the laser absorption system
through the outlet stream of the SPP. 共2兲 A separate laser absorp- Io
tion system is used to investigate the mixedness of the fuel vapor- log10
air stream as it departs the SPP injector. The laser absorption I 110,000
system is a relatively simple diagnostic tool based on the absorp- ⑀⫽ ⫽ (3)
C•L P T0.302
tion of electromagnetic radiation of 3.39 ␮m wavelength by hy-
drocarbons, 关6兴. The major components of the system were pro- where C 共gmol/cm3兲 is the concentration of the absorbing species
vided by the University of California at Berkeley, 关7兴. Mongia 关8兴, and P T 共atm兲 is the total pressure.
in his Ph.D. thesis, describes the technique and its application to Equation 共1兲 is rearranged for the case of fluctuations in the
premixer outlet streams. absorbing medium and laser/detector system. This is done as fol-
Figure 2 is a schematic drawing of the laser absorption setup lows:
used. The major components are the 50 mW, 3.39 ␮m He-Ne laser
with power supply and the infrared 共IR兲 detector that is thermal • I o , I, and kPL are assumed to have time-mean ( m ) and time-
electrically cooled and maintained at 5°C temperature. The neutral fluctuating ( ⬘ ) components: I o ⫽I om ⫹I ⬘o , I⫽I m ⫹I ⬘ and
density filter shown is used to attenuate the laser beam, since the k PL⫽(k PL) m ⫹(k PL) ⬘ .
power laser is too great for the IR detector. The plano-convex
• It is assumed that (k PL) ⬘ Ⰶ1. Thus, e ⫺k PL ⬘ ⬵1⫺(k PL) ⬘ .
focusing lenses are required in order to minimize the effects of
• The fluctuations in I o 共i.e., the signal without absorbing spe-
beam steering caused by differences in index of refraction as the
cies present兲 and k PL 共i.e., the absorbing medium兲 are as-
laser beam passes through the heated outlet stream of the SPP. The
sumed uncorrelated. Thus, the mean value of the product
narrow bandwidth IR band pass filter is used to remove any ther-
mal radiation emitted from the metal surfaces of the SPP. Signal I ⬘o (k PL) ⬘ is zero. Also, terms of order higher than two in the
display and processing are accomplished with the digitizing fluctuations are assumed negligible.
oscilloscope. Substitution into Eq. 共1兲 and rearrangement gives the following
Temporal fluctuations spatially averaged over the line-of-sight equation linking the standard deviation ( sd ) over the mean ( m ) of
path length are measured with the laser absorption system. The the variables:
line-of-sight measurements are made along a 12.7-mm diametric
path through the SPP outlet stream at a location 2 mm above the
SPP outlet. The SPP outlet nozzle and JSR are removed.
By the Beer-Lambert law:
I sd
Im
⫽ 冑冋共 k PL 兲 sd
共 k PL 兲 m
•ln 冉冊册冉冊
I om
Im
2
⫹
I osd
I om
2
. (4)
I The transmission data (I) are collected with the digitizing oscil-
⫽e ⫺k PL (1) loscope over a period of 50 ms at a rate of 10 samples per milli-
Io second for a total of 500 samples. For each case of 500 samples,
the mean (I m ) and standard deviation (I sd ) are determined. The
where I is the beam intensity 共or the transmission signal兲 leaving noise base is determined in the absence of fuel flow with only the
the absorbing medium, I o is the beam intensity entering the ab- flow of air. The measurements indicate the base noise inherent in
sorbing medium 共or the signal in the absence of the absorbing the laser/electronic system is independent of the inlet temperature
species兲, k 共cm⫺1atm⫺1兲 is the absorption coefficient at the given 共from 293 to 773 K兲 and is constant with a value of I osd /I om
wavelength, P 共atm兲 is partial pressure of the absorbing species ⫽0.023. In the results section, I sd /I m is plotted as a function of
共fuel兲, and L 共cm兲 is the path length. Perrin and Hartmann 关9兴 I m /I om for different values of (k PL) sd /(k PL) m and the measure-
determined the absorption coefficient for the 3.39 ␮m laser wave- ments with the fuel present are superimposed, thereby permitting
length for methane diluted in nitrogen at temperatures of 293 to (k PL) sd /(k PL) m to be determined.
787 K and 1 atm total pressure. Their data may be expressed as Fluctuations in k PL can arise in at least three ways: 共1兲 the
follows: absorption coefficient (k) can fluctuate through variations in tem-
perature, 共2兲 the partial pressure of the fuel ( P) can fluctuate
k⫽22.89e ⫺0.003033T . (2) through variations in the fuel/air ratio 共F/A兲, and 共3兲 the path
length (L) can fluctuate through fluctuations in the thickness of
Yoshiyama et al. 关10兴 examined absorption at the 3.39 ␮m the mixing layer along the outer edge of the stream. Additionally,
wavelength by methane and propane. Their results are expressed for the case of a blended fuel, differential diffusion of the fuel
in terms of an extinction coefficient, ⑀ 共cm2/gmol兲. The definition components could cause k PL to vary. Fluctuations in the trans-
of ⑀ and its expression for CH4 are given as follows: mission signal would also occur in the case of incomplete vapor-
ization, though, as stated above, the evidence from the laser scat- • The mean residence time of the JSR is 2.3⫾0.1 ms, based on
tering tests indicates this effect is absent from the present the air and fuel mass flow rates, the combustion temperature
experiments. of 1790 K, the molecular weight of the combustion gases and
With increasing temperature, the absorption decreases, note Eq. the JSR volume.
共2兲. Thus, I m approaches I om and the effect of the kPL fluctuations • The inlet total temperature of the JSR is the second stage
on I sd /I m decreases as the temperature increases. This effect is temperature of the SPP. For the combustion experiments
illustrated in the results section. shown, it is varied from 423 to 723 K. An inlet total tempera-
ture of 623 K is used for the experiments of NOx as a func-
tion of fuel type.
Jet-Stirred Reactor „JSR…, JSR Operating Conditions,
• The fuel-air equivalence ratio 共␾兲 is adjusted to maintain the
Diagnostics, and Fuels constant combustion temperature. For the experiments run at
Jet-Stirred Reactor. The jet-stirred reactor used in this study 1790 K combustion temperature and 623 K inlet total tem-
is described by Lee et al. 关4兴 and detailed in the Ph.D. thesis of perature, this results in ␾⫽0.61⫾0.02 for most of the fuels.
Lee 关11兴. The internal dimensions of the JSR are identical to the The exceptions are natural gas 共␾⫽0.65兲, benzene 共␾⫽0.57兲,
JSR used by Steele et al. 关12兴. The JSR is depicted in the upper and toluene 共␾⫽0.58兲.
part of Fig. 1. The internal volume is 15.8 cm3, the internal height JSR Characterization. Calculations of the Damköhler num-
is 45 mm, and the internal diameter 共at the widest point兲 is 25 ber 共Da兲 and the turbulence Reynolds number ReT ) are performed
mm. The JSR is fed by reactants and stirred by the outlet stream for the JSR operated with the 623 K inlet temperature. With meth-
of the SPP accelerated through a 4-mm diameter nozzle. The
ane selected as the example fuel, the laminar flame speed (S L ) is
nozzle is located in the Hastelloy® nozzle block. Exhaust gases
estimated using the method of Göttgens et al. 关13兴. The laminar
leave the JSR through four drain holes located 90 deg apart near
flame speed leads to the characteristic chemical time ( ␶ L ), assum-
the bottom of the reactor. Temperature measurements, gas sam-
pling, and viewing of the JSR cavity are conducted at 2/3rd reac- ing ␶ L ⬵D L /S L2 . The laminar diffusivity (D L ) is estimated based
tor height. There are four ports located at this height, 90 deg apart. on the inlet temperature 关14兴, using the air value for thermal dif-
Two of these are blocked during operation. One of the remaining fusivity. The result is ␶ L ⬇0.15 ms. The characteristic mixing time
open ports accepts the thermocouple probe and the other accepts ( ␶ T ) is evaluated assuming ␶ T ⬵d/u ⬘ , where d is taken to be the
the gas sample probe. inlet jet diameter 共4 mm兲 and the RMS turbulence velocity (u ⬘ ) is
assumed to be 10% of the inlet jet velocity. This gives the char-
JSR Operating Conditions. The JSR operating conditions acteristic mixing time as 0.2 ms. Hence, Da⫽ ␶ T / ␶ L ⬇1.3. The
are listed as follows:
turbulence Reynolds number (u ⬘ d/ ␯ ) is 1350, where the kine-
• The combustion temperature 共i.e., the measured, corrected matic viscosity 共␯兲 is based on air at the jet inlet temperature.
temperature of the recirculation zone兲 is normally maintained Under these conditions, the JSR operates as a two-zone reactor.
at 1790 K. Studies by Rutar et al. 关15兴 indicate the Damköhler number must
• The air mass flow rate 共given above兲 is maintained at 4.7 be reduced to about 0.15 before the JSR can operate as a single-
kg/hr. zone, well-stirred reactor. In the present experiments, the flame
Table 1 List of fuels used in experiments „FBS: fuel bound sulfur, FBN: fuel bound nitrogen…
Chemical FBS FBN

Fuel Purity Formula 共ppmw兲共ppmw兲
natural ¯ 93% CH4 negligible negligible
gas
industrial ¯ 95% C3H8 negligible negligible
propane
methanol 99% CH3OH 0 0
ethane 99% C2H6 0 0
propane 99% C3H8 0 0
n-pentane 99% C5H12 0 0
n-hexane 99% C6H14 0 0
n-heptane 99% C7H16 0 0
n-dodecane 99% C12H26 0 0
n-hexadecane 99% C16H34 0 0
benzene 99% C6H6 0 0
toluene 99% C7H8 0 0
light naphtha ¯ C5.9H12.5 negligible ⬍1
共K-LN兲
supplied by K
light naphtha ¯ C5.9H13.4 negligible ⬍3
共U-LN兲
supplied by U
low sulfur DF2 ¯ C13.8H26.3 ⬇200 46
共C-LSD兲
supplied by C
low sulfur DF2 ¯ C13.8H24.2 ⬇200 91
共T-LSD兲
supplied by T
high sulfur DF2 ¯ C13.8H23.7 ⬇3000 84
共T-HSD兲
supplied by T
low sulfur DF2 ¯ C13.8H24.2 ⬇200 49
共U-LSD兲
supplied by U
zone is wrapped around the inlet jets and represents about 5–10% uncooled tip of the sample probe is an extension of the JSR. From
of the reactor volume. The large recirculation zone serves as the time to time, gas chromatography is employed to check for light
post-flame zone. Because of the stirring within the recirculation hydrocarbon gases in the JSR. Typically, total measured hydrocar-
zone, combined with the relatively short residence time 共2.3 ms兲 bon for the JSR recirculation zone is small 共i.e., less than 100
and relatively low pressure 共1 atm兲, free radical and intermediate ppmv C1 basis兲.
species, in particular O, H, OH, H2, and CO, remain at relatively
large concentrations throughout the reactor. Both zones contribute Fuels. The fuels used in this study are listed in Table 1. The
NOx as discussed in the ‘‘modeling and interpretation’’ section empirical chemical formulas listed for the naphtha and diesel fuels
below. are based on elemental analyses performed at commercial fuel
analysis laboratories. Compound information obtained with the
Combustion Temperature and Species Concentration Mea- analyses indicates the light naphtha fuels are composed of
surements. For the work reported herein, all probe measure- n-alkanes 共50%兲, iso-alkanes 共30%兲, naphthenes 共15%兲, and aro-
ments of temperature and species concentrations are obtained in matics 共5%兲, all in the C5 –C6 range.
the recirculation zone of the JSR. This zone comprises the bulk of
the JSR and exhibits nearly uniform temperature and composition.
Combustion temperature is measured with a 0.127-mm Experimental Results
type-R thermocouple coated with a ceramic compound 共to prevent
oxidation of CO, H2, and hydrocarbons at the thermocouple Laser Absorption Measurements. Percent mean transmis-
surface兲 and corrected for radiation losses 共typically, a ⫹40° sion (I m /I om ⫻100) as a function of the temperature for the outlet
correction兲. stream of the SPP is plotted in Fig. 3. Comparison to the literature,
Gas composition is measured by drawing combustion gas that is, to Eqs. 共2兲 and 共3兲 applied to the present experimental
through a small, water-cooled, quartz sampling probe and routing conditions, is also plotted. The fuels are natural gas and industrial
the gas through water removal impingers and into a bank of gas propane, each at a mole fraction of 0.0277 in the stream. The
analyzers for CO, CO2, O2, NO, and NOx determination. A metal corresponding fuel-air equivalence ratios are 0.27 for natural gas
bellows sample pump is located between the impingers and the and 0.68 for industrial propane. The figure clearly shows the in-
bank of gas analyzers. In order to minimize the disturbance of the crease in transmission with increasing temperature. The figure
JSR flow field as well as to minimize internal probe reactions, the also indicates a good agreement between the measurements for
uncooled tip of the probe is made as small in diameter and as natural gas and Eq. 共2兲. On the other hand, Eq. 共3兲 predicts lower
short as practical, i.e., 3-mm O.D. by 30-mm length. The I.D. of transmission than measured, though the trend with respect to tem-
the probe is maintained constant, that is, the entry is not restricted. perature is consistent with the other results.
The remainder of the probe, which sits outside the JSR cavity, is Percent standard deviation in transmission ((I sd /I m )⫻100) is
jacketed with water flow. The water is made sufficiently warm to plotted versus the temperature of the outlet stream of the SPP in
prevent condensation within the probe sample. The overall length Figs. 4 and 5. The fuels are natural gas 共Fig. 4兲 and No. 2 diesel
of the sample probe is 300 mm. A 6-mm O.D. Teflon tube, also fuel 共Fig. 5兲. For all data, ␾⫽0.68. The legend of each figure
maintained warm, connects the sample probe outlet to the im- indicates the ratio of the first and second stage airflow rates and
pingers. By the sampling technique used, most of the NOx mea- the first stage temperature of the SPP. The measurements for
sured is NO, indicating conversion of NO to NO2 in the sample methane exhibit a falloff in the percent standard deviation as the
probe is small. Hence, loss of NOx by absorption of NO2 in water temperature of the SPP outlet stream is increased. Also seen is a
is taken to be a minor-to-negligible effect in the present work. decrease in the percent standard deviation as the airflow rate of
Gasdynamic calculations indicate the residence time of the the first stage of the SPP is increased relative to the airflow rate of
sampled gas within the hot uncooled tip of the probe is approxi- the second stage. All of the diesel fuel measurements 共Fig. 5兲 lie
mately 0.1 ms. Although this time is small, some loss of CO by close to the noise base of the experiment.
oxidation and a relatively small production of NO within the hot Analysis of the transmission signals of these measurements in
uncooled tip of the probe cannot be ruled out. That is, the hot the frequency domain indicates no predominant frequencies.
Fig. 3 Effect of temperature on the transmission of the 3.39 ␮m He-Ne laser for
methaneÕair and propaneÕair mixtures at 1 Atm
Fig. 4 Effect of temperature and air split on the percent standard deviation in the
laser transmission for the outlet stream of the SPP with natural gas-air mixture at
␾Ä0.68. Legend: First stage airflow rate divided by the second stage airflow rate,
and nominal first stage temperature in Kelvin. Long second stage used.
In Fig. 6, results from the laser absorption experiments are dis- standard deviation over mean of the laser transmission changes by
played on the basis of Eq. 共4兲. The fuel-air equivalence ratio is two effects: 共1兲 temperature and 共2兲 fluctuations in kPL.
␾⫽0.68, and the first to second stage ratio of the SPP airflow rates With natural gas as fuel, Fig. 6 shows the percent mean trans-
is 1 and 2. Plotted as the dependent variable is I sd /I m . The inde- mission increase from 46% 共corresponding to the stream tempera-
pendent variable is I m /I om . The parameter is the ratio of the ture of 293 K兲 to 89% 共corresponding to 723 K兲. The measure-
standard deviation of kPL divided by the mean of the term. The ments at 293 K have the highest sensitivity of the experiment and
curves are calculated from Eq. 共4兲. The measurements are super- indicate a percent standard deviation in k PL 共i.e.,
imposed on the curves. The asymptotic value of I sd /I m is 0.023, (k PL) sd /(k PL) m ⫻100) of 5 to 6%. Because of the 293 K tem-
the noise of the laser/detector system. perature, there is temperature uniformity, and thus, negligible
From left to right in Fig. 6, the temperature of the outlet stream variation in k. The standard deviation in k PL represents the varia-
of the SPP increases and, thus, the transmission increases. The tion in the fuel partial pressure 共or fuel/air ratio兲 integrated over
Fig. 5 Effect of temperature and air split on the percent standard deviation in the
laser transmission for the outlet stream of the SPP with No. 2 low sulfur diesel
fuel-air mixture at ␾Ä0.68. Legend: First stage airflow rate divided by the second
stage airflow rate, and nominal first stage temperature in Kelvin. Long second
stage used.
Fig. 6 Standard deviation in transmission versus mean transmission. Long sec-
ond stage for the SPP.
the width of the stream. The 5– 6% standard deviation indicates a With diesel fuel, a further increase in mean transmission occurs.
small variation in the cross-stream averaged fuel/air leaving One measurement point is shown in Fig. 6. This indicates a per-
the SPP. cent mean transmission of about 85% and a value of no more than
With increasing temperature, I m /I om increases and I sd /I m de- about 5% in the percent standard deviation in kPL.
creases. The I sd /I m measurement becomes less sensitive to the Measurements with a multiple-pass reflected laser beam or at
variation in kPL, and thus, there is more uncertainty in the values
greater total pressure would be useful for both improving the sen-
of (k PL) sd /(k PL) m inferred from the experiment. The maximum
percent standard deviation in kPL is about 10%, or double that of sitivity of the experiment and permitting enhanced spatial resolu-
the 293 K data. tion. Nonetheless, the measurements indicate the standard devia-
With the industrial propane, the mean transmission increases, tion of the cross-stream averaged fuel-air ratio of the outlet stream
since fewer fuel molecules are present with ␾ maintained con- of the SPP is no more than about 10% of the mean fuel-air ratio.
stant. The temperature range is 293 to 623 K, corresponding to an
increase in percent mean transmission from 73 to 87%. The maxi- NOx as a Function of the Outlet Temperature of the SPP .
mum value of the percent standard deviation in kPL is again The amount of NOx formed in the jet-stirred reactor is first exam-
about 10%. ined as a function of the outlet temperature of the SPP. In order to
Fig. 7 Effect of SPP temperatures and air split on NOx for natural gas-firing of the
JSR. Legend: First stage airflow rate divided by the second stage airflow rate and
nominal first stage temperature in Kelvin. Long second stage used.
Fig. 8 Effect of fuel type on NOx JSR combustion temperature of 1790 K and
residence time of 2.3Á0.1 ms. The SPP conditions are outlet temperature of 623 K,
and first to second stage airflow ratio of 1.
maintain a constant combustion temperature in the JSR, the fuel- The decrease in NOx with increasing outlet temperature is prob-
air equivalence ratio is decreased as the outlet temperature of the ably caused mainly by a decrease in Fenimore prompt NOx as the
SPP is increased. mixture is leaned. This behavior is suggested in the work of Rutar
Results for natural gas are shown in Fig. 7. 共These experiments et al. 关16兴, in which a high-pressure jet-stirred reactor was oper-
were run at a combustion temperature of about 1850 K, rather ated for both preheated and unheated inlet streams. For running at
than the normal value of 1790 K for this study.兲 Two trends are short residence times, the NOx dropped when the reactor was
noted in the figure. 共1兲 A mild decrease in NOx, from about 8.5 to leaned upon preheating the inlet stream of premixed methane and
7.0 ppmv, occurs as the outlet temperature increases from 423 to air. Chemical kinetic calculations of the CH radical showed a
773 K and 共2兲 a slight decrease in NOx is noted when the SPP is significant decrease upon preheating, indicating a decrease in the
adjusted to give equal airflow rates for the first and second stages. formation of prompt NOx. Improvement in fuel-air mixedness
Results for ethane and industrial propane also indicate mini- could in principle also explain the trend in NOx. However, the
mum NOx with the SPP set for the equal first and second stage measurements in Fig. 6 do not support this hypothesis, since the
airflow rates. Interestingly, however, these fuels do not exhibit a ratio (k PL) sd /(k PL) m does not decrease with increasing SPP
dependence of NOx on the SPP outlet temperature. outlet temperature.
Fig. 9 Same conditions as Fig. 8, except the fuel NOx formed through 100% con-
version of FBN is deducted for the diesel fuels
Fig. 10 Effect of fuel type on CO for JSR combustion temperature of 1790 K and
residence time of 2.3Á0.1 ms. The SPP conditions are outlet temperature of 623 K,
and first to second stage airflow ratio of 1.
NOx and CO as a Function of Fuel Type. The measurement Modeling and Interpretation of NOx
of NOx for the full range of the fuels used in this study is plotted
Reactor modeling is conducted to help explain the increase in
in Fig. 8. In this case, the SPP outlet temperature is 623 K and the
NOx with increasing C/H ratio. Chemical kinetic modeling is per-
ratio of the first to second stage SPP airflow rates is 1.0. The
combustion temperature of the JSR is 1790 K. The residence formed using simple, though realistic, chemical reactor arrange-
times are: SPP first stage: 12 ms, SPP second stage: 5 or 12 ms ments to simulate the JSR. Thus, as a first step, the structure and
and JSR: 2.3⫾0.1 ms. Lowest NOx 共3.5 ppmv兲 is measured for behavior of the JSR is examined.
methanol, while highest NOx 共11.4 ppmv兲 is found for the diesel The combustion field of the atmospheric pressure jet-stirred re-
fuels containing 84 –91 ppmw FBN. These measurements of NOx, actor consists of a turbulent flame zone on and near the jet-
obtained in the recirculation zone of the JSR, closely represent centerline of the reactor, surrounded by the recirculating post-
exhaust emissions, since the recirculation zone is nearly uniform flame zone, which accounts for the bulk of the reactor. The flame
in NOx and the exhaust leaves the JSR from the recirculation zone volume is about 5–10% of the JSR volume, contains signifi-
zone. cant chemical energy release and exhibits peak concentrations of
In Fig. 9, the NOx data are adjusted by subtracting the contri- CO 共of about 1–2% by mole兲. These points are known from sam-
bution of the FBN for the diesel fuels. This is done assuming pling measurements for CO, CO2, and O2 on a path across the
complete conversion of the FBN to NOx. Noted now is a rela- reactor and through the centerline 共see 关4兴兲. Because the oxidation
tively well-behaved increase in NOx with increasing molar of CO is relatively slow at 1 atm pressure and because the back-
carbon-to-hydrogen ratio of the fuels from methane through the mixing delays CO burnoff, the CO levels of the recirculating post-
diesel fuels. Over this range of fuels, it appears possible to rea- flame zone only fall to about 0.15 to 0.4% by mole. These data are
sonably predict the NOx emission 共from fixation of the air- plotted in Fig. 10. On the other hand, equilibrium levels for CO
nitrogen兲 based on the fuel carbon-to-hydrogen ratio. The blended for the temperature and stoichiometry conditions of the recirculat-
fuels, i.e., the naphthas and diesel fuels, are essentially indistin- ing post-flame zone are much lower—about 0.005 to 0.007% by
guishable from the pure alkanes tested on the basis of NOx as a mole. Recombination of the active species O, H, and OH is also
function of C/H ratio. However, methanol exhibits unique NOx relatively slow in the 1 atm reactor. The estimated chemical time
behavior. This is most likely the result of very little Fenimore for O-atom recombination under the present conditions is 10–20
prompt NOx formed by the combustion of methanol. ms, 关17兴. This is significantly longer than the 2.3 ms residence
The carbon monoxide 共CO兲 measurements obtained for the re- time of the JSR. Thus, the O-atom remains at super-equilibrium
circulation zone of the JSR also exhibit a significant dependence levels throughout the reactor. This indicates significant formation
on the fuel carbon-to-hydrogen ratio 共see Fig. 10兲. A relatively of NO by O-atom attack on N2 for most points in the reactor. The
well-behaved increase in CO with C/H ratio is noted for ethane mechanisms at work are the nitrous oxide and Zeldovich mecha-
through the diesel fuels. Methane 共i.e., natural gas兲 and methanol, nisms.
being relatively slow burning fuels, exhibit more CO than indi- The CH radical, on the other hand, is short lived, resides mainly
cated by the extrapolation of data for ethane-to-diesel. The CO in the flame zone and does not survive into the post-flame zone.
measurements for the recirculation zone of the JSR are 80 to 90% Thus, prompt NO formation caused by the attack of the CH radi-
lower than the peak CO levels measured in the flame zone of the cal on N2 mainly occurs in the flame zone.
reactor that surrounds the inlet jet of the JSR. Nonetheless, the CO Studies by Steel et al. 关18兴, Bengtsson et al. 关19兴, and Rutar
concentrations for the recirculation zone are significantly higher et al. 关15兴 used a model of two perfectly stirred reactors 共PSRs兲 in
than exit plane emissions acceptable for a commercial combustor. series to chemical kinetically study JSRs. This is the 2-PSR
This is caused because the JSR does not have a plug flow burnout model. The first PSR simulates the flame zone, that is, the region
zone for CO. of rapid oxidation and maximum free radical concentrations,
Table 2 Results for NOx from the 2-PSR modeling Table 4 Modeling of NOx as a function of COÕH2 ratio; also
listed are the NOx measurements of this study „all NOx ex-
% Difference % of NOx pressed as ppmv „15% of O2 dry……
in NOx of % NOx by as
Model to Zone Prompt Fuel C/H⫽ 2-PSR 3-PSR Measured
Fuel Measurement PSR1:PSR2 NOx 2CO/H2 NOx NOx NOx
Methanol ⫺33% 20%:80% 5% Methane 0.25 3.0 3.7 4.7–5.1
Methane ⫹12 to ⫹21% 60%:40% 34% n-Hexadecane 0.47 4.2 4.7 6.8 –7.1
Ethane ⫺6 to ⫺19% 50%:50% 30% Benzene 1.00 6.4 6.5 10.2
while the second PSR simulates the recirculating post-flame zone. and the blended fuels studied, no satisfactory chemical kinetic
In Rutar et al. 关15兴, two independent methods are used to deter- mechanism for fuel oxidation with nitric oxide formation has been
mine the fraction of the reactor that is the flame zone. In one identified 共though some have been tested兲 in the present work.
method, the thickness of the turbulent flame zone is estimated In order to chemically kinetically model the full range of fuels
from the literature results for premixed turbulent flame speed, tested, fast oxidative pyrolysis is assumed. That is, CO and H2 are
while in the second method, the modeled values of both NOx and the fuels fed into the models and the O2 input is debited, per the
CO are matched to the experimental values by adjusting the rela- following chemical equation:
tive sizes of the two PSRs. The two methods show agreement. For
the Damköhler number of 1.3 of the present research, the results x y
Cx Hy ⫹ O2→xCO⫹ H2 (5)
of Rutar et al. 关15兴 indicate a flame zone size of about 10% of the 2 2
reactor volume.
The first PSR of the two and three PSR models is assigned the
The chemical kinetic mechanism used for the present modeling
average temperature found in the methanol, methane, ethane, and
of the methane, methanol, ethane, and propane flames is GRI 3.0,
propane computations above and the second and third PSRs are
关20兴. The computations are performed for the measured fuel and
again assigned the experimental temperature of 1790 K. The
air flow rates of the experimental reactor. The first PSR is as-
mechanism used is that due to Miller and Bowman 关21兴. This
sumed adiabatic, since the rate of chemical energy release is large
mechanism is used so that the NOx resulting from the attack of
relative to the rate of heat loss for this zone. The second PSR is
O-atom on N2 may be exclusively studied. Prompt NOx is not a
assigned the corrected measured temperature of 1790 K for the
factor, since no hydrocarbon material is injected into the model
recirculation zone. Results from the 2-PSR modeling of NOx are
reactor. Additionally, NOx formed from NNH, which is a factor in
listed in Table 2. In this modeling, the first PSR has 5% of the
the GRI 3.0 mechanism, is not present in the Miller-Bowman
reactor total residence time, and the second PSR has 95%.
mechanism. The only mechanisms forming NOx are the Zeldovich
The results in Table 2 indicate reasonable agreement of the
and nitrous oxide mechanisms.
modeled to the measured NOx. The modeling shows the distinct
The results are shown in Table 4 below. The modeling repli-
difference between NOx formation for methanol and methane-
cates the trend of the measurements, i.e., the increase in NOx with
ethane. Little prompt NOx is formed during the combustion of
increasing C/H ratio. The modeling shows an approximate dou-
methanol. This is responsible for the relatively small percentage
bling in the NOx from methane to benzene. The doubling is also
of NOx formed in the methanol flame zone. On the other hand, the
seen in the experiment.
modeling for methane and ethane shows about a 50:50 division in
The results provide evidence for the finding that NOx increases
the NOx formed in the flame and post-flame zones. Also, prompt
with C/H ratio, because of increasing attack of O-atom on N2
NOx 共formed exclusively in PSR1兲 accounts for about 1/3rd of the
leading to increasing amounts of NOx formed by the Zeldovich
total NOx modeled.
and nitrous oxide mechanisms.
In subsequent modeling, the first PSR is split into two
The increasing amounts of O-atom result from the increasing
zones—an initial PSR operating at incipient blowout, followed by
amounts of CO oxidized. Key reactions are
a second PSR. The initial PSR is meant to simulate the thin flame
fronts, while the second PSR is meant to simulate reaction in the CO⫹OH→CO2⫹H (6)
balance of the flame zone. The concluding PSR is the recirculating
post-flame zone, as above. The initial PSR is assumed to be adia- H⫹O2→O⫹OH. (7)
batic, while the second and third PSRs are assigned the measured The impact of O-atom on NO formation by the Zeldovich and
temperature of 1790 K. The initial PSR has a very short residence nitrous oxide mechanism is contained in the following reactions:
time. The second PSR is assigned 10% of the reactor residence
time, while the third PSR is assigned the balance of the residence O⫹N2→NO⫹N (8)
共i.e., almost 90%兲.
N⫹O2→NO⫹O (9)
The three PSRs in series model shows good agreement to the
measurements and simulates the increase in NOx from methanol O⫹N2⫹M→N2O⫹M (10)
to methane and with increasing C/H ratio. The results are shown
in Table 3. N2O⫹O→NO⫹NO. (11)
The GRI 3.0 mechanism does not consider hydrocarbon fuels of In the case of the experiment, complex hydrocarbon-nitrogen
order greater than C3. For the C5 to C16 alkanes, toluene, benzene, chemistry, leading to variable prompt NO formation is also occur-
ring. How prompt NOx changes with increasing C/H ratio under
lean premixed combustion is unclear and must await chemical
Table 3 Comparison of NOx by the 3-PSR model to the mea- mechanism developments for large fuel molecules.
sured NOx „all NOx expressed as ppmv „15% O2 dry…
Fuel Modeled Measured Conclusions and Recommendations

Methanol 2.8 3.3 The major findings of this study are as follows:
Methane 5.2 4.7–5.1
Ethane 5.2 5.3– 6.2 1 Low NOx emissions are obtained for pure and practical
Propane 5.9 5.8 – 6.5 blended liquid fuels burned under lean, fully vaporized, well
premixed, high intensity combustion conditions. The NOx
emission for No. 2 diesel fuel is about double that of natural Absorption by Hydrocarbon Molecules at 3.392 䊐 of He-Ne Laser,’’ Jpn. J.
Appl. Phys., 24, No. 1, pp. 8 –13.
gas, given the same combustion temperature and residence 关7兴 Dibble, R. W., 1999, personal communication, University of California, Ber-
time. keley, CA.
2 The staged prevaporizing-premixing injector, that relies on 关8兴 Mongia, R. K., 1998, ‘‘Optical Probe for Measuring the Extent of Air and Fuel
two stages of air temperature, is described and used to obtain Mixing in Lean Premixed Combustors and the Effect of Air and Fuel Mixing
the fuel-air stream fed into the combustor. The operating con- on Combustor Performance,’’ Ph.D. dissertation, University of California-
Berkeley, Berkeley, CA.
ditions of the injector as used in the present research are 关9兴 Perrin, M. Y., and Hartmann, J. M., 1989, ‘‘High Temperature Absorption of
discussed. Measurements and inspection of the injector indi- the 3.39 䊐m He-Ne Laser Line by Methane,’’ J. Quant. Spectrosc. Radiat.
cate the NOx measurements were obtained for a fully pre- Transf., 42共6兲, pp. 459– 464.
vaporized, well premixed inlet stream devoid of pre-flame 关10兴 Yoshiyama, S., Hamamoto, Y., Tomita, E., and Minami, K., 1996, ‘‘Measure-
ment of Hydrocarbon Fuel Concentration by Means of Infrared Absorption
reaction of the mixture.
Technique with 3.39 䊐m He-Ne Laser,’’ JSAE Review, 17, pp. 339–345.
Recommendations for future work on the staged prevaporizing- 关11兴 Lee, J. C. Y., 2000, ‘‘Reduction of NOx Emission for Lean Prevaporized-
premixing injector are the following: Premixed Combustors,’’ Ph.D. dissertation, University of Washington, Seattle,
WA.
1 Operation for significantly reduced residence times within 关12兴 Steele, R. C., Malte, P. C., Nicol, D. G., and Kramlich, J. C., 1995, ‘‘NOx and
the injector. 共Preliminary testing at double the flow rate of N2O in Lean-Premixed Jet-Stirred Flames,’’ Combust. Flame, 100共3兲, pp.
440– 449.
the present paper has already been conducted. This showed 关13兴 Göttgens, J., Mauss, F., and Peters, N., 1992, ‘‘Analytical Approximations of
no significant increase in NOx, indicating little if any degra- Burning Velocities and Flame Thicknesses of Lean Hydrogen, Methane, Eth-
dation in the fuel vaporization and fuel-air mixing within the ylene, Ethane, Acetylene and Propane Flames,’’ Twenty-Fourth Symposium
injector.兲 (International) on Combustion, The Combustion Institute, Pittsburgh, PA, pp.
2 Testing of the injector at elevated pressure, beginning at 129–135.
关14兴 Abraham, J., Williams, F. A., and Bracco, F. V., 1985, ‘‘A Discussion of Tur-
about 10 atm and ultimately approaching 30 atm. bulent Flame Structure in Premixed Charges,’’ Engine Combustion Analysis:
New Approaches, P-156, SAE, Warrendale, PA, pp. 27– 43.
Acknowledgments 关15兴 Rutar, T., Malte, P. C., and Kramlich, J. C., 2000, ‘‘Investigation of NOx and
CO Formation in Lean Premixed, Methane-Air, High-Intensity, Confined
The research has been made possible through a Graduate Fel- Flames at Elevated Pressures,’’ Twenty-Eighth Symposium (International) on
lowship from the Parker Hannifin Corporation, through equipment Combustion, The Combustion Institute, Pittsburgh, PA.
关16兴 Rutar, T., and Malte, P. C., 2001, ‘‘NOx Formation in High-Pressure Jet-Stirred
donation from the John Fluke Corporation and through the loan of Reactors With Significance to Lean-Premixed Combustion Turbines,’’ paper
laser system components from the University of California, submitted to ASME Turbo Expo 2001.
Berkeley. 关17兴 Nicol, D. G., 1995, ‘‘A Chemical Kinetic and Numerical Study of NOx and
Pollutant Formation in Low-Emission Combustion,’’ Ph.D. dissertation, Uni-
versity of Washington, Seattle, WA.
References 关18兴 Steele, R. C., Tonouchi, J. H., Nicol, D. G., Horning, D. C., Malte, P. C., and
关1兴 Lee, J. C. Y., and Malte, P. C., 2000, ‘‘Staged Prevaporizer-Premixer,’’ U.S. Pratt, D. T., 1998, ‘‘Characterization of NOx, N2O, and CO for Lean-Premixed
Patent No. 6,174,160B1, Jan. 16, 2001. Combustion in a High-Pressure Jet-Stirred Reactor,’’ ASME J. Eng. Gas Tur-
关2兴 Capehart, S. A., Lee, J. C. Y., Williams, J. T., and Malte, P. C., 1997, ‘‘Effect bines Power, 120, pp. 303–310.
of Fuel Composition on NOx Formation in Lean Premixed Prevaporized Com- 关19兴 Bengtsson, K. U. M., Benz, P., Schaeren, R., and Frouzakis, C. E., 1998,
bustion,’’ ASME Paper No. 97-GT-336. ‘‘NyOx Formation in Lean Premixed Combustion of Methane in a High-
关3兴 Lefebvre, A. H., 1989, Atomization and Sprays, Taylor & Francis, Bristol, PA. Pressure Jet-Stirred Reactor,’’ Proc. Combust. Inst., 27, pp. 1393–1401.
关4兴 Lee, J. C. Y., Malte, P. C., and Nicol, D. G., 1999, ‘‘NOx as a Function of Fuel 关20兴 Smith, G. P., Golden, D. M., Frenklach, M., Moriarty, N. W., Eiteneer, B.,
Type: C1-to-C16 Hydrocarbons and Methanol,’’ ASME Paper No. 99-GT-270. Goldenberg, M., Bowman, C. T., Hanson, R., Song, S., Gardiner, W. C. Jr.,
关5兴 Mansour, A., Benjamin, M. A., Straub, D. L., and Richards, G. A., 2000, Lissianski, V., and Qin, Z., 1999, GRI-Mechanism 3.0, http://
‘‘Application of Macrolamination Technology to Lean, Premix Combustion,’’ www.me.berkeley.edu/gri_mech.
ASME Paper No. 2000-GT-0115. 关21兴 Miller, J. A., and Bowman, C. T., 1989, ‘‘Mechanism and Modeling of Nitro-
关6兴 Tsuboi, T., Inomata, K., Tsunoda, Y., Isobe, A., and Nagaya, K., 1985, ‘‘Light gen Chemistry in Combustion,’’ Prog. Energy Combust. Sci., 15, pp. 287–338.
Investigation of the Response of
an Air Blast Atomizer Combustion
Chamber Configuration on Forced
Modulation of Air Feed at
C. Hassa
e-mail: [email protected] Realistic Operating Conditions
J. Heinze DLR investigated forced combustion oscillations of two liquid fuel burners in a research
combustion chamber at elevated pressures simulating idle conditions of aircraft engine
K. Stursberg combustors. The work was performed in collaboration with MTU Munich. An existing
combustion chamber with optical access, capable to operate up to 20 bar, was upgraded
Institute of Propulsion Technology, with an air flow pulsator, that bypasses air from the combustor plenum to the exhaust with
German Aerospace Center, a sinusoidal massflow variation up to 700 Hz. Pressure transducers in the plenum and the
Linder Höhe flame tube monitored the forced disturbances. A photomultiplier recorded the OH* chemi-
51147 Köln, Germany luminescence of the flame. For the agreed operating conditions frequency scans of these
values were registered. Additionally images of the OH* chemiluminescence were taken at
selected frequencies and evaluated in a statistical manner, to separate turbulent and
periodic behavior. From the analysis of the pressure data, it can be concluded, that
serious thermoacoustic feedback was not observed for both burners. However, burner 2
with the flame detached from the wall exhibited a higher fluctuation level as burner 1 with
the wall attached flame. A resonant behavior was observed near the characteristic fre-
quency of the sound room comprised of plenum, flame tube, and burner nozzle as con-
necting passage. The chemiluminescence images show different modes of spatial fluctua-
tion for the burners and for burner 2 they also vary with the operating condition.
关DOI: 10.1115/1.1584478兴
Introduction In the European project, DLR had the task to investigate the
response of a generic aeroengine combustor on forced modulation
In the ACIACOC project, European manufacturers of industrial
of the air feed to practical airblast atomizer nozzles furnished by
gas turbines and aeroengines have been working together with MTU at realistic operating conditions. The response of the spray
research establishments and universities towards the understand- combustion was to be studied by monitoring flame tube pressure
ing of combustion instabilities and their reduction by active con- and total OH* emission in their dependence on the applied forcing
trol 共CORDIS RTD 关1兴兲. In power generation, lean premixed com- frequency versus the steady state. Having found frequency ranges
bustion is widely implemented in gas turbines fueled by natural of interest, further two-dimensional imaging of OH* was to be
gas. However, many of these engines experience potentially fatal performed with conditional averaging through the cycle to acquire
combustion instabilities, those are difficult to predict and limit the the time response of the combusting system and to understand the
power output or a further reduction of unwanted emissions. Hence influence of the most important geometric and thermodynamic
active control is desirable 共Berenbrink and Hoffmann 关2兴兲. For parameters.
future generations of aeroengines implementation of lean pre-
mixed prevaporized 共LPP兲 combustion is also considered as a
means to reduce NOx emission 共IPCC Report 关3兴兲. Aviation, hav- Experimental Set-up
ing higher safety standards, must therefore develop solutions for To meet the requirements of the experimental program, the
combustor oscillations concurrently with LPP development to single sector combustor 共SSC test rig兲 was chosen for the experi-
avoid problems from day one of service. Therefore comparatively ments, because of its operating range and its excellent optical
more work has to be invested in the analysis of the problems also access for laser diagnostics. The test rig is shown in Fig. 1 as
including the effect of the liquid fuel. In part load conditions, schematic representation. Electrically preheated air enters the
combustor oscillations are sometimes also observed for the diffu- combustor plenum via a choked nozzle. It is then fed to the flame
sion burners in use today 共Konrad et al. 关4兴兲. Hence it seems tube through the atomizer and the heat shield of the combustor
worthwhile to gain understanding of the influence of burner de- head plate from where it is issued as cooling film along the quartz
sign parameters on combustion oscillations of diffusion burners windows in the primary zone. 20% of the primary air is bypassed
whose steady-state behavior is well understood. This is especially through the pulsator, a rotating valve with a controllable fre-
true for the higher lean blowout limits needed for future engines quency range between 0 an 700 Hz, that creates a sinusoidal
with increased inlet pressures and turbine inlet temperatures. modulation of the combustion air.
The investigated airblast atomizers are shown in Fig. 2. Liquid
Contributed by the International Gas Turbine Institute 共IGTI兲 of the THE AMERI- kerosene is sprayed from the central pressure atomizer onto the
CAN SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL atomizer lip and is atomized between the two co-swirling air
OF ENGINEERING FOR GAS TURBINES AND POWER . Paper presented at the Inter-
national Gas Turbine and Aeroengine Congress and Exhibition, Amsterdam, The
streams of the nozzle. The only geometrical difference between
Netherlands, June 3– 6, 2002; Paper No. 2002-GT-30059. Manuscript received by the two atomizers consists in the contour of the outer swirl chan-
IGTI, Dec. 2001, final revision Mar. 2002. Associate Editor: E. Benvenuti. nel, which plays a decisive role for the flame shape and flame
872 Õ Vol. 125, OCTOBER 2003 Copyright © 2003 by ASME Transactions of the ASME
Fig. 1 Schematic representation of the single sector com- Fig. 3 FFT peak amplitudes at forced frequencies of plenum
bustor and flame tube pressure
position within the combustor. The diffuser formed as quarter-

circle 共MTU 1兲 causes a strong opening of the flow field and a The signals continuously produced by the sensors are sampled
flame burning near the combustor entrance plane, while the cylin- with an AD converter card on a PC and processed by ’’Flex Pro
drically ending swirl channel 共MTU 2兲 positions a slim flame Control,‘‘ a software for data recording, analysis and graphic rep-
more downstream. resentation. At constant flow conditions eight batches of each of
Ignition is achieved with a hydrogen pilot flame which in turn these signals are taken in a time interval of 4 s. Each batch com-
is ignited by a spark ignitor. The primary zone recirculation is prises 4000 individual samples with a time between samples of
closed by a row of opposed, preheated secondary air jets. Air at 0.125 ms, resulting in a frequency resolution of 1 Hz in the Fou-
ambient temperature is used to convectively cool the flame tube rier transform. Its maximum amplitude is usually situated at the
and control the combustor pressure via the choked nozzle at the forcing frequency.
combustor exit. Optical access to the primary zone is achieved by The variation of modulation frequency results in a variation of
three quartz windows in the side walls of the pressure casing and the FFT amplitudes at forced frequency. Figure 3 presents these
four quartz windows in the flame tube. peak values. The amplitude of the plenum pressure response de-
creases rapidly with increasing modulation frequency at low fre-
quencies, reaches in this case a minimum at 232 Hz and increases
Diagnostics and Data Analysis with a moderate slope, strongly fluctuating at higher frequencies.
Characteristic pressures, temperatures, air mass flow rates, air The flame tube pressure shows a similar behavior, but with less
fuel ratio, cooling water mass flow rates and modulation fre- fluctuations at higher frequencies and with a relative maximum at
quency are continuously recorded. The operating parameters 232 Hz. The PMT signal not has shown here exhibits a strong
maximum at the same frequency.
• air preheat temperature, To show the response of the flame tube pressure and combus-
• pressure loss of the atomizer, tion intensity on the plenum pressure modulation more clearly, the
• combustion chamber pressure, ratio of the flame tube pressure and PMT signal to the plenum
• air fuel ratio, and pressure at the forcing freqiuency are used for the evaluation of
• modulation frequency the experimental data. The amplitude ratios are shown in Fig. 4;
are controlled and varied by a computer program. The global view both plots result in the same resonance frequency of 232.8 Hz.
of the flame is recorded with a video camera. During the frequency variation the following parameters were
kept constant:
FFT-Analysis of Pressure and OH* Chemiluminescence Os-
• combustion air preheat temperature,
cillations. As indicated in Fig. 1 two transient wall pressures are
• flame tube pressure,
detected by water cooled quartz pressure sensors with small di-
• pressure loss of the airblast atomizer, and
mensions: the plenum pressure 共Kistler 1兲 and the flame tube pres-
• fuel flow rate.
sure 共Kistler 2兲. The OH* chemiluminescence intensity fluctua-
tion of the fluctuating part of the flame is synchronously recorded It was assumed that the air inflow into the plenum via critical
by a photo multiplier tube 共PMT兲. nozzles is not influenced by pressure oscillations in the combustor.
But this assumption was slightly wrong. The flame tube pressure
p, the air preheat temperature T, and the fuel flow rate m F vary
during the test duration of 20 minutes by less than 1% of the
nominal value. Only the deviation from nominal value of the
pressure loss ⌬p/p of the airblast atomizer is more pronounced:
up to 5%.
In spite of the constant pressure and temperature in front of the
airblast atomizer and the constant pressure loss of the atomizer
during frequency variation, the air mass flow rate is not constant
共Fig. 5兲. The measured air flow rate 共foxboro vortex flowmeter兲
and the mass flow rate calculated from measured pressure and
temperature in front of the critical nozzles show the same charac-
teristic decrease with increasing frequency. It is assumed, that the
Fig. 2 Airblast atomizers; on the left: MTU 1 „for wall attached oscillating velocity portion increases the pressure loss of the air-
flames…; on the right: MTU 2 „for lifted flames… blast atomizer with increasing frequency in a nonlinear manner.
Fig. 6 Measured volumetric OH* distributions „top… and de-
convoluted spatially resolved OH* distributions „bottom… of a
wall attached flame „left… and of a lifted flame „right… at
T preheat Ä700 K, AFRÄ25 and pÄ9 bar
PMT signal shows significant contributions only at the fundamen-

tal frequency and the harmonics of the forced combustion oscil-
lation.
For the spatially resolved OH* imaging an intensified CCD
camera is used. The integration time of 5 ␮ s for a single image is
Fig. 4 FFT amplitudes ratio of OH* emission „top… and flame sufficiently short to freeze the motion of all movements. Because
tube pressure „bottom… related to plenum pressure of the highly turbulent fluctuations of the combustion normally
only phase averaged images are digitized and stored on a PC for
further image processing. To enable the synchronization with the
combustion oscillation and to define the phase angle, the modula-
Two-Dimensional OH* Chemiluminescence Imaging . For tor of the primary air flow is equipped with an optical switch. This
the indication of the reaction zone the radiation of electronically switch generates a trigger pulse at fully opened by pass valve
excited hydroxyl radicals OH* at 312 ⫹⫺7 nm was used 共Haber 共minimal combustion air flow rate兲 which defines the phase angle
et al. 关5兴兲. From a measurement of the OH* chemiluminescence of 0 deg. In general, a combustion oscillation is monitored by a
only the volumetric information of the heat release can be de- series of nine subsequent phase averaged images taken with an
duced. The intensity of an OH* image is always the integral of all equidistant phase angle increment of 45 deg. The periodic fluctua-
contributions towards the camera view. tion of the phase averaged images is generally small, two to three
On the additional assumption of axial symmetry it is possible to times smaller than the turbulent fluctuations of single images.
deconvolute the OH* image to deduce the spatially resolved heat Creating a short movie of the phase averaged images the
release 共Dasch 关6兴兲. To illustrate the difference, Fig. 6 shows the periodic motion of the OH* chemiluminescence of the reaction
measured and deconvoluted OH* distributions of highly axially zone during the combustion oscillation can be visualized in an
symmetric wall attached and detached flames. instructive way.
For time resolved measurements of the combustion oscillations For a fixed operating condition 共nozzle type, pressure, AFR, air
a representative area of the flame of about 3 cm is imaged on the preheating兲, phase averaged images averaged over the oscillation
photocathode of the PMT. Figure 7 show these areas marked by a period are equal to within 5% with respect to their integral inten-
circle for the MTU1 air blast injector. The Fourier analysis of the sity for all examined forced oscillation frequencies from 22 Hz up
to 696 Hz. But the periodic fluctuations are quite different for
these frequencies. To enable a systematic overview of the periodic
OH* fluctuations of the different oscillation frequencies, the peri-
odic fluctuations of the oscillation frequencies from 72 Hz to 696
Hz together with the phase averaged OH* distribution averaged
Fig. 7 OH* chemiluminescence of a forced combustion oscil-

lation at 295 Hz of the wall attached flame at 9 bar. The average
Fig. 5 Measured primary air flow rate as function of forcing and the periodic fluctuation of all phase resolved images are
frequency shown.
Table 1 ACIACOC test conditions
Test objects MTU–Burner 1

共for wall attached flames兲
MTU–Burner 2
共for lifted flames
Flametube pressure 3 bar, 6 bar, 9 bar
Air preheat 450 K, 700 K
temperature
Burner pressure loss 3%
Air/fuel ratio 25 and near lean
extinction
over the oscillation period are shown. To separate the periodic

fluctuation from the influence of the turbulent fluctuation, the pe-
riodic fluctuations are deduced from phase averaged images com- Fig. 8 Resonance frequencies according to PMT signal for all
operating conditions; marked are the different injectors and
prising 600 single shots. This means, that 4.800 phase resolved preheat temperatures
images are used for the determination of each periodic fluctuation
image.
Operating Conditions From these results and the identity of the swirlers within the
tested airblast atomizers it can be concluded:
As previously expressed, the industrial interest focused on the
frequency range of 0 to 700 Hz at idle conditions. A fluctuation of • there is no influence of the airblast atomizer’s diffuser shape
20% RMS was seen as the maximum value necessary to cover on the examined resonance frequency of the system,
oscillations induced by compressor instabilities. MTU was spe- • there is no distinguishable influence of the flame shape and
cifically interested to see behavioral differences between the two the thermal conditions within the flame tube on the examined
air blast nozzle designs that are respectively used in different ap- resonance frequency, and
plications. Thermodynamic parameters were to be varied around a • at a constant geometry of the test section the resonance fre-
typical idle condition in order to understand their influence on the quency depends only on air preheat temperature.
oscillatory behavior of the combustor. Thus after consultation of
partners MTU and University of Karlsruhe, the following mutu- Therefore the plenum/flame-tube combination can be treated as
ally agreed test conditions were determined 共Table 1兲. a sound room, consisting of two hollow spaces and a connecting
Pressure was varied between 3 and 9 bar, covering idle pres- passage 共airblast atomizer兲. The condition, that all dimensions of
sures for small and large engines, preheat temperature between the sound room are smaller than ␭/4 of the stimulating oscillation,
450 K being below and 700 K being above the final boiling point is fulfilled.
of the fuel, to see the eventual influence of prevaporization of the The resonance frequency of the system f 0 can then be calcu-
liquid fuel and AFR between 25 and near lean blow out to register lated by the equation deduced from 共Veit 关7兴兲
the influence of eventual partial extinction during the modulation
period on combustion oscillation. 冑
f 0 ⫽c/ 共 2 ␲ 兲 * 2 * 冑A e f f / ␲ * 共 V 1 ⫹V 2 兲 / 共 V 1 * V 2 兲
Experimental Results with

FFT-Analysis of Pressure and OH* Chemiluminescence Os- c ⫽ speed of sound of the medium within the passage
cillations. At all test conditions the pulsation frequency was var- Aef f ⫽ cross-section of the connecting passage
ied between 0 and 700 Hz in steps of 12 Hz. After a short period V1 ⫽ plenum volume
for adjustment, the signal samples were taken at every frequency V2 ⫽ flame tube volume.
stop. All recorded data were analyzed as indicated above. The oscillating gas volume within the connecting passage—
Each frequency scan results in spectral amplitude distributions driven by the changing plenum pressure—is the deciding factor in
resembling qualitatively Fig. 4. The forcing frequencies of the the acoustic coupling. The gas in the passage 共atomizer兲 has al-
spectral peaks of the scans are considered as resonance frequen- ways preheat temperature; no hot gas is passing upstream. Thus
cies of the system. These frequencies and the FFT amplitudes of only the preheat temperature influences the resonance frequency;
the PMT signal at these spectral positions 共in arbitrary, but always there is no influence of the backside temperature. In this case the
the same units兲 for all test conditions are plotted in Fig. 8. With analogue model results in a remarkably good agreement with the
respect to frequency, the data exhibit the following trends 共note above given experimental findings:
that there is a small difference between peak frequency in the
absolute and in the relative spectral amplitude distributions兲:
T⫽450 K→ f 0 ⫽259.9 Hz
• all measured resonance frequencies are found in the fre-
quency range 256 –326 Hz 共because of the frequency steps of
12 Hz the uncertainty amounts to ⫹/- 6 Hz兲, T⫽700 K→ f 0 ⫽324.2 Hz.
• the frequencies at low preheat temperature 共450 K兲 tend to
the lower limit and at high preheat temperature 共700 K兲 to the Nearly the same result as with the PMT signal 共Fig. 8兲 is ob-
upper limit, tained, if the ratio of flame tube pressure/plenum pressure is ex-
• the ratio between the limiting frequencies 共1.27兲 is equal to amined 共Fig. 9兲.
the root extracted from the preheat temperature ratio 共1.25兲, The chemiluminescence images were evaluated for the periodic
and fluctuation of the intensity in the phase-averaged images in the
• there is no distinct difference in the resonance frequencies for resonant frequency band at 295 Hz. The experimental results
both test objects. show the influence of the nozzle configuration, pressure, tempera-
The burner geometry is by far the most sensible factor, with
configuration 2 producing about twice as much pressure and
chemiluminescence fluctuation as configuration 1. The geometry
effect was expected by the industrial partner, however since
acoustic feedback was not registered and fluctuating pressure am-
plitudes remained small in this combustor, there does not seem to
be cause for serious concern.
The preheat temperature has a different influence on pressure
and rate of heat release. The lower temperature gives slightly
higher heat release fluctuations but roughly a twofold increase in
relative pressure fluctuation. This is explained with the help of
Fig. 10. It represents the deconvoluted image of the average
chemiluminescence and it’s periodic fluctuation at resonance for
the lower preheat. The fluctuation shows two different burning
zones, which we believe to be connected to flame stabilization by
Fig. 9 Resonance frequencies according to ratio
p flame tube Õ p plenum for all operating conditions; marked are the
inner and outer recirculation. Whereas the photomultiplier regis-
different preheat temperatures ters luminescence irrespective of it’s origin, the pressure sensor
will more likely be influenced by the change of the location of
heat release.
With rising pressure, the amplitudes of the flame tube pressure
ture and AFR on the flame tube and flame response. Their se-
fluctuation normalized with the plenum fluctuation remain rather
quence in decreasing order of importance turns out to be
constant, whereas the chemiluminescence increases less than pro-
• burner geometry, portional with pressure due to the more important quenching at
• preheat temperature, higher pressure. The two facts suggest that the oscillation is pro-
• pressure, and portional to the gas density or the number of molecules taking part
• AFR. in the combustion.
The differences with respect to air/fuel ratio are also small with
a slightly higher fluctuation for the richer AFR 25 case. Here, the
higher reaction rate seems to outweigh the influence of partial
extinction for the AFR near blowoff. This is probably a feature
which is specific to the nozzle design and method of fuel prepa-
ration. During the lean blowout trials preceding the choice of the
lean AFR, it was noted that the nozzle exhibits a rather sharp
extinction with good circumferential homogeneity close to
blowoff.
Summarizing the influence of the operating conditions, the pa-
rameters leading to a more compact reaction zone, higher pres-
sure, temperature and equivalence ratio, display lower fluctuation
Fig. 10 Averaged OH* chemiluminescence of the detached levels.
flame at 9 bar, AFR 25, and 450 K preheat temperature
Fig. 11 Phase resolved OH* chemiluminescence of a forced combustion oscillation at 295 Hz of a detached flame at
9 bar „AFR 25, 700 K air preheat temperature… and averaged over 600 single images
Fig. 12 Periodic fluctuations of the OH* chemiluminescence for a detached flame at 9 bar for the different oscillation
frequencies of 22 Hz–696 Hz. The detached flame „MTU 2 nozzle… burns at an AFR of 25 and an air preheating of 700 K.
Two-Dimensional OH* Chemiluminescence Imaging. riodicity of the induced mass flow or AFR variation this is not the
Compared to the wall attached flame 共MTU 1 nozzle, Fig. 6 and case for the ‘‘waving’’ motion. Most probably, it is connected to a
Fig. 7兲 the wall detached flames generated by the MTU 2 nozzle residence time effect.
共Fig. 10兲 show a higher sensitivity to forced combustion oscilla- On the basis of a calculation of the isothermal flow field and a
tions. Both the integral intensity and the spatial distribution of Lagrangian particle tracking routine, the residence time of a fluid
OH* vary significantly during a combustion oscillation. For 3 bar, volume in the luminescent part of the images can be estimated
700 K, and AFR 25 the flame exhibits a periodic lifting from and between 1 and 2 ms. Hence, exchange of the charge of the heat
reattachment to the burner mouth. Figure 11 shows for an example release zone between two successive phase averaged images can
the strong flame zone variation of the combustion oscillation at be expected for frequencies above 162 Hz. Since the wave effect
295 Hz of the detached flame at 9 bar. The picture with the peri- becomes apparent in the phase averaged images, it must be a
odic fluctuations brings out, that they are most intensive at the periodic effect, not connected to a time or length scale of statisti-
root of the flame and in the burn out zone. However comparison cal turbulence.
with the average shows the flame burning rather stable in the Different mechanisms could be responsible for the formation of
region of maximum heat release. Since the location and fluctua- that movement of the chemiluminescent zone:
tion intensity of the region of maximum heat release is most im-
portant for the thermo-acoustic oscillation, the comparison of • AFR variations, caused by periodic accumulation of liquid at
mean and fluctuating parts shows a favorable behavior even in the the atomizer lip could also lead to a periodic variation of the
resonance frequency of the sound room. dispersion of the fuel,
Figure 12 shows the systematic overview of the periodic oscil- • periodic vortex detachment could be triggered by the forced
lations of the wall detached flame at 9 bar. Because the intensities modulation of the airflow 共Büchner 关8兴兲, and
averaged over a period are equal within 5% for all frequencies, • periodic disturbances of the nozzle through flow could lead to
only once the OH* distribution averaged over all phase angles a loss of self similarity for the gas flow. Such an effect was
and frequencies 共‘‘OH* average’’兲 is shown. For the lowest fre- recently observed in an isothermal investigation of a sinusoi-
quencies 共⭐72 Hz 兲 the combustion oscillations of the detached dally modulated air blast nozzle 共Giuliani et al. 关9兴兲.
flame are very intense, but the increase with respect to decreasing For the investigation of these effects, different diagnostics
frequencies is less pronounced as it is for the wall attached flame. would be needed. The Schlieren technique was tried to detect
The integral value of the periodic combustion fluctuation of the coherent vortices but failed due to the three-dimensional nature of
detached flame is about three times higher for frequencies exclud- the flow and strong thermal boundary layers from the window
ing the lowest modulation frequency of 22 Hz. At that frequency cooling.
the highest variation of the expansion of the hot flow, probably
connected to the highest AFR fluctuation, with the ensuing radial
displacement of the reaction zone is observed. Conclusion
At frequencies above 162 Hz, the periodic fluctuations often From the analysis of the pressure data, it can be concluded that
show two separate zones of high periodic fluctuation. In the video serious thermoacoustic feedback was not observed for both burn-
sequences the frequencies above 295 Hz create a ‘‘waving’’ mo- ers. The separation of turbulent and periodic fluctuations by phase
tion of the chemiluminescence distribution as opposed to the averaging of chemiluminescence showed the turbulent fluctuation
‘‘breathing’’ motion of the lower frequencies. While the ‘‘breath- to be higher than the periodic part.
ing’’ behavior can be understood as the quasi-steady response of Burner 2 with the flame detached from the wall exhibited a
chemistry and flow having short time scales compared to the pe- higher fluctuation level as burner 1 with the wall attached flame.
A resonant behavior was observed near the characteristic fre- The authors thank K. Katheder of MTU Munich for the use of
quency of the sound room comprised of plenum, flame tube, and the data of his isothermal flowfield calculation and T. Behrendt for
burner nozzle as connecting passage. the use of his deconvolution algorithm.
The importance of the varied parameters in decreasing order
was: burner geometry, preheat temperature, pressure and AFR. References
The chemiluminescence images show different modes of spatial 关1兴 CORDIS RTD-PROJECTS/© European Communities, 2000, ‘‘Active Control
of Instabilities in Advanced Combustion Chambers,’’ http://dbs.cordis.lu.
fluctuation for the burners. Burner 2 exhibits a ‘‘breathing’’ mo- 关2兴 Berenbrink, S., and Hoffmann, S., 2000, ‘‘Suppression of Dynamic Combus-
tion of the heat release zone at lower frequencies and a ‘‘waving’’ tion Instabilities by Passive and Active Means,’’ ASME-2000-GT- 0079.
motion at frequencies above 295 Hz. 关3兴 IPCC Special Report, 1999, Aviation and the Global Atmosphere, Cambridge
University Press, Cambridge, UK.
关4兴 Konrad, W., et al., 1998, ‘‘Combustion Instability Investigations on the BR
Outlook 710 Jet Engine,’’ ASME J. Eng. Gas Turbines Power, 120, pp. 34 – 40.
关5兴 Haber, L., Vandsburger, U., Saunders, W., Khanna, V., 2000, ‘‘An Examination
In a follow-up project with MTU we will investigate more spe- of the Relationship Between Chemiluminescent Light Emissions and Heat Re-
cifically the role of the liquid fuel, since this is the missing link lease Rate Under Non Adiabatic Conditions’’, 2000-GT-0121.
between the forced airflow and the heat release. It is planned to 关6兴 Dasch, C. J., 1992, ‘‘One-dimensional Tomography: A Comparison of Abel,
observe simultaneously the Mie scattering and the chemilumines- Onion-Peeling, and Filtered Backprojection Methods,’’ Appl. Opt., 31共8兲,
1146 –1152.
cence to clarify the extent of AFR variations with forced oscilla- 关7兴 Veit, I., 1974, Technische Akustik, Kamprath-Reihe kurz und bündig, Vogel-
tion frequency and the influence of the fuel contained in the fuel Verlag, Würtzburg.
film prior to atomization. 关8兴 Büchner, H., 1992, ‘‘Experimentelle und theoretische Untersuchungen der
Entstehungsmechanismen selbsterregter Druckschwingungen in technischen
Vormisch-Verbrennungssystemen,’’ Diss., Univ. Karlsruhe, Shaker-Verlag
Acknowledgments Aachen.
关9兴 Giuliani, F., Diers, O., Gajan, P., and Ledoux, M., 2002, ‘‘Characterization of
Financial support of the ACIACOC Project by the European an Air-Blast Injection Device With Forced Periodic Entries,’’ Proceedings of
Commission, BE 97 4324 BRPR CT97 0506 is gratefully the IUTAM Symposium on Turbulent Mixing and Combustion, Pollard and
acknowledged. Candel, eds., Kluwer, Dordrecht, The Netherlands.
H. Karim Advanced Catalytic Pilot for Low
K. Lyle1 NOx Industrial Gas Turbines
S. Etemad This paper describes the design and testing of a catalytically stabilized pilot burner for
current and advanced Dry Low NO x (DLN) gas turbine combustors. In this paper, appli-
L. L. Smith cation of the catalytic pilot technology to industrial engines is described using Solar
Turbines’ Taurus 70 engine. The objective of the work described is to develop the catalytic
W. C. Pfefferle pilot technology and document the emission benefits of catalytic pilot technology when
compared to higher, NO x producing pilots. The catalytic pilot was designed to replace the
Precision Combustion, Inc.,
existing pilot in the existing DLN injector without major modification to the injector.
410 Sackett Point Road,
During high-pressure testing, the catalytic pilot showed no incidence of flashback or
North Haven, CT 06473
autoignition while operating over wide range of combustion temperatures. The catalytic
reactor lit off at a temperature of approximately 598 K (325°C/617°F) and operated at
simulated 100% and 50% load conditions without a preburner. At high pressure, the
P. Dutta maximum catalyst surface temperature was similar to that observed during atmospheric
pressure testing and considerably lower than the surface temperature expected in lean-
K. Smith burn catalytic devices. In single-injector rig testing, the integrated assembly of the cata-
lytic pilot and Taurus 70 injector demonstrated NO x and CO emission less than 5 ppm @
Solar Turbines, Inc.,
15% O 2 for 100% and 50% load conditions along with low acoustics. The results dem-
2200 Pacific Highway,
onstrate that a catalytic pilot burner replacing a diffusion flame or partially premixed
San Diego, CA 92186
pilot in an otherwise DLN combustor can enable operation at conditions with substan-
tially reduced NO x emissions. 关DOI: 10.1115/1.1586313兴
Introduction lean limit and achieve 9–25 ppm NOx . Nevertheless the reliance
on a high NOx pilot can be a barrier to further NOx reductions.
Government regulations are forcing both industrial and utility
gas turbine manufacturers to improve engine designs for achiev-
ing lower possible CO and NOx emissions, 关1,2兴. In this quest to Catalytic CombustionÕCatalytic Pilot
achieve lower emissions, lean-premixed combustion technology is Catalytic combustion has been known to provide combustion
preferred by both the industrial and utility engine manufacturers. stability to very lean fuel air mixtures 关6 – 8兴. Catalytically stabi-
Lean-premixed combustion technology has demonstrated the abil- lized combustion was first demonstrated by Pfefferle 关6兴 and was
ity to achieve NOx concentrations as low as 9 ppm corrected to shown to be an efficient method for burning fuels in lean fuel air
15% O2 , during operation on natural gas, 关1,3兴. mixtures without significant formation of pollutants. Dalla Betta
At low-NOx operating conditions, however, flame temperatures 关7兴 demonstrated NOx⬍3 ppm from a small-scale staged catalytic
are reduced and flame stability issues arise. In addition, combus- combustion system operated at outlet temperatures from 1225°C
tion of uniform, premixed fuel/air mixtures, with little additional to 1500°C and pressure from 12 to 20 atmosphere. Current devel-
combustor air, can lead to combustion-induced pressure oscilla- opment status and design challenges for application of full cata-
tions 共noise兲. To mitigate these combustion instability issues with lytic combustion that have been addressed by Siemens Westing-
house Power Corp for large utility engines and Solar Turbine for
lean-premixed flames, gas turbine manufacturers frequently use
smaller industrial engines are discussed in the reference by Fant
higher temperature pilot flames to impart stability to the main 关9兴. Dalla Betta 关10兴 discusses application of full catalytic com-
combustion process. One such technique is the use of either a bustion for a 1.5 MW engine.
diffusion flame pilot or a partially premixed pilot, 关3–5兴. Tradi- This paper presents an alternative concept to a full catalytic
tionally the pilot is another fuel injector in which a small portion combustion system. Instead of replacing the entire DLN combus-
of the fuel is combusted in a diffusion flame or partially premixed tor with catalytic combustion, the catalytic combustion technology
mode. The rest of the fuel is combusted in a lean-premixed flame discussed in this paper only replaces the diffusion flame pilot.
while the pilot maintains the stability of this main flame. In addi- Thus the catalytic pilot concept focuses on replacing the highest
tion to providing stability at baseload, the pilot also provides com- temperature portion of DLN combustor with catalytic combus-
bustion stability during engine startup, load ramping, transients, tion technology. This makes the catalytic pilot a good candidate
and fuel transfer operation. for both new engines and retrofit applications without major
Depending on the design of the combustor, 2–10% of the fuel modification and cost. The concept uses the best features of cata-
can be used for the piloting at baseload. If more fuel is used for lytic combustion 共increased stability through prereaction兲 and
piloting, more NOx is produced due to the higher temperatures conventional 共proven兲 aerodynamically stabilized combustion
associated with pilot flame. Thus with such conventional pilots, technology.
dry low NOx 共DLN兲 combustors can operate close to the overall The motivation of using a catalytic pilot is that the catalytic
prereaction in the pilot will provide enhanced reactivity to the gas
1
Currently Research Assistant at the Department of Mechanical Engineering,
mixture exiting the pilot and thereby obviate the need to use high
Stanford University, Building 520, Stanford, CA 94305. temperature flames to stabilize the combustion of the primary fuel
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN air mixture exiting the swirler.
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF This paper discusses the application and results from integra-
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna-
tional Gas Turbine and Aeroengine Congress and Exhibition, Amsterdam, The Neth-
tion of a catalytic pilot with a lean-premixed fuel injector from a
erlands, June 3– 6, 2002; Paper No. 2002-GT-30083. Manuscript received by IGTI, DLN gas-turbine combustor. Figure 1 shows a sketch of the cata-
December 2001, final revision, March 2002. Associate Editor: E. Benvenuti. lytic pilot concept and how it integrates within a main injector
Fig. 2 Sketch of the catalytic reactor for the pilot
now-lean reactive mixture are then achieved in the downstream

combustion section to produce the pilot flame. This is called
‘‘Rich Catalytic/Lean-burn,’’ or RCL™ combustion. This ap-
proach avoids both soot formation as discussed in literature,
关14,15兴, and the high temperatures, which in noncatalytic RQL
共rich-burn/quench/lean-burn兲 designs lead to high NOx formation.
In the RCL™ system, fuel-rich reaction occurs at moderate tem-
peratures on the catalyst surface. The catalyst also allows fuel-rich
reaction outside the gas-phase flammability limits in the tempera-
ture range of 700–900°C 共973 K/1533°F–1173 K/1652°F兲. Thus,
the high temperatures present in the fuel-rich stage of noncatalytic
systems 共e.g., RQL兲 are simply not present in the RCL device. As
a result, of long autoignition delay time, which is shown in Fig. 3,
for mixing catalytically reacted natural-gas with air after the cata-
Fig. 1 Sketch of the catalytic pilot concept within a lean pre- lytic reactor, mixing can be completed without autoignition. Sub-
mixed injector „swirler¿pilot… sequent to mixing, a low temperature lean premixed pilot flame
can be stabilized and as a result this combustion approach yields
low NOx .
共injector consist of an annular swirler and a pilot in the center of In Fig. 3, the autoignition delay time is shown as calculated for
it兲 for a lean premixed combustion system. The figure shows that a representative gas-mixture 共in the post-mix region after the cata-
a portion of the fuel air/mixture passes through the catalytic pilot lytic section兲 as a function of mixed gas temperature and it can be
and prereacts within the catalytic reactor before a pilot flame is observed that at 973 K 共700°C/1292°F兲, the autoignition delay
stabilized downstream of the reactor. The rest of the fuel and air time is approximately 0.045 seconds, which is sufficiently long to
passes through the swirler portion of the injector and the catalytic permit mixing.
pilot flame anchors a swirl-stabilized flame. This reactor design approach can provide several advantages
At low equivalence ratios, which are required for low NOx over fuel lean catalytic reactor design and these are
operation of the combustor, the main swirler flame depends on the
catalytic pilot flame for its stability and the pilot’s flame depends • Low temperature lightoff: Significantly lower temperature
on the catalytic prereaction for its stability. Depending on design lightoff 共as discussed in the results section兲 than can be
and load condition of the engine, the equivalence ratios of the achieved in fuel-lean catalysis.
main swirler flame and the pilot flame can be different. • No preignition or flashback danger in the catalytic reactor:
Unlike fuel-lean catalyst systems, there is no fuel in the cool-
Catalytic Reactor Design ing air stream to ignite. At the same time, reaction of the
As shown in Fig. 1, a catalytic pilot consists of a catalytic
reactor and a downstream homogeneous combustion zone. There
are several approaches currently practiced in design of a catalytic
reactor for the gas turbine and example of these approaches can be
found in the literature, 关11–13兴. In the subject approach, the tem-
perature rise in the catalytic reactor is limited by limiting the
extent of fuel oxidation that occurs within the catalyst itself. Fig-
ure 2 shows a sketch of the catalytic reactor concept.
The catalytic reactor of Precision Combustion Inc.’s 共PCI兲 pilot
utilizes fuel-rich partial oxidation to limit the extent of reaction in
the reactor. As shown schematically in Fig. 2, all of the pilot fuel
and a fraction of the pilot air are mixed before contacting a cata-
lyst under fuel-rich conditions. The balance of the pilot air pro-
vides catalyst cooling. The cooling air and catalyzed fuel/air mix-
ture are subsequently rapidly mixed after the catalyst section to
produce a fuel-lean, reactive mixture. The mixing is complete
before the mixture exits the pilot and comes in contact with the
fuel/air mixture from the swirler. Thus even though the catalyst
section of the reactor uses fuel-rich combustion, overall the reac-
tor for the pilot operates fuel lean. Ignition and combustion of the Fig. 3 Prediction of autoignition delay
Fig. 4 Photograph of modified Solar Taurus 70 premixed fuelÕ
air injector hardware, with PCI’s catalytic pilot
fuel-rich mixture in contact with the catalyst is oxygen-

limited, and gas-phase combustion within the catalyst bed
poses no problem.
• Tolerant to fuel/air unmixedness: In addition to reducing the
potential for flashback/preignition, the fuel rich catalyst de-
sign reduces the sensitivity of catalyst temperature to fuel/air
ratio, allowing safe operation with up to ⫾10% uniformity
共compared to the required ⫾2% as reported in Yee 关16兴兲.
• Improved catalyst durability: Fuel-rich operation reduces
catalyst volatilization and substrate oxidation, enhancing
catalyst life relative to lean burn systems.
Single-Injector Catalytic Pilot

To verify and demonstrate the low NOx potential of the cata- Fig. 5 Top panel: single-injector atmospheric pressure test fa-
lytic pilot for industrial engines, a pilot was fabricated for a single cility at PCI. botom panel: single-injector high-pressure test fa-
injector that is used in Solar Turbines’ Taurus 70 turbine. The cility at Solar Turbine.
Taurus 70 is an industrial engine with a design rating of 7.2 MW
@ 1394 K 共1121°C/2050°F兲 turbine rotor inlet temperature, 关17兴.
The engine has 12 injectors (swirler⫹pilot) arranged circumfer-
entially around an annular combustor liner.
Figure 4 shows a hardware picture of the catalytic pilot inte- configuration was first tested at PCI, in an atmospheric pressure
grated inside a Taurus 70 injector The catalytic pilot replaced the rig, a photograph of which is shown in the top panel of Fig. 5.
existing partial premixed pilot without modification to the sur- Testing was conducted both for the catalytic reactor only and
rounding injector geometry. for the catalytic pilot and injector together. These tests were
As shown in Fig. 4 the catalytic pilot is installed on the injector conducted at the simulated baseload condition of T-70 engine.
centerline within the swirler hub. The orientation of the injector in This implies that the pilot and injector were run at the same
the photograph shows the mounting flange at upper left, and flow velocities in the atmospheric pressure rig as they would at
swirler and catalytic reactor exits at lower right. All the testing baseload condition in an engine. The velocities through the com-
reported in this paper was conducted with the modified Taurus 70 ponent were identical to their values at baseload when the inlet
injector shown in Fig. 4. temperature used at atmospheric condition was same as that of the
The modified Taurus 70 injector assembly was integrated with baseload.
an augmented backside cooled 共ABC兲 combustor liner, 关18兴. Ap-
High-Pressure Testing Facility at Solar Turbines. After
proximately 50% of the combustion system airflow enters the
completion of testing at atmospheric pressure conditions, the
liner through the catalytic pilot and injectors. The remaining 50%
modified injector assembly was tested at Solar’s high-pressure
of the air is used to cool the liner and enters the combustion
single-injector test facility; a photograph of the facility is shown
zone 共liner兲 either through the dome cooling holes in the liner at
above in the bottom panel of Fig. 5. The test facility permitted
the front or through dilution holes in the post-primary combus-
independent control of the airflow, natural gas 共CH4: 95.82%,
tion zone.
C2H6: 1.8%, C3H8: 0.32%, C4H10: 0.13%, CO2: 0.89%, N2:
1.0兲 flows to the catalytic pilot and swirler, and total pressure of
Test Facilities the system. At this facility, the modified injector was tested at
Two different test facilities were used during the evaluation of close to full load 共limited by available air flow兲 and half-load
the catalytic pilot performance. conditions, which are shown in Table 1.
During the several days of operation at Solar’s high-pressure
Atmospheric Pressure Testing Facility at PCI. The inte- facility no autoignition or flashback in the catalytic pilot was
grated assembly of the modified T-70 injector and the ABC liner observed.
Table 1 Nominal Taurus 70 full-load combustor operating
conditions
Baseload inlet temperature 705 K 共432°C/810°F兲

Baseload inlet pressure 1.7 Mpa 共250 psia兲
Baseload pressure drop 4.0%
Half-load inlet temperature 636 K 共363°C/687°F兲
Half-load inlet pressure 1.02 Mpa 共153 psia兲
Half-load pressure drop 3.5%
Discussion of Results
Results have been obtained for both atmospheric pressure and a
pressure of 15 atm. operation. Only high pressure results will be
discussed as these are more relevant to the Taurus 70 engine.
Catalyst Lightoff. The lightoff characteristics of the catalytic
reactor at 15 atm. are shown in Fig. 6. A heater was used to heat
the air to the rig. The abscissa shows the inlet air temperature and
the ordinate shows both the catalyst surface temperature, and the
temperature of the fuel-lean gas exiting the catalytic reactor. As
expected, prior to lightoff, the catalyst temperature and the tem-
perature of the gas exiting the reactor were both the same as that Fig. 7 Catalyst surface temperatures at a pressure of 15 atm
of the inlet air temperature. It can be observed from Fig. 6 that
lightoff of the reactor occurred when the inlet air temperature
reached a temperature of approximately 325°C 共598 K/617°F兲.
Lightoff is indicated by the separation of the catalyst temperature reactor, no reaction will occur inside the catalytic reactor. The
and the exit gas temperature, and also by the difference of these unreacted fuel air mixture exiting the catalyst will then be com-
two temperatures from that of the inlet air. As the catalyst lights busted in the downstream liner 共combustor兲 section along with the
off, it becomes sufficiently active that fuel was oxidized and heat fuel/air mixture from the swirler albeit with higher emissions.
release occurred on the surface. The gas in the catalytic reactor Once the combustor ramps up and the compressor discharge tem-
was heated by the convection of heat from the catalyst surface. perature reaches the lightoff temperature of the reactor, fuel oxi-
Due to resistances in heat transfer mechanisms, the exit gas tem- dation in the reactor starts to occur. This catalytic reaction pro-
perature was less than the maximum surface temperature. vides further stability to the overall combustion process. The
The measured lightoff temperature of 325°C 共598 K/617°F兲 is higher fuel-rich equivalence ratios which will result from such a
substantially lower than the 450°C 共723 K/842°F兲 lightoff tem- fuel split are within the safe operating range of the reactor, since
perature reported in the literature for catalytic combustion of natu- higher fuel rich equivalence ratios produce less heat. When 50%
ral gas using fuel lean catalysis, 关19兴. This lower lightoff tempera- load is reached, fuel flow to the pilot is readjusted for operating in
ture is a direct consequence of using fuel rich oxidation in the the low NOx regime
catalytic reactor and has shown no dependence on pressure. Fig- Surface Temperature of the Catalyst. Figure 7 shows cata-
ure 6 also shows that the lightoff temperature is lower than the lyst surface temperatures at a pressure of 15 atmospheres as func-
compressor outlet temperatures both at 100% and 50% load con- tion of nondimensional reactor length. The maximum catalyst sur-
ditions of the Taurus 70 machine. Thus the catalytic reactor of the face temperature was below 850°C, a temperature much lower
pilot does not need a preburner for operation from 50% load to than the reported maximum surface temperature for fuel lean cata-
100% load. lytic combustion by 共Dalla Betta, 关20兴兲. This reduced temperature
During startup, the catalytic pilot can be used as a noncatalytic operation decreases catalyst volatilization, and sintering, which
fuel stage to provide 50% to 60% of the fuel for the combustor are factors judged potentially life limiting for catalytic combustors
thereby creating a rich primary zone. Since compressor exit air and permits the use of metal for catalyst substrate.
temperature will be lower than the lightoff temperature for the The maximum catalyst surface temperature was similar to the
maximum surface temperature during atmospheric operation. As a
consequence of this, catalytic reactor operation at atmospheric
pressure gives a good estimate to the maximum temperature at
high pressure.
Emission Results. Due to air flow limitation in the high pres-
sure rig, emission data were collected at 15 atm. instead of 17
atm, which is the baseload pressure for Taurus 70 engine. As
discussed earlier in the ‘‘single-injector catalytic pilot’’ section,
the modified Taurus 70 injector was integrated with an ABC liner
and installed in the high-pressure rig. As shown in Table 2, the
Table 2 Air splits in the combustor
Catalytic Pilot Configuration Standard Pilot Configuration

Catalytic pilot 12.2% Pilot 3.3%
Swirler 40.6% Swirler 45.0%
Liner 共dome cooling兲 12.2% Liner 共dome cool.兲 13.4%
Liner 共dilution holes兲 35.0% Liner 共dilution holes兲 38.3%
Fig. 6 Lightoff for the catalytic reactor at a pressure of 15 atm
Fig. 8 NOx emission dependence on catalytic pilot equiva-
lence ratios
replacement of the standard diffusion pilot with the catalytic pilot

increased the air split to the pilot to 12.2% from 3.3%.
All emission results will be discussed in terms of pilot and Fig. 9 Measured NOx and CO emission for the modified Taurus
swirler equivalence ratios and front end adiabatic flame tempera- 70 injector at close to 100% load condition
tures 共calculated based on pilot and swirler air and fuel flow兲
rather than primary zone flame temperatures, which are difficult to
measure because of entrainment of dome cooling air and liner
through the NOx data set. Figure 9 also shows the operating tem-
dilution air.
Figure 8 shows the sensitivity of NOx emission to catalytic pilot perature for this modified configuration at 100% load to be around
equivalence ratios at a pressure of 15 atm. The symbols show the 1950 K 共1677°C/3050°F兲.
experimental data obtained after correction to 15% O2 in a dry We observe that with this modified configuration less than 5
sample, and the lines represent trend lines drawn through the data ppm NOx was achieved for the operating condition. Also, less than
points with constant swirler equivalence ratio. For a constant 5 ppm NOx was also achieved in the front end temperature range
swirler equivalence ratio ( ␾ swirler), emission data were first col- of 1900–2000 K 共1627–1732°C/3000–3150°F兲. In diffusion
lected with the highest pilot equivalence ratio (NOx⬍20). Then flame pilot tests with a similar 共same effective area兲 swirler, NOx
gradually the pilot equivalence ratio ( ␾ pilot) was dropped until emission of 15 ppm resulted for an operating front end tempera-
blowoff occurred or CO exceeded 20 ppm. CO exceeded 10 ppm ture of approximately 2140 K 共1868°C/3394°F兲. Clearly, the use
in only the first two data points ( ␾ pilot⫽0.55, 0.61兲 from the low- of larger air split for the catalytic pilot 共can be observed from
est swirler equivalence ( ␾ swirler⫽0.61) ratio series shown in Fig. flow-splits shown in Table 2兲 has permitted operating at lower
8. CO was lower than 10 ppm at all other data points shown in front end operating temperature and thereby decreased NOx .
Fig. 8. Figure 9 also shows that at a front end temperature of 1960 K
We observe from this figure that for higher swirler equivalence 共1687°C/3070°F兲, measured NOx values of 1.9 ppm and 5.2 ppm
ratio less piloting is necessary and blowoff occurred at lower pilot were observed. This discrepancy in NOx data is due to the use of
equivalence ratio. This is expected and the increased stability is very different pilot and swirler equivalence ratios for the two
achieved at the expense of higher NOx emission. Below ␾ of 0.65, cases. If these data are compared to the data of Fig. 8, it can be
NOx dependence on the pilot equivalence ratio was linear.
Figure 8 also shows that the best NOx emission (NOx
⬍5 ppm) can be obtained for this configuration of modified Tau-
rus 70 injector and ABC liner when the swirler equivalence ratio
is between 0.5 to 0.65 and the pilot equivalence ratio is between
0.55 to 0.65. Thus the modified injector showed a wide operability
range. If the fuel to the catalytic pilot was shut-off while operating
at these low swirler equivalence ratios 共0.5–0.65兲, blowoff would
occur. This indicated that the catalytic pilot provided stability to
the main flame.
Figure 9 shows the same emission data 共CO and NOx) of Fig. 8
when plotted as a function of front end temperature of the com-
bustor. The front-end temperature was calculated based on the air
flows to the pilot, and injector, and fuel flows to the swirler and
pilot. Note that the calculated front end temperature does not take
into account any reverse flow dilution air 共liner cooling兲 or dome
cooling that may participate in the combustion process occurring
in the primary zone of the combustor. As such, the temperature
shown in the x-axis is not the primary zone temperature during
combustion. Comparison of these data with data from lean-
premixed combustion of Kendric 关4兴 and Snyder 关3兴 suggest that
the primary zone temperatures may be 200 K lower than the cal-
culated front-end temperature shown in the abscissa of Fig. 9.
The symbols represent measured emission data 共closed symbols
represent NOx data and open symbols represent CO data兲 after Fig. 10 NOx emission „100% load… as function of front end
correction and the solid line is an exponential trend line drawn temperature with percent pilot as parameter
load to 100% load condition. High-pressure testing for catalytic
pilot has demonstrated single digit (⬍5 ppm) NOx and CO emis-
sions along with low acoustics at 50% and 100% load conditions
for Solar Turbines Taurus 70 engine.
Acknowledgments
Precision Combustion Inc. gratefully acknowledges the support
of the U.S. Department of Energy SBIR program and Energy
Efficiency & Renewable Energy program. DOE SBIR and DOE
EREN/Office of Power Technologies funding supported develop-
ment of the catalytic reactor technology. In particular we want to
recognize Mr. Stephen Waslo and Ms. Patricia Hoffman from the
DOE OPT program for supporting the development of the cata-
lytic reactor technology, and Dr. Richard Johnson and Ms. Abbie
Layne from the National Energy Technology Laboratory, NETL,
for supporting the catalytic pilot application. PCI also gratefully
acknowledges the facilities, equipments, support, and technical
discussions provided by Solar Turbines.
References
关1兴 Vandervort, C. L., 2000, ‘‘9 PPM NOx /CO Combustion System for ‘‘F’’ Class
Industrial Gas Turbines,’’ ASME Paper No. 2000-GT-0086.
Fig. 11 Measured NOx and CO emission for the modified Tau- 关2兴 Davis, L. B., and Black, S. H., 2000, ‘‘Dry Low NOx Combustion Systems for
GE Heavy-Duty Gas Turbines,’’ General Electric Technical Report, GER-
rus 70 injector at 50% load condition
3568G, GE Power Systems, Schenectady, NY.
关3兴 Snyder, T. S., Rosfjord, T. J., Mcvey, J. B., Hu, A. S., and Schlein, B. C., 1996,
‘‘Emission and Performance of a Lean-Premixed Gas Fuel Injection System
observed that a NOx level of 1.9 ppm was observed for swirler for Aeroderivative Gas Turbine Engines,’’ Trans. ASME, 118, pp. 38 – 45.
关4兴 Kendrick, D. W., Bhargava, A., Colket, M. B., and Sowa, W. A., 2000, ‘‘NOx
and pilot equivalence ratios of 0.66 and 0.51, respectively, Scaling Characteristics for Industrial Gas Turbine Fuel Injectors,’’ ASME Pa-
whereas for the higher NOx case ␾ swirler and ␾ pilot were 0.61 and per No. 2000-GT-1998.
0.71, respectively. The use of a higher pilot equivalence ratio was 关5兴 Prade, B., Streb, H., Berenbrink, P., Schetter, B., and Pyka, G., 1996, ‘‘Devel-
responsible for the increase in NOx . Thus there is an optimum opment of an Improved Hybrid Burner—Initial Operating Experience in a Gas
percent of pilot fuel flow above which increasing the percentage Turbine,’’ Proceedings from the 41st ASME IGTA Congress and Exhibition,
Birmingham, UK.
has detrimental effect on NOx emission. This can be observed in 关6兴 Pfefferle, W., 1978, ‘‘The Catalytic Combustor: An Approach to Cleaner Com-
Fig. 10, which shows the emission data plotted as function of bustion,’’ J. Energy, 2, pp. 142–146.
front end temperature with percent pilot as a parameter. The data 关7兴 Dalla Betta, R. A., 1997, ‘‘Catalytic Combustion Gas Turbine Systems: The
shows that 18 –20% pilot to be the optimum percentage for lowest Preferred Technology for Low Emissions Electric Power Production and Co-
NOx emission. It can also be observed from Fig. 10 that for a generation,’’ Catalysis Today, 35, pp. 129–135.
lower percent pilot of 12%, blowoff occurred at a higher front end 关8兴 Kraemer, G., Strickland, T. R., Pfefferle, W. C., and Ritter, J., 1998, ‘‘A Com-
pact Catalytic Combustor System for Small Turbogenerators,’’ Proceeding of
flame temperature of approximately 3100°F. the Joint Power Generation Conference, EC-Vol. 5, ASME New York.
In addition to the testing at close to 100% load condition, emis- 关9兴 Fant, D. B., Jackson, G. S., Karim, H., Newburry, D. M., Dutta, P., Smith, K.
sion data were also obtained at 50% load condition. The 50% load O., and Dibble, R. W., 2000, ‘‘Status of Catalytic Combustion R&D for the
condition is shown in Table 1. The catalytic reactor of the pilot Department of Energy Advanced Turbine Systems Program,’’ ASME J. Eng.
was active even when the inlet temperature was dropped to the Gas Turbines Power, 122, pp. 293–300.
关10兴 Dalla Betta, R. A., and Rostrup-Nielsen, T., 1999, ‘‘Application of Catalytic
50% load inlet temperature of 636 K 共363°C/687°F兲. Combustion to a 1.5 MW industrial Gas Turbine’’ Catalysis Today, 47, pp.
The pressure of the system was also brought down to a pressure 369–375.
of 153 psia, which is the inlet pressure for Taurus 70 engine at 关11兴 Sadamori, H., Tanioka, T., and Matsuhisa, T., 1994, ‘‘Development of a High
50% load condition. Figure 11 shows data similar to Fig. 9 but for Temperature Combustion Catalyst System and Prototype Catalytic Combustor
Turbine Test Results,’’ Proc. of 1st Workshop on Catalytic Combustion, H.
50% load condition. In an effort to control emission over the Arai, ed., The Research Association for Catalytic Combustion of Catalysis
entire operating range, lean-premixed combustion systems are Society of Japan, Tokyo, pp. 18 –20.
frequently designed to maintain constant flame temperature as 关12兴 Ozawa Y., Tochihara, Y., Mori, N., Yuri, I., Kanazawa, T., and Sagimori, K.,
the engine changes power level. Thus for an operating point of 1998, ‘‘High Pressure Test Results of Catalytically Assisted Ceramic Combus-
tor for a Gas Turbine,’’ ASME Paper No. 98-GT-381.
1950 K, which is the same as that of the 100% load condition, it 关13兴 Furuya, T., Sasaki, K., Hanakata, Y., Mitsuya, K., Yamada, M., Tsuchiya, T.,
can be observed from Fig. 10, that emissions less than 5 ppm are and Furuse, Y., 1995, ‘‘Development of a Hybrid Catalytic Combustor for a
achievable. 1300 Class Gas Turbine,’’ Catal. Today, 26, pp. 345–3501.
关14兴 Rollbuhler, J. R., 1989, ‘‘Fuel-Rich Catalytic Reaction Experimental Results,’’
Combustion Noise. During the testing of the modified Taurus AIAA/SAE/ASME/ASEE, 27th Joint Propulsion Conference, Sacramento,
70 injector at Solar’s facility, pressure fluctuations in the combus- CA.
关15兴 Brabbs, T. A., and Olson, S. L., 1985, ‘‘A Soot Free Technique for In Situ
tor were recorded. The magnitude of the pressure fluctuations Hydrogen-Like Enrichment,’’ NASA Technical Paper #2498.
never exceeded 0.5 psi 共rms兲 during testing including 100% and 关16兴 Yee, D., Lundberg, K., and Weakley, C. K., 2000, ‘‘Field Demonstration of a
50% load conditions. 1.5 MW Industrial Gas Turbine With a Low Emission Catalytic Combustion
System,’’ ASME Paper No. 2000-GT-088.
关17兴 Hoshizaki, J. A., 1997, ‘‘Operating Experience of the First Solar Taurus 70S
Conclusions Mechanical Drive Installation,’’ ASME Paper No. 97-GT-354.
关18兴 Smith, K., and Fahme, A., 1999, ‘‘Back Side-Cooled Combustor Liner for
A prototype catalytic pilot was fitted into an existing Taurus 70 Lean-Premixed Combustion,’’ ASME Paper No. 99-GT-239.
injector without major modifications to the lean premixed portion 关19兴 Beebe, K. W., Cutrone, M. B., Matthews, R. N., Dalla Betta, R. A., Schlatter,
of the injector. Testing of a single Taurus 70 modified injector J. C., Furuse, Y., and Tsuchiya, T., 1995, ‘‘Design and Test of a Catalytic
共representative of the 12 injectors in an engine兲 has been com- Combustor for a Heavy Duty Industrial Gas Turbine,’’ ASME Paper No. 95-
GT-137.
pleted. The catalytic pilot operated robustly without any flashback 关20兴 Dalla Betta, R. A., Schlatter, J. C., Nickolas, S. G., Razdan, M. K., and Smith,
or autoignition even with wide variation of fuel/air ratio. No pre- D. A., 1995, ‘‘Application of Catalytic Combustion Technology to Industrial
burner was required for the operation of catalytic pilot from 50% Gas Turbines for Ultra-Low NOx Emissions,’’ ASME Paper No. 95-GT-65.
Combustion System Development
D. W. Kendrick1 for the Ramgen Engine
B. C. Chenevert
The research and development of a unique combustion engine is presented. The engine
B. Trueblood converts the thrust from ramjet modules located on the rim of a disk into shaft torque,
which in turn can be used for electrical power generation or mechanical drive applica-
J. Tonouchi tions. A test program was undertaken that included evaluation of the pre-prototype engine
and incorporation of improvements to the thrust modules and supporting systems. Fuel
S. P. Lawlor mixing studies with vortex generators and bluff-body flame holders demonstrated the
importance of increasing the shear-layer area and spreading angle to augment flame
R. Steele volume. Evaluation of flame-holding configurations (with variable fuel injection methods)
concluded that the heat release zone, and therefore combustion efficiency, could be ma-
Ramgen Power Systems, nipulated by judicious selection of bluff-body geometry, and is less influenced by fuel
Bellevue, WA 98005 injection distribution. Finally, successful operation of novel fuel and cooling air delivery
systems have resolved issues of gas (fuel and air) delivery to the individual rotor seg-
ments. The lessons learned from the pre-prototype engine are currently being applied to
the development of a 2.8MW engine. 关DOI: 10.1115/1.1586314兴
Introduction Theory of Operation: The Ramgen Engine

The paper presents the design and development of a fundamen- Ramjet Technology. The engine is based on ramjet technol-
tally new kind of power generation device known as the Ramgen ogy which produces propulsive force by increasing the momen-
engine. These efforts have been aided by the Department of En- tum of the working fluid via combustion and expansion through a
ergy, as part of an initiative to bring viable and innovative energy supersonic nozzle, 关1兴. In contrast to other air-breathing engines,
technologies to market. the working cycle of the ramjet engine is not accompanied by
The design of the pre-prototype engine represents the applica- mechanical compression of the working fluid but rather through a
tion of well-established ramjet principles to power generation. In- shock system located within the inlet portion of the engine. This
makes the ramjet engine the simplest mechanical air-breathing
stead of using a ramjet to provide thrust for a supersonic aircraft
engine to date through its minimal use of moving parts. Figure 1
or missile, the engine converts the linear thrust from a ramjet into illustrates the working cycle of the ramjet engine and demon-
shaft torque, which in turn drives a generator which provides elec- strates the four fundamental processes: ingestion, compression,
tricity or shaft power for a mechanical drive applications. At ma- heat release, and expansion. Flame stabilization is typically
turity, the production engine will have a thermal efficiency rival- achieved from a simple flame holder 共V-Gutter or equivalent兲 lo-
ing and surpassing conventional gas turbines, with highly cated at the end of the inlet passage or dump plane and sized for
competitive CO and NOx emissions through its operation in a lean the appropriate flow velocities. The air-fuel mixture entering the
premixed mode. The unique engine design also permits the use of combustion chamber is ignited by recirculating hot combustion
waste fuels, including landfill gas and coal bed methane as well as products comprising the recirculation zone. The flow within the
synthetic 共‘‘Syn Gas’’兲 and biomass fuels, for energy production. combustor is subsonic; hence, an initial contraction of the flow
area 共throat兲 is necessary to both reduce the internal pressure and
Such atypical gaseous fuels cannot be currently utilized in con-
accelerate the flow to sonic conditions, 关2兴. After attaining sonic
ventional gas turbine systems due to the risks of autoignition, velocities, the flow encounters a diverging passage to continue the
flashback, engine wear and inability to ingest atmospheric fuels acceleration process. The addition of the nozzle allows the flow to
directly. continue to expand to supersonic exit velocities, until ambient
The pre-prototype engine was fully assembled in Tacoma, WA pressure is attained—‘‘ideal expansion.’’
in July 1998. Tests occurring in 1999 validated ignition and flame
holding at idle rotor speeds, the mechanical integrity of the rotor The Ramgen Engine. The essential concept of the engine is
the incorporation of one or more ramjets onto the rim of a rotor
at supersonic tip speeds, and overall system integrity. Develop-
such that the thrust from the ramjets acts tangentially, causing it to
ment work conducted between late 1999 to September 2000 fo-
rotate at supersonic rim speeds. In this way, the thrust from the
cused on improving combustion stability, fuel delivery methods, ramjets is converted into shaft torque. Figure 2 is an isometric
and air film cooling systems. These efforts resulted in substantial illustration of the high-speed rotor and the stationary main engine
progress toward operation at full rotor speeds 共4300 rpm, Mach case from the pre-prototype engine. The figure illustrates how the
1.1 inflow conditions兲. two ramjet flow-paths are incorporated into the rim of the high-
speed rotor. As shown, the two-dimensional ramjet flowpaths are
1
open on their exterior, or radially outermost, surface. Closure of
Corresponding author, Technology Leader Aero-Thermal. E-mail:
[email protected]. the ramjet flowpath is facilitated by the stationary surface of the
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN main engine case. The combination of a ramjet flowpath where
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF one of the ramjet surfaces is moving relative to the other is analo-
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- gous to the moving projectile and stationary barrel wall of the
erlands, June 3– 6, 2002; Paper No. 2002-GT-30084. Manuscript received by IGTI, ram-accelerator, 关3– 6兴. The ramjet flow-paths are mounted on the
December 2001, final revision, March 2002. Associate Editor: E. Benvenuti. rim of the rotor as intertwined helixes at an optimized angle to
ables the engine to be self-pumping, drawing in the required flow
as determined by the effective flow area of the inlet system.
Figure 3 is a cut-away view depicting the flow path including
the supersonic inlet, diffuser, combustor and exit nozzle for the
pre-prototype engine. As shown in the figure, supersonic inflow
共Station 0兲 meets the compression surface of the inlet. Through a
series of Mach waves and strongly reflected shock, the required
inlet compression is achieved. The gas is then diffused through a
Fig. 1 Ramjet engine working cycle two dimensional diffuser whereby the static pressure rises to attain
required combustor inflow conditions. A simple rearward-facing
step flameholder is employed as the baseline flameholder to an-
chor the combustion activity. Standard diffusion piloting schemes
共to be discussed兲 serve to anchor the flame and ensure acceptable
combustion efficiency and acoustic levels. The hot combustion
gases are then accelerated to supersonic velocities in the nozzle.
Experimental Results: The Pre-Prototype Engine
Experimental Setup
Engine Layout. Figure 2 illustrates the layout of the pre-
prototype engine. The primary fuel employed was industrial grade
methane 共NG兲, supplied at a line pressure of two atmospheres and
ambient temperature. The fuel system was manifolded and accu-
rately measured by Worcester Model 20-755X I/P valves and
Sponsler Model SP1-1/2-CB-PH7-8-4RFX turbine meters. This
base fuel was supplemented with pure hydrogen to assess alter-
ations in the ignition delay times and flame speeds on combustor
performance. Hydrogen was supplied by a 1200 m3 tube trailer,
with an initial supply pressure of 136 atmospheres. Consistent
Fig. 2 The pre-prototype ramgen engine with the methane supply, the hydrogen was regulated and con-
trolled via I/P control valves and turbine meters. Downstream of
the metering systems, the hydrogen and methane lines were joined
ensure sufficient inlet length, combustor velocities, residence and routed to the main engine fuel line. To eliminate any hydro-
times and optimized thrust vectors. Combustion products from gen backflow into the lower pressure methane line, a check valve
one ramjet are not ingested into its neighbor due to a mechanical was installed in the methane line.
separation known as a strake which wraps around the rotor’s pe- Three fuel systems were installed in the engine. The first or
riphery 共Fig. 2兲. Note that the helix/strake angle dictates the inlet, primary fuel system was via radial fuel struts located inside the
diffuser, combustor, and nozzle lengths, which in turn dictates the inlet air annulus 共see Fig. 2兲. Due to their upstream location, fuel
combustor velocity, residence time, and thrust vectors. delivery struts induced a high level of mixedness by the time the
The shallow strake angle has a number of implications. First, mixture reached the rotor. The second or tertiary fuel system was
despite the supersonic rim speeds, the axial velocity of the premix via a series of porous metal plates circumferentially arranged
ingested is very low. As a result, the inflow can be delivered to the around the inlet, downstream of the primary fuel struts. This also
ramjets with minimal pressure and viscous losses, facilitating a achieved a high degree of mixedness due to its upstream location.
simple premixer system. As a consequence, the engine is capable Both of the fuel systems supply the bulk of the gaseous fuel to the
of burning a wide variety of low pressure and subquality fuels, system and required only slightly above atmospheric pressures to
including coal-bed methane, low-BTU natural gas, biomass fuels, operate effectively. The third or pilot fuel system supplied pure
and hydrogen. Secondly, the ‘‘auguring’’ nature of the rotor en- fuel to the flame stabilization device 共to be discussed兲 and paral-
Fig. 3 Ramgen engine flow details
leled similar diffusion piloting schemes, 关7,8兴. Injection of the Table 1 Engine conditions „per ramjet…
pilot fuel was either through film-type slots located 1 cm upstream
Pilot H2 AFC
of the dump plane 共interfacial plane dividing the end of the dif-
Mair Mfuel 共%, by Fuel 共%, by 共%, mass
fuser with the front of the combustor兲 and/or through the bluff- 共kg/s兲共kg/s兲 volume兲 - volume兲 total flow兲
body flame stabilizer itself. For all of the reported tests, only the
diffusion piloting system was employed. 4 ⭐.095 0–100% NG/H2 0–70% 10%
As with typical combustion systems, the engine was outfitted
with a variety of instrumentation and diagnostic features to maxi-
mize output data from each combustion run. Combustor dynamic
pressures and floor temperatures were measured using PCB Model design point speeds was accomplished with a 447 kW AC starter
112A04 piezoelectric pressure transducers and a Land Infrared IR motor connected to a regenerative variable frequency drive unit
Thermometer Model X0144IRT, respectively. The IR detector 共VFD兲. The VFD also compensates for ramjet thrust generation.
viewed the combustor through a 1 cm diameter quartz window The starting motor is required to ‘‘start,’’ 关5兴, the inlet shock sys-
and had a spectral range from 700–2000 nm to accurately mea- tem to initiate positive thrust generation. Combustion was initi-
sure the liner floor temperature without significant contamination ated at low rotor speeds via a pulse discharged igniter timed ap-
from such combustion byproducts as CO2 and H2 O, 关9兴. Two propriately to the location of the dump plane via magnetic
additional IR detectors were mounted on the engine inlet and ex- pickups.
haust, and aligned with the axis of engine rotation, to monitor the Flame Stabilizers. Three flame stabilizers were employed in
combustor and nozzle strake wall temperatures. A 3 cm diameter this investigation 共Figs. 4 and 5兲. The first is a 2.5 cm high,
quartz port was also added to the exhaust duct in order to view the rearward-facing step flameholder built into the engine’s floor. The
broadband chemiluminescence via color video cameras. second builds from the first in that it employs vortex generators
Instrumentation interfaced with the System Control and Data 共‘‘VG’’s兲 1 cm upstream of the step. This configuration was cho-
Acquisition System 共SCADA兲, and included emergency shutoff sen to enhance mixing of the pilot fuel and core airflow and en-
control logic where appropriate. All low-speed data 共gas flow courage a radial growth in combustion activity. Four sizes were
rates, static pressures and temperatures兲 were monitored via RS- investigated: 1, 1.3, 1.91, and 3.81 cm high. The final flame holder
View 共Allen-Bradley兲, while control logic was managed by concept, ‘‘pylon,’’ was again an adaptation of the rearward facing
RSLogics software 共Allen-Bradley兲. All high-speed data 共dynamic step and was chosen to increase flame surface area through cre-
combustor pressures and liner temperatures兲 were similarly moni- ation of multiple shear layers issuing from the dump plane.
tored in real time via a PC running a National Instruments Wave-
View7 system sampling up to 100 kHz per channel. Post process- Experimental Results. A timed spark from the pulsed dis-
ing was facilitated via in-house Excel macros and documented in charge ignitors resulted in the simultaneous ignition of the com-
Excel workbooks. bustor once pilot fuel had been initiated and engine speed had
Table 1 details the range of flow conditions for a standard run. been stabilized at 800 rpm. The fuel/oxidizer mixture ratio, igni-
The ‘‘AFC’’ depicts the level of air film cooling 共liner cooling兲 tion timing and AFC flow rates were optimized for greater than
required to maintain acceptable combustor floor temperatures and 95% ignition effectiveness irrespective of the flame stabilizer em-
hence mechanical system integrity. Engine starting and ramp up to ployed. Ignition was verified by monitoring the high speed data
Fig. 4 Vortex generator flameholder
Fig. 5 Pylon and rearward-facing step
acquisition system and ensuring that both combustors revealed Evident is that the first half of the combustion chamber encom-
typical floor temperature traces as shown in Fig. 6 below. This passes the recirculation zone and furthermore, that the location of
particular test was conducted with the 1.9 cm tall VGs in place, the shear layer reattachment point is located approximately 20 cm
located between the pilot fuel slot and the dump plane. Both com- downstream of the step 共note the location of the maximum floor
bustors were lit as noted by the successive temperature pulses temperature兲. A relatively stationary thermal trace develops in the
seen in the plot 共top curve兲. It should be noted that due to the forward half of the combustor, fluctuating within 27 K from trace
single location of the IR detector at the 12 o’clock position on the to trace. The latter half of the combustor, however, appears domi-
engine case, adjacent pulses denote opposite combustors. Due to nated by more incomplete combustion zones as evident by sub-
the sampling rate of the detector, the trace gives an accurate mea- stantially greater temperature fluctuations 共⫾120 K兲 between suc-
sure of the floor temperature as the rotor moves relative to the cessive cycles.
fixed position of the Land IR. In fact, taking into consideration the After stabilization at 800 rpm for two minutes, an automated
rotor’s speed and sampling rate, the rotor travels approximately 3 fuel, AFC and rotor ramp schedule was initiated via the Allen-
mm between successive samples at the design speed 共4300 rpm兲. Bradley control logic 共RSLogics兲 to attain the desired rim Mach
Also noted in the figure is the location of the beginning and end of number of approximately 1.1 共4300 rpm兲. This inlet Mach number
the 33 cm long combustor as determined by magnetic pickups in permits the exit nozzle to choke, and maximizes available thrust.
the stationary engine case—see the square pulse 共dump plane兲 and In practice, the ramp is broken down into three separate ramps,
the triangle pulse 共throat—middle curve兲. Also shown is the com- which are individually optimized for adequate pilot fuel, AFC and
bustor pressure versus rotor position 共bottom curve兲. rotor acceleration: 800–2000 rpm, 2000–3000 rpm, and 3000–
4300 rpm. The main issue of using a single ramp for all the quan-
tities was evident from the video and temperature traces during
the ramping process. The fluid dynamics and combustion clearly
alter as inlet/rotor speeds increase—most likely the combined ef-
fects of the nonuniform boundary conditions, increasing flow ve-
locities, centrifugal effects and stratification with increasing cen-
trifugal loadings. The AFC flow rates were found to be critical in
ensuring a sufficiently robust flame during the ramping process. If
not incremented correctly, it would eventually quench the entire
combustion process.
Above 2000 rpm, flame stabilization was generally restricted to
the recirculation zone. The shear layer combustion, as indicated
by the IR traces whose peaks move progressively upstream, was
reduced as the overall combustor residence time dropped. Simple
reactor modeling of the combustion zone 共to be discussed兲 re-
vealed that above 2000 rpm, combustion is only possible in the
recirculation zone where local residence times are conducive to
methane oxidation. It appears, therefore, that there is a critical
rotor speed above which the shear layer combustion is replaced by
Fig. 6 Single rotor rotation showing combustor floor tempera- combustion within the recirculation zone.
ture and pressure traces The VFD is shown to hold the rotor at 4300 rpm after some
Fig. 7 NG and H2 addition test plan with 1.3 cm VGs
perturbations at the top of the ramp 共Fig. 7兲. This transient is due with stoichiometric conditions. The figure was then generated as-
to imbalance in required motor power transitioning from ramping suming the entrainment rate was preserved throughout the exer-
to steady-state operation. The power output required to maintain cise.
the engine’s speed is a critical factor in the overall performance of Also shown in the figure are three curves estimating the floor
the two thrust modules 共combustor/nozzle systems兲. A typical test temperature based on a simple, one dimensional model which
matrix 共varying fuel loading between the three fuel circuits, H2 includes the heat transfer from the gas to the floor, 关10兴, and
addition, etc.兲 is effectuated after the rotor speed and drive stabi- equilibrium calculations 共STANJAN兲 of flame temperatures for
lizes. each fuel mixture. As demonstrated by the model, it is evident that
the local equivalence ratio is a significant factor in determining
Typical Test Sequence. Depicted in Table 2 is a sample test
the gas temperature.
run whereby the methane flow rate was incrementally ramped to a
Details of the above combustor are further explored through
prescribed point, whereafter the hydrogen level was then incre-
chemical reactor modeling of the events depicted in Fig. 7. It
mented. Also included on the table is the VFD power required to
should be noted that at each of the denoted points, high-speed data
stabilize the engine rpm at 4300 rpm. Decreases in the required
共IR and combustor pressure traces兲 were taken. A five-element
VFD level is a direct measure of the effectiveness of the thrust
reactor network was used and is shown in Fig. 9. Two reactor
modules and hence system efficiency. Points labeled 1 through 5
blocks 共#1 and 2兲 were employed to model the shear layer which
on the table refer to low-speed data points 共see Fig. 7兲 at pre-
scribed operating conditions. Shown in Fig. 7 is a marked drop in communicated with the recirculation zone 共#5兲. The shear layer
the combustor floor temperature between points 1 and 3. If it is half-angle was assumed to be 5°, 关11兴, and extended five duct
assumed that the floor temperature is linked to the bulk average heights downstream, 关12兴. The shear layer entrainment rate 共as
flame temperature, it would seem reasonable that the combustion discussed above兲 was maintained at 3% and all the fuel was as-
zone is leaning out with this increase in NG flow rate and hence sumed to enter element 1 to simulate the thin layer of fuel issuing
has passed through the stoichiometric fuel-air ratio. The argument from the fuel slot. The recirculation zone element received fuel
can be extended through all points at a constant rotor speed of through entrainment from neighboring elements 1 and 2. The
4300 rpm. Figure 8 depicts the indicated combustor floor tempera- GRI-MECH 2.12 was used to model the chemical kinetics mecha-
ture versus local equivalence ratio based on an estimate of the nism for NG/H2 oxidation. It will be shown that the model accu-
entrained airflow rate into the recirculation zone per the rationale rately predicted the observed trends of Fig. 7.
above. The baseline air entrainment rate was determined by as- Figure 10 shows the high speed data trace at point 1 for one
suming that the maximum combustor temperature was coincident rotation of the rotor. The top curve depicts the temperature while
the bottom is the combustor pressure. Apparent is the forward
location of the peak heat release within the combustor. A rapid
drop ensues in the IR trace as the gas is unable to sustain com-
Table 2 1.3 cm tall VGs test plan bustion outside the recirculation zone.
Figure 11 depicts the modeling results corresponding to point 1.
NG H2 VFD AFC T
Pt. 共Kg/s兲共Kg/s兲共KW兲共kg/s兲共K兲 It reveals the predicted reactor temperatures based on the assumed
flow rates and residence times determined experimentally. The
0.041 0.0034 361 0.41 991 highest temperature is seen to occupy reactor element 2 共2908°F兲
0.054 0.0034 345 0.41 1020
1 0.061 0.0034 342 0.41 1020 which corresponds to the shear layer near the middle and end of
0.071 0.0034 335 0.41 1014 the recirculation zone, consistent with the high-speed trace 共Fig.
0.078 0.0034 335 0.41 1009 10兲. The recirculation zone 共element 5兲 predicts a lower tempera-
2, 3 0.084 0.0034 335 0.41 969 ture 共1303°F兲, most likely attributed to the reduced fuel-air ratio
4 0.081 0.0066 320 0.41 964
0.071 0.0061 310 0.41 1011 and increased wall heat transfer at this location. Element 1 ap-
5 0.078 0.0061 310 0.41 989 pears to be insensitive to local equivalence ratio and behaves more
as a mixing element since its residence time is below the required
Fig. 8 Combustor floor temperatures versus equivalence ratio „data and
model predictions…
Fig. 9 Chemical reactor modeling network
Fig. 10 High-speed data trace for point 1
blowoff residence time for its equivalence ratio (⌽⫽1.65). The layer, which ignites the incoming cold charge. The success of this
temperature in the combustor is shown to drop off after the cen- flameholder, therefore, depends on the spreading rate of the shear
trally located peak, most likely attributed to entrainment of the layer, which must extend to the opposite wall to ensure high com-
large, unreacted core flow and AFC flows with the products of bustion efficiency. The inflow shear spreading half-angle is ap-
combustion. proximately 5°, 关11兴. This angle 共␾兲 can be augmented though
A similar analysis is made at point 2. The high speed data trace combustion, 关14兴, by the following relation, where S T is the tur-
and reactor network modeling results are shown in Figs. 12 and bulent flame speed and U is the approach velocity:
冉冊
13, respectively. As evidenced in both figures, there is a down-
stream migration in the peak heat release location. The model ST
results show that the fuel addition has quenched most combustion
␾ ⫽sin⫺1 . (1)
U
activity in the recirculation zone and shear layer and shifted it
downstream. Residence times in the upstream elements 共1, 2, and Recognizing that the spreading angle can be augmented with a
5兲 are below the required blowout times required to sustain com- fuel with a higher turbulent flame speed, hydrogen was blended
bustion. The high-speed data traces confirm these modeling results with the baseline methane fuel. Note that hydrogen’s turbulent
as evidenced by the combustion activity near the nozzle. Contrary flame speed is up to four times methane’s flame speed, 关14兴. Tests
to the model results is the reduced participation of the recircula- with the baseline rearward-facing step revealed poor mixing be-
tion zone in the high-speed data. This is most likely due to the tween the pilot fuel and mainstream airflow. As such, most of the
model’s overprediction of fuel entering the recirculation zone. pilot fuel directly entered the recirculation zone, resulting in over-
At the termination of the experiment, the maximum combustor rich conditions. These disadvantageous situations resulted in sig-
floor temperature is 1413°F. Combustion modeling results for this nificant quenching of the combustion directly behind the step, and
condition appear in Fig. 14 below. It appears that the recirculation often shifted the combustion zone towards the center of the com-
zone residence times are adequate to sustain combustion with the bustion chamber. Efforts were focused on enhancing the mixing of
hydrogen addition 共compare Figs. 13 and 14兲 and such predictions the pilot fuel stream with the main core flow with the implemen-
are aligned with the temperature rise at the end of the test 共Fig. 7兲. tation of the VGs.
Vortex generators of various heights were examined with the
Comparisons in Flame Stabilization Configurations. As noted intention of creating sufficient streamwise vorticity to enhance
previously, three flame stabilization devices were tested: a base- mixing of the pilot and core air flows to radially extend the com-
line, rearward-facing step, a rearward-facing step and VGs, and a bustion zone. Throughout the series of tests, it was found that
rearward-facing step and pylon. The mechanism for fuel-air mix- there was a delicate balance between increased mixing 共taller
ing and flame stabilization with the baseline rearward facing step VGs兲 and reduced combustion stability. The larger the VGs, the
flameholder is conventional shear layer mixing and creation of a less stable the combustion zone, most likely due to the increased
recirculation zone, 关13兴. The combustion is sustained by the turbulence and quenching activity promoted by the larger struts,
steady entrainment of the combustion byproducts into the shear 关15兴. A sample high speed trace is shown in Fig. 15 for the opti-
mal VG configuration 共1 cm long兲.
Figure 15 shows that there is a greater streamwise distribution
in temperature relative to the baseline step and secondly, the com-
bustor pressure is more uniform. This is important since it is this
pressure which forces the exhaust nozzle to choke, maximizing
system performance.
The VGs provided some improvement with regards to heat re-
lease distribution and fuel-air mixing but failed to demonstrate a
dramatic decrease in VFD power over the baseline configuration
with fuel loading variations. The next series of tests investigated
the Pylon design with various configurations in fuel injection 共Fig.
Fig. 11 Gas temperature modeling results for point 1 5兲. This concept provides the greatest flexibility in fuel placement,
Fig. 12 High-speed data trace for point 2
Fig. 13 Gas temperature modeling results for point 2
affords the combustor with the largest possible combustion vol-

ume to maximize heat release and provides the largest flame sur-
face area. Such attributes favor an increase in overall combustion
efficiency.
Preliminary CFD simulations using Star-CD showed encourag-
ing results 共Fig. 16兲. The pressure based solver was employed
using the standard k-␧ turbulence model and included the effects
of rotation. The figure depicts total temperature contours through
various axial slices of the combustor. Apparent is the radial
growth in combustion activity near the center line due to the pres-
ence of the pylon. Unfortunately, due to the insufficient blockage
afforded by the pylon and the high momentum of the approach Fig. 16 Temperature contours from a CFD simulation of the
flow, it was ineffective in extending combustion in the axial di- pylon flame stabilizer
rection. The design did, however, provide a more homogenous
temperature distribution relative to the proceeding designs, irre-
spective of the fuel injection configuration 共Fig. 17兲. This test was floor. Due to this more uniform and elevated temperature distri-
conducted with pilot fuel issuing from a single, 32 mm diameter bution, combustion intensity and pressures were increased.
hole from the pylon’s faceplate and from the fuel slot along the An overall comparison between VG and pylon test data is made
by comparing corresponding fuel loadings 共BTUs of fuel兲 to VFD
power for a rotor speed of 4300 rpm. Lower VFD levels for the
same fuel 共BTU兲 loading would therefore signify higher system
effectiveness. A summary plot appears in Fig. 18. All VG tests are
denoted by square symbols while their pylon counterparts are de-
noted by diamonds. Apparent is the more compact grouping of the
pylon data. The dotted box, which captures all such runs, further
illustrates the fact that increasing fuel loading 共expressed in
BTU/s兲 lowers required VFD power. The greater slope and com-
pactness of the pylon data illustrates that it is more effective in
flame holding, providing increased flame surface area and hence
higher burner pressures through increased combustion activity.
Fig. 14 Gas temperature modeling results for gas off The net effect is higher ramjet thrust and overall cycle efficiency.
Fig. 15 High-speed trace for the 1 cm tall VGs
Fig. 17 High-speed trace for the pylon flame stabilizer with 32 mm fuel hole
Fig. 18 VFD comparisons between all VG and pylon runs
Conclusions References
The preliminary results from a pre-prototype engine are pre- 关1兴 Heiser, W. H., and Pratt, D. T., 1994, Hypersonic Airbreathing Propulsion
共AIAA Ed. Series兲, AIAA, Washington, DC.
sented. Initial combustion tests reveal the importance of flame 关2兴 Oates, G. C., 1997, Aerothermodynamics of Gas Turbine and Rocket Propul-
volume and surface area for increased combustion efficiency, ir- sion 共AIAA Ed. Series兲 AIAA, Washington, DC.
respective of fuel injection schemes. Chemical kinetic modeling 关3兴 Bogdanoff, D. W., 1992, ‘‘Ram Accelerator Direct Space Launch System: New
was shown to accurately depict the essential flow features of a Concepts,’’ J. Propul. Power, 8共2兲, pp. 481– 490.
sample test matrix whereby stepwise increases in NG and hydro- 关4兴 Harsha, P. T., and Edelman, R. B., 1981, ‘‘Assessment of a Modular Ramjet
Combustion Model,’’ J. Spacecraft, 19共3兲, pp. 430– 436.
gen flow rates were separately investigated. The results confirm
关5兴 Hertzberg, A., Bruckner, A. P., and Bogdanoff, D. W., 1988, ‘‘Ram Accelera-
the visual observations that dramatic flow field variations occur tor: A New Chemical Method for Accelerating Projectiles to Ultrahigh Veloci-
both during ramp up and pilot fuel excursions due to the nature of ties,’’ AIAA J., 26共2兲, pp. 195–203.
the engine. 关6兴 Yungster, S., Eberhardt, S., and Bruckner, A. P., 1991, ‘‘Numerical Simulation
of Hypervelocity Projectiles in Detonable Gases,’’ AIAA J., 29共2兲, pp. 187–
193.
关7兴 Kendrick, D. W., Anderson, T. J., Sowa, W. A., and Snyder, T. S., 1999,
Acknowledgments ‘‘Acoustic Sensitivities of Lean-Premixed Fuel Injectors in a Single Nozzle
This effort was funded in part by the Department of Energy Rig,’’ ASME J. Eng. Gas Turbines Power, 121, pp. 429– 436.
under the cooperative agreement #DE-FC26-00NT40, awarded 关8兴 Snyder, T. S., Rosfjord, T. J., McVey, J., Hu, A., and Schlein, B., 1994, ‘‘Emis-
sions and Performance of a Lean-Premixed GasFuel Injection System for
between the period of Sept. 29, 2000 and Aug. 31, 2001. The Aeroderivative Gas Turbine Engines,’’ ASME Paper No. 94-GT-234.
computational support 共Star CD兲 was provided by Combustion 关9兴 Eckbreth, A. C., 1988, Laser Diagnostics for Combustion Temperature and
Science and Engineering 共Columbia, MD兲. Species, Turnbridge Wells, Kent, UK.
关10兴 Kreith, F., and Bohn, M. S., 1986, The Principals of Heat Transfer, Harper and 关13兴 Zukoski, E. E., 1978, ‘‘Afterburners,’’ The Aerothermodynamics of Aircraft
Row, Cambridge, UK. Gas Turbine Engines, Gordon C. Oates, ed., Air Force Propulsion Laboratory,
关11兴 Schlichting, H., 1979, Boundary-Layer Theory, McGraw-Hill, New York. AFAPL-TR-78-52, Chap. 21.
关12兴 Ozawa, R. I., 1981, ‘‘Survey of Basic Data on Flame Stabilization and 关14兴 Glassman, I., 1987, Combustion, Academic Press, San Diego, CA.
Propagation for High Speed Combustion Systems,’’ Paper No. AFAPL-TR- 关15兴 Lefebvre, A. H., 1983, Gas Turbine Combustion, Hemisphere, Washington,
70-81. DC.
S. M. Martin
17136 Bak Road,
The Premixed Conditional
Belleville, MI 48111-3525
e-mail: [email protected]
Moment Closure Method Applied
J. C. Kramlich
to Idealized Lean Premixed Gas
Turbine Combustors
G. Kosály
e-mail: [email protected] This paper presents the premixed conditional moment closure (CMC) method as a new
tool for modeling turbulent premixed combustion with detailed chemistry. By using con-
J. J. Riley ditional averages the CMC method can more accurately model the affects of the turbulent
e-mail: [email protected] fluctuations of the temperature on the reaction rates. This provides an improved means of
solving a major problem with traditional turbulent reacting flow models, namely how to
Department of Mechanical Engineering, close the reaction rate source term. Combined with a commercial CFD code this model
University of Washington, provides insight into the emission formation pathways with reasonable runtimes. Results
Box 352600, using the full GRI2.11 methane kinetic mechanism are compared to experimental data for
Seattle, WA 98195-2600 a backward-facing step burning premixed methane. This model holds promise as a design
tool for lean premixed gas turbine combustors. 关DOI: 10.1115/1.1587740兴
Introduction turbine combustors fall into this regime. The premixed CMC
method had runtimes approximately twice as long as a three-step
The premixed conditional moment closure 共CMC兲 method is
global mechanism using the same CFD program, while providing
developed and coupled to a commercial CFD program to model
a better match of the data and details of the intermediate species.
an idealized lean premixed 共LMP兲 gas turbine 共GT兲 combustor, a
The goal of this research is to develop the model and validate it
two-dimensional backward-facing step burning premixed methane
against a simplified reactor that has many of the same features as
at an equivalence ratio of 0.9. Using the full GRI2.11 kinetic
a LPM GT combustor. Further research will enhance the model
mechanism, temperature, velocity, and major species are com- and validate it against LPM GT combustor data.
pared to data with good agreement. Intermediate species that con-
tribute to the three NOx formation pathways are calculated, but are
not presented here. The CMC method is a new theory that has Formulation
been applied to nonpremixed combustion with good success 共Kli- For a turbulent premixed combustion model to accurately pre-
menko and Bilger 关1兴 and Smith 关2兴兲. The application of the CMC dict the emissions from a LPM GT combustor it must account for
method to premixed systems has been proposed, but has not yet the affects of temperature fluctuations on the reaction rates, model
been done. The premixed CMC method replaces the species mass the affects of micromixing on the chemical kinetics and allow
fractions as independent variables with species mass fractions that detailed chemistry. Most premixed turbulent combustion models
are conditioned on a reaction progress variable 共RPV兲. Conserva- used to date have been based on chemical reactor modeling or the
tion equations for these new variables are then derived and solved. eddy breakup model, mainly because of their relatively short run-
The general idea behind the premixed CMC method is that the times. Chemical reactor models break the flow into a series of
behavior of the chemical species is closely coupled to the reaction perfectly stirred reactors and plug flow reactors, which allow full
progress variable 共Klimenko and Bilger 关1兴兲. Thus, species con- kinetic mechanisms, but are limited to the infinite mixing and no
servation equations that are conditioned on the RPV will have mixing regimes, respectively. The eddy breakup 共EBU兲 model and
terms involving the fluctuation quantities that are much more its variant, the eddy dissipation model assume the reaction rates
likely to be negligible, hence improving the accuracy of the are mixing limited and base the reaction rates on ad hoc mixing
chemical kinetic rates. In essence, knowledge of the statistics of models. While these models are popular for their runtimes, they
the RPV contains most of the information needed to calculate do not model the interaction between the small scale mixing and
species mass fractions. In return for this better conditioned prob- the chemical kinetics, so they are limited in their capability to
lem, an increase in dimensionality occurs and a new conservation predict the emissions of LPM GT combustors. Recently PDF
equation 共for the RPV兲 must be solved, along with a conservation methods have been used to model LPM GT combustors. These
equation for the variance of the RPV. The CMC method accounts models account for the temperature fluctuation in the reaction
for the interaction between scalar dissipation 共micromixing兲 and rates and allow larger kinetic models, but require a model for the
the chemical kinetics, while decoupling the kinetics from the bulk micromixing. So far a universally accepted micromixing model
flow 共macromixing兲. The RPV and its variance are used to gener- has not been developed and it is still open to debate whether the
ate an assumed shape beta function PDF that is used to integrate improved accuracy over earlier models justifies the longer run-
the conditioned mass fractions to obtain the species mass fractions times. The flamelet and g-equation models are only valid for
in physical space. In certain cases the conditioned mass fractions flames in the wrinkled and corrugated flamelet regimes and are
are uniform over the computational domain, reducing the dimen- not valid for the high intensity mixing of a LPM GT combustor.
sionality of the problem. It will be shown that lean premixed gas The CMC method was independently developed by Klimenko
关3兴 and Bilger 关4,5兴 from first principles as a model for turbulent
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN non-premixed flames. Rather than using conventional averages,
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF the reactive scalars are conditionally averaged on the mixture
fraction. Klimenko 关3兴 has emphasized that turbulent diffusion in
erlands, June 3– 6, 2002; Paper No. 2002-GT-30094. Manuscript received by IGTI, mixture fraction space can be modeled more rigorously than in
Dec. 2001, final revision Mar. 2002. Associate Editor: E. Benvenuti. physical space. Bilger 关4,5兴 observed that most fluctuations in
mass fraction were associated with fluctuations of the mixture aged statistics. The conditioning variable used here is the reaction
fraction. Rather than considering flame surface statistics and the progress variable 共RPV兲 based on the sensible enthalpy defined in
laminar reactive-diffusive structure attached to the flame surface Eq. 共10兲.
as in the flamelet model, the CMC model is based on conditional The premixed CMC equations using Smith’s 关2兴 derivation
moments at a fixed location and time within the flow field. 共which follows Klimenko 关3兴兲 starts from the conservation of spe-
A conditional average is an average taken with only those mem- cies mass fraction equation given below. Bilger’s 关4,5兴 derivation
bers of the whole ensemble that meet a specified condition 共Bilger uses different equations and assumptions then Klimenko, but de-
关4兴兲. For the non-premixed method the conditioning variable is velops the same equations.
normally the mixture fraction and the conditional average is taken
with the mixture fraction within an infinitesimal range of a set ⳵Y i
␳ ⫹ ␳ u•ⵜY i ⫽ⵜ• 共 ␳ D i ⵜY i 兲 ⫹ ␳␻ i (5)
value. For the premixed CMC method the conditioning variable is ⳵t
the reaction progress variable 共RPV兲, defined later. A conditioned
variable is not only a function of position in space 共and time in Here ␳ is the density, Y i is the species mass fraction, u is the
unsteady problems兲 but also of the mixture fraction 共or RPV兲. The velocity vector, D i is the species diffusivity, ␻ i is the production
dimension of the problem is increased and experimental data must rate 共1/s兲 of species Y i , and i⫽1,ns where ns is the number of
be recorded over the additional dimensions 共Bilger 关4兴兲. However, species.
experiments have shown that for some problems the conditioned The enthalpy conservation equation is given as follows:
variables are nearly constant in some coordinates, so the spatial
⳵h ⳵P
derivatives can be ignored. Crocco 关6兴 was the first to use a nor- ␳ ⫹ ␳ u•ⵜh⫽ⵜ• 共 ␳ D h ⵜh 兲 ⫺ ␳ W R ⫹ . (6)
mally dependent variable as an independent variable in boundary ⳵t ⳵t
layer theory. A simplification of the conservation equations for the
Here h is enthalpy, D h is the enthalpy diffusivity, ␳ W R is the heat
remaining dependent variables results when they are transformed
loss due to radiation 共it is positive for heat loss兲, and P is the
into the new coordinate system 共Bilger 关4兴兲.
pressure. Klimenko and Bilger 关1兴 and Bilger 关7兴 assume that Le
Before developing the premixed CMC equations it is useful to
⫽1, so all the diffusivities are equal. Below is the sensible en-
define a unconditional and conditional averaged variable. An un-
thalpy equation, which is derived in Martin 关8兴.
conditioned average is defined as follows 共also termed ensemble
average兲: ⳵hs ⳵P
1
N
␳
⳵t
⫹ ␳ u•ⵜh s ⫽ⵜ• 共 ␳ D h ⵜh s 兲 ⫺ ␳ W R ⫹
⳵t
⫺␳ 兺 ␻h
i
i f ,i
具 Y 共 x,t 兲典 ⫽
N 兺 Y 共 x,t 兲 .
i
i
(1) (7)
s
where h is the sensible enthalpy and h f ,i is the species heat of
The 具典 brackets denote an averaged value. The instantaneous val- formation, defined as follows:
ues are defined as follows:
ns
Y 共 x,t 兲 ⫽ 具 Y 共 x,t 兲典 ⫹y 共 x,t 兲 (2) h⬅h s ⫹ 兺h i

f ,i Y i (8)
where Y is the instantaneous value at a given location and time,
冕
ns
具 Y 典 is the unconditioned average, and y is the fluctuating 共devia- T
toric兲 term. The governing conservation equations are usually h s⫽ 兺 Y i C p i dT. (9)
based on the unconditional average values 共due to Reynolds or i T0
Favre averaging兲 and require closure models for the fluctuating The reaction progress variable 共RPV兲 is defined as
terms. Normally first-order closure models are used, which are
based on the average values. If the fluctuating terms are large h s ⫺h su
relative to the average values the model will be inaccurate. This is c 共 x,t 兲 ⬅ . (10)
⌬h ad⫺u
s
one problem with traditional CFD combustion modeling. A con-
ditioned average is defined as follows 共Bilger 关4兴兲: The u subscript signifies the unburned state. The denominator is
N the difference in sensible enthalpy between the adiabatic equilib-
1
具 Y 共 x,t 兲兩 x⫽x a 典 ⫽
N 兺 Y 共 x,t 兲 ␦ 共 x ⫺x 兲 .
i
i i
a (3)
rium and unburned states, which is fixed for a given mechanism
and initial conditions. C will have values between 0.0 and 1.0.
The conservation equation for c is derived in Martin 关8兴 and is
Only values of Y are averaged that meet a given condition, in this shown as follows:
冋兺册
case x⫽x a . A shorthand notation for the conditional average is
具 Y 兩 x a 典 . The vertical bar means that only those values that meet
冉冊
关 ¯␳ W R 兴 ⫺ ¯␳ ␻
˜ i h f ,i
the conditions to the right of the bar are used in the averaging. For ⳵␳
¯ c̃ ␮t i
every value of the conditioning variable, x a , there will be a dif- ⫹ⵜ• 关 ¯␳ ũc̃ 兴 ⫽ⵜ• ⵜc̃ ⫹
⳵t ␴c ⌬h ad⫺u
s
ferent average value 具 Y 兩 x a 典 , thus increasing the dimensionality of
the problem. The instantaneous value is similar to the uncondi- ⬅ⵜ• 共 ¯␳ D h ⵜc̃ 兲 ⫹¯␳ S̃ c . (11)
tional form.
Equation 共11兲 is essentially a nondimensional energy 共enthalpy兲
Y i 共 x,t 兲 ⫽ 具 Y 兩 x a 典 ⫹y ix 共 x,t 兲 (4) equation.
The premixed CMC equation is derived in Martin 关8兴 using the
With the correct choice of conditioning variable the fluctuating above relations to give
term in Eq. 共4兲 will be smaller then in Eq. 共2兲. In essence, by
increasing the dimensionality of the problem, the fluctuating terms ⳵Qi ⳵ 2Q i
are reduced, which should allow more accurate prediction of the 具␳兩␨典 ⫹ 具 ␳ u 兩 ␨ 典 •ⵜQ i ⫽ 具 ␳␻ i 兩 ␨ 典 ⫹ 具 ␳ D 共 ⵜc 兲 2 兩 ␨ 典
⳵t ⳵␨ 2
average species mass fractions. Additional conditioning con-
straints can be added to further reduce the fluctuating terms. ⳵Qi
The conditioning variable should be selected to describe the ⫺ 具 ␳ S c兩 ␨ 典 ⫹ 具 e y兩 ␨ 典 ⫹ 具 e Q兩 ␨ 典
⳵␨
instantaneous degree of reactedness, so scatter caused by varia-
tions in the reactedness is eliminated from the conditionally aver- (12a)
具 e y兩␨ 典 ⬅
1
P␨ 再冋
2 具 ␳ D 共 ⵜc•ⵜy i 兲兩 ␨ 典
⳵ P␨
⳵␨
⳵
册
⫺ 关具 ␳ D 共 ⵜc 兲 2 y i 兩 ␨ 典 P ␨ 兴
⳵␨
to model. This leaves just three terms in Eq. 共14兲, which are the
production, diffusion and convection in c space, respectively.
Equation 共14兲 is a set of ns⫹1 (ns is the number of species兲
冎
⫺ⵜ• 共具 ␳ uy i 兩 ␨ 典 P ␨ 兲 ⫺ 共具 ␳ S c y i 兩 ␨ 典兲 (12b)
second order ordinary differential equations in c space. Here only
adiabatic problems were investigated, so the conditional enthalpy
equation is not needed. For a given chemical kinetic mechanism
the only inputs are the initial conditions and the scalar dissipation
具 e Q 兩 ␨ 典 ⬅ 关具 ␳ 共 ⵜc•ⵜQ i 兲兩 ␨ 典兴 ⫹ⵜ• 共具 ␳ D 兩 ␨ 典 ⵜ 共 Q i P ␨ 兲兲 (12c)
共micromixing frequency兲 given as follows:
where e Y and e Q are error terms that require models to close. The
terms in Eq. (12a) are the time rate of change of Q, the convec- 兵兵具 N 兩 ␨ 典其其 ⬅ 兵兵具 ␳ D 共 ⵜc 兲 2 兩 ␨ 典其其 ⬇ 兵兵 ␳ D 共 ⵜc 兲 2 其其 (15)
tion in physical space, the production due to reactions, turbulent N is calculated from the flow field and used in the table lookup to
diffusion in c space and convection in c space. This equation obtain the conditioned species and temperature. The sensitivity of
describes the conditional averages of the species as a function of the scalar dissipation on the solution will be explored in the next
space and c. It would nominally be solved as a coupled system section. The conditioned reaction rate is approximated as
with equations for momentum, turbulence model and c 共mean and
variance兲. Q is defined as the value of the species mass fraction or 具 ␳␻ 兩 ␩ 典 ⬇ ␻ 共 Q 1 ,Q 2 .......Q ns ,Q h 兲 . (16)
enthalpy conditioned on the RPV 共␨兲 being equal to a given value
␩, Bilger 关4兴. The fluctuating components of the species and enthalpy are much
smaller in conditioned space, so this will give a much better ap-
Q⬅ 具 Y 共 x,t 兲兩 ␨ ⫽ ␩ 典 (13) proximation of the reaction rate than in normal unconditioned
methods. This set of equations was solved over a range of N and
One of the main differences between the premixed and non- used as a table lookup for the results discussed in the next section.
premixed CMC methods is that the premixed version is based Equation 共16兲 is also used to calculate S c in Eq. 共14兲.
on a nonconserved scalar, which adds an additional term to the To convert the conditioned data to unconditioned data an as-
equations 共the third term on the RHS of Eq. 12(a)). This extra sumed shape beta function PDF is used based on the RPV from
term makes the equations more difficult to solve as explained Eq. 共11兲 and its variance calculated from the following equation.
冋册
below. The only known attempt to solve the premixed CMC equa-
tions was by Smith 关2兴, who was not able to obtain a converged ¯ c̃ ⬙ 2
⳵␳ ␮t
solution. ⫹ⵜ• 关 ¯␳ ũc̃ ⬙ 2 兴 ⫺ⵜ• ⵜc̃ ⬙ 2
⳵t ␴c
冋兺册
Non-premixed jet flame data have suggested that the spatial
derivatives in Eq. (12a) can often be neglected 共Klimenko and
Bilger 关1兴, Smith 关2兴, and Bilger 关9兴兲. This is particularly true for ¯␳ c̃ ⬙ ␻
˜ ⬙ h f ,i
˜␧ i
the radial dependence in jet flames, but it is often also true for the ⫽C c1 ␮ t 共 ⵜc̃ 兲 2 ⫺C c2¯␳ ˜c ⬙ 2 ⫺2 (17)
axial dependence in jet flames with fast chemistry 共e.g., hydrogen ˜␬ ⌬h ad⫺u
s
flames兲. Note that this does not imply that unconditional mean
The last term is modeled with the following relationship, see Mar-
species mass fractions are independent of axial or radial position.
tin 关8兴 for the derivation.
Instead, it implies that the conditional mass fractions are indepen-
再冎
dent of spatial location, while the PFD depends on spatial location NS NS
˜ i⬙ h f ,i
˜c ⬙ ␻ c̃ ⬙ 2 ␻
˜ i h f ,i
through the RPV and its variance. If the spatial dependence can
indeed be neglected, a substantial reduction in computational 兺 i
s
h ad⫺u
⫽100
T̄ 共 c̃⫹0.167兲
兺 i
s
h ad⫺u
(18)
complexity can be obtained. Specifically, the behavior of the con-
ditioned variables depend only on the conditioning variable 共mix-
ture fraction or c) and its scalar dissipation, which will vary with
Results and Discussions
position. Thus, the calculation of the conditioned variables can be The test problem used to validate the model was an experiment
decoupled from the flow field calculation, since the only interlink- by El Banhawy et al. 关10兴 that consists of a flame stabilized by a
age will be through c and its scalar dissipation. This opens the backward-facing step. Premixed natural gas 共94% CH4兲 and air
possibility of doing a detailed chemistry calculation off line and are flowed across the backward-facing step, which establishes a
using a lookup table to feed information back to the fluid dynam- recirculation zone that stabilizes the flame. A flame sheet is an-
ics calculation. chored to the lip of the step, from which it progresses across the
For premixed problems with high mixing intensities 共i.e., in the free portion of the duct. The flame sheet touches the wall at a
distributed reaction regime兲 the conditioned species may be suffi- specific axial location, and matching this ‘‘touching distance’’ is a
ciently uniform within the reactor to allow the equations to be challenging test of the various modeling approaches. The Rey-
simplified by neglecting spatial derivatives 共Smith 关2兴兲. This is nolds number based on step height was approximately 13,500 and
along the lines of a partially stirred reactor, which has strong back the airflow rate was 125 kg/hr giving an inlet velocity of 10.5
mixing giving nearly uniform properties throughout the reactor. m/sec at 300 K. Velocity measurements of nonreacting 共not shown
This is a generalization of a perfectly stirred reactor, and is similar in their paper兲 and reacting flows were measured with a laser-
to a LPM GT combustor. Here this hypothesis will be used and the Doppler anemometer. Both mean and RMS axial velocities are
validity of this assumption will be discussed in the next section. reported. Temperature measurements were taken with a bare ther-
With this assumption the equations are volume averaged over the mocouple and have an error of up to 40 K due to radiative heat
entire reactor volume to give 共see Martin 关8兴 for the derivation兲. transfer between the thermocouple and the cold walls. The instru-
ments included a flame ionization detector for unburned hydrocar-
⳵Qi ⳵ 2Q i bons 共UHC兲, infrared detectors for CO and CO2 and a paramag-
兵兵具 ␳␻ i 兩 ␨ 典其其 ⫺ 兵兵具 ␳ S c 兩 ␨ 典其其 ⫹ 兵兵具 ␳ D 共 ⵜc 兲 2 兩 ␨ 典其其 ⫽0 netic analyzer for O2 . CO, CO2 and O2 samples had the UHC and
⳵␨ ⳵␨ 2
(14) H2 O removed before being measured. The UHC samples retained
the H2 O. This test case was chosen because it has many of the
The double curly brackets, 兵兵其其, signify volume averaging. The characteristics of LPM GT combustors, which have lean, pre-
volume averaging will cause the error terms in Eq. (12a) to be mixed fuel and air burning in a flame that is stabilized by a recir-
negligible. Besides the advantage of removing spatial dependence culation zone. This is generally easier to model than an actual
from the equation, the other advantage of volume averaging is that burner, which has a complex three-dimensional, swirl induced re-
it also eliminates the conditioned velocity term, which is difficult circulation zone and normally burns natural gas. The purpose of
Fig. 2 Fluent and CMC with GRI2.11 N Ä1, RSM turbulence
model, equivalence ratioÄ0.9 axial velocity „mÕsec…, axial ve-
locity rms „mÕsec…, temperature K, CH4 -wet, O2 -dry, CO2 -dry,
and CO-dry, species are mole percentages, „O2 , CO2 , and CO
Fig. 1 El Banhawy et al. †10‡ data, equivalence ratioÄ0.9, axial have the UHC removed…
velocity mÕsec, axial velocity rms, temperature K, unburned
hydrocarbons-wet, O2 -dry, CO2 -dry, and CO-dry, species are
mole percentages „O2 , CO2 , and CO have the UHC removed….
Reprinted by permission of Elsevier Science from Premixed, 共UHC兲 as mole percentage-wet, O2 as mole percentage-dry, CO2
Turbulent Combustion of a Sudden-Expansion Flow, by Y. El as mole percentage-dry, and CO as mole percentage-dry 共the O2 ,
Banhawy et al., Combustion & Flame, 50, pp. 153–165, copy- CO and CO2 also had the UHC removed兲 from El Banhawy et al.
right 1983 by The Combustion Institute. 关10兴 for an equivalence ratio of 0.9 and a step height of 0.02 m
共giving an expansion ratio of 2.0兲. Figure 2 shows the same vari-
ables calculated from the volume-averaged premixed CMC
this research was to show the potential of the premixed CMC method using the full GRI2.11 kinetic mechanism with a scalar
combustion model for modeling LPM GT combustors, so simpli- dissipation (N) of 1.0 1/sec. This value was found by trial and
fications of the fluid mechanics that do not greatly affect the com- error by solving the equations with a guessed value of N and
bustion are desirable. These simplifications will reduce the errors using Eq. 共15兲 to calculate a new value of N. This was continued
caused by the turbulence model and reduce runtimes. The fuel was until both values agreed. It was found that N could be varied by
modeled as pure methane instead of natural gas. an order of magnitude from this value with minimal affect on the
The volume averaged premixed CMC equations, which were flow field and major species solution, so small errors in calculat-
described in the previous section, were added to the Fluent CFD ing the scalar dissipation will not adversely affect the solution.
The intermediate species and radicals do vary with changes in
program 共Fluent 关11兴兲 with the Reynolds stress turbulence model scalar dissipation. As N is increased from a value of unity there is
共RSM兲 and the nonequilibrium wall boundary conditions. The full less intermediates species, such as C2 H4 , CH2 O, C2 H6 , H2 ,
GRI2.11 kinetic mechanism 共Bowman et al. 关12兴兲 was used over a H2 O2 , and CH2 . As N increases the mixing becomes more in-
range of scalar dissipation values. The current formulation of the tense, decreasing the Damköhler number and eventually becoming
premixed CMC method assumes adiabatic walls, while the data a perfectly stirred reactor. The free radicals are relatively constant
had water cooled walls, so there will be some errors in the model with small changes in scalar dissipation around a value of unity,
near the walls. but the NO increases as N increases at larger values of c. It is
Figure 1 shows contour plots of axial velocity m/sec, the axial seen from the temperature and species plots that the flame starts at
velocity rms m/sec, temperature K, unburned hydrocarbons about the same location and initially burns at about the same rate.
Figures 2 and 3 show that both combustion models predict the
turbulence intensity of the recirculation zone 共shown as the axial
velocity rms兲 well, but through the flame zone the match is poor.
Both combustion models 共using the same turbulence model兲 pre-
dict a nearly constant turbulence intensity, while the data shows
an order of magnitude increase through the flame. One explana-
tion for this is that there is no direct link from the reactions to the
turbulence generation. The CMC model provides a direct link for
the turbulence 共small-scale turbulence兲 to feed into the reaction
rates, but there is not a source term in the turbulent kinetic energy
equation for the reactions to generate turbulence, which seems
physically possible as the flow accelerates through the reactor.
This is a short-coming of all turbulence models, since they were
developed for incompressible flows. Improving this aspect of tur-
bulence models is beyond the scope of this work.
The premixed CMC method had runtimes approximately twice
as long as the three-step global EDM CH4 mechanism using the
same CFD program, while providing a much better match of the
data and details of the intermediate species. The runtimes were
about two hours using an 800MHz Intel PC with the Linux oper-
ating system. Using a previous solution as the initial conditions
reduced the runtimes to about half an hour. A structured 110 by 20
cell grid was used. To generate the CMC table lookup database
required 10 seconds to 5 minutes for each value of N. The smaller
values of N gave a stiffer system, which required longer to solve.
These initial results from the premixed CMC gave a good
match of the experimental data for the backward-facing step prob-
lem. To fully validate the model, more test cases will need to be
run over a range of problem types. Additional runs will also help
determine the range of validity of the volume averaged CMC
method. When measurements of conditioned species in premixed
flames becomes available in the future, they can also be used to
determine the range of validity of the model. Since LPM GT
combustors have more intense mixing then the backward facing
step used here, this method should be applicable for that class of
problems. More detailed experimental data will also help estimate
the size of the error terms that were neglected in the derivation of
the volume-averaged equations.
Earlier is was stated that for a turbulent premixed combustion
Fig. 3 Fluent three-step EDM, Reynolds stress model equiva- model to accurately predict the emissions of a LPM GT combus-
lence ratioÄ0.9 axial velocity „mÕsec…, axial velocity rms „mÕ
tor it needed three features, namely it must account for the affects
sec…, temperature K, CH4 -wet, O2 -dry, CO2 -dry, and CO-dry, spe-
cies are mole percentages, „O2 , CO2 , and CO have the CH4 of temperature fluctuations on the reaction rates, model the affects
removed… of micromixing on the chemical kinetics and allow detailed chem-
istry. The CMC method has all three features. By doing condi-
tional averages the temperature fluctuations are accounted for, the
scalar dissipation models the small scale mixing and full kinetic
Approximately two thirds of the way through the flame the mod- mechanisms can be used with reasonable runtimes.
eled results appear to burn slightly slower then the data. From the
velocity contours it is seen that the reattachment point of the re-
circulation zone is close to the data, but the peak velocity is far- Conclusions
ther downstream. The turbulence intensity in the recirculation It was shown that the premixed CMC method in its volume-
zone closely matches the data, but down stream it is much lower, averaged form gave a good match to the experimental data and
the reason for this will be discussed later. The peak temperature is also provided details of the intermediate species that are very
about 200 K higher than the data, which has heat loss through he
walls. In previous runs using Fluent’s built in three-step EDM difficult to measure. The results were much closer to the data than
mechanism with adiabatic and nonadiabatic walls 共not shown the three-step global mechanism that is similar to the ones used in
here兲 it was seen that the nonadiabatic walls reduced the peak most industrial reacting CFD work with only a small increase in
temperature by about 200 K. Based on this the temperature is a runtimes. The premixed CMC method shows good promise as a
good match. The UHC 共plotted here as CH4 ), O2 , and CO2 match future design tool for lean premixed gas turbine combustors.
the data very well, except for the slower burning rate late in the More work will need to be done to determine the size of the
flame. The CO gradients are in the correct location, but the mag- error terms that were neglected in Eq. 共14兲 and to determine the
nitude is high in the modeled results. The experimental data has a limits of applicability of the volume averaged formulation of the
peak CO concentration of about 0.50%, while the calculated peak premixed CMC method.
is about 4.3%.
Figure 3 shows the same conditions as the previous figure cal-
culated using the built in three-step kinetic mechanism and EDM
Nomenclature
model in Fluent instead of the CMC method. Here the flame starts c ⫽ reaction progress variable
to burn slightly earlier then the data, but burns much slower with Cp ⫽ specific heat at constant pressure
a large amount of unburned fuel exiting the burner. Overall this D ⫽ diffusivity
gave a poor match. No attempt was made to adjust the parameters eY ⫽ error term
of the EDM model to improve the match, because this defeats the eQ ⫽ error term
purpose of a predictive model. h ⫽ enthalpy
N ⫽ scalar dissipation 共micromixing frequency兲 Superscripts
N ⫽ total number of values to be averaged i ⫽ species number
ns ⫽ number of species s ⫽ denotes sensible enthalpy
P ⫽ probability density function
P ⫽ pressure References
Qi ⫽ conditioned mass fraction or enthalpy 关1兴 Klimenko, A. Y., and Bilger, R. W., 1999, ‘‘Conditional Moment Closure for
Sc ⫽ source term for c equation Turbulent Combustion,’’ Prog. Energy Combust. Sci., 25, pp. 595– 687.
关2兴 Smith N. S. A., 1994, ‘‘Development of the Conditional Moment Closure
t ⫽ time Method for Modeling Turbulent Combustion,’’ Ph.D. thesis, University of Sid-
u ⫽ velocity ney.
WR ⫽ radiation heat loss 关3兴 Klimenko, A. Y., 1990, ‘‘Multicomponent Diffusion of Various Admixtures in
x ⫽ space dimension Turbulent Flow,’’ Fluid Dyn., 25, pp. 327–334.
关4兴 Bilger, R. W., 1991, ‘‘Conditional Moment Methods for Turbulent Reacting
x ⫽ conditioning variable Flow Using Crocco Variable Conditions,’’ Charles Kolling Report, Department
Y ⫽ species mass fraction of Mechanical Engineering, the University of Sydney, TN F-99.
y ⫽ fluctuating species mass fraction 关5兴 Bilger, R. W., 1993, ‘‘Conditional Moment Closure for Turbulent Reacting
Flow,’’ Phys. Fluids A, A5共2兲, pg. 436 – 444.
Greek Letters 关6兴 Crocco, 1948, ‘‘Lo Stato Limite Laminare Neigas,’’ Monographie Scientifiche
di Aeronautica No. 3, Rome, 1946, Translated as Report No. AL684 Aerophys
␨ ⫽ conditioning variable Lab, North American Aviation Inc., Los Angles, CA.
␩ ⫽ value of conditioning variable 关7兴 Bilger, R. W., 1993, ‘‘Conditional Moment Closure Modeling and Advanced
␳ ⫽ density Laser Measurements,’’ Turbulence and Molecular Processes in Combustion, T.
Takeno, ed., The 6th Toyota Conference, Elsevier, New York, pp. 267–285.
␻ ⫽ reaction rate 关8兴 Martin, S. M., 2003, ‘‘The Conditional Moment Closure Method for Lean
Premixed Turbulent Combustion,’’ Ph.D. thesis, University of Washington, to
Subscripts be published.
a ⫽ conditioning variable 关9兴 Bilger, R. W., 2000, ‘‘Future Progress in Turbulent Combustion Research,’’
Prog. Energy Combust. Sci., 26, pp. 367–380.
ad-u ⫽ difference between adiabatic and unburned 关10兴 El Banhawy, Y., Sivasegaram, S., and Whitelaw, J. H., 1983, ‘‘Premixed, Tur-
f ⫽ heat of formation bulent Combustion of a Sudden-Expansion Flow,’’ Combust. Flame, 50, pp.
h ⫽ enthalpy 153–165.
i ⫽ species number 关11兴 Fluent, 2001, www.fluent.com.
关12兴 Bowman, C. T., Hanson, R. K., Davidson, D. F., Gardiner, Jr., W. C., Lissian-
ns ⫽ number of species ski, V., Smith, G. P., Golden, D. M., Frenklach, M. and Goldenberg, M., 1998,
u ⫽ unburned state www.me.berkeley.edu/gri – mech/GRI 2.11.
Liquid Fuel Placement and Mixing
of Generic Aeroengine Premix
Module at Different Operating
J. Becker Conditions
C. Hassa Fuel placement and air-fuel mixing in a generic aeroengine premix module employing
plain jet liquid fuel injection into a counter-swirling double-annular crossflow were in-
DLR–German Aerospace Center,
vestigated at different values of air inlet pressure (6 bar, 700 K and 12 bar, 700 K) and
Institute of Propulsion Technology,
51170 Cologne, Germany
liquid-to-air momentum flux ratio, both parameters being a function of engine power.
Kerosene Jet A-1 was used as liquid fuel. Measurement techniques included LDA for
investigation of the airflow and Mie-scattering laser light sheets and PDA for investiga-
tion of the two-phase flow. Measurements were taken at various axial distances from the
fuel nozzle equivalent to mean residence times of up to 0.47 ms. It was found that the
initial fuel placement reacts very sensitively to a variation of liquid-to-air momentum flux
ratio. Susceptibility of the spray to dispersion due to centrifugal forces and to turbulent
mixing is primarily a function of the fuel droplet diameters, which in turn depend on
operating pressure. The data are interpreted by evaluation of the corresponding Stokes
numbers. 关DOI: 10.1115/1.1587741兴
Introduction flow, changing by about the same factor as air density, which
profoundly affects fuel jet penetration and hence the initial fuel
The injection of kerosene fuel through plain jet nozzles into an
placement.
annular crossflow is one of several possible concepts of a premix
Measurement techniques employed included laser-Doppler an-
module for lean premixed prevaporized 共LPP兲 combustion for
emometry 共LDA兲 for an investigation of the airflow, in the ab-
aviation gas turbines, 关1兴. The motivation for LPP technology, 关2兴,
sence of the liquid fuel, concerning the mean and rms values of
is the fact that under typical aeroengine combustion conditions,
the three velocity components. Furthermore, additional high data
harmful oxides of nitrogen (NOx) are primarily formed according rate LDA measurements were analyzed by fast Fourier transfor-
to the thermal or Zeldovich mechanism, which is characterized by mation 共FFT兲, which allowed the estimation of the turbulence
an exponential dependence of the NO formation rates on tempera- length scale. In the presence of the liquid fuel flow, Mie-scattering
ture. Since the chemical reaction of NO formation remains far laser light sheets were employed for visualization of the dense
from its thermodynamic equilibrium during typical combustor spray in the vicinity of the fuel nozzle and phase-Doppler an-
residence times on the order of 5 ms, NOx emissions can be ef- emometry 共PDA兲 was applied for a detailed investigation of the
fectively reduced by lowering the rate of NO formation. This is fuel spray, particularly the spatial distribution of the liquid fuel
achieved by feeding the combustor with an air-fuel mixture of flux and the representative droplet diameters, such as the SMD.
lean composition, i.e., by using a significant amount of excess air
which acts as thermal ballast and hence keeps the combustion Experiment
temperature low. In order to provide this mixture, premix mod-
ules, which are mounted at the combustor inlet, must be optimized Generic Premix Module. Figure 1 shows the generic premix
concerning their mixing performance, keeping in mind that the module investigated in this work, consisting of a cylindrical center
parameters that affect mixing performance can change signifi- body, an outer cylinder, and an annulus with two concentrical
cantly over an engine flight cycle and that the risk of autoignition axial swirlers in between. Looking upstream, the airflow passing
limits the acceptable residence time to less than 1 ms. The benefits through the inner swirler rotates in the clockwise direction,
of the above-mentioned premix module concept employing radial whereas the airflow passing through the outer swirler rotates in the
plain jet fuel injection into an annular swirling airflow are thermal counterclockwise direction. Both swirlers are designed to generate
management considerations and the fact that the very fuel rich a swirl angle of 30 deg. During operation, fuel is injected out of
core associated with axial fuel injection through pressure swirl the center body in the radial direction through small plain jet
atomizers at high air inlet pressure is avoided. nozzles into the counter-swirling double-annular airflow. A
The objective of the present work is the investigation of the supplementary airflow passes through the center body for cooling
process of air-fuel mixing inside a generic premix module of this of the fuel lines. Whereas in regular operation, several fuel
type as a function of those operating parameters whose variation nozzles equispaced about the circumference of the center body
over a flight cycle have the greatest effect on air-fuel mixing per- would be used, measurements were made with only one active
formance. These are 共1兲 air density, typically varying by a factor fuel nozzle. The single-nozzle data was later extrapolated to more
between 3 and 4, which has a pronounced effect on liquid fuel than one fuel nozzle by means of a superposition algorithm. How-
atomization and fuel spray dispersion, and 共2兲 liquid fuel mass ever, the premix module was equipped with identical fuel nozzles
at various axial stations, all of them oriented as shown in Fig. 1.
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN Hence by taking measurements at a fixed location 1 mm down-
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF stream from the outlet of the premix module and switching from
one fuel nozzle to another, different downstream distances of the
erlands, June 3– 6, 2002; Paper No. 2002-GT-30102. Manuscript received by IGTI, measurement planes were realized. The swirler kit could be
Dec. 2001, final revision, Mar. 2002. Associate Editor: E. Benvenuti. moved in the axial direction, so that the axial distance between the
Fig. 1 Schematic of premix module „not to scale…. White arrow
indicates active fuel nozzle. Fig. 2 PDA measurement grid
swirler outlet plane and the active fuel nozzle was kept constant. light scattered by reflection, hence satisfying the requirement of
Reference measurements by PDA at ambient temperature, hence clear dominance of one single scattering mechanism. In order to
precluding evaporation, but elevated pressure for realistic atomi- be able to access the full plane just 1 mm downstream of the
zation behavior, showed that the difference in SMD of the fuel outlet of the premix module, the beam separation at the 310 mm
spray droplets generated by any two fuel nozzles was between 6% front lens of the transmitting optics was set to just 18 mm. The
and 11% of the mean of all fuel nozzles, so that for engineering resulting small number of fringes in the measurement volume was
purposes, the fuel nozzles can be considered to be identical. accounted for by setting a SNR requirement of ⫹3dB for burst
Test Conditions. The primary goal of the present work was validation. For good dense-spray performance, the measurement
to elucidate how the considerable variations of inlet air density volume had rather small dimensions of about 70 ␮m in diameter
and of fuel mass flow that are characteristic of a flight cycle of an and 100 ␮m in length due to a slit in the receiving optics. The
aviation gas turbine affect the mixing performance of the present measurement grid consisting of 198 positions is shown together
generic premix module. Air inlet conditions were chosen to be 6 with the contour of the annulus in Fig. 2. At each node, the mea-
bar, 700 K, resulting in actual values of density and kinematic surement was stopped after acquisition of 20000 validated
viscosity of ␳ air⫽3.0 kg/m3 , ␯ air⫽1.1e-5 m2 /s, and 12 bar, 700 K, samples or after 10 seconds, whichever came first. If fewer than
resulting in ␳ air⫽6.0 kg/m3 , ␯ air⫽5.6e-6 m2 /s. The total velocity 200 validated samples were recorded in 10 seconds, the measure-
of the swirling airflow was kept constant. Kerosene Jet A-1 with a ment was discarded altogether and the measured liquid flux was
set to zero. Otherwise, the raw data was processed using an in-
density of 795 kg/m3 was used as liquid fuel, so that the effect of
house algorithm, 关7兴. The uncertainty of the SMD measurements
realistic evaporation behavior on spray dispersion is taken into
is about ⫾5%, which is the aggregate error due to possible opti-
account. Since previous research has shown that the jet penetra-
cal misalignment, changes of droplet refractive index during heat-
tion and hence the initial fuel placement of a liquid jet in cross-
up, and photomultiplier voltage setting. The uncertainty of the
flow is governed by the liquid-to-air momentum flux ratio q
representative diameters D10% and D50%, where, e.g., D10% is
⫽( ␳ liq•u liq
2
)/( ␳ air•u air
2
), 关3– 6兴, this parameter will be used the droplet diameter such that 10% of the total liquid volume of
throughout this work instead of the fuel mass flow. The air veloc- the spray is present in droplets below this diameter, is estimated to
ity is calculated from the volumetric airflow at the inlet of the be on a similar level. The measurement of the D90%, which can
premix module and the cross-sectional area of the annulus, and react very sensitively to just a few very large droplets, is less
the fuel velocity is calculated from the volumetric fuel flow and reliable.
the cross-sectional area of the fuel nozzle orifice. For the investigation of the airflow by LDA in the absence of
Measurement Techniques. In a first step, the initial fuel the fuel, the air was seeded with submicron particles of a mixture
placement was visualized by laser light sheets. The beam of a of ethanol and glycerin, necessitating tests at ambient temperature.
continuous wave 400 mW Nd:YAG laser (␭⫽532 nm) was Since the Reynolds number of the airflow at 6 bar and 700 K,
formed into a sheet by means of a set of lenses. The sheet had a using the hydraulic diameter of the annulus, is already 0.9•105 , so
nominal height 共i.e., based on the Gaussian beam diameter兲 of that the region of Reynolds number invariance has basically been
about 48 mm, of which only the central 30 mm were used, so that reached, the test conditions were scaled to 1 bar, 280 K at constant
the illumination intensity at the edge of the sheet was still 68% of air velocity, resulting again in a Reynolds number of about 0.9
the peak value in the center. The nominal thickness of the waist of •105 . The LDA system consisted of the two-dimensional PDA
the light sheet was 110 ␮m. The Mie-scattering image of the spray transmitting optics mentioned above, an in-house high sensitivity,
illuminated by the laser light sheet was recorded by an intensify- wide aperture receiving optics positioned at an off-axis angle of
ing CCD camera equipped with an interference filter using a gate 15° for maximum data rates, and a Dantec Burst Spectrum Ana-
time 共effectively an exposure time兲 of 2 ␮s. Due to restrictions of lyzer. The uncertainty of the velocity measurements is estimated
optical access, the camera could not be positioned perpendicular to be less than ⫾2%.
to the light sheet. The resulting distortions were corrected for by If LDA data rates are sufficiently high and if the sample size is
the camera control software after calibration with reference grid sufficiently large, then the turbulence time scale can be derived by
positioned in the light sheet plane. Each measurement consisted of spectral analysis, 关8兴. Application of the Taylor hypothesis then
the acquisition of 100 single images and subsequent averaging. leads to the turbulence length scale L t . In the present case, data
PDA measurements were conducted focusing on the spatial dis- rates of 120 kHz to 240 kHz were achieved by using three particle
tribution of liquid fuel flux and representative droplet diameters. A seeders in parallel. Based on these data rates, the density of the
Dantec 2-D PDA system was used to measure the axial and the particles of about 800 kg/m3 and the upper bound of the particle
vertical droplet velocities and the droplet diameters. The receiving size of 1 ␮m, which is ensured by an impactor at the outlet of the
optics were positioned at an off-axis angle of 50°, where light particle generators, a mass loading ratio of approximately 1e-4 is
scattered by first-order refraction is about 150 times stronger than calculated, so that feedback of the seeding on the flow itself is
Table 1 Measurement planes
Mean Residence Times

Measurement Corresponding to
Techniques Measurement Planes 共ms兲
Laser light sheets 0.09; 0.16; 0.35
PDA 0.27; 共0.35;兲 0.47
LDA 共0.07;兲 0.27
negligible. The required sample size was increased from 10,000 to

100,000 validated events. By variation of the signal gain, it was
usually possible to ensure that the velocity, i.e., turbulence, spec-
trum was not significantly corrupted by the noise spectrum. The
uncertainty of the results amounts to a factor of about 2, so that
the true value of L t is estimated to be between approximately 1/2
and 2 times the measured value. Details of the measurements are
given by Becker and Hassa 关9兴. Similar measurements are re-
ported by Hassa et al. 关10兴.
Measurements were conducted at various axial distances from
the fuel nozzle corresponding to mean residence times of up to
0.47 ms. This value is sufficiently low to offer acceptable safety
margins regarding autoignition even at high power operating con-
ditions. According to the correlation for autoignition times in pre-
mix modules derived by Guin 关11兴, the autoignition time expected
for 35 bar and 850 K is 1.1 ms, implying a safety factor of greater
Fig. 3 Laser light sheet images at 6 bar
than 2 if the most downstream measurement plane is taken as the
exit plane of the premix module. Table 1 lists the measurement
planes for each measurement technique. The data of the measure-
ment planes put in parentheses is not discussed explicitly in this core of the spray cloud is found to remain slightly closer to the
paper. center body at 12 bar. For the subsequent detailed investigation of
the spray development in the counter-swirling annular airflow by
Results combined analysis of the LDA data of the airflow and the PDA
data of the fuel spray, a constant value of q⫽3 was considered to
Laser Light Sheet Images. Different values of the liquid-to- offer the best compromise. Test conditions with q⬎3 were ruled
air momentum flux ratio q between q⫽2 and q⫽6 were realized, out due to significant impingement of the liquid fuel on the outer
because liquid fuel penetration at a fixed location is primarily cylinder at t res⫽0.35 ms. At q⫽2, on the other hand, the bulk of
a function of q and the fuel nozzle diameter, 关3– 6兴. Particularly the spray remains close to the center body, resulting in poor inter-
at elevated air density and q up to q⫽12, it had been found action with the shear layer and even partial film formation on the
that the outer contour of the liquid fuel jet can be described by the
correlation
z
D 冉
⫽1.57•q 0.36• ln 1⫹3.81•
x
D 冊 (1)
between x/D⫽2 and x/D⫽22, where x is the streamwise dis-

tance, z is the penetration depth, and D is the diameter of the fuel
nozzle, 关6兴. Fuel film formation on the center body or the inner
surface of the outer cylinder was to be avoided in order to control
the risk of autoignition due to possibly very long residence times
of the fuel contained in the film and also to avoid coke and gum
formation by thermal decomposition of the fuel. Film formation
on the center body is a consequence of insufficiently high values
of q, whereas exceedingly high values of q 共relative to the width
of the annulus兲 lead to equally undesirable impingement of the
fuel jet on the outer cylinder. Film formation on the outer cylinder
can also be brought about by strong centrifugal forces acting on
the fuel spray.
The Mie-scattering laser light sheet images are presented in
Figs. 3 and 4. The intensity of the images has been normalized so
that the highest pixel count in each image corresponds to the
brightest shade of red. Analysis of the results revealed that film
formation is difficult to avoid completely in the present geometry.
Since the premix module had to be scaled to fit into the pressure
housing of an existing test rig, the annular width is rather small,
resulting in a narrow allowable range of q. Greater annular widths
will be able to accommodate a wider range of q without the oc-
currence of unacceptable fuel film formation on either surface.
The effect of q on the initial fuel placement can be seen clearly in
Figs. 3 and 4. The effect of air density seems to be small, but the Fig. 4 Laser light sheet images at 12 bar
Table 2 Representative droplet diameters
Meas. SMD D10% D50% D90%

Test Condition Plane 共␮m兲共␮m兲共␮m兲共␮m兲
6 bar, 700 K, 0.27 ms 18.2 11.4 18.7 37.4
q⫽3 0.47 ms 19.2 11.9 20.0 38.2
12 bar, 700 K, 0.27 ms 16.2 10.1 16.4 34.8
q⫽3 0.47 ms 16.0 10.2 16.3 31.7
the spatial distribution of SMD at 12 bar is very homogeneous, a

layered structure is found at 6 bar, where larger droplets are found
Fig. 5 Measured fuel flux „cm3 Õcm2 "s… at t resÄ0.47 ms, q Ä3, at greater radii. This suggests that at 12 bar, the spray is subjected
left: 6 bar; right: 12 bar to a stochastic mixing process due to turbulence. At 6 bar, on the
other hand, the spray dispersion appears to be governed by inertial
forces resulting from the initial radial velocity of the liquid fuel
center body. Film formation was not observed at q⫽3. The fuel and centrifugal forces resulting from the tangential velocity im-
mass flow rates at q⫽3 based on a single nozzle were 1.38 g/s at parted on the fuel by the airflow. These effects will be discussed in
6 bar, 700 K and 1.95 g/s at 12 bar, 700 K. more detail below. Table 2 summarizes the measured representa-
If the premix module is to be operated at constant AFR tive diameters of the spray at two different PDA measurement
throughout the flight cycle, then this implies that q is proportional planes. The values for the downstream measurement plane given
to the air density. If AFR decreases as engine power is increased, in the table have been computed by weighting the representative
then the variation of q becomes even greater. The narrow range of diameter considered at each node of the grid with the measured
acceptable variation of q in the present geometry hence can prob- liquid fuel flux at the same node. At the more upstream measure-
ably only be achieved by internal circumferential staging of the ment plane, the spray was extremely dense, yielding raw data
fuel nozzles or external staging of the premix modules, or a com- rates above 150 kHz and validation rates below 50% in the spray
bination of both. core. Under these adverse conditions, the fuel flux cannot be reli-
Phase-Doppler Anemometry „PDA… Data „Single Nozzle…. ably determined, so that the local representative diameters had to
Results, all of which have been obtained at q⫽3, are presented in be weighted with the raw data rate instead.
Figs. 5 and 6. Figure 5 consists of the contour plots of the mea- As is to be expected, the airblast atomization of the liquid fuel
sured liquid fuel flux at 6 bar and 12 bar at the measurement plane jet yields smaller droplets if the air density is increased. At 12 bar,
corresponding to 0.47 ms. The color map is adjusted to the maxi- the representative diameters remain approximately constant, but at
mum measured value at each test condition and consists of ten 6 bar, an increase is observed as the smallest droplets present in
colors. Hence the deepest red represents fuel flux values between the upstream measurement plane have disappeared so quickly due
90% and 100% of the maximum, orange represents values be- to evaporation that shrinking of slightly larger droplets could not
tween 80% and 90% of the maximum et cetera. At 6 bar, the core compensate the loss by the time the spray reaches the downstream
of the spray cloud is found almost exactly in the center of the measurement plane.
annulus, enabling close interaction of the spray with the shear Laser-Doppler Anemometry „LDA… Data. In order to obtain
layer. As a consequence, a significant portion of the spray has the full velocity information including rms values of all three
been displaced from the core in the counterclockwise direction by velocity components using a two-dimensional LDA system, radial
the swirl component of the flow in the outer part of the annulus, velocity profiles were recorded at three different tangential posi-
while the core itself was still under the influence of the clockwise tions. Explorative measurements had indicated wakes resulting
swirl in the inner part of the annulus. At 12 bar, the spray is from the swirler vanes. Hence, in each measurement plane, tan-
located closer to the center body and its interaction with the shear gential profiles of the axial velocity were obtained first in order to
layer has been somewhat poorer. The fuel flux in the outer part of identify the locations of the wakes. Then, for a comprehensive
the annulus is noticeably lower than in the inner part. characterization of the airflow, measurements were taken in a
Contour plots of the measured SMD at 6 bar and 12 bar are wake as well as in the region between two adjacent wakes, result-
given in Fig. 6. The color map is identical for both plots. Whereas ing in the need to record six profiles each in the inner and outer
part of the annulus.
Figures 7 and 8 show these data after transposition into the
cylindrical coordinate system. In the measurement plane corre-
sponding to 0.27 ms, the peak tangential velocity has already de-
cayed considerably from its values further upstream, and the
wakes and passages have largely disappeared. The rms-values of
all three velocity components are around 15 m/s, with slightly
higher values in the center of the annulus where the shear layer of
the counter-rotating swirling flow is located.
Results of measurements of L t in the plane corresponding to
0.27 ms are given in Fig. 9. The data pertaining to the direction
perpendicular to the axial direction was not decomposed into the
radial and tangential component due to the high uncertainty of the
measurement. Clearly, the turbulence length scale in the axial di-
rection is much larger than the length scale in the perpendicular
direction, which can be attributed to the ratio of length to hydrau-
Fig. 6 Measured SMD „␮m… at t resÄ0.47 ms, q Ä3, left: 6 bar; lic diameter of the passages of the axial swirlers of approximately
right: 12 bar three.
Fig. 9 Radial profiles of turbulence length scale at t res
Ä0.27 ms
Fig. 7 Radial profiles of mean velocity components at t res

Ä0.27 ms, normalized with nominal total velocity and the assumption that the liquid jet maintains its exit velocity
u liq until breakup. If q is held constant, then u liq⬀ ␳ air
0.5
, so that the
penetration depth of the liquid jet u liq•t b is not a function of air
density.
Analysis The stop distance s stop in this case is the distance covered by a
In order to understand the fuel spray dispersion and hence the droplet until its velocity relative to the surrounding air has van-
air-fuel mixing, the PDA data was analyzed together with the ished. s stop is the product of the initial droplet velocity and the
LDA data of the airflow, with the laser light sheet images provid- velocity relaxation or response time of the droplet, given by 关6兴:
ing information concerning the initial fuel placement. Primarily, 4 ␳ liq D 2p
the observed differences between the results obtained at 6 bar and ␶ relax⫽ • • (3)
12 bar concerning liquid fuel placement and spatial distribution of 3 ␳ air c D • Rep • ␯ air
SMD will be analyzed. 24
Liquid fuel placement is the aggregate result of three contribu- c D⫽ • 共 1⫹0.15• Re0.687
p 兲 (4)
tions: 共1兲 penetration of the liquid jet, 共2兲 the stop distance of the Rep
individual droplets after breakup of the liquid jet, and 共3兲 radial 兩共 uair⫺up 兲兩 •D p
displacement of the spray due to centrifugal forces. Rep ⫽ . (5)
The penetration of the liquid jet until breakup can be estimated ␯ air
from the equation for the time required for completion of breakup, For creeping flow (Rep⬍1) only, Eqs. 共3兲–共5兲 reduce to the better
关5兴, which was originally derived for the secondary breakup of known form ␶ relax⫽ ␳ liqD 2p /18 ␮ air and ␶ relax is the time constant of
droplets the differential equation describing droplet motion. In a noncreep-
t b ⫽C•D 0 • 冑␳ liq 1
␳ air u air
(2)
ing flow, as is the case in the present investigation, ␶ relax depends
on the relative droplet velocity, so that the concept of ␶ relax as a
constant is—strictly speaking—invalidated and a comparison of
␶ relax of droplets with different relative velocities leads to unphysi-
cal results. Therefore, the creeping flow equation for ␶ relax is often
found in the literature even if Rep⬎1. Numerically, the error thus
incurred is often not large. However, the effect of air density on
droplet motion, which is non-negligible in the present context, can
only be taken into consideration if the more cumbersome Eqs.
共3兲–共5兲 are used, but care must be taken not to evaluate the effect
of different values of relative velocity on the relaxation time. Re-
turning now to the task of evaluating the stop distance and assum-
ing that the initial droplet velocity is identical to the velocity of
the liquid jet u liq , Eqs. 共3兲–共5兲 applied to the D50% of the two
different test conditions yield s stop⫽0.9 mm at 6 bar and 0.8 mm
at 12 bar. This small difference could not have been detected
based on the PDA measurement grid. Hence neither liquid jet
penetration nor the stop distance can explain the differences in
fuel placement between 6 bar and 12 bar, which must therefore be
due to the effect of centrifugal forces acting on the spray.
Susceptibility to radial displacement due to centrifugal forces is
represented by the value of the swirl Stokes number, 关12兴. The
definition of the Stokes number in general is 关13兴:
␶ relax
St⫽ . (6)
␶ flow
␶ flow is the time scale of the feature of the flow that is to be
considered. In the context of the swirl Stokes number, ␶ flow
⫽r/u tan(r). The value of the swirl Stokes number indicates
Fig. 8 Radial profiles of velocity fluctuation at t resÄ0.27 ms whether droplet movement in the swirling airflow is dominated by
Table 3 Stokes numbers at t resÄ0.27 ms
Test Cond. Repr. Droplet D 共␮m兲 Stswirl Stturb

6 bar, D10% 11.4 0.2 1.1
700 K, D50% 18.7 0.4 2.5
q⫽3 D90% 37.4 1.1 7.0
12 bar, D10% 10.1 0.1 0.7
700 K, D50% 16.4 0.2 1.4
q⫽3 D90% 34.8 0.7 4.3
centrifugal forces, resulting in radial displacement of the droplet,

or whether it is dominated by the drag force, causing the droplet
to faithfully follow the streamlines of the airflow. This latter case Fig. 10 Fuel flux „cm3 Õcm2 "s… at t resÄ0.47 ms, q Ä3, extrapo-
is expected for St⬍0.1. If St⬎0.1, then the drag force is not lated to four fuel nozzles, left: 6 bar; right: 12 bar
clearly dominant and the droplet will be centrifuged out of the
swirling flow sooner or later. In the range 0.1⬍St⬍1, even small
variations of St can lead to markedly different droplet trajectories.
If the swirl Stokes number is very large, i.e., St⬎10, then the become serious at high pressure, initial fuel placement at high
droplet will hardly follow the curved streamlines of the airflow to power conditions should be such that swirl-induced radial dis-
begin with 关12兴. placement of the spray is not needed to transport liquid fuel into
Correspondingly, the turbulence Stokes number, for which the outer part of the annulus.
␶ flow⫽L t /u ⬘ , indicates to which degree the motion of a given Regarding the values of the turbulence Stokes number, almost
droplet can be affected by the large turbulent eddies of the airflow. all of the liquid fuel at 6 bar is contained in droplets that are too
While large droplets may be too heavy to interact with a turbulent heavy to react to the turbulent eddies of the airflow, because the
eddy over the eddy’s lifetime, as indicated by a large value of the droplet response time is significantly longer than the large eddy
turbulence Stokes number, smaller droplets may be subjected to lifetime. At 12 bar, however, interaction with turbulent eddies
intense turbulent mixing due to their shorter relaxation times and should still be possible for the D50%, so that about half of the
correspondingly small Stokes numbers. liquid fuel is subject to turbulent mixing. These trends are borne
For both test conditions, the swirl Stokes number and the tur- out by the spatial distributions of SMD presented in Fig. 6. Tur-
bulence Stokes number were calculated for the three representa- bulent mixing is not fast enough, though, to move the bulk of the
tive diameters D10%, D50%, and D90% in the measurement spray due the brevity of the residence time of the premix module.
plane t res⫽0.27 ms. The Stokes numbers were evaluated for the
inner part of the annulus, where almost all of the liquid fuel is
located in this measurement plane 共see Figs. 2 and 3兲. The liquid Extrapolation to N Fuel Nozzles
density of the droplets was estimated to be around 650 kg/m3 ,
Extrapolation of the fuel flux data to more than one fuel nozzle
based on a CFD analysis of a similar test case, 关7兴. A representa-
was accomplished by copying, rotating and superpositioning the
tive relative droplet velocity of 28 m/s was derived from a com-
spatially resolved single-nozzle fuel flux data recorded in the most
parison of the velocity information of the PDA data and the air
downstream PDA measurement plane. The results presented in
velocity data of Fig. 7. The turbulence data used for Stturb was
Figs. 10–12 show that for eight equispaced fuel nozzles, the cir-
u ⬘ ⫽15 m/s, L t ⫽0.75 mm 共see Figs. 8 and 9兲. Results are given in cumferential homogeneity of the fuel spray has basically reached
Table 3. Based on the assessment of experimental errors given its asymptotic value, as the maximum fuel flux varies by less than
above and the fact that the liquid density of the droplets is not 10% in the circumferential direction. However, this value of the
known exactly, the uncertainty of the swirl Stokes number is recommended number of fuel nozzles is not necessarily applicable
about ⫾20%. The uncertainty of the turbulence Stokes number, to other premix modules, as it probably depends on the circum-
which due to the uncertainty of the turbulence length scale mea- ferential length as well as on the swirl angle. It is interesting to
surement alone amounts to a factor of about 2, is hence increased note that the particular, highly asymmetric shape of the spatial
to a factor of roughly 2.5. fuel flux distribution at 6 bar involving a long, one-sided tail
Since the Stokes number in general depends on the droplet actually causes a decrease in fuel spray homogeneity as the num-
response or droplet velocity relaxation time and hence on the ber of fuel nozzles is increased from 4 to 6.
square of the droplet diameter, the values of both Stokes numbers
considered here are noticeably lower for the 12 bar test case than
for the 6 bar test case. The higher air density per se at 12 bar also
lowers the Stokes number, but to a lesser degree than the droplet
diameters. Based on the data in Table 3, about half of the liquid
fuel of the spray at 12 bar is contained in droplets that are ex-
pected to be subject to only very slow radial displacement due to
centrifugal forces, whereas at 6 bar, significantly faster radial dis-
placement is expected. This is in agreement with the spatial dis-
tribution of the liquid fuel flux shown in Fig. 5. In this context it
should be noted that in the center of the annulus, the tangential
velocity of the airflow vanishes, which is an inherent feature of a
counter-swirling flow field. This constitutes a barrier for small
droplets in the inner part of the annulus that may drift in the radial
direction due to centrifugal forces, but then lose their tangential
momentum—and hence the centrifugal forces acting on them—
during their slow drift through the zone of negligible tangential air
velocity. These droplets hence become trapped in the inner annu- Fig. 11 Fuel flux „cm3 Õcm2 "s… at t resÄ0.47 ms, q Ä3, extrapo-
lus. In order to overcome this trapping effect, which is expected to lated to six fuel nozzles, left: 6 bar; right: 12 bar
Nomenclature
cD ⫽ drag coefficient
D ⫽ diameter
Lt ⫽ turbulence length scale
p ⫽ pressure
q ⫽ fuel-to-air momentum flux ratio (⫽( ␳ liq•u liq
2
)/
( ␳ air•u air))
2
r ⫽ radial coordinate
Re ⫽ Reynolds number
s ⫽ distance
St ⫽ Stokes number
t ⫽ time
Fig. 12 Fuel flux „cm3 Õcm2 "s… at t resÄ0.47 ms, q Ä3, extrapo- T ⫽ temperature
lated to eight fuel nozzles, left: 6 bar; right: 12 bar
u ⫽ velocity
u⬘ ⫽ rms-value of velocity fluctuation
Greek Letters
Conclusions ␯ ⫽ kinematic viscosity

␳ ⫽ density
Considering the practical implications of this research, it ap-
␶ ⫽ time scale
pears to be important to carefully match, by way of the number of
fuel nozzles, the maximum value of q occurring during a flight Subscripts
cycle to the width of the annulus of the premixer. Usually, q will 0 ⫽ initial value
reach its maximum value at the highest power setting. Based on
b ⫽ breakup
correlations for the droplet sizes resulting from atomization of a
plain liquid jet in crossflow at elevated pressure, 关6,14,15兴, and the res ⫽ residence
swirl Stokes number analysis performed in the present work, the p ⫽ particle 共used synonymously with droplet兲
very small droplets generated at a typical takeoff condition will be Abbreviations
highly able to follow the spiral path of the airflow, so that no
radial displacement of the spray is expected. Therefore, q can, and AFR ⫽ air-to-fuel ratio 共by mass兲
should, be selected so that initial fuel placement makes use of LDA ⫽ laser doppler anemometry
nearly the entire width of the annulus at the highest power setting. PDA ⫽ phase doppler anemometry
At lower power operating conditions, q will also be lower and rms ⫽ root mean square
thus initial fuel placement will be disproportionately in the inner SMD ⫽ Sauter mean diameter
part of the annulus. However, the larger droplets generated due to
the lower air density associated with lower power conditions will
have larger swirl Stokes numbers and hence be susceptible to
centrifugal forces. Hence the spray will be distributed across the
width of the annulus as it travels through the premix module. References
Obviously, a Stokes number analysis requires a considerable 关1兴 Bittlinger, G., and Brehm, N., 1999, ‘‘High Pressure Combustion Test of LPP
amount of input data and therefore mainly serves to increase un- Modules in an Axially Staged Combustor Using a Multisector Rig,’’ 14th
derstanding of data already acquired. A detailed predictive calcu- International Symp. on Airbreathing Engines 共ISABE兲, Florence, Italy, paper
lation of the spray dispersion appears to be possible only by CFD IS-007.
simulation. This in turn requires accurate spray initial conditions, 关2兴 Lefebvre, A. H., 1998, Gas Turbine Combustion, 2nd Ed., Taylor & Francis,
Philadelphia, PA, Chap. 9.
which usually have to be acquired by time-consuming PDA mea- 关3兴 Inamura, T., Nagai, N., Hirai, T., and Asano, H., 1991, ‘‘Disintegration Phe-
surements. Recently, however, a comprehensive model of the at- nomena of Metalized Slurry Fuel Jets in High Speed Air Stream,’’ Proceed-
omization of a plain liquid jet in crossflow has been developed by ings, 5th Int’l Conference on Liquid Atomization and Spray Systems (ICLASS-
Rachner et al. 关16兴. The results of the present investigation may 91), National Institute of Standards and Technology, Gaithersburg, MD, pp.
serve as input data for a design optimization by CFD, in the 839– 846.
course of which the inner and outer diameters of the annulus, the 关4兴 Chen T. H., Smith, C. R., Schommer, D. G., and Nejad, A. S., 1993, ‘‘Multi-
swirl angles, and the number of fuel nozzles could be varied as far zone Behavior of Transverse Liquid Jet in High-Speed Flow,’’ AIAA Paper No.
as the boundary conditions set by the engine as a whole permit. In 93-0453.
关5兴 Wu, P. K., Kirkendall, K. A., Fuller, R. P., and Nejad, A. S., 1997, ‘‘Breakup
order to accommodate the wide variation of q occurring in an
Processes of Liquid Jets in Subsonic Crossflows,’’ J. Propul. Power, 13, pp.
engine cycle without resorting to staging, it appears promising to 64 –73.
increase the number of concentrical axial swirlers from 2 to 3 and 关6兴 Becker, J., and Hassa, C., 2002, ‘‘Breakup and Atomization of a Kerosene Jet
include a concentrical filmer surface between the outer two swirl- in Crossflow at Elevated Pressure,’’ Atomization Sprays, 12, pp. 49– 67.
ers. This would avoid impingement of the fuel jet on the outer 关7兴 Brandt, M., Schmitz, G., and Rachner, M., 1998, ‘‘An Experimental and Nu-
cylinder at all values of q and simultaneously ensure that the merical Study of Kerosine Spray Evaporation in a Premix Duct for Gas Tur-
spray is exposed to the highly beneficial shear layer generated by bine Combustors at High Pressure,’’ Combust. Sci. Technol., 138, pp. 313–
the inner two swirlers. 348.
关8兴 Adrian, R. J., and Yao, C. S., 1987, ‘‘Power Spectra of Fluid Velocities Mea-
sured by Laser Doppler Velocimetry,’’ Exp. Fluids, 5, pp. 17–29.
关9兴 Becker, J., and Hassa, C., 2001, ‘‘Messung des turbulenten Längenmaßes in
einem generischen Vormischmodul für Flugtriebwerke,’’ Proceedings, 9,
Acknowledgments GALA Fachtagung, Winterthur, Switzerland, paper 25 共in German兲.
关10兴 Hassa, C., Blümcke, E., and Eickhoff, H., 1990, ‘‘Measurements of Eulerian
The authors wish to thank Sebastian Cordes for his design work Macro Timescales in Highly Swirling Flows and Comparison With Computa-
related to this investigation and Dr. Ralf von der Bank of Rolls- tional Model,’’ Proceedings, 5th Int’l Symp. on Application of Laser Tech-
Royce Deutschland for his comments and cooperation. niques to Fluid Mechanics, Lisbon, Portugal, paper 10.1.
关11兴 Guin, C., 1998, ‘‘Characterization of Autoignition and Flashback in Premixed- 关14兴 Jasuja, A. K., 1982, ‘‘Plain-Jet Airblast Atomization of Alternative Liquid
Injection Systems,’’ RTO Meeting Proceedings 14, Gas Turbine Engine Com- Petroleum Fuels Under High Ambient Pressure Conditions,’’ ASME Paper No.
bustion, Emissions and Alternative Fuels, NATO Research and Technology 82-GT-32.
Organization 共RTO兲, Neuilly-sur-Seine, France, pp. 30/11–30/19. 关15兴 Hautman, D. J., and Rosfjord, T. J., 1990, ‘‘Transverse Liquid Injection Stud-
关12兴 Dring, R. P., and Suo, M., 1978, ‘‘Particle Trajectories in Swirling Flows,’’ J. ies,’’ AIAA Paper No. 90-1965.
Energy, 2, pp. 232–237. 关16兴 Rachner, M., Becker, J., Hassa, C., and Doerr, T., 2002, ‘‘Modelling of the
关13兴 Chung, J. N., and Troutt, T. R., 1987, ‘‘Simulation of Particle Dispersion in an Atomization of a Plain Liquid Jet in Crossflow at Gas Turbine Conditions,’’
Axisymmetric Jet,’’ J. Fluid Mech., 186, pp. 199–222. Aerospace Science and Technology, 6, pp. 495–506.
Fuzzy Fuel Flow Selection Logic
A. Martucci
for a Real Time Embedded Full
Senior Design Engineer
A. J. Volponi
Authority Digital Engine Control
Fellow The control logic of the modern full authority digital engine control is comprised of many
control loops, each of which has a specific purpose. Typical control loops include (but are
Pratt & Whitney, not limited to) a high or low rotor speed governor, an acceleration and deceleration loop,
400 Main Street, and various limiting loops for temperature, speed, fuel flow, and rate of change of fuel
East Hartford, CT 06108 flow. The logic that determines which of these loops is ‘‘in control’’ at any particular time
has a history of being very simplistic. This selection logic is usually nothing more that a
cascade of minimum and maximum selection logic gates. Since this logic is so simplistic,
the control engineer often has to ‘‘fine-tune’’ the compensator design for each loop, in
order to achieve proper performance. 关DOI: 10.1115/1.1365396兴
Introduction the conservative loop to keep it from being selected under certain
circumstances. This, of course, changes the stability design of the
The purpose of this paper is to a present a means of achieving
previously optimized conservative loop.
a selection logic design based on fuzzy logic principles. Hereafter,
This paper provides an alternate means to ‘‘fine-tune’’ the con-
this shall be referred to as a fuzzy fuel flow selector. The paper trol. The means of fine tuning, however, will not affect the control
provides a brief overview of the most typical architecture for real loops that the control engineers went through such trouble to op-
time embedded full authority digital engine control 共FADEC兲 con- timize. Instead, fuzzy logic principles will be applied to the selec-
trol logic, rule development for the fuzzy fuel flow selector, and a tion logic. The end result will be referred to as fuzzy fuel flow
realistic implementation of the fuzzy fuel flow selector. Also pro- selection logic and will minimize the common control design
vided is a typical application that is used to contrast the perfor- problems from loop interaction, while maximizing performance
mance between traditional selection logic and the fuzzy fuel flow for flight maneuvers that require maximum performance.
selector. An overview of the fuzzy logic process is also included This paper is organized as follows:
for completeness.
It is certainly an understatement to say that it is difficult to 1. A brief overview of the fuzzy logic process will be provided.
design the control logic for the modern FADEC. After all, it is not 2. The typical control loop architecture, used in a modern FA-
like the control designers have only one control loop to worry DEC, will be described 共including the traditional fuel flow selec-
about. On the contrary, many control loops need to be designed in tion logic兲. This architecture will be the basis for the design of the
order to meet the overall system requirements. Naturally, each fuzzy fuel flow selection logic.
loop has its own particular function and unique set of require- 3. The details of the fuzzy fuel flow selection logic will be
ments. developed and explained.
For reasons of simplicity, the control design engineers typically 4. An example implementation will be provided. The system
design and optimize all control loops as stand-alone systems using performance using the traditional fuel flow selection logic, and the
piecewise linear models. Many control loops are designed conser- fuzzy fuel flow selection logic, will be compared.
vatively. Typically, loops that maintain a fixed set point 共so-called 5. The issue of fuzzy fuel flow selection and control loop sta-
governor loops兲 are among these. Some loops need to be designed bility will be discussed.
more aggressively. Loops, which cause the engine to transition
from one power setting to another, are good examples of aggres- The Fuzzy Logic Process
sively designed control loops. The notion of a fuzzy set was introduced in 1965 by Lotfi
The logic that decides which of these loops control at any one Zadeh 关1兴, who was then a professor of Electrical Engineering at
time is referred to as fuel flow selection logic 共hereafter called the University of California at Berkeley. His idea was to expand
selection logic兲. Historically, selection logic is simplistic, com- upon the general definition of a set in order to allow for the treat-
prised of cascades of select min and select max logic gates. Inter- ment of uncertainty in the classification of objects. A fuzzy set is
esting things begin to happen when all of these individually de- a collection of objects whose membership, unlike those in a crisp
signed 共and optimized兲 control loops are connected to the set, is a matter of degree. This is expressed in terms of a mem-
selection logic. It is very common, for example, for the selection bership function that assigns a number between 0 and 1 to each
logic to transfer control prematurely from an aggressively de- object, where 1 denotes absolute membership in the set and 0
signed loop to a conservatively designed loop. The end result is denotes exclusion from the set. An object assigned a value some-
performance that is not quite as good as it could have been had the where in between 0.6 is said to belong to the set with degree 共or
selection logic allowed the aggressive loop to control a little membership兲 0.6. This also implies that it is excluded from the set
longer. In order to avoid such situations, the control designers will with degree 1⫺0.6⫽0.4. Thus, by virtue of this fuzzy classifica-
‘‘fine-tune’’ the individual loops to achieve the desired perfor- tion, it is possible to deal with sets which are not 共mathematically兲
mance. This fine tuning usually involves adjusting the loop gain of well defined, but rather, have been defined in linguistic terms,
such as the set of numbers that are approximately equal to ␲, or
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN the set of jet engines exhibiting low exhaust gas temperature mar-
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF gin.
tional Gas Turbine and Aeroengine Congress and Exhibition, Munich, Germany, May
Inasmuch as ordinary set theory serves as a foundation for clas-
8 –11, 2000; Paper 2000-GT-0046. Manuscript received by IGTI October 1999; final sical logic, fuzzy set theory has likewise aided the development of
revision received by ASME Headquarters October 2000. Associate Editor: D. Wisler. fuzzy logic 共关2兴兲. Fuzzy logic combines notions from ordinary Ar-
Table 1 MinÕMax and Bayesian methods
Min/Max
Operation Method Bayesian Method
Fig. 1 Fuzzy logic processes X 1 AND X 2 min (W 1 ,W 2 ) W*1 W2
X 1 OR X 2 max (W 1 ,W 2 ) W 1 ⫹W 2 ⫺W 1* W 2
NOT X 1 1⫺W 1 1⫺W 1
istotelian logic and fuzzy set theory, in that the truth of a given
proposition 共like a fuzzy set兲 can be something between 0 共repre-
senting false兲 and 1 共representing true兲. Logical operations and This process is carried out for each input variable X i and their
rules of inference have also been defined, which allows for the respective beliefs are forwarded to the rulebase for evaluation. A
description 共and analysis兲 of complex systems in a natural lan- typical rule may, for example, take the form:
guage setting. This process mimics the way humans convey and
process information, and thus provides a mathematical vehicle for IF 共 X 1 is small兲 AND 共 X 2 is medium兲 THEN 共 Y is positive兲 .
the representation of imprecise rules and heuristics. It wasn’t long The first step in evaluating this rule is to determine the belief
after Zadeh’s introduction of fuzzy logic that applications arose in value of the rule which is, by definition, the belief value of the
the area of dynamic control. Mandami 关3兴 demonstrated its utility antecedent 共the IF part兲. This can be accomplished in many ways.
by successfully constructing a fuzzy logic control system for a The two most popular methods are referred to as the Min/Max
steam engine. This first application heralded a new era for fuzzy method and the Bayesian method. The former method takes the
logic as a control paradigm, with what seems an innumerable Min belief of the arguments that are combined by AND and the
number of applications to follow, ranging from focus systems in Max belief of arguments that are combined by an OR. In a similar
camcorders to automated control of railway cars. fashion, the Bayesian method uses product and generalized sum to
Generally speaking, there are three major processes common to accomplish these operations. These relations are summarized in
all fuzzy control systems: fuzzification, rulebase inference, and Table 1, where w 1 and w 2 are the beliefs assigned to antecedent
defuzzification 共Fig. 1兲. variables X 1 and X 2 , respectively.
1. Fuzzification is the process that converts a crisp input value The second step is to transfer this composite belief to the con-
to fuzzy set values, thereby forming the interface between the real sequent variable 共in this case Y兲. This transfer can also be accom-
world and the fuzzy inference process. plished in many ways. Once again there are three methods which
2. Rulebase inference is the application of a series of rules are in popular use: the Mandami method, the Larsen method, and
given in terms of fuzzy set values and producing for output vari- the Sugeno method. The first two of these methods essentially
able共s兲, fuzzy set values for each rule fired. These rules usually reconfigure the membership sets associated with the output vari-
take the form IF 共X1 is big兲 AND 共X2 is small兲, THEN 共Y is able共s兲. In our example above, if we assume that X 1 is small and
medium兲. X 2 is medium has belief 0.6 and 0.7 respectively, then if we were
3. Defuzzification converts the fuzzy set values 关of the output using the Min/Max method, the rule would assign a belief of 0.6
variable共s兲兴 back to a crisp value so that it may be utilized in the to the classification Y is positive 共it would be 0.42 if we used the
control. Bayesian method兲. Figure 3 illustrates the effect of the Mandami
and Larsen methods in this example.
The process of fuzzification consists of assigning linguistic la- The bold lines indicate how the membership of positive is
bels 共e.g. large, small, etc.兲 to a crisp input variable and determin- changed using both methods. The Mandami method uses the Min
ing the level of belief for that label for each point in the domain of of the incoming belief level 共0.6兲 and the membership curve and
the input. Unlike traditional logic, we may have levels of belief the Larsen method uses the product of the incoming belief and the
that range somewhere between 0 共false兲 and 1 共true兲. Figure 2, for membership curve.
example, illustrates three linguistic labels 共small, medium, and In the Sugeno method, the output is not a fuzzy set but rather a
large兲 being applied to the crisp variable X, which takes on values crisp numerical value. For example, a rule in this type of system
from 0 to 1000. The membership curves depicted for small, me- would have the form
dium, and large provide the level of belief 共or truth level兲 that the
variable X falls in these three categories 共or classifications兲 at each IF共 X 1 is small兲 AND 共 X 2 is medum兲 THEN 共 Y ⫽0.75兲 .
value of X throughout its domain. For example, at the value of Since the output parameter 共Y兲 is given an exact numerical value
325, X is considered small with belief w small and is considered to as a result of satisfying the rule, there is no need for assigning
be medium with belief w medium . Since the value 325 intersects membership functions to this variable.
large at an ordinate value of zero, its level of belief is 0, i.e., X The last step in the general process is to defuzzify the results of
⫽325 is definitely NOT large. each rule to form a crisp value and to aggregate all of these crisp
results to form a single crisp output to pass to the control. Since
the Sugeno systems already accomplish this directly, this step is
not required for such systems. Once again, there are many meth-
ods available to accomplish this. We will describe only two,
Fig. 2 Example membership curves Fig. 3 Mandami and Larsen methods
Fig. 4 Centroid and mean of max methods Fig. 5 Typical control law architecture
namely the Centroid and Mean of Max 共MOM兲 methods. The mon practice to govern on one rotor speed for ‘‘above idle’’ con-
centroid method takes the adjusted membership 共bold line兲 and ditions 共usually N1兲 and to idle govern on the N2 rotor speed. The
computes the centroid or first moment. This involves computing idle governor can be thought of as a limiting loop that will not let
兰 x f (x)dx, where f (x) denotes the membership function in ques- the engine drop below the minimum speed required to maintain
tion. This will produce a crisp value on the horizontal axis that the idle operating point. Since the engine Wf to rotor speed trans-
geometrically represents the balancing point for that membership fer function can usually be reduced to a first order lag with a gain,
area. The MOM method picks those abscissa values which admit the typical compensation method for the rotor speed governor is a
maximum belief points on the adjusted membership and then lead lag compensator. Once again, the output of the rotor speed
computes the mean of these points 共see Fig. 4兲. Because of the
computational complexity of the centroid method, the MOM governor is a rate of change of fuel flow.
method is typically used in real time applications. The acceleration loop 共accel loop for short兲 and deceleration
The final step is to aggregate these crisp values to form a single loop 共decel兲 are both designed to transition the engine from one
crisp output value. The method which we will employ is to take a operating point to another as quickly as possible and not exceed
weighted average where the weights will be equal to the level of any physical engine operating parameters. As such, these loops
belief in the corresponding rules 共Tuskamoto’s method兲. Thus, if are designed to be as aggressive as possible 共i.e., 4 –5 rad cross-
we had n rules R 1 ,R 2 ,...,R n with respective beliefs over兲. It is common practice that the accel and decel loops control
w 1 ,w 2 ,...,w n yielding crisp values 共after defuzzification兲 on the rate of change of N2 共N2 dot兲. Since the calculation of N2
y 1 ,y 2 ,...,y n , then the final output would be dot involves the use of some sort of derivative lag, the typical
compensation method, for the accel or decel loops, is a propor-
n tional plus integral compensator.
兺 wy
i⫽1
i i The limiting loops act as bounds on the Wfdot to the common
integrator. They represent either physical engine and/or actuator
y⫽ n .
limitations of the real engine hardware, or safety limits to avoid
兺w
i⫽1
i flameout or surge. It is common practice for a FADEC to contain
the following limiting loops.
N2 and N1 redline loops: These loops place limitations on how
It should be evident from the above descriptions that once the fast the high and low rotor spools are allowed to rotate. These
membership curves are defined and fixed and the rulebase speci- loops represent allowable maximums. The compensation for these
fied, then the fuzzy system defines a deterministic mapping be- loops is typically the same as for any other speed control loop.
tween the input variables and the output variable共s兲, and that this Temperature limit loop: This loop limits the temperature of the
relationship is neither capricious nor random. It will, like any parameter in question, usually exhaust gas temperature 共EGT兲.
mathematical relationship, yield the same output for the same in- This loop represents an allowable maximum. Since the engine Wf
put. The utility of this approach is that we have employed only to temperature transfer function can usually be reduced to a single
natural language rules and have taken advantage of the impreci- pole and zero, the typical compensation method for the tempera-
sion in the classification of the input/output variables to arrive at a ture limiter is a lead lag compensator.
quantifiable 共and easily modifiable兲 algorithm. In the fuel flow Ratio unit limit loop: This loop limits the Wf/P3 ratio units, and
selection logic described in the sequel, we will employ the Sugeno represents both maximum and minimum allowable limits. This
method because of its effectiveness in real-time applications. loop contains no compensation in the classic sense. Rather, max
and min ratio units are calculated as a function of corrected N2,
multiplied by burner pressure, and compared to the actual fuel
Typical Control Loop Architecture flow through a gain.
A high level overview of typical FADEC control law architec- Wfdot limits: The Wfdot limits are not really control loops.
ture is provided in Fig. 5. Inspection of Fig. 5 shows three cat- They represent scheduled physical limits on the rate of change of
egories of control loops, selection logic, and a common integrator. fuel flow to the engine. They represent both maximum and mini-
The reader should note that the output of each control loop is a mum allowable limits, and are usually scheduled as a function of
rate of change of fuel flow 共hereafter called Wfdot兲 which, when corrected N2.
integrated, will produce the fuel flow required to meet the perfor- The reader should note that care must be taken to ensure that a
mance criterion demanded by the engine operating condition. loop’s compensator does not ‘‘wind up’’ when it is not being se-
Control Loops. The three categories of Control loops are as lected by the selection logic. There are various techniques to ac-
follows. complish this, but these methods are not the subject of this paper
The rotor speed governing loops are designed to hold a particu- and, as such, will not be presented.
lar engine rotor speed 共N1 or N2兲 at a particular operating point. Selection Logic. This section will discuss the traditional selec-
Since the only function of these loops is to maintain an operating tion logic used in a typical FADEC. In order to facilitate this
point, they are traditionally designed very conservatively 共i.e., discussion, a simplified FADEC control law architecture will be
2–3 rad crossover兲, and to have hardly any overshoot. It is com- defined with the following loops:
Fuzzy Fuel Flow Selection Logic
The selection logic, in the present FADEC designs, grants ab-
solute authority to the selected control loop. One and only one
loop is in control at any given moment. The approach taken in the
fuzzy fuel flow selector is a departure from this classical para-
digm. In this design, each loop has a level of authority 共LA兲 which
is computed in terms of fuzzy rules and membership functions.
The sum of the LAs over all loops is unity. Thus, at any given
moment, several loops may be in control to differing levels of
authority. This approach provides a smoother transition between
loops competing for control with an attendant increase in system
performance.
Fig. 6 Traditional selection logic The fuzzy selector will be designed to perform loop selection
for the performance loops of the FADEC. These include the N1
speed governing loop, the accel loop, and the decel loop. Limiting
loops such as N2 redline and min/max fuel flow rates of change
limiting are best performed as crisp limiters 共i.e., the simplistic
• An N2 Idle Governing Loop cascade of min and max select gates兲, and are not included in this
• An N1 Governing Loop design. A simplified block diagram is provided in Fig. 7.
• Acceleration and Deceleration Loops The fuzzy selector is segmented into two stages. The first com-
• An N2 Red Line Limiting Loop putes a degree of authority for each control loop under consider-
• A Wfdot Limiting Loop ation. These values will be numbers between 0 and 1. The second
• A Ratio Units Limiting Loop stage computes a rate of change of fuel 共Wfdot兲 which combines
The simplified architecture defined above also will be used as the all of the control loop outputs in accordance to their computed
basis for the design of the fuzzy selection logic. This design will degree of authority.
be presented in the next section. The fuzzy selector inputs will consist of the loop errors and the
The traditional selection logic is a simplistic cascade of min Wfdots for each of the loops in question. Dividing each of the
select and max select logic gates. This is illustrated in Fig. 6, with loop errors by the max error for that loop will normalize these
explanations to follow. Figure 6 is provided to show the selection errors. This will provide a non-dimensional loop error in the range
in functional form. The intent here is to facilitate better under- of 关⫺1, 1兴. A similar method is used for the Wfdots.
standing. The actual implementation may be different to optimize The inputs are passed through the fuzzy rulebase, a sample of
code. which is depicted in Table 2. This table is a succinct 共though
Selection begins with a min select between the calculated Wf- functional兲 representation of the rules. The qualifiers are chosen as
dots from the N1 governor loop and the accel loop. This is more follows: large positive 共LP兲, small positive 共SP兲, zero 共ZE兲, small
easily explained if we consider the situation at steady state. At negative 共SN兲, large negative 共LN兲, positive 共POS兲, and negative
steady state the N1 request and the actual N1 are equal. The N1 共NEG兲.
loop error is zero and the Wfdot from the N1 loop is also zero. The rulebase is a combination of two sets of rules. The first set
The accel loop error, however, is very large since an accel refer- 共rules 1 through 3兲 mimics the rules from the traditional selector,
ence is being calculated, but the rate of change of N2 is zero. In but with the aid of the fuzzy qualifiers. For example, rule 1 states:
this situation the N1 loop will be chosen by the min select and, If the loop error from the N1 Governor is SN or ZE or SP and...
barring any limit exceedence, the N1 governing loop will be in ...The difference between the Wfdots from the N1 Governor Loop
control. and the Accel Loop is ZE or NEG (the N1 Governor Wfdot is
A different situation will occur if a transient to full power is smaller than the Accel Wfdot) and...
suddenly requested. In this situation the N1 loop error, and the
Wfdot from the N1 loop, will become very large. When the Wfdot ...The difference between the Wfdots from the N1 Governor Loop
from the N1 loop becomes larger than the Wfdot from the accel and the Decel Loop is ZE or POS (the N1 Governor Wfdot is
loop, the accel loop will win the min select and assume control. larger than the Decel Wfdot), then...
The accel loop will remain in control until the actual N1 gets ...The Level of Authority for the N1 Governor Loop is 1, and the
close enough to the N1 reference such that the Wfdot from the N1 Level of Authority for the Accel and Decel Loops is 0.
loop drops below the Wfdot from the accel loop. This rule 共longwinded though it may sound兲 would be exactly
The output from the first min select is sent to a max select with the same as choosing the N1 governor loop through the min select
Wfdots from the N2 idle loop and the decel loop. The N2 idle with accel and the max select with decel, if the qualifiers were
loop acts to ensure that the minimum Wfdot required to keep the crisp.
engine at idle is satisfied. This max select will not let the Wfdot to The second set 共rules 4 through 7兲 defines rules which are de-
the common integrator fall below this minimum idle limit. The signed to enhance performance. For example, rule 4 states:
reasoning for the max select with the decel loop is similar to that
of the min select with the accel loop, but for transients in the
reverse direction.
The remaining min select and max select gates are utilized to
protect against physical limit exceedence. The winning Wfdot,
from the first two selections, is min selected with the maximum
allowable Wfdots from the N2, Wfdot, and ratio units loops. The
winner of that selection is max selected with the minimum allow-
able Wfdots from the Wfdot, and ratio units loops.
Common Integrator. The purpose of the common integrator is
to integrate the winning Wfdot from the selection logic into the
Wf fuel flow command to the engine. This integrator is very often
implemented as a limited integrator to eliminate the possibility of
integrator windup. Fig. 7 Overview of fuzzy selector
Table 2 Fuzzy selector rule base
Loop errors WfDot errors

Rule Loop
# N1 Accel Decel N1-Accel N1-Decel LA⫽1
1 SN or ZE or ZE or N1Gov
ZE or NEG POS
SP
2 SN or POS Accel
ZE or
SP
3 SN or NEG Decel
ZE or
SP
4 LP SN or Accel
ZE or
SP or Fig. 9 N 1-accel „decel… Wfdot membership functions
LP
5 LP LN Decel
6 LN LN or Decel
SN or The qualifiers LN, SN, ZE, SP, LP, POS, and NEG are defined
ZE or by means of membership functions for each loop. The member-
SP ship functions for the N1 governor, and accel loops, are provided
7 LN LP Accel
in Figs. 8共a兲 and 8共b兲, respectively. The membership function for
decel is identical to that of accel. The membership functions for
the N1-accel Wfdot is provided in Fig. 9. The membership func-
tions for the N1-decel is identical to that of N1-accel.
If the loop error from the N1 Governor is LP and... The performance of the fuzzy selector will be demonstrated
with an example implementation.
The loop error from Accel is LN or SN or ZE or SP then...
The Level of Authority for the Accel Loop is 1, and the Level of An Example Implementation
Authority for the N1 governor and Decel Loops is 0.
This rule effectively keeps the control on the accel loop when The performance of the fuzzy fuel flow selection logic will be
the N1 governor position loop error is considered to be large compared to that of the traditional selection logic. This will be
positive 共according to the fuzzy definition of a large positive er- accomplished by using a non-real-time simulation model of a
ror兲. The accel loop is allowed to do its job as long as this rule is small engine control system of Hamilton Sundstrand manufacture.
satisfied. This model uses a three state, state variable engine model 共SVM兲
as the representation of the plant. The simulation of the FADEC
contains all of the control loops already presented in the sections
on selection logic. The simulation was implemented using an in-
dustry standard simulation tool.
A loop indicator is included in the simulation. This loop indi-
cator will indicate which loop is currently being chosen by the
selection logic. Table 3 is provided to define the loop indicator.
A series of test cases will provide the means of comparison of
the two selection methods. The test cases shall be limited to sea
level static, standard day conditions for brevity.
First and foremost is the case that represents a snap accel from
idle to max take-off conditions. Figures 10 and 11 show the time
response of the N1 spool speed, and the loop indicator, when TLA
is moved from the idle detent to the max take off detent in 1 sec.
The solid line represents the response when the traditional selector
is used, and the dashed line represents the response when the
fuzzy selector is used. The reader will note the performance in-
crease when the fuzzy selector is used. The typical specification
for a 1-sec acceleration from idle to max take off, at sea level
static standard day conditions, is 5 sec to reach 98 percent of the
N1 reference. When considered against this specification, the per-
Table 3 Control loop indicator definition
Loop indicator # Loop in control

1 N1 Governor
2 Acceleration
3 Deceleration
4 Not used
5 N2 idle governor
6 N2 red line
7 Not used
8 Max Wfdot limit
9 Min Wfdot limit
10 Max ratio unit limit
Fig. 8 „a… N 1 governor error membership functions; „b… accel 11 Min ratio unit limit
„decel… error membership functions.
Fig. 10 N 1 time response for idleÕMTO snap accel Fig. 13 Wfdot and limits for idleÕMTO snap accel
Fig. 11 Loop indicator for idleÕMTO snap accel Fig. 14 N 1 time response for idleÕMTO slow accel
Fig. 12 N 2 dot and accel reference for idleÕMTO snap accel Fig. 15 Loop indicator for idelÕMTO slow accel
formance increase is slightly greater than 2 percent. This is a crease. Figure 13 provides the winning Wfdot plotted alongside
modest increase, but is considered promising given the simplicity the max Wfdot limit. The reader will note that the fuzzy selector
of the rulebase. causes the winning Wfdot to come closer to the Wfdot max limit
The performance increase is realized because the control stays than the traditional selector. Indeed, this higher Wfdot is the rea-
on the accel loop longer, and smoothly transitions from the accel son any performance increase was realized at all. If the allowable
loop to the N1 governor loop according to the fuzzy ruleset. This maximum Wfdot 共the physical limitations兲 were higher, it would
is in sharp contrast to the abrupt 共and premature兲 change dictated have been possible to tune the fuzzy selector to realize an even
by the traditional selection logic. Figure 12 provides the time greater performance increase.
response of the rate of change of N2 dot and the accel reference. The same test case can be run, but this time a 10-sec transient
The reader will note that no potentially harmful overshoot of the will be used. Examination of Figs. 14, 15, and 16 共analogous to
accel reference is observed. Figs. 10, 11, and 12兲 shows virtually identical results between the
A word needs to be said about the modest performance in- traditional selector and the fuzzy selector.
Table 4 Additional test cases for fuzzy selector
Test case共s兲 Expected results

Wfdot max 共and min兲 limits Both selectors choose Wfdot
exceeded max 共or min兲 limiting loop
N2 red line limit exceeded Both selectors choose N2 red
line limiting loop
Ratio units max 共and min兲 limits Both selectors choose ratio
exceeded units max 共or min兲 limiting
loop
presence of white noise, the fuzzy selector has a tendency to stay

on the N1 governor. This additional noise immunity is gained
because the fuzzy selector is based on a series of weighted rules,
rather than simplistic max and min logic select gates.
Fig. 16 N 2 dot and accel reference for idleÕMTO slow accel
Additional test cases were run whose results will not be pre-
sented here. A description of the additional test cases, and the
expected results, are tabulated in Table 4. All expected results
were realized.
The reader will note that, in all test cases, the transient began at
the same N2 idle governing point.
Stability
The stability of a control loop is usually specified in terms of its
relative stability characteristics. The most classic of these is the
gain margin 共GM兲 and the phase margin 共PM兲. These relative
stability measures are obtained from frequency response plots of
the loop while it is in a steady-state operating condition. It is
possible, for example, to determine the relative stability of the N1
governing loop, stand alone, when it is in steady state. The rela-
tive stability for each of the control loops of the example system
presented in this paper is well known and considered sufficient.
Fig. 17 N 1 time response for idleÕMTO snap accel wÕnoise It is not possible, however, to determine the relative stability
characteristics of the fuzzy selector system while it is transitioning
from one control loop to another since, during the transition, the
system is not in steady state. The authors have chosen a very
simple test, however, to determine the stability of the entire sys-
tem at discrete transition points. A simulation of an impulse was
input into the N1 governing loop and the accel loop at three
transition points during a 1-sec accel from idle to maximum take
off. The transition points were chosen as 25 percent N1 governing
共i.e., LA for the N1 governor is 0.25兲–75 percent accel, 50 per-
cent N1 governing–50 percent accel, and 75 percent N1
governing–25 percent accel. The resulting impulse responses
demonstrate stability at all three transition points. Figure 19 pro-
vides the N1 time response for idle/MTO snap accel when an
impulse is input to accel loop at the 25 percent N1 governing–75
Fig. 18 Loop indicator for idleÕMTO snap accel wÕnoise
The reason the results are so similar is that, for this particular
case, the error in the N1 governor loop never gets sufficiently
large to trigger the performance enhancement rules of the fuzzy
selector. Therefore, the fuzzy selector exhibits behavior very simi-
lar to that of the traditional selector.
The fuzzy selector seems to be more tolerant to system noise
than the traditional selector. Figures 17 and 18 show the time
response of the N1 spool speed, and the loop indicator, when the
same 1-sec transient, outlined above, is run with white noise ap-
plied to the N1 speed signal.
Examination of Fig. 18 shows that while the traditional selector
has a tendency to momentarily jump onto the decel loop in the Fig. 19 N 1 impulse Response for idleÕMTO snap accel
percent accel transition point 共42.66 sec兲. The conclusion from LA ⫽ level of authority
Fig. 19, and the other responses not provided, is that the fuzzy LN ⫽ large negative
selector is stable. LP ⫽ large positive
The reader should note that the approximation of the impulse MTO ⫽ max take off power
function that was used for this demonstration was 10,000 percent NEG ⫽ Negative
N1 共or 10,000 percent/sec N2 dot兲 for one update of the N1 N2 ⫽ high rotor spool speed
governing and accel control logic 共33 millisec兲. N2dot ⫽ rate of change of N2
N1 ⫽ low rotor spool speed
Future Considerations N1Gov ⫽ N1 governor control loop
The authors intend to pursue the following three areas regard- POS ⫽ positive
ing the fuzzy fuel flow selector: PM ⫽ phase margin
1. The example implementation used in this paper was simplis- P3 ⫽ burner pressure
tic. Its purpose was to demonstrate the feasibility of the fuzzy fuel SN ⫽ small negative
flow selector. Future work will integrate the fuzzy fuel flow selec- SP ⫽ small positive
tor into a simulation that represents the complete implementation SVM ⫽ state variable engine model
of a typical FADEC. Since this will require the fuzzy fuel flow TLA ⫽ throttle lever angle
selector to work in the presence of additional control loops, there Wf ⫽ gas generator fuel flow
is little doubt the rulebase will need to be expanded. Wfdot ⫽ rate of change of Wf
2. The amount of program memory required in implementing ZE ⫽ zero
the fuzzy fuel flow selector, and the program execution time, re-
quires investigation.
3. The authors would like to establish a technique whereby the References
relative stability of the fuzzy fuel flow selector, during transitions, 关1兴 Zadeh, L. A., 1965, Fuzzy sets, Inform. Control Vol. 8.
may be established. 关2兴 Zadeh, L. A., 1973, Outline of a New Approach to the Analysis of Complex
Systems and Decision Processes, IEEE Transactions on Systems, Man, and
Nomenclature Cybernetics, SMC-3,共1兲共1973兲.
关3兴 Mandami, E. H., 1974, Applications of Fuzzy Algorithms for Simple Dynamic
Accel ⫽ acceleration control loop Plant, Proc. IEEE, Vol. 121, No. 12.
Decel ⫽ deceleration control loop 关4兴 Ross, Timothy J., 1995, Fuzzy Logic with Engineering Applications, McGraw-
Hill, Inc., ISBN 0-07-053917-0
EGT ⫽ exhaust gas temperature 关5兴 Volponi, Allan J. and Solomonides, Jason B., 1994, Advanced Engine Control
FADEC ⫽ full authority digital engine control Development Utilizing Fuzzy Logic and Rapid Prototyping, Royal Aeronauti-
GM ⫽ gain margin cal Society Proceedings ISBN 1 85768 021 9, November 1994.
The Use of Kalman Filter
and Neural Network
A. J. Volponi
H. DePold
Methodologies in Gas Turbine
R. Ganguli Performance Diagnostics:
Pratt & Whitney,
400 Main Street,
A Comparative Study
East Hartford, CT 06108
The goal of gas turbine performance diagnositcs is to accurately detect, isolate, and
assess the changes in engine module performance, engine system malfunctions and in-
C. Daguang strumentation problems from knowledge of measured parameters taken along the engine’s
Beijing University of Aeronautics gas path. The method has been applied to a wide variety of commercial and military
and Astronautics, engines in the three decades since its inception as a diagnostic tool and has enjoyed a
Beijing, China reasonable degree of success. During that time many methodologies and implementations
of the basic concept have been investigated ranging from the statistically based methods
to those employing elements from the field of artificial intelligence. The two most publi-
cized methods involve the use of either Kalman filters or artificial neural networks (ANN)
as the primary vehicle for the fault isolation process. The present paper makes a com-
parison of these two techniques. 关DOI: 10.1115/1.1419016兴
Introduction will briefly describe this procedure, however, for a more detailed
discussion we refer the reader to the following sources: 关4 –13兴.
The goal of gas turbine performance diagnostics is to accurately
The general approach taken for engine fault diagnostics typi-
detect, isolate, and assess the changes in engine module perfor-
cally involves the use of a linearized model approximation evalu-
mance, engine system malfunctions, and instrumentation prob-
ated at a selected engine operating point. This provides a matrix
lems from knowledge of measured parameters taken along the
relationship between changes in engine component performance
engine’s gas path. Discernable shifts in engine speeds, tempera-
共independent parameters兲 and the attendant changes in typically
tures, pressures, fuel flow, etc., provide the requisite information
measured engine parameters such as spool speeds, internal tem-
for determining the underlying shift in engine operation from a
peratures and pressures, fuel flow, etc. 共dependent parameters兲.
presumed nominal state. Historically, this type of analysis was
This relationship may be succinctly represented as
performed through the use of a Kalman filter or one of its deriva-
tives to simultaneously estimate a plurality of engine faults. In the z⫽Hx⫹ ␪ (1)
past decade, artificial neural networks 共ANN兲 have been em-
ployed as a pattern recognition device to accomplish the same where z is a vector of measured parameter deltas, x is a vector of
task. Both methods have enjoyed a reasonable success. fault deltas, H is a matrix of fault influence coefficients, and ␪ is
The purpose of this paper is to outline the two methodologies, a random vector representing the uncertainties inherent in the
discuss their relative merits and weaknesses, and provide a direct measurement process. In addition to the precision of the indi-
comparison of the two techniques via a controlled computer simu- vidual sensors, it has been customary to address the potential for
lation study. In the sequel, we will provide a brief general descrip- sensor bias and drift. Consequently, the fault vector given in the
tion of both the Kalman filter and ANN as applied to the diagnos- model above is often configured to contain components directly
tic problem. For the purpose of conducting a comparison, we will related to sensor error in addition to engine fault deltas.
limit the framework of the diagnostic system to the problem of The fault vector x given in the model can be thought of as the
isolating a single fault to the component level. The single faults concatenation of an engine fault vector (x e ) and a sensor 共error兲
under consideration will be comprised of engine module, engine fault vector x S , i.e., x⫽ 关 x e ]x s 兴 T where
冋册冋册
system, and instrumentation faults.
⌬ ␩ FAN ⌬N1 err
Kalman Filter Approach ⌬⌫ FAN ⌬ P3 err
Kalman filter methods were introduced as a fault isolation and x e ⫽ ⌬ ␩ CH and x s ⫽ ⌬W f err . (2)
assessment technique for relative engine performance diagnostics ] ]
in the late 1970s and early 1980s 共关1–3兴兲. The success enjoyed in ⌬A 5 ⌬T49err
these early programs promoted the use of these techniques in sub-
sequent years to become the central methodology utilized in many We may rewrite Eq. 共1兲 as
冋册
current engine performance analysis programs. In the sequel we
xe
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN z⫽H e x e ⫹H s x s ⫹ ␪ ⫽ 关 H e ] H s 兴 ¯ ⫹ ␪ ⫽Hx⫹ ␪ . (3)
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF xs
tional Gas Turbine and Aeroengine Congress and Exhibition, Munich, Germany, May
8 –11, 2000; Paper 2000-GT-0547. Manuscript received by IGTI November 1999;
The matrix H has been partitioned into two parts; a matrix of
final revision received by ASME Headquarters February 2000. Associate Editor: D. engine fault influence coefficients (H e ) and a matrix of sensor
Wisler. fault influence coefficients (H S ). The generation of these matrices
Table 1 Sample root cause influences
T49C2 WF N2C2 N1C2 P25Q2 T25C2 T3C2 P3Q2

°C percent percent percent percent °C °C percent
FAN 3.86 0.70 0.30 ⫺0.68 ⫺2.00 ⫺1.95 ⫺1.58 ⫺0.03
LPC ⫺4.54 ⫺0.66 ⫺0.29 ⫺0.14 1.18 0.11 ⫺2.62 0.01
HPC ⫺6.80 ⫺0.80 0.06 ⫺0.05 ⫺0.83 ⫺0.71 ⫺3.66 0.17
HPT ⫺10.88 ⫺1.29 0.57 ⫺0.08 ⫺1.29 ⫺1.14 4.03 1.26
LPT ⫺1.19 0.96 ⫺0.63 0.98 3.40 3.45 ⫺1.42 0.11
2.5 BLD ⫺3.07 ⫺0.49 ⫺0.16 0.00 1.04 0.85 ⫺0.86 0.00
FP14 1.22 0.21 0.07 ⫺0.24 ⫺0.67 ⫺0.73 0.15 ⫺0.01
FP8 ⫺0.61 ⫺1.39 ⫺0.17 ⫺0.64 ⫺1.06 ⫺1.31 ⫺2.62 ⫺1.09
2.9 BLD ⫺4.22 ⫺1.06 ⫺0.29 ⫺0.06 0.68 0.63 0.60 0.02
TCC 17.75 2.10 ⫺0.90 0.12 2.14 1.86 ⫺4.38 ⫺1.11
HPCSVM ⫺0.95 ⫺0.11 0.39 0.00 ⫺0.08 ⫺0.09 0.34 0.09
P49 Error ⫺0.33 ⫺1.70 ⫺0.25 ⫺0.46 ⫺0.55 0.63 ⫺0.21 ⫺1.22
and their interpretation have been discussed in great detail in Single Fault Isolation. The Kalman filter can be configured
关4,14兴 and elsewhere in the literature and will not be pursued in to emulate a single fault isolator (SFI). This is a snapshot type of
this paper. analysis in the sense that it operates on a set of measurement
The model, as configured above, has often been used for the deltas without any a priori information or pre-history. The object
purpose of tracking slowly occurring changes in engine perfor- of the analysis is to flag a root cause on the basis of a single
mance from revenue flight data, through the use of a Kalman measurement delta set. Root causes are single fault occurrences
filter-based methodology. An estimate for these performance and are pre-defined for the system. They consist of coupled faults
shifts, x̂, would be given by within the major modules of the engine, certain system faults such
冋册冋册冋册冉冋册冊
as handling and ECS bleed leaks and failures, variable stator vane
x̂ e x̄ e De x̄ e malfunctions, TCC malfunctions as well as certain instrumenta-
¯ ⫽x̂⫽x̄⫹D 共 z⫺Hx̄ 兲 ⫽ ¯ ⫹ ¯ z⫺ 关 H e ]H s 兴 ¯ tion faults. These faults are assumed to occur in isolation, i.e.,
x̂ s x̄ s Ds x̄ s there will be one and only one root cause occurrence at any given
time. The purpose of the single fault isolator is to identify the
correct root cause once a trend shift is detected.
x̂⫽prediction⫹GAIN共residual兲⫽prediction⫹correction (4)
Root Causes. Root Causes can be thought of as state vari-
where x̄ represents an a priori estimate of the engine/sensor fault ables, x 1 ,x 2 , . . . ,x n . They are represented within this system as
deltas and D is the 共Kalman兲 gain matrix referred to as the diag- vectors of measurement deltas, z * 1 ,z *
2 , . . . ,z n* , which are calcu-
nostic matrix. The diagnostic matrix is computed as a function of lated by applying influence coefficients. As an illustration, we can
several quantities; the engine/sensor influence coefficients H, the consider the following 12 root causes. These root causes may exist
measurement covariance matrix R, and a positive semidefinite
at varying levels. The state representation of the root cause will be
weighting matrix P0 . The diagnostic matrix is computed as
in the form of a one percent cause to be consistent with other
冋册冋册
influences.
De P e0 H Te 共 H e P e0 H Te ⫹R 兲 ⫺1 1. FAN coupled FAN 共⫺1 percent ␩, ⫺1.25 percent FC兲
D⫽ ¯ ⫽ P 0 H 共 H P 0 H ⫹R 兲 ⫽
T T ⫺1 ¯¯ 2. LPC coupled LPC 共⫺1 percent ␩, ⫺1.10 percent FC兲
3. HPC coupled HPC 共⫺1 percent ␩, ⫺0.80 percent FC兲
Ds P s0 H sT 共 H s P s0 H sT ⫹R 兲 ⫺1 4. HPT coupled HPT 共⫺1 percent ␩, ⫹0.75 percent FP4兲
(5) 5. LPT coupled LPT 共⫺1 percent ␩, ⫹1.65 percent FP45兲
6. 2.5 BLD stability bleed leak 共one percent兲
7. 2.9 BLD start bleed 共one percent兲
where P (0e )andP 0(s )
are weighting submatrices for the engine and 8. FP14 fan discharge area 共one percent兲
sensor fault estimation, respectively. An in-depth report on the 9. FP8 core discharge area 共one percent兲
generation of the P 0 and R matrices can be found in references 10. TCC turbine case cooling 共on兲
11. HPCSVM HPC stator vane misrigging
关4,5兴. 12. P49 Error P49 indication problem 共two percent兲
The use of the predictor/corrector methods like the Kalman An example of the influences for these root causes is depicted in
filter to estimate sensor error has made possible a more reliable
Table 1.
and consistent gas turbine module performance analysis. The pro-
cedure outlined above can be applied in a snapshot analysis or as The actual root cause may appear as some multiple of the in-
a continuing recursive analysis as new engine data are made avail- fluences represented in Table 1. For instance, a 2.5 bleed root
able over time. In either scenario, the simultaneous determination cause may manifest itself as a stuck open bleed 共⬇15 percent 2.5
of both engine faults and measurement errors by this methodology bleed fault兲 or a partial bleed leak, say two percent 2.5 bleed fault.
has been successfully applied to a large number of commercial These two faults will be treated as the same by the Kalman esti-
and military families of engines with varying instrumentation mator. The difference between the two faults is one of magnitude.
suites for two decades. The magnitude is estimated by the Kalman filter. Its ability to
In the sequel, we shall alter our emphasis to consider the prob- estimate correctly will depend on the signal to noise ratio for the
lem of fault isolation, given the premise that a fault event has been given fault. In this particular case, a two percent 2.5 bleed is
detected. The problem of detection becomes one of recognizing a
sometimes confused with an LPC fault or a 2.9 Bleed fault. On the
step or rate change in a gas path parameter or a collection of
parameters. The problems associated with fault detection and the other hand, a stuck 2.5 Bleed 共14.77 percent兲 has a significantly
mechanisms which can be applied to accomplish this task have higher S/N ratio and is estimated correctly. Another factor which
been reported by 关15兴. The types of faults that will be considered impacts accuracy of the estimator is the number of measured pa-
in this discussion include engine performance faults, engine sys- rameters. We will consider systems that have between four and
tem faults, and instrumentation faults. eight measurements in flight.
Given the definitions above for root cause influences 共one per- Table 2 Single fault problem set
cent兲, a typical set of possible single faults may be constructed.
Fault percent
For purposes of this study, we will consider the following single
fault definitions, depicted in Table 2. FAN ⫺2
If we represent these 11 faults as, x 1 ,x 2 , . . . ,x 11 , and the ma- LPC ⫺2
HPC ⫺2
trix of influences by H * ,(9⫻8) 1 matrix depicted above then a set HPT ⫺2
of expected measurement delta vectors, z * 1 ,z 2* , . . . ,z *
11 , would LPT ⫺2
be calculated by 2.5 BLD 共low and high兲 2 and 14.77
2.9 BLD 共low and high兲 6.74 and 15.45
z i* ⫽ 共 H * 兲 T x i for i⫽1,2, . . . ,11. (6) HPCSVM ⫺6
P49 Error 2
Each z i* would be an 8⫻1 vector of measurement deltas.
The general form of the discrete Kalman filter estimator at time
k⫹1 is as follows:
State extrapolation: As mentioned above, the normalized measurement error will
take into consideration the measurement nonrepeatabilities of the
x̂ 共 k⫹1 兩 k 兲 ⫽⌽ 共 k⫹1 兲 x̂ k . system. It is assumed that these are known a priori or computed
Covariance extrapolation: from data during initialization of the diagnostic system. However
the values are obtained, they are assumed to be known and are
P 共 k⫹1 兩 k 兲 ⫽⌽ 共 k⫹1 兲 P k ⌽ T 共 k⫹1 兲 ⫹Q k⫹1 . passed to the Kalman filter in the form of a positive definite ma-
trix R. The diagonal elements of this matrix represent the vari-
Kalman gain:
ances of the measurement deltas 共corrected quantities兲. Thus, if
D k⫹1 ⫽ P 共 k⫹1 兩 k 兲 H k⫹1
T
共 H k⫹1 P 共 k⫹1 兩 k 兲 H k⫹1
T
⫹R k⫹1 兲 ⫺1 . we represent the measurement delta-delta vector 共between time k
and k⫹1, assuming that a trend shift has been detected during this
State update: time period兲 by Z⫽ 关 z 1 ,z 2 , . . . ,z 8 兴 , then the diag(R)
x̂ k⫹1 ⫽x̂ 共 k⫹1 兩 k 兲 ⫹D k⫹1 关 z k⫹1 ⫺H k⫹1 x̂ 共 k⫹1 兩 k 兲兴 ⫽关␴21,␴22, . . . , ␴ 28 兴 , represents the individual variances. For defi-
niteness, assume we have calculated the nth single fault estimate.
Covariance update: The associated normalized measurement error norm, e n , is calcu-
冋册
lated as follows:x̂ n ⫽SFI estimate for the nth root cause
P k⫹1 ⫽ 关 I⫺D k⫹1 H k⫹1 兴 P 共 k⫹1 兩 k 兲 .
冉冊
8 1/2
z k ⫺z k* 2
For the single fault isolator we make the following adjustments:
兺 ␴k
⌽ k⫹1 ⫽I e n⫽
k⫽1
(7)
8
Q k⫹1 ⫽0 兺
k⫽1
z 2k
H k⫹1 ⫽M c 共 H * 兲 T
k ⫽kth element of the vector M c (H * ) x̂ n .
T
z*
M c ⫽measurement configuration vector⫽diag共 m 1 ,m 2 , . . . ,m 8 兲 It is possible for the estimated values for some single faults to
be opposite in polarity than what would be reasonably expected.2
共diagonal matrix assuming eight potential measurements兲 For example, an HPT SF of magnitude ⫺2 percent might yield an
where m j ⫽ 再 1
0
if jth measurement is available
otherwise.
estimated value of ⫹0.5 percent for an LPC SFI. Given that a
sudden shift in observed gas path parameters has taken place dur-
ing engine operation, it is not likely that the condition of any
The effect of the measurement configuration matrix is to zero given module has improved and thus a positive shift in perfor-
the rows of the root cause influence matrix corresponding to the mance would not be given serious consideration. Thus, some pre-
measurements that are NOT available. The resulting matrix of processing is mandated before the error ranking is performed.
influences is 8⫻9 for this example. This takes the form of perusing each of the SFI estimates and
The single fault isolation is obtained by processing the general considering only those which admit a reasonable polarity. Order-
Kalman equations iteratively to provide a snapshot analysis for ing the errors from min to max we obtain
each of the root causes under consideration 共11 in this example兲. e i 1 ⭐e i 2 ⭐¯⭐e i n (8)
Each ‘‘call’’ to the Kalman filter will be made with a different P 0
matrix chosen to accentuate the kth root cause. Since these are corresponding to the single fault estimates
snapshot analyses, the covariance update calculation is not re-
quired. The a priori state estimate is also assumed to be zero. An x̂ i 1 ⭐x̂ i 2 ⭐¯⭐x̂ i n (9)
SFI analysis is performed typically after a trend shift has been in order of likelihood. In most cases, the first of these ranked
detected in the measurement deltas at some discrete time, say, single faults is deemed to be the underlying fault and is reported
between time k and k⫹1. The ‘‘delta-delta’’ (Z k⫹1 ⫺Z k ) consti- to the user. In some instances, however, it may be prudent to
tutes the input measurement delta, Z, to the SFI. report the first and second SFs, since there exists the possibility
The SFI is evaluated iteratively for each single fault 共11 in the for an erroneous fault identification, especially if the associated
above example兲. This process will yield estimates for each single error norms are ‘‘close.’’
fault under consideration. It is necessary to rank each of these Depending on the circumstances, the SFI may experience a
estimates and determine the top two or three single faults. The confusion between two single faults. The aliasing of SFs depends
measure used to compare estimates for this purpose of ranking is on the single faults themselves, their relative magnitude with re-
a normalized measurement error norm which we will describe spect to the nonrepeatability of the measurements and the number
below. The single fault admitting the minimum error is deemed of measurements available for the analysis. All of these factors
the most likely, the second smallest error, the next likely, and so can contribute to a confounding of the underlying SF with another
forth.
2
Although this is not a common occurrence, it is possible for an SF estimate
1
The row dimension of H * is 9 共and not 11兲 since the BLD25 and BLD29 root indicating an improvement in performance to admit the minimum measurement error
causes are used to model the two different magnitude faults being considered. norm and hence win the isolation selection.
Table 3 SFI accuracy: Kalman filter, 8 measurements entries indicate the percentage of time the implanted fault is cor-
rectly annunciated as first, second, and third choice.
8 Measurement Set
The results demonstrate a 96.9 percent and 91.1 percent accu-
FAULT 1st Top 2 Top 3 racy for the first choice fault isolation for eight and four measure-
ment systems, respectively. If we weaken the accuracy criteria to
FAN 100% 100% 100%
LPC 90% 100% 100% that of being correct within the top two choices, the relative pre-
HPC 100% 100% 100% cision of the isolation increases to 100 percent and 96.1 percent.
HPT 100% 100% 100% In either case, the isolation accuracy is quite good.
LPT 100% 100% 100% The fact that we can obtain a 91 percent hit rate with only four
2.5 BLD 85% 100% 100%
2.9 BLD 96.7% 100% 100% measurements raises the question as to why one would ever con-
HPCSVM 100% 100% 100% sider adding more instrumentation. There are several answers to
P49err 100% 100% 100% this question, one being that a multiple fault performance assess-
Total 96.9% 100% 100% ment would not perform as well with fewer measurements. But
even within the constraints of this study 共i.e., single fault annun-
ciation兲, the ‘‘more the better’’ argument still holds if we consider
Table 4 SFI accuracy: Kalman filter, 4 measurements the robustness of the system. One measure of robustness would be
to vary the module couplings 共flow capacity versus efficiency,
4 Measurement Set hard coupled relationship兲 in the implanted faults beyond that
FAULT 1st Top 2 Top 3 which is assumed in the numerical model 共i.e., the influences co-
efficients H * ). Figures 1 and 2 depict the impact on first choice
FAN 100% 100% 100% accuracy for eight and four measurement systems, respectively,
LPC 90% 90% 100%
HPC 100% 100% 100% when the coupling factor for the compression modules 共FAN,
HPT 100% 100% 100% LPC, HPC兲 are randomly varied.4
LPT 100% 100% 100% The LPC and HPC clearly exhibit greater robustness to model-
2.5 BLD 50% 75% 95% ing assumptions in the 8 measurement system.
2.9 BLD 80.0% 100.0% 100%
HPCSVM 100% 100% 100% In the sequel, we will briefly discuss artificial neural networks
P49err 100% 100% 100% 共ANNs兲, and their usage in engine performance diagnostics. The
Total 91.1% 96.1% 99.4% discussion will be intentionally brief since there already exists
ample documentation of this methodology in the literature, and
the reader is directed to the references for additional detail. For
the purpose of making a direct comparison with the Kalman
SF. To mitigate the possibility of a false identification, some post- filter methodology, we will confine much of the discussion to the
processing of the SFI results may be required. This will improve single fault isolation problem introduced in the prelude. The iden-
the overall accuracy of the system. The rules to be applied would tical computer simulation test will be used as the vehicle for the
be empirically motivated and would be suggested by computer evaluation.
simulation test cases. For the problem set considered in this study
共Table 2兲, it was not necessary to undergo this type of post- Artificial Neural Network Approach
processing.
The fault isolation problem can be considered to be a pattern
Computer Simulation Results. Using Equation 共1兲, we can classification problem. N-dimensional vectors in an
generate a set of hypothetical ‘‘noisy’’ measurements for each of N-dimensional space represent the system response. The system
the 11 single faults under consideration (x i ,i⫽1,2, . . . 11). These response for different faults tends to be partitioned into different
noisy measurement vectors (z k ) are then passed through the SFI regions of this space and can be regarded as patterns. Pattern
process, the results ranked and tabulated for first, second, and recognition involves learning these partitions, from simulated or
third ranked faults. These results appear in Table 3 and 4 for real data, so that a given system response can be classified as a
eightand four measurement set3 systems, respectively, where the particular fault. Neural networks represent a powerful pattern rec-
four bleed leak faults have been combined into two faults. The ognition technique, and have been applied for fault detection of
3
4
The turbine modules’ 共HPT and LPT兲 isolation accuracy was essentially unaf-
The four measurements set used in this study consisted of the flight required fected and hence were not plotted.
parameters (T49, Wf, N1 and N2).
Fig. 1 Coupling factor impact on 8 measurement system
Fig. 2 Coupling factor impact on 4 measurement system
complex systems such as aerospace vehicles 共关16兴兲, nuclear power The discussion below involves the use of two types of neural
plants 共关17兴兲, and chemical process facilities 共关18兴兲 among others. networks for engine fault diagnostics, a feed forward ANN trained
A key advantage of neural networks over other methods is their using back-propagation 共BP兲 algorithm and a hybrid neural
ability to recognize relationships between patterns despite the network.
presence of noise contamination and/or partial information 共关19兴兲.
Most applications of ANN to fault diagnostics follow a com- Back-Propagation „BP… Algorithm. While there are several
mon process. The ANN is trained off-line on fault signatures re- types of neural networks, multilayer feedforward networks trained
lating the changes in system measurements from a ‘‘good’’ base- using the backpropagation algorithm have emerged as the most
line to system faults. Typically, faults are embedded into the widely used. Figure 3 illustrates the schematic of a feedforward
computer simulations, or real fault data is used, or a combination neural network, which consists of an input layer, an output layer,
of both. In case simulated data is used, noise must be added to and one or more hidden layers. The number of neurons in the
make the simulations realistic. Such a training process where the input and output layers is determined by the number of input
ANN is presented input and output data by the system designer is measurements and output parameters. The number of hidden layer
known as supervised learning. Once the ANN has been properly nodes is selected based on convergence criterion and the charac-
trained using this process of supervised learning, it can analyze teristic of the input-output mapping relationship.
data that are different from those it was originally exposed to A three-layer feedforward network is used for the present work,
during the training sessions. When the trained ANN is placed as shown in Fig. 1. The feedforward network is trained using
on-line, it recognizes a similar response from the actual system. supervised learning, which involves presenting input-output pairs
to the ANN and then using the BP algorithm to learn the relation-
ships between the inputs and outputs by minimizing the following
error measure:
N
E⫽ 兺E
k⫽1
k (10)
in which E k represents the root mean square error associated with

the kth training sample and where N represents the number of
samples that are used for training the network. The BP algorithm
uses a gradient search to perform the nonlinear optimization
needed to minimize the error. Further details about the BP algo-
rithm are outlined in standard texts on neural networks 共关20兴兲.
To improve the ANN’s ability to deal with data scatter, the input
data were normalized using the following formula:
Table 5 SFI accuracy: ANN, 8 measurements
8 Measurement Set
FAULT 1st Top 2 Top 3
FAN 100% 100% 100%
LPC 60% 80% 90%
HPC 100% 100% 100%
HPT 100% 100% 100%
LPT 100% 100% 100%
2.5 BLD 80% 85% 85%
2.9 BLD 86.7% 86.7% 86.7%
HPCSVM 90% 100% 100%
P49err 100% 100% 100%
Total 90.7% 94.6% 95.7%
Fig. 3 Architecture of three layer feedforward ANN
Table 6 SFI accuracy: ANN, 4 measurements Table 8 SFI accuracy: hybrid NN, 4 measurements
4 Measurement Set 4 Measurement Set

FAULT 1st Top 2 Top 3 FAULT 1st Top 2 Top 3
FAN 100% 100% 100% FAN 90% 100% 100%
LPC 90% 90% 90% LPC 90% 90% 100%
HPC 90% 100% 100% HPC 100% 100% 100%
HPT 100% 100% 100% HPT 100% 100% 100%
LPT 100% 100% 100% LPT 100% 100% 100%
2.5 BLD 75% 75% 75% 2.5 BLD 70% 90% 100%
2.9 BLD 86.7% 86.7% 86.7% 2.9 BLD 86.7% 86.7% 86.7%
HPCSVM 100% 100% 100% HPCSVM 70% 100% 100%
P49err 100% 100% 100% P49err 100% 100% 100%
Total 93.5% 94.6% 94.6% Total 91.1% 97.8% 100%
Y in ⫽Y im / 共 Y i max• ␴ i 兲 training data based on influence coefficients, the Hybrid ANN

uses the influence coefficients as part of the network model.
where Y i is the ith monitoring parameter; n, m, and max are the A Gaussian nearest neighbor function was substituted for the
normalized, measured, and maximum possible value, respectively; ANN feature identification function and the fault mapping func-
␴ i is the standard deviation of the ith monitoring parameter. tion was ignored for this Hybrid NN algorithm. The output of the
The standard deviations and the influence coefficients used for network was the root-sum-of-the-squares number of standard de-
the ANN testing are the same as those used for the Kalman filter viations from a perfect match of the fault pattern. The output of
described previously. For comparison of fault isolation results the network was used directly to rank the faults.
with four inputs and eight inputs, 20 training cases and 50 testing A test was made to determine if the neural network mapping to
cases were generated. The training cases were used for the BP the faults could also be optimized. Up to 72 weightings were
algorithm to train the neural network. Once the neural network available between the eight measurement features and the nine
was trained, the test cases were used to evaluate the performance faults. Thirty-six weightings were available with four measure-
of the neural network. Data used for training was not used for ments. While it was demonstrated that the weightings could be
testing the neural network. The diagnostic results were considered optimized, the tests were run with all the weightings set to unity.
for the three highest outputs, which represent the three most likely The results from the Hybrid neural net are shown in Tables 7
faults. and 8, for eight and four measurements, respectively. The accu-
The test results for eight and four measurements using the BP racy of the Hybrid neural network compares favorably with the
ANN are shown in Table 5 and Table 6, respectively. For both the Kalman SFI. For the eight measurements case, both the Kalman
eight and four measurement cases, the Kalman SFI results shown SFI and the Hybrid ANN show a fault isolation accuracy among
in Tables 5 and 6 are better than the BP ANN results. For the eight the top three choices of 100 percent. For the four measurement
measurements case, the Kalman SFI has a 100 percent accuracy in case, the Kalman SFI shows a fault isolation accuracy among the
fault isolation among the top three choices, compared to 95.7 top three choices of 99.4 percent compared to 100 percent for the
percent for the BP ANN. For the four measurements case, the Hybrid ANN.
Kalman SFI has an accuracy of 99.4 percent in fault isolation In general, the Hybrid ANN gives better results than the BP
among the top three choices, compared to 94.6 percent for the ANN. This may be because the hybrid uses the influence coeffi-
BP ANN. cients for each fault within the network, whereas the BP ANN has
Hybrid Neural Network Algorithm. The BP ANN is handi- to learn the influence coefficients from the training data.
capped relative to the Kalman SFI in some ways. For example,
while the Kalman SFI uses influence coefficients in the form of
the H matrix to define the model, the BP ANN uses data generated Kalman Filter and Neural Network Methods
from influence coefficients to ‘‘learn’’ the model. Feedforward neural networks are typically made of intercon-
A Hybrid NN is a network architecture where one or more nected nonlinear neurons and are particularly useful where the
ANN functions are replaced by an algorithm that includes domain input-output relationship is nonlinear. The Kalman filter and the
knowledge 共关21兴兲. The objective is to substitute features in the Hybrid neural network assume a linearized model of the system.
neural network architecture that are already analytically under- In the present application, the linear model takes the form of
stood. This avoids the need for training the ANN to learn infor- influence coefficients. For a feedforward ANN, BP learning for
mation which is already known. For example, instead of using linear problems in engine fault diagnostics will generally result in
poorer performance compared to a Kalman filter or hybrid ANN
approach.
Table 7 SFI accuracy: hybrid NN, 8 measurements Learning in neural networks involves mapping an input to an
output. The input and outputs can be generated from models, from
8 Measurement Set real data, or a combination of both. ANNs can therefore also be
model-free estimators, a quality that is very useful if modeling
FAULT 1st Top 2 Top 3
information such as influence coefficients are not available. Kal-
FAN 100% 100% 100% man filters, on the other hand, are model-based estimators, and are
LPC 90% 100% 100% suitable for problems in engine performance diagnostics where
HPC 100% 100% 100%
HPT 100% 100% 100% influence coefficients are available as the model.
LPT 100% 100% 100% Neural networks are not limited to multilayer neurons with BP
2.5 BLD 90% 100% 100% training. Self-organizing maps based on competitive learning
2.9 BLD 100% 100% 100% 共关22兴兲, simulated annealing based on statistical thermodynamics
HPCSVM 100% 100% 100%
P49err 100% 100% 100% 共关23兴兲, Boltzmann learning 共关24兴兲 and radial basis functions 共关25兴兲
Total 97.8% 100% 100% are some of the other developments in ANN which may be appli-
cable to engine diagnostics. Combining the Kalman filter ap-
proach with some of the ANN methods may yield superior results P3 ⫽ HPC exit pressure
for the engine diagnostic problem, than those obtainable within P49 ⫽ exhaust pressure
either methodology acting alone. HPCSVM ⫽ variable stator vane
SFI ⫽ single fault isolator
Concluding Remarks H ⫽ influence coefficient matrix
Test results have suggested that the back-propagation neural D ⫽ Kalman gain matrix
network, the hybrid neural network, and the Kalman filter method R ⫽ measurement covariance matrix
are highly accurate for isolating single gas turbine fault symp- P ⫽ state covariance matrix
toms. Furthermore, the results also indicate that these methodolo- ⌽ ⫽ state transition matrix
gies compare favorably in terms of accuracy, with a very slight x ⫽ state ⌬ vector
advantage going to the Kalman filter approach. z ⫽ measurement ⌬ vector
Each method has its own advantages and disadvantages. ␩ ⫽ efficiency ⌬
The ANN is inherently nonlinear and can be used in applica-
⌫ ⫽ flow capacity ⌬
tions where model information is scarce or lacking altogether. The
ANNs are, however, data driven and therefore must be trained. FC ⫽ flow capacity
The training is typically performed offline in a supervised fashion, A ⫽ turbine nozzle area ⌬
meaning that the input-output relationship is known. In the present Subscripts
application this means that the underlying fault共s兲 in the training
data are already known. This could be a drawback if real engine e ⫽ relating to engine parameters
data was used for training, since the precise disposition of the s ⫽ relating to sensor parameters
fault may or may not be known. If the engine configuration and/or 0 ⫽ initial condition
instrumentation noise levels change, the ANN approach would k ⫽ discrete time k
require that a re-training be performed. Once trained, the ANN Superscripts
architecture provides a numerically simple, and hence fast diag-
nostic operator suitable for real-time application. T ⫽ matrix transpose
The Kalman filter is a linear model-based estimator and is suit- ⫺1 ⫽ matrix inverse
able in those instances where a linear model is available and is
known to be a reasonably accurate representation of the input-
output relationship. In the engine performance diagnostics appli-
cation, influence coefficients have, historically, been fairly accu-
rate and robust linear models. The Kalman filter approach utilizes
References
all model information available 共a priori estimate information, 关1兴 Urban, L. A., 1974, ‘‘Parameter Selection for Multiple Fault Diagnostics of
measurement noise information, etc.兲 and can be easily configured Gas Turbine Engines,’’ AGARD Conference Proceedings, No. 165, NATO,
Neuilly-sur-Seine, France.
to operate with different measurement suites, and fault configura-
关2兴 Urban, L. A., 1972, ‘‘Gas Path Analysis Applied to Turbine Engine Condition-
tions: single fault 共as in this present study兲 or multiple fault iso- ing Monitoring,’’ AIAA/SAE Paper 72-1082.
lation systems. Adaptive measures are also available to allow real- 关3兴 Volponi, A., 1983, ‘‘Gas Path Analysis: An Approach to Engine Diagnostics,’’
time reconfiguration of the Kalman filter to changing Time-Dependent Failure Mechanisms and Assessment Methodologies, Cam-
measurement noise levels. bridge University Press, Cambridge, UK.
The hybrid neural network, like the Kalman filter, is a model- 关4兴 Volponi, A. J., and Urban, L. A., 1992, ‘‘Mathematical Methods of Relative
based method utilizing influence coefficients as the primary linear Engine Performance Diagnostics,’’ SAE Trans., 101; Journal of Aerospace,
Technical Paper 922048.
model in a neural network architecture framework. For the single-
关5兴 Doel, D. L., 1992, ‘‘TEMPER—A Gas Path Analysis Tool for Commercial Jet
fault isolation problem referenced in this study, the hybrid NN Engines,’’ ASME Paper 92-GT-315.
closely resembles a weighted least squares solution. This explains 关6兴 Doel, D. L., 1993, ‘‘An Assessment of Weighted-Least-Squares Based Gas
the close agreement between the hybrid NN and Kalman filter in Path Analysis,’’ ASME Paper 93-GT-119.
the single-fault isolation computer simulation study. 关7兴 Stamatis, A. et al., 1991, ‘‘Jet Engine Fault Detection With Discrete Operating
Points Gas Path Analysis,’’ J. Propul. Power, 7, No. 6.
关8兴 Merrington, G. L., 1993, ‘‘Fault Diagnosis in Gas Turbines Using a Model
Acknowledgment Based Technique,’’ ASME Paper 93-GT-13.
The authors wish to thank Zhang Jin, Zhang Mingchuan, and 关9兴 Glenny, D. E., 1988, ‘‘Gas Path Analysis and Engine Performance Monitoring
In A Chinook Helicopter,’’ AGARD: Engine Condition Monitoring, NATO,
Zhu Zhili of the Department of Jet Propulsion and Power at the Neuilly-sur-Seine, France.
Beijing University of Aeronautics and Astronautics as well as Lu 关10兴 Winston, H., et al., 1991, ‘‘Integrating Numeric and Symbolic Processing For
Pong-jeu and Hsu Tzu-cheng of the Institute of Aeronautics and Gas Path Maintenance,’’ AIAA Paper 91-0501.
Astronautics, National Cheng Kung University in Taiwan for their 关11兴 Luppold, R. H., et al., 1989, ‘‘Estimating In-Flight Engine performance Varia-
work in developing the ANN Fault Analysis Application under tions Using Kalman Filter Concepts,’’ AIAA Paper 89-2584.
contract with Pratt & Whitney. 关12兴 Gallops, G. W., et al., 1992, ‘‘In-Flight Performance Diagnostic Capability Of
An Adaptive Engine Model,’’ AIAA Paper 92-3746.
关13兴 Kerr, L. J., et al., 1991, ‘‘Real-Time Estimation of Gas Turbine Engine Dam-
Nomenclature age Using a Control Based Kalman Filter Algorithm,’’ ASME Paper 91-GT-
ANN ⫽ artificial neural network 216.
关14兴 ‘‘Sensor Error Compensation in Engine Performance Diagnostics,’’ 1994,
FAN ⫽ fan module ASME Paper No. 94-GT-058.
LPC ⫽ low pressure compressor 关15兴 DePold, H., and Gass, F. D., 1998, ‘‘The Application of Expert Systems and
HPC ⫽ high pressure compressor Neural Networks to Gas Turbine Prognostics and Diagnostics,’’ ASME Paper
HPT ⫽ high pressure turbine 98-GT-101.
LPT ⫽ low pressure turbine 关16兴 McDuff, R. J., and Simpson, P. K., 1990, ‘‘An Investigation of Neural Network
2.5 BLD ⫽ stability bleed for F-16 Fault Diagnosis,’’ Proceedings of the SPIE Technical Symposium on
Intelligent Information Systems, The International Society for Optical Engi-
2.9 BLD ⫽ start bleed
neering, Bellingham, WA.
TCC ⫽ turbine case cooling 关17兴 Guo, Z., and Uhhrig, R. E., 1992, ‘‘Using Modular Neural Networks to Moni-
N1 ⫽ low spool speed tor Accident Conditions in Nuclear Power Plants,’’ Proceedings of the SPIE
N2 ⫽ high spool speed Technical Symposium on Intelligent Information Systems, The International
T49 ⫽ exhaust gas temperature Society for Optical Engineering, Bellingham, WA.
关18兴 Wantanabe, K. et al., 1989, ‘‘Incipient Fault Diagnosis of Chemical Processes 关23兴 Kirkpatrick, S., 1984, ‘‘Optimization by Simulated Annealing: Quantitative
via Artificial Neural Network,’’ AIChE J., 35, No. 11. Studies,’’ J. Stat. Phys., 34.
关19兴 Holmstrom, L., and Koistinen, P., 1992, ‘‘Using Additive Noise in Back Propa- 关24兴 Ackley, D. H., Hinton, G. E., and Sejnowski, T. J., 1985, ‘‘A Learning Algo-
gation Training,’’ IEEE Trans. Neural Netw., 3, No. 1. rithm for Boltzmann Machines,’’ Cognitive Science, Vol. 9, Cognitive Science
关20兴 Haykin, S., 1994, Neural Networks—A Comprehensive Foundation, Macmil- Society, Cincinnati, OH.
lian, New York. 关25兴 Broomhead, D. S., and Lowe, D., 1988, ‘‘Multivariate Function Interpolation
关21兴 Dasarathy, B. V., 1991, Nearest Neighbor (NN) Norms: NN Pattern Classifi- and Adaptive Networks,’’ Complex Systems, Vol. 2, Complex Systems Publi-
cation Techniques, IEEE Computer Society Press, New York. cations, Champaign, IL.
关22兴 Kohonen, T., 1990, ‘‘The Self-Organizing Map,’’ Proc. IEEE, 78, IEEE, New 关26兴 Cybenko, G., 1989, ‘‘Approximations of Superposition of a Sigmoidal Func-
York. tion,’’ Mathematics of Control, Signals and Systems, Vol. 2.
Active Combustion Instability
Control With Spinning Valve
Actuator
Active combustion control has been accomplished in many laboratory and real-world
P. Barooah1 combustion systems by fuel modulation as the control input. The modulation is commonly
achieved using reciprocating flow control devices. These demonstrations have been suc-
T. J. Anderson2 cessful because the instabilities have been at relatively low frequencies (⬃200 Hz) or the
scale of demonstration has been small enough to require very small levels of modulation.
J. M. Cohen3 A number of real-world instabilities in gas turbine engines involve higher frequencies
(200–500 Hz) and attenuation requires the modulation of large fractions of the engine
United Technologies Research Center, fuel flow rate (hundreds of pounds per hour). A spinning drum valve was built to modulate
East Hartford, CT 06108 fuel for these applications. Tests showed that this device provided more than 30% flow
modulation up to 800 Hz for liquid fuel flows of greater than 400 lbm/hr. This paper
describes the performance of the valve in flow bench tests, open-loop forcing, and closed-
loop instability control tests. The closed-loop tests were done on a single-nozzle combus-
tor rig which exhibited a limit-cycling instability at a frequency of ⬃280 Hz with an
amplitude of ⬃7 psi. It also encounters an instability at 575 Hz under a different set up of
the rig, though active control on that instability has not been investigated so far. The test
results show that the spinning valve could be effectively used for active instability control,
though the control algorithms need to be developed which will deal with or account for
actuator phase drift/error. 关DOI: 10.1115/1.1582495兴
1 Introduction heat release aeroengines. Two different instabilities are observed

As the environmental pressures to further reduce NOx, CO and under different rig setups. One is a limit cycling instability at
smoke continue, combustion chambers in gas turbines have be- around 280 Hz and the other is a resonant peak at around 575 Hz.
come more susceptible to combustion instabilities. The traditional So far the spinning valve has been used in active control of only
design methodology for suppressing combustion instabilities is the 280 Hz instability, though it was developed with instabilities
proving inadequate; therefore new techniques, such as active con- at higher frequencies in mind.
trol approaches are sought. This paper reports active control work for the 280 Hz instability
Combustion instabilities are caused by the coupling of acoustic using the spinning valve as the fuel modulation actuator. Section
velocity/pressure fluctuations with unsteady heat release, 关1兴. One 1.2 describes the experimental facility set up for combustion in-
method of active control that has been widely used in recent times stability control. Section 2 describes the operating principles of
is modulation of fuel supply to the combustor. If the heat release the spinning valve and the results of fuel modulation tests done in
fluctuation due to the modulating fuel supply is in opposite phase a cold-flow test loop. Section 3 briefly describes the phase shifting
to the natural instability driven heat release fluctuation, the two control algorithm, as it was applied in combustion control tests. A
will cancel each other out and the instability will be suppressed. valve controller was employed to make the spinning valve modu-
The simplest implementation of this method is the phase shifting late fuel in phase with a controller command. Section 4 describes
controller, in which the fuel supply is modulated at a certain phase the valve controller development work. The results of active con-
shift from the main harmonic of the pressure oscillations inside trol tests with the spinning valve in the single nozzle combustor
the combustor. rig are described in Section 5 and finally, Section 6 summarizes
One impediment to the successful application of such active the lessons learned from the exercise.
control schemes is the lack of actuators capable of providing large
modulation levels at frequencies in the range of these instabilities, 1.2 Single-Nozzle Combustor. As part of a NASA-
i.e., in the 250–1000 Hz range. A rotary fuel modulation device, sponsored active instability control program, a single-nozzle com-
or ‘‘spinning valve’’ was developed which could modulate fuel bustor rig was constructed to replicate combustion instabilities
with an amplitude of more than 30% of mean flow rate up to 1 seen in full-scale aeroengines, 关2兴. The active combustion control
kHz. The valve was used for active combustion control in a tests with the spinning valve were conducted using this combus-
NASA program in which the goal was the suppression of high- tor. Figure 1 shows a schematic of the combustor test section
frequency longitudinal modes, which exist in higher-volumetric assembly. Jet-A fuel was supplied to the injector by a high-
1
pressure fuel supply. The fuel injector was a prefilming, high-
Current address: Departmnet of Electrical and Computer Engineering, University
of California, Santa Barbara, CA.
shear airblast injector that was traceable to full-scale engine hard-
2
Current address: Institute for Defense Analyses, Alexandria, VA. ware. The combustor operated at realistic engine conditions and
produced ⬃4 MW at peak power. All the active control tests de-
3
Corresponding author. Current address: Pratt & Whitney, 411 Main Street, MS
163-03, East Hartford, CT 06108.
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN scribed in this paper were done at the operating conditions shown
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF in Table 1. These values are for a mid-power condition.
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- Dynamic measurements of the unsteady component of pressure
erlands, June 3– 6, 2002; Paper No. 2002-GT-30042. Manuscript received by IGTI, were made at a number of locations within the combustor. The
Dec. 2001, final revision, Mar. 2002. Associate Editor: E. Benvenuti. sensor labeled ‘‘PLA1C1’’ in Fig. 1 is used within this paper as a
Fig. 1 Single-nozzle combustor assembly, showing choked inlet „left… and
outlet „right… boundaries, fuel injector and combustor liner. Unsteady pres-
sure measurement „PLA1C1… located at 1.865 in. downstream of combustor
bulkhead. Dimensions shown are in inches.
pressure representative of the unsteady combustor pressure. This sure, respectively. The fuel flow rate was modulated with the spin-
sensor was mounted on the liner at 1.865 in. downstream of the ning valve, which was controlled by a PC-based DSP 共dSpace
combustor bulkhead. Data were acquired with a dSpace data ac- data acquisition/ control system, ds1003 processor board兲 through
quisition system typically sampling at 5 kHz. Low-pass 共2 kHz兲 a Moog T200 motor controller.
anti-aliasing filters were installed on all channels. Fast response
pressure measurements were also made in the fuel line down- 2 Spinning Valve Actuator
stream and upstream of the fuel modulating valve to measure
modulation levels and check steadiness of the fuel supply pres- The spinning valve design was based on a rotary concept rather
than a conventional reciprocating-spool configuration in order to
generate the maximum frequency response 共Fig. 2兲. The concept
Table 1 Combustor operating conditions and uncontrolled in- used a rotating drum with 12 regularly spaced holes around the
stability characteristics for 300 Hz instability circumference, which aligned with holes in the surrounding hous-
ing to pass flow. By minimizing the clearance between the hous-
Test Variable Mean Value ing and the drum, leakage was reduced when the holes in the
Inlet air pressure, P 3 (psia) 110 drum and housing were misaligned. Exit holes in the housing
Inlet air temperature, T 3 (°F) 610 were radially opposed to balance pressure and minimize trans-
Fuel flow rate, W f (lbm/hr) 207 verse loads. A design with two exit holes was chosen for this
Air flow rate, W a (lbm/sec) 2.55 application to balance radial pressure loads on the shaft and pro-
Unsteady pressure 6.5
⬘ (psi)
Amplitude, P comb
vide sufficient flow capacity in a reasonably sized package.
Mean fuel/air ratio 0.022 It should be noted that additional exit holes could have been
Instability frequency 共Hz兲 280 built into the housing, reducing the valve size for a given flow
capacity in exchange for added plumbing complexity. Additional
Fig. 2 Schematic of the spinning valve concept, using a rotating drum with multiple
holes, which align with exit holes in the case. A close tolerance was maintained to
produce the maximum level of modulation and to minimize leakage. An even number
of exit holes provided a pressure balance.
Fig. 3 Schematic of the cold flow test facility. A cooling system maintained a constant tem-
perature despite heat added from pump work at high-pressure differentials. A high-pressure
nitrogen system was used to pressurize the fuel system to realistic operating pressures.
manifold complexity can, however, introduce unexpected systems operate at frequencies up to at least 1 kHz. It was designed to
response characteristics due to the acoustics of the plumbing sys- provide on/off flow modulation. The level of modulation could be
tem. reduced, if necessary, with the installation of a valve bypass line.
Unlike a reciprocating device, the upper frequency limit of the The performance metric used to evaluate this valve system was
spinning valve was not due to the spool inertia or power required its ability to modulate the fuel flow at the point of injection: the
to accelerate it. A relatively low power was required to run this fuel nozzle. It was difficult to directly make high-response mass
device at all frequencies to overcome motor losses, shaft seal flow rate measurements and, as such, the results presented here
friction and limited fluid friction in the valve chamber. infer flow modulation from fluctuations in pressure differential,
The valve built for this program used a Moog G413-404 motor, which were easier to measure. With up and downstream measure-
a readily available motor capable of meeting the speed, accelera- ments close-coupled to the simulated injector, a reasonably accu-
tion, and precision requirements. Runout and radial clearance tol- rate measure of flow modulation could be made. The three fuel
erances were better than 0.001 in., making it possible to design for pressure sensors 共upstream of valve, downstream of valve, and
clearances that minimized leakage when the valve is closed. downstream of the orifice兲 showed that fluctuations downstream
Twelve holes around the spool circumference allowed a fuel of the injector were negligible and could be ignored and that fluc-
modulation frequency of 1 kHz to be reached at 5000 rpm, well tuations upstream of the valve were negligible for small levels of
within the capability of this motor. modulations. However, these did not hold for large levels of
modulations and so mass flow calculations done for the high
2.1 Fuel Flow Rate Modulation Authority. The valve sys- modulation cases have some error. Figure 4 shows a sample time
tem and its associated plumbing were tested in a small Jet-A flow trace from a pressure transducer downstream of the valve when it
loop with a capacity similar to that required for the single nozzle was running at 500 Hz. Flow modulation was inferred from pres-
combustor tests 共⬃500 lbm/hr兲. The flow-loop facility configura- sure oscillations by a simple analysis based on the momentum
tion is shown in Fig. 3. The valve was tested with a downstream equation for incompressible mass flow through an orifice in terms
orifice with a flow capacity equal to that of the fuel nozzle. A of the pressure differential. From these basic equations, it can be
bypass line upstream of the test valve could be opened to set the shown that
upstream supply pressure. An accumulator or pulse damper was
installed upstream of the spinning valve to reduce pressure oscil-
lations at the inlet. This device was pressurized to approximately ⳵W f 1 ⳵共 ⌬ P 兲
⫽ • . (1)
60% of the operating pressure at the valve inlet and, in lower Wf 2 ⌬P
modulation levels, could damp out almost all upstream oscilla-
tions. However, the modulation levels achieved using the spinning This equation assumes that ⌬P remains approximately constant
valve were so large that the pulse damper was less effective with ( ␦ (⌬ P)Ⰶ⌬ P) which was not always the case for the modulation
this valve. The fuel reservoir could be pressurized with nitrogen to levels associated with this valve. Nonetheless, this equation was
simulate operation at elevated fuel system pressures 共such as used to evaluate modulation and the results compared with other
would be encountered during combustion tests兲. methods of measurement. Despite this discrepancy, the use of
The modulating valve must have a flow capacity approximately pressure measurements and this correlation seem to provide rea-
equal to or greater than the injector flow capacity when open, and sonably accurate results, based on the results shown by Anderson
substantially lower 共or almost zero兲 than the orifice flow capacity 关3兴.
when closed, to allow the device to effectively modulate the flow. Forced-response sine-sweep experiments were done from
The valve was required to have sufficient response at the frequen- which pressure amplitude modulation levels were determined as a
cies expected in aeroengine instabilities, so it was designed to function of frequency. These were used to compute mass flow
Fig. 4 Time traces showing response of fuel pressure „mea-
sured downstream of the valve and upstream of fuel nozzle
orifice… to 500 Hz flow modulation command „cold flow test… to
the spinning valve
Fig. 6 Modulated fuel pressure response curves for the spin-

ning valve as the base pressure was changed between atmo-
modulation with Eq. 共1兲. Figure 5 shows a sample mass-flow spheric and combustor pressure „flow bench tests…
frequency-response curve based on pressure data for the spinning
valve over a range of 20 Hz to 800 Hz.
Figure 5 plots the amplitude of fuel modulation normalized loop tests at cold flow facility it was seen that the spinning valve
with the mean flow rate as a function of frequency. Several char- has considerable mass flow modulation capacity up to 1 kHz.
acteristics stand out in this response curve. The first is that the
modulation levels are about 60% of mean fuel flow or more up to 3 Phase-Shifting Instability Control Algorithm
400 Hz. However, they did decay at the higher frequencies. Also, The control algorithm used to suppress instability in this pro-
a repeatable dip occurred at 500 Hz; which might be due to fuel gram was of a phase-shifting type. This is one of the simplest
manifold acoustics. Lengthening of the line between the valve and control algorithms that can be used to suppress instabilities and
accumulator did show a reduction in the resonant dip frequency, has been successfully used in the past for this purpose, 关4兴.
but because of the configuration of the accumulator plumbing, it In this scheme, the main harmonic of the combustor pressure
was difficult to determine the exact line length associated with this signal near the instability frequency 共⬃280 Hz兲 was extracted
problem. When the system was operated at higher pressures, the with a filter, scaled by a control gain, and shifted by a control
frequency and level of attenuation were reduced. Figure 6 shows phase which was then sent to the secondary fuel modulation ac-
modulated fuel mass flow response curves 共amplitude versus fre- tuator as the driving control signal. A modified form of an ex-
quency兲 as the base pressure was changed. It was observed that tended Kalman filter 共EKF兲 based frequency tracking algorithm
elevated base pressures improved the modulation level and shifted proposed by LaScala 关5兴 was used to construct the phase-shifted
the 500-Hz resonant dip to a lower frequency. It is not known why control signal.
the response changes with base pressure. Note that Fig. 6 shows A disadvantage of both a fixed-gain observer-based and delay-
only the modulation pressure amplitude measured downstream of based phase-shifting controllers is that the bandpass filters in-
the valve and not the mass flow rate amplitudes. From these open- volved in these implementations are nominally fixed. If the fre-
quency of the pressure oscillations varies over a large interval for
a narrow bandpass filter, the filtered signal will be significantly
attenuated and the control signal will be dominated by noise. Wid-
ening the filer bandwidth prevents attenuation of the pressure sig-
nal, but more noise is accepted. More dynamical modes can then
affect the actuator action, and destabilization of these modes is
possible.
The advantage of the frequency tracking EKF observer was that
even if the main mode frequency of the oscillations changed from
its nominal value, so long as the rate of change was slow, the
observer kept track of the main mode frequency and the signal to
noise ratio of the filtered signal was maintained. It was observed
that the maximum rate of change in the peak frequency the ob-
server could track was about 5 Hz/s.
4 Spinning Valve Controller Development

The spinning valve presented a unique challenge in adapting it
to closed-loop control. The shaft position had to be controlled
such that it followed a certain trajectory yielding the fuel modu-
lation with a desired frequency and phase. This was done by run-
ning the motor in the ‘‘torque mode’’ where the command to the
T200 motor controller controlled the motor armature current. That
Fig. 5 Valve flow modulation capacity as a function of command was determined by a PID controller. Figure 7 shows the
frequency block diagram of the spinning valve control system. V c 共command
Fig. 7 Block diagram of the spinning valve controller
to the T200 motor controller兲 was the control input and ␾ m shaft
position was the output. ␾ m was measured with an encoder and
had to follow a desired trajectory, which was determined from the
phase of the combustor pressure 共described below兲. Fig. 9 Stacking of combustor pressure phase to create a
single trajectory for the spinning valve motor shaft to follow
Determining the Desired Trajectory. The phase of the fuel
modulation and the phase of the motor shaft were therefore related
by a factor of N H 共number of holes on the drum, namely 12兲. If
the phase of the motor shaft was, say, ␾ m , and the location of the Measuring the Motor Shaft Position. An incremental three-
hole nearest the 0 point of the shaft in the direction of rotation was channel encoder measured the position of the motor shaft. The
␾ h , the phase of the fuel modulation measured immediately encoder used was a HP HEDS 6540 three-channel encoder with a
downstream of the valve would be 12␾ m ⫹ ␾ h . Due to the feed resolution of 1024 counts per revolution. The output of the circuit
line and injector dynamics, the phase of the fuel modulation into was the measured motor shaft position in volts 共⫺9.89 V to 9.86
the combustor was not the same as that at the outlet of the valve. V for 0 to 2␲兲. The encoder’s two quadrature outputs were used to
If the phase difference introduced by the feed line and injector produce the incremental position advance measurement, and the
was ␾ f , then the phase of fuel modulation into the combustor absolute position information was inferred with the help of the
would be 12␾ m ⫹ ␾ h ⫹ ␾ f . For positive voltage command, the third index channel, which produced a high signal once per
shaft always rotated in a CW direction looking from the shaft end. revolution—always at the same position of the codewheel.
The direction of rotation was kept always the same by allowing
only positive commands. If ␾ c was the phase of the combustion Valve Control Algorithm. A PID controller was used to con-
pressure at any time, and ␥ was the control phase shift, the desired trol the motor shaft angle by commanding the required command
fuel pressure phase was ␾ c ⫹ ␥ . Hence the problem reduced to input V c to the spinning valve motor. Figure 7 shows the block
making 12␾ m ⫹ ␾ h ⫹ ␾ f follow ␾ c ⫹ ␥ as closely as possible. diagram of the control system. It should be noted that part of the
This was simplified to a tracking problem of ␾ m following loop, from the motor shaft rotation to the fuel modulation into the
( ␾ c ⫹ ␥ )/12 as closely as possible. This simplification was pos- combustor, was run open loop. The motor shaft position was not
sible since all the other phases are constant with time and was controlled to produce minimum error between the phase of modu-
taken care of by the control phase ␥. The phase ( ␾ c ⫹ ␥ )/12 was lated fuel in the combustor and the phase of combustion pressure,
the desired motor shaft phase, ␾ d . but to minimize error between the motor shaft position and a
Another issue that had to be resolved was that since the mea- calculated desired position which was expected to yield a fuel
sured phase was between 0 and 2␲, at any point it was also to be modulation that follows the phase of the combustion pressure. The
decided which ramp was to be tracked 共Fig. 8兲. To get around this tracking error was defined as the desired minus the actual shaft
problem, the phase angles were added to create a single phase position.
ramp instead of a saw-tooth phase trajectory. However, since the
phases cannot be stacked infinitely due to finite floating point e⫽ ␾ d ⫺ ␾ m
number sizes, after a while the value had to be reduced by an
integral multiple of 2␲ so as to make it small again. Some book- The control command was computed as a combination of the er-
keeping had to be done at this point to ensure all variables were ror, the derivative error and the integral of the error
modified uniformly 共especially the derivative of error, etc.兲. The
result was that the desired phase trajectory was a saw-tooth func-
tion with period much larger than 2␲, but an integral multiple of it
共Fig. 9兲.
V c ⫽K p e⫹K I 冕
to
t
edt⫹K D
de
dt
,
Fig. 8 Schematic of change of shaft position during several cycles of rotation.

This behavior led to an ambiguity in deciding which branch to track.
Fig. 10 Response of the spinning valve in a step-input test Fig. 11 Bode plot „magnitude and phase… of the identified
valve transfer function and loop transfer function „product of
valve and controller transfer functions G loopÄ Gv * Gc …
where the parameters K p , K I , and K D are to be determined ap-

propriately to make the shaft position ␾ m follow the desired phase identified from the step response test described earlier 共Eq. 共2兲兲
␾ d as closely as possible. and the controller was the one designed according to the above
procedure with a gain crossover set to 100 Hz. The PID gains
4.1 System Identification and Control Design. To design were K P ⫽1.1, K I ⫽6.8, and K D ⫽0.001. Since the fuel modula-
an effective PID controller, an accurate model of the spinning tion frequency was 12 times that of the motor, the bandwidth of
valve was needed. For a DC motor, the relation between motor the system was 1200 Hz, well beyond the nominal operating point
armature current I m and shaft position ␾ m 共in radian兲 is given by of 280 Hz for this program. The phase margin was about 120 deg.
a first-order transfer function. Since the valve is a combination of Since the identified valve models were recognized as not very
a DC motor and a spool rotating on a fluid film, its overall dy- useful in representing the system, evaluating the controller’s per-
namics can be expected to be more complicated than that of just formance in simulation using these models was not useful. One
the motor. point should be noted here that since the fuel modulation pro-
Step input tests were done with the spinning valve where step duced by the valve is 12 times the motor shaft frequency; small
inputs of voltage were supplied to the motor in the ‘‘torque mode’’ errors in the shaft position results in large errors between the
and resulting shaft rotation was examined to identify the transfer commanded and actual fuel pressure phase.
function of the valve. Tests were done in the cold flow bench
under various operating conditions 共i.e., different fuel system base
pressures and different levels of step inputs兲 to investigate if there
5 Active Instability Control Test Results
were nonlinearities in the system, and if any, how strong those The spinning valve was used for closed-loop combustion con-
effects were. trol of the single nozzle combustor. The combustor was operated
Figure 10 shows the result of such a test 共step in V c ⫽ at the conditions shown in Table 1. The nominal operating point
⫹0.65 V). The best-fit second-order system to match the response was established first by setting the air inlet temperature, fuel flow
of the valve 共with a sampling rate of 4 kHz兲 was rate, and air flow rate. Baseline, uncontrolled measurements of the
instability were acquired. Figure 12 shows a short time trace, the
0.156共 z⫹0.687兲 amplitude spectrum 共amplitude as a function of frequency兲 and
G V2 f ⫽ . (2)
共 z⫹0.823 兲共 z⫺0.991兲 PDF 共probability density function兲 of fluctuating combustion pres-
Step input tests with different command steps yielded very similar ⬘ 共measurement from transducer PLA1C1兲. The shape
sure, P comb
results when it was a positive step. For negative steps, i.e., when of the PDF shows this to be a limit-cycling instability according to
the voltage command was reduced, the valve responses were con-
siderably different from those for positive steps. The best fit trans-
fer function for a step input test 共step in V c ⫽⫺0.6 V) was
0.053共 z⫺0.835兲
G V2 f ⫽ (3)
共 z⫺0.941兲共 z⫺0.991兲
共4 kHz sampling rate兲, which was considerably different from the
transfer function in Eq. 共2兲.
The system transfer function depended on the direction of con-
trol input 共command Vc兲; hence a controller designed on the basis
of one such identified model would not work well under different
circumstances. Hence it was decided to tune the PID gains for a
nominal model to get a good phase margin with the required band-
width; and use those values as initial guesses in real application.
The PID gains were tuned during debugging of the control system
by commanding the fuel modulation to track a sinusoidal signal
from an analyzer. The choice of a PID controller can itself be
questioned for such a nonlinear plant. PID was chosen over other
more sophisticated techniques due to its ease of implementation. Fig. 12 Time trace, amplitude spectrum and PDF of fluctuating
Figure 11 shows the Bode plot of the loop transfer function of combustor pressure in the 280 Hz instability condition „mid-
the plant and controller, the plant was the spinning valve model as power…
Fig. 14 Comparison of combustion pressure amplitude spec-
tra for uncontrolled and the best controlled point
Fig. 13 Peak amplitude near 280 Hz and r.m.s. value of com-

bustor pressure as a function of control phase, showing dis- control algorithm must ameliorate the adverse effect of time delay
crepancy between amplitude and r.m.s. attenuation and noise in the plant. The control algorithm used in this demon-
stration was not sufficiently sophisticated to account for these ef-
fects.
criteria given by Banaszuk et al. 关6兴. The amplitude spectrum is 5.1 Spinning Valve Performance. Control authority may
computed by computing the power spectrum with Welch’s aver- also have been affected by the tracking performance of the spin-
aged periodogram method 共using Matlab’s PSD function兲 and then ning valve. Figure 15 shows the difference between the instanta-
using the following scaling formula to compute amplitude as a neous phases of the combustion pressure and the fuel pressure
function of frequency 共measured downstream of the valve兲 for the best-controlled case.
The commanded control phase shift was constant, so the valve
A p 共 f 兲 ⫽2 * 冑
PSD共 p ⬘ 兲
储w储2
共兺 w 兲2
(4)
should modulate fuel with a constant phase difference from the
combustor pressure. However, Fig. 15 shows that the valve peri-
odically lost tracking by as much as 250 deg.
where A p ( f ) is the amplitude, p ⬘ is the fluctuating combustion It was found that this periodic change in the phase difference
pressure and w is a hanning window of length equal to the number between combustor and fuel line pressures was due to the loss of
of fft points chosen. The number of fft points was 4096, yielding tracking by the spinning valve motor. Since the spinning valve
a frequency resolution of 1.2 Hz for the 5 kHz sampling rate used drum had 12 holes, the fuel modulation frequency was 12 times
in data acquisition. the motor shaft frequency. Therefore, the difference between fuel
Closed-loop control was applied by modulating the fuel flow pressure phase and 12␾ m should be constant.
rate and varying the phase shift of the control signal through a Figure 16 shows the difference between the fuel line pressure
complete cycle of 360 deg with 30 deg increments and measuring 共downstream of valve兲 and 12␾ m . The phase difference was not
combustor pressure response at those points. Inlet temperature, constant, but was slowly drifting in one direction. This could have
fuel flow rate, and air flow rate were held constant through the been caused by one of two things: either the encoder codewheel
test. The PID gains for the valve controller were K P ⫽2, K I ⫽1, was slipping on the motor shaft, or the encoder electronics were
and K D ⫽0.005. These were empirically determined by trial and
error. Figure 13 shows the peak amplitude of combustor pressure
⬘ as a function of control
near 280 Hz and also the r.m.s. of P comb
phase shift. The largest suppression of the peak amplitude 共⬃3
times兲 was achieved at a control phase shift of 180 deg. The
attenuation of r.m.s. value was smaller 共⬃15%兲, and occurred at a
phase shift of about 270 deg. Note that the r.m.s value was com-
puted from the time trace data directly and the peak amplitude
near 280 Hz was extracted from the amplitude spectrum calcula-
tion as explained previously.
Figure 14 shows the comparison of uncontrolled and controlled
amplitude spectra of the fluctuating combustor pressure for the
point where best attenuation was achieved. The amplitude at the
instability frequency 共280 Hz兲 was brought down by a factor of
10. The time traces show that, during control, although the ampli-
tude of oscillations near the instability frequency was low, the
oscillations were not uniformly low over time. A ‘‘peak-splitting’’
phenomenon and excitation of low-frequency modes, as evident in
Fig. 14, contributed to this. Peak splitting was the primary reason
why attenuation of the r.m.s was considerably less than for the
tonal peak. Peak splitting has been observed in other demonstra-
tions with other types of actuators, 关6兴. As explained by Banaszuk
et al. 关6兴 peak splitting is primarily due to large time delays in the Fig. 15 Instantaneous phase difference between P fuel and
combustor and actuation systems. In order to be effective, the ⬘
P comb during closed-loop control
Acknowledgments
This work was funded by the NASA Glenn Research Center.
The NASA Glenn Program Manager was Mr. John DeLaat. The
contributions of Messrs. Daniel Paxson and Clarence Chang, also
of NASA Glenn, were notable. Dr. Jeffery Lovett was the Pratt
and Whitney Program Manager. At UTRC, Messrs. William Pros-
cia, Scott Bortoff, Joseph Haley, and John Miano made significant
contributions that affected the outcome of this work.
Nomenclature
P3 ⫽ inlet air pressure
T3 ⫽ inlet air temperature
Wf ⫽ fuel flow rate
Wa ⫽ air flow rate
P⬘ ⫽ unsteady pressure
⬘
P comb ⫽ unsteady combustor pressure
f ⫽ frequency
Fig. 16 Instantaneous difference between the phase of fuel ␾m ⫽ motor shaft phase
pressure and 12* „motor shaft position… from during closed- ␾d ⫽ desired motor shaft phase
loop control, showing ‘‘phase drift’’ of spinning valve control ␾c ⫽ combustion pressure phase
␥ ⫽ control phase shift
␾f ⫽ feed line phase shift
not measuring the position accurately. Unfortunately, none of Vc ⫽ command to valve
these hypotheses can be checked with available data, more experi- e ⫽ shaft position error
ments will have to be done. This ‘‘phase drift’’ contributed to the KP , KI , KD ⫽ PID controller gains
poor control performance. Future work will have to address this Gv ⫽ motor transfer function
issue. Gc ⫽ valve controller transfer function
6 Summary
The spinning valve was capable of modulating fuel flow rate 共at References
least 30% of mean fuel flow up to 1 kHz兲 at high frequencies and 关1兴 Kiel, B., 2001, ‘‘Review of Advances in Combustion Control, Actuation, Sens-
ing, Modeling and Related Technologies for Air Breathing Gas Turbines,’’
was used for closed-loop control of combustion instabilities. The AIAA Paper No. AIAA 2001-0481.
spinning valve’s modulation capability depended on the feed line 关2兴 Cohen, J. M. et al., 2000, ‘‘Longitudinal-Mode Combustion Instabilities:
configurations and operating pressures. These features may be ex- Modeling and Experiments,’’ presented at the ‘‘NATO RTO Symposium on
ploited to design the feed line plumbing so that the valve has Active Control Technology for Enhanced Performance Operational Capabili-
ties of Military Aircraft, Land Vehicles and Sea Vehicles,’’ Braunschweig,
maximum authority in the frequency range of interest. Germany, May 8 –11.
This valve was used to control a ⬃280-Hz combustion instabil- 关3兴 Anderson, T. J., Proscia, W., and Cohen, J. M., 2001, ‘‘Control of a Liquid-
ity in a 4 MW single-nozzle combustor with realistic engine hard- Fuel Jet in an Unsteady Cross-Flow,’’ ASME Paper No. 2001-GT-0048.
ware. A three times reduction of maximum pressure amplitude 关4兴 Cohen, J. M., Rey, N. M., Jacobson, C. A., and Anderson, T., 1998, ‘‘Active
Control of Combustion Instability in a Liquid Fueled Low-NOx Combustor,’’
was achieved. However, reduction of r.m.s. pressure was small ASME/IGTI Gas Turbine Expo and Congress, Stockholm, June.
共⬃15%兲, due to side-band amplification 共‘‘peak splitting’’兲 and 关5兴 LaScala, B., 1994, ‘‘Approaches to Frequency Tracking and Vibration Con-
phase drift of the spinning valve controller. trol,’’ Ph.D. Thesis, Department of Systems Engineering, The Australian Na-
The drift of the spinning valve phase relative to the control tional University, Dec.
关6兴 Banaszuk, A., Mehta, P., Jacobson, C. A., and Khibnik, A., 2003, ‘‘Limits of
phase may be addressed through investigating and improving the Achievable Performance of Controlled Combustion Processes,’’ IEEE Trans.
ability of this system to track a fixed command over time. The Autom. Control, submitted for publication.
phase-ramping strategy, illustrated in Fig. 9, may have resulted in 关7兴 Muruguppan, S., Park, S., Annaswamy, A. M., Ghoniem, A. F., Acharya, S.,
an accumulation of some fixed error, leading to this phase drift. and Allgood, D. C., 2001, ‘‘Optimal Control of a Swirl Stabilized Spray Com-
bustor Using System Identification Approach,’’ Paper No. AIAA 2001-0779.
Avoidance of side-band amplification or ‘‘peak splitting’’ may 关8兴 Paschereit, C. O., Schuermans, B., and Campos-Delgado, D., 2001, ‘‘Active
require the application of more sophisticated control techniques, Combustion Control Using an Evolution Algorithm,’’ Paper No. AIAA 2001-
such as LQG-LTR controllers, 关7兴, or evolution algorithms, 关8兴. 0783.
The Recuperative-Auto Thermal
Reforming and the Recuperative-
Reforming Gas Turbine Power
Cycles With CO2 Removal—Part
I: The Recuperative-Auto
Thermal Reforming Cycle
D. Fiaschi The relatively innovative gas turbine based power cycles R-ATR and R-REF
e-mail: [email protected] (Recuperative–Auto Thermal Reforming GT cycle and Recuperative–Reforming GT
cycle) here proposed are mainly aimed to allow the upstream CO 2 removal by the way of
L. Lombardi natural gas fuel reforming. The power unit is a gas turbine (GT), fueled with reformed
and CO 2 cleaned syngas produced by adding some basic sections to the simple GT cycle:
L. Tapinassi • auto thermal reforming (ATR) for the R-ATR solution, where the natural gas is reformed
into CO, H 2 , CO 2 , H 2 O, and CH 4 ; this endothermic process is completely sustained by
Dipartimento di Energetica ‘‘Sergio Stecco,’’ the heat released from the reactions between the primary fuel 共 CH 4 兲 , exhausts and steam.
University of Florence, • water gas shift reactor (WGSR), where the reformed fuel is, as far as possible, shifted
Via Santa Marta, 3, into CO 2 and H 2 by the addition of water. • water condensation, in order to remove a
50139 Firenze, Italy great part of the fuel gas humidity content (this water is totally reintegrated into the
WGSR). • CO 2 removal unit for the CO 2 capture from the reformed fuel. Among these
main components, several heat recovery units are inserted, together with GT cycle recu-
perator, compressor intercooler, and steam injection in combustion chamber. The CO 2
removal potential is close to 90% with chemical scrubbing using an accurate choice of
amine solution blend: the heat demand is completely provided by the power cycle itself.
The possibility of applying steam blade cooling by partially using the water released from
the dehumidifier downstream the WGSR has been investigated: in these conditions, the
R-ATR has shown an efficiency range of 44–46%. High specific work levels have also
been observed (around 450–550 kJ/kg). These efficiency values are satisfactory, espe-
cially if compared with ATR combined cycles with CO 2 removal, more complex due to the
steam power section. If regarded as an improvement to the simple GT cycle, R-ATR shows
an interesting potential if directly applied to a current GT model; however, partial rede-
sign with respect to the commercially available version is required. Finally, the effects of
the reformed fuel gas composition and conditions on the amine CO 2 absorption system
have been investigated, showing the beneficial effects of increasing pressure (i.e., pressure
ratio) on the thermal load per kg of removed CO 2 . 关DOI: 10.1115/1.1587743兴
Introduction The power plants proposed in both parts of the present work be-
long to the second category, consisting in two GT power cycles
During the last ten years, special attention was devoted by re-
with integrated fuel decarbonization and subsequent CO2 capture.
search boards and industry to the study and development of power
The CO2 removal from the fuel is an increasingly investigated
cycles with low CO2 emissions. The activities in this field regis-
opportunity: the general approaches are those linked with CRGT
tered an important increase during the last five years, due to Kyoto
共chemically recuperated gas turbine兲 and POGT 共partial oxidation
Protocol, subscribed by many industrialized countries.
gas turbine兲 cycles, 关11,13兴. In the CRGT, the original methane
Many of the proposals for new-concept environmentally
fuel is converted into a mixture of H2 , CO2 , CO and steam by the
friendly plants involved GT-based power cycles. Two basic solu-
steam reforming: The heat due to the related endothermic reac-
tions seem to be practicable in short-medium term:
tions can be provided by thermal recuperation from the exhausts
1. semi-closed GT cycles with downstream CO2 removal from 关10,11兴 and/or by a partial combustion into the reformer itself
the CO2 enriched exhausts, 关1–10兴; 共auto thermal reforming, ATR, 关12兴兲. The recently studied CRGTs
2. GT-based power cycles with upstream CO2 removal, ob- represent an attractive alternative to the classic GT recuperative
tained by the fuel decarbonization amount of the combustion cycles, resulting in higher levels of heat recovery, thanks to
chamber, 关10–12兴. chemical enhancement, 关14兴. When considering its application in
the field of power plants with low CO2 emissions, a high level of
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN methane fuel conversion into H2 becomes mandatory. To this aim,
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF the operating temperature and pressure of the reformer play a key
role: temperatures of 800–1000°C allow a conversion level of
erlands, June 3– 6, 2002; Paper No. 2002-GT-30116. Manuscript received by IGTI, about 60 to 95%, increasing with the steam/methane ratio and
Dec. 2001, final revision, Mar. 2002. Associate Editor: E. Benvenuti. with lowering reformer pressure, 关11兴.
Fig. 1 Schematic of the steam cooled R-ATR cycle
Since GT exhaust temperatures range between 500 and 600°C, fuel is premixed with steam produced by thermal recuperation
it is impossible to achieve high levels of conversion by simple from the hot reformed fuel at the ATR exit 共streams 51–10兲. The
reforming. This is the reason for adding the water gas shift reactor following reactions take place into the ATR:
共WGSR兲 downstream the reformer, which allows the conversion
of the reformed gas into a CO2 and H2 rich mixture, ready to be CH4 ⫹H2 O↔CO⫹3H2
treated by the CO2 removal sections.
The upstream CO2 removal allows a notable size reduction of CO⫹H2 O↔CO2 ⫹H2
the CO2 separation equipment with respect to the downstream
option, due to the very different mass flow rates. Chemical CO2 CH4 ⫹1/2O2 ↔CO⫹2H2 .
separation is here investigated and discussed. Physical separation
by membranes is a continuously evolving alternative technology The gas composition downstream the ATR can be considered,
and has reached commercial levels, even if the presence of the with a satisfactory approximation, to be that resulting from chemi-
hydrogen is not, at present, well studied; the separation efficiency cal equilibrium if the correct catalyzers are used, 关12兴, and an
is, however, much lower than in chemical scrubbing, 关15兴. More- adequate reactor catalyst bed volume, with a heat exchanger of
over, the membranes require high pressure levels, which reduce three to four tube rows per cold side pass is adopted, 关14,19,20兴.
the methane conversion in the reforming process. Consequently, The partial combustion reaction of methane can be, in practice,
chemical CO2 separation by solution of ethanolamines was se- considered completely at equilibrium.
lected as more promising, also in the field of upstream CO2 On the whole, the ATR input and output parameters can be
capture, 关1,2兴. summarized as shown in Table 1. The nine output parameters are
Chemical absorption has been selected for a number of reasons: determined by the overall energy and mass balances, the chemical
species balances, and the chemical equilibrium of the three con-
1. information about properties of physical solvents for CO2 , sidered reactions. If the flow rate of the oxygen input corresponds
such as Selexol, Rectisol, etc., is proprietary and reliability of the to the stoichiometric amount, there is no oxygen left at the ATR
data available is disputable. exit.
2. n/desorption of physical solvents by pressure swinging is The reformed fuel is then sent, after having provided the nec-
theoretically effective, but in practice a system including as many essary heat to produce the steam needed by the reforming process,
as six small turbine/generator sets to recover the energy of the to the water gas shift reactor 共WGSR兲, where the main shift reac-
expansion, 关16兴, must be considered. In such conditions the power
expenditure for CO2 capture exceeds the values of 4 MJ/kg of
captured CO2 which is considered the good practice for MEA/ Table 1 Summary of ATR inputs and outputs
DEA chemical systems, 关1兴. With more optimized solutions of
amines in water 共as were assumed in the present work兲, values as Input Parameters Output Parameters
low as 1.8– 2 MJ/kg CO2 have been calculated.
Variable Description Variable Description
3. Use of amine solutions in chemical industry is very common
共e.g., CLAUS processes for H2 S; LPG purification, ammonia pro- p Reformer pressure m9 ATR exit
cesses兲 so the know-how is widespread and there are no propri- gas mass
etary rights, so that the documentation about thermochemical data flow
m4 Air Flow rate T9 ATR exit
is wide. The typical range of heat consumption in modern chemi- gas
cal absorption processes for ammonia production is temperature
0.7– 1.4 MJ/kg CO2 . The physical absorption processes may be T4 Inlet air xCH4,9 ATR exit
designed for zero heat consumption, but for comparison with the temperature gas
chemical processes, the mechanical energy requirements have also xN2,4 Inlet air xCO,9 composition
xO2,4 composition xCO2,9
to be considered, 关17,18兴. xCO2,4 xH2,9
xH2 O,4 xH2 O,9
Description of the R-ATR Power Cycle m5 H2 O-CH4 mixture xN2,9
mass flow rate
The R-ATR cycle 共recuperated auto thermal reforming兲 is sche- xCH4,5 CH4 mass fraction xO2,9
matically represented in Fig. 1: The heat demand due to the meth- at point 5
ane fuel reforming is provided by the partial combustion reactions T5 H2 O-CH4 mixture
temperature
of compressed air and methane into the ATR. The primary CH4
Table 2 Main R-ATR cycle operating data
␤1 First Compressor Pressure Ratio 冑␤

m comp 共kg/s) Compressor mass flow 100
T max (K) Maximum cycle temperature 1500
T b (K) Turbine blade metal temperature 1100
␧H Turbine blade cooling efficiency 0.4
ef Turbine film cooling effectiveness 0.2
␩ SC Compressor isentropic efficiency 0.89
␩ SC Turbine isentropic efficiency 0.91
Lp CC Combustion chamber pressure loss ⌬p/p 3%
Lp RIG Recuperator pressure loss ⌬p/p 5%
Lp ATR ATR pressure loss ⌬p/p 10%
DT RIG 共K) Recuperator approach temperature difference 30
␩ CO2 CO2 removal efficiency of the chemical absorber 90%
tion CO⫹H2 O⫽CO2 ⫹H2 takes place. In this component, the

concentration of CO2 and H2 in the fuel gas is enhanced, in order Fig. 2 Performance maps of R-ATR cycle
to facilitate their separation. The WGSR reaction is slightly exo-
thermic, with surplus heat used to preheat the external water en-
tering at point 5. The fuel coming out from the WGSR 共point 11兲
has a relevant concentration in H2 and CO2 , but a consistent cooling, two fundamental effects influence the efficiency: the re-
fraction of water vapor is still present: This is removed by the cuperation and the steam injection in combustion chamber. At low
subsequent condensation in the condensing heat exchanger 共point pressure ratios, high levels of recuperation can be applied, which
12, SEP兲. Part of the condensed water 共stream 19兲 is recovered leads to lower T 41 temperature 共Fig. 3兲 and then to lower steam
and integrated with the external input stream 6, and directed to the production 共Fig. 4兲, that is almost entirely due to blade cooling.
ATR section. The heat of condensation released by the SEP pro- With increasing pressure ratio, the lower recuperation brings to
cess is used to preheat the external water 共point 17兲, before pro-
ducing steam for the turbine blade cooling. The fraction of water/ higher T 41 ; under such conditions, an amount of steam exceeding
steam eventually exceeding the amount due to blade cooling is that necessary for blade cooling is produced and the steam injec-
sent to the compressor intercooler 共point 33兲, after having pro- tion in combustion chamber can be applied. Above ␤⫽8, the pe-
vided part of its sensible heat to the CO2 removal section, and/or nalization due to lower recuperation is dominant and the effi-
to the combustion chamber steam injection 共downstream the ciency drops. The steam injection has, however, positive effects
HRSG兲. on the specific work W sp , which is further increased after the best
Subsequently, the dehumidified fuel coming from SEP is sent to
the amine CO2 removal section (CO2 REM, Fig. 1兲, where the
90% of CO2 is separated and prepared for storage and disposal
共point 23兲. The decarbonized fuel is sent to the combustion cham-
ber 共point 13兲. The heat content of the exhausts is used for GT
cycle recuperation 共points 31– 41兲, for the production of the steam
eventually due to blade cooling 共HRSG, points 41–20兲 and, fi-
nally, to sustain the amine regeneration of the CO2 removal sec-
tion upstream the stack, 关1兴. The air due to the ATR is taken in
point 40, downstream the GT recuperator.
The remaining heat can cover some cogeneration utilities
共points 52–53兲. The small compressor C3 is needed to move the
clean fuel gas to the GT combustor, against the friction losses
encountered along the ATR, WGSR, CO2 removal and the other
heat exchangers.
The basic cycle operating data on which the whole analysis is
carried out are reported in Table 2.
Performance Analysis of the R-ATR Cycle

The R-ATR power cycle has been simulated using a dedicated Fig. 3 Recuperator end temperatures versus ␤ for the steam
program, developed by the authors using EES. The adopted blade cooled R-ATR
cooling model has been recently developed and tested on some
current production GTs, 关21兴. Since the GT cycle is the core of the
proposed powerplant and the exhaust temperature is a strong func-
tion of pressure ratio, an extensive analysis of the performance
and CO2 removal potential versus pressure ratio ␤ has been car-
ried out. The effects of maximum cycle temperature T max have not
been discussed, because an increase of this parameter enhances
the performance of both gas cycle and reforming process, leading
to lower CO2 emissions.
The performance map exploited by the R-ATR power cycle
共primary fuel LHV efficiency ␩ GT versus specific power W sp )
with variable ␤ is represented in Fig. 2, for air and steam blade
cooling. The efficiency levels are interesting, showing a consistent
margin of improvement with the adoption of blade steam cooling.
Significantly different optimizing pressure ratios can be observed:
9 for steam cooling and 17 for air cooling. In the case of steam Fig. 4 Steam production versus ␤ for the steam cooled R-ATR
efficiency point. On the whole, the variation in efficiency is lim-
ited to about 1 point on the whole inquired field for the steam-
cooled case.
In the air cooled R-ATR, a more than double flow rate of cool-
ant has to be supplied to the turbine with respect to the steam
cooling case, 关21兴, which entails the displayed negative effects on
the efficiency.
The amount of steam production is at the same level of the
steam cooled R-ATR. The specific work is higher, due to the
larger amount of allowable steam injection in combustion cham-
ber 共Fig. 4兲 which cancels the negative effect of the increased
compressor work for extensive air cooling. The compressor inter-
cooling levels are very similar at pressure ratios higher than the
optimizing value, since the spray injected water m ws is set to be
the saturating amount. At ␤ lower than the ␩ GT optimizing value, Fig. 5 CO2 removal and cycle efficiency versus K steam
the amount of intercooling water for the steam-cooled cycle is not
the saturating one, but the maximum possible to allow the produc-
tion of the necessary steam for blade cooling. Some beneficial
effects of pressure ratio on performance are due also to the re-
duced heat demand of CO2 sequestration unit, with increasing
absorber operating pressure, 关15,22兴, as will be remarked in the
section dedicated to the CO2 removal system.
The sharpened shape of the performance and steam production
curves related to the steam cooling cases around the optimization
point is directly linked with the intercooler mass flow rate: This is
set to be the saturating amount only starting from pressure ratios
that allow the production of the necessary steam for blade cooling.
At very low ␤s, the recuperation level of the cycles is limited by
the reduction of the spray intercooling and the subsequent in-
crease of T 41 to allow the production of the required steam mass
flow due to blade cooling. Two different operating regimes are
thus present for ␤ lower and higher than the optimizing value. In
other words, the efficiency of the steam cooled configurations is
optimized when m ws is a little higher than the amount required for
blade cooling.
Fig. 6 R-ATR Specific CO2 emission versus ␤ and K steam
The Reforming Section: Parametric Analysis of the
R-ATR Power Cycle
The fuel gas composition from the ATR to the combustion
chamber is shown in Table 3 for the ␩ GT optimizing ␤. One of the
main parameters influencing the ATR performance is the K steam
ratio, defined as K steam⫽m s5 /m p f 共symbols are referred to the
cycle points of Fig. 1兲. Its allowable range is limited by the tem-
perature T 7 at the WGSR exit 共cooling water side兲 plus an ap-
proach DT⫽10 to 30 K, since the steam directed to the ATR is
produced in recuperative way by cooling of the reformed fuel.
This influences both the cycle efficiency 共especially at low and
high ␤s, Fig. 5兲 and CO2 capture potential 共Fig. 6兲. A rather im-
portant effect on W s p has been observed, due to the decreasing
steam production with increasing K steam . It can be explained by
observing that the reduction of T 10 entails the decay of T 11 and
then T 18 共Fig. 7兲, which leads to the subsequent drop in HRSG
steam production. The lower bound of K steam , is about 1.5, to
avoid the formation of solid carbon in the ATR, 关10兴. The pressure
Fig. 7 Influence of K steam on the main fuel reforming and

Table 3 Fuel gas composition at the main exits of the R-ATR deCO2 chain components
deCO2 chain devices
Composition 关%兴共Mass/Volume兲

ATR exit WGSR exit SEP exit deCO2 exit ratio has a marginal influence on the CO2 removal but deter-
SPECIE T⫽1154 K T⫽489 K T⫽395.6 K T⫽316.7 K mines the upper limit for K steam in order to satisfy the condition
H2 4.09/35.6 5.17/44.5 5.96/51.1 9.67/60.2 T 10⬎T 7 .
H2 O 23.8/21.7 14.2/13.7 0.97/0.94 1.40/0.975 The specific CO2 emission of R-ATR versus ␤ 共around the op-
CO 15.4/9.99 0.44/0.27 0.51/0.31 0.82/0.37 timizing values兲 and K steam is shown in Fig. 6: A low sensitivity to
CO2 13.3/5.10 36.8/14.5 42.5/16.7 6.90/1.97 ␤ is confirmed, whereas the K steam ratio plays a key role in this
CH4 0.18/0.21 0.18/0.19 0.20/0.22 0.30/0.23
N2 43.2/27.3 43.2/26.8 49.9/30.8 80.9/36.2 way. In particular, it must be kept at the maximum allowable
O2 0/0 0/0 0/0 0/0 values, in order to maintain the lowest possible CO2 emission
level.
Fig. 8 CO2 Clean fuel gas composition at the combustor inlet
Fig. 9 CO2 Amine absorption system heat demand for the
R-ATR cycle
The reformed CO2 clean fuel gas composition at the combus-
tion chamber inlet of GT versus pressure ratio 共related to the
steam cooling configuration at fixed K steam⫽1.8) is reported in The purified gas leaves the column at the top and the solution—
Fig. 8: The species H2 , CO2 , and CH4 are practically unchanged loaded with CO2 —leaves at the bottom. This liquid stream is
with increasing ␤, whereas the N2 increases due to the decrement thermally regenerated in the stripper, the previous formed CO2
of the H2 O and CO. This suggests, coupled with the results shown compounds are broken and gaseous CO2 is produced. The heat
in Fig. 8, a wide margin for the operative range of R-ATR in terms requirement for the thermal regeneration is the energetic price to
of CO2 emissions, and consequently a wide adaptability to many pay for the reduction of CO2 . The CO2 amine absorption system
different GT models. has been simulated by the ASPEN⫹code, 关26兴.
The temperature range of the hot side 共9–51兲 of the steam/CH4 The stream exiting from the top of the stripper, hence, contains
superheater, shown in Fig. 3 versus ␤, can bring to metal dusting essentially carbon dioxide and it needs compression up to at least
conditions, a well-known phenomena involved in reforming pro- 80 bars for transportation and final storage. In calculation of the
cesses. An alloy is likely to be carburised if the carbon activity of powerplant efficiency, the energetic expense for CO2 compression
the environment is higher than that of metal, 关23兴. The carbon 共which can be estimated in about 355 kJ/kgCO2 from 关22兴兲 has not
activity depends on C, CO, CO2 , CH4 , and H2 O partial pressures. been directly accounted as it can depend on several variables 共type
Referring to the average composition of the reformed fuel at the of disposal, transportation system, etc.兲.
ATR exit 共Table 3兲, the following values have been calculated: Including this reference value, and considering that, in the op-
timized case 4.94 kg/s of CO2 are separated for the R-ATR cycle,
p COp H2 /p H2 O⫽1.80 the efficiency changes from 45.0% to 43.4%.
The same kind of system was studied for removal of CO2 from
2
p CO /p CO2 ⫽2.15 the exhaust of power cycles such as natural gas combined cycle
共NGCC兲 or semi-closed gas turbine combined cycle 共SCGT/CC,
2
p CH4 /p H 2
⫽0.0015. 关22兴兲. However, in the present application there are relevant dif-
ferences in design and performance.
Referring to the graphs of carbon activity as a function of gas
The first difference lays in the operating pressure of the absorp-
composition and temperature 共originally from 关24兴 and reported
tion unit: While in this case one must work at a high pressure level
on 关23兴兲, the operating field of the 9–51/8 –5 heat exchanger falls
depending on the pressure ratio, in the case of flue gas treatment
in a dangerous zone, where, depending on the alloy used, metal
the operating pressure is atmospheric. Moreover, the mass flow
dusting is likely to proceed, especially into the 500–700°C range.
rate to be processed is definitely less than exhausts from a gas
Practical experience from various reforming plants around the
turbine: in the R-ATR case the mass flow rate to be treated is 14.9
world appears to be contradictory concerning the success or fail-
kg/s with reference to a compressor inlet flow rate of 100 kg/s,
ure of specific materials in resisting metal dusting attack. Expo-
while, with the same reference, in a conventional gas turbine it
sure trials of various highly alloyed metals in the Mossgas reform-
would be about 102 kg/s. Absorption of CO2 is favored by in-
ing plant have consistently demonstrated that high Cr containing
creasing pressure, as the mass transfer from gaseous to liquid
materials 共50/50 Cr-Ni alloy兲 appeared to be totally immune to
phase grows with pressure, allowing a better removal efficiency at
metal dusting after 17 months 关23,25兴. The experience on these
constant solution mass flow rate or, vice versa, constant efficiency
chemical plants can be effectively extended to the reforming sec-
with reduced mass flow rate. As a direct consequence, working at
tions of the R-ATR.
higher pressure allows a reduction of the specific thermal request
for the regeneration of amine solution.
Detailed Evaluation of CO2 Absorption Section Heat The specific thermal request of the CO2 removal on the R-ATR
Demand cycle—reported in Fig. 9 versus pressure ratio and for different
The carbon dioxide removal section consists substantially in an temperature levels of the inlet gas to the absorber—ranges be-
absorption unit and a regeneration unit 共a schematic of this sec- tween 1700 and 2300 kJ/kg. Considering an SCGT/CC with CO2
tion, which is common to many absorption/desorption applica- amine chemical absorption from the exhaust, the minimal specific
tions, can be found in 关4兴兲. The CO2 rich stream enters the ab- thermal request is about 3590 kJ/kg, 关22兴. The comparison is, of
sorber from the bottom and is scrubbed by the descending solution course, favorable to the present system, considering that the spe-
of amines. Here, CO2 passes into liquid phase and reacts with the cific emission factors are almost comparable: between 70 and
amines forming new compounds that subtract CO2 from liquid 100 gCO2 /kWh for the present type of cycle and about
phase, enhancing thus its further transfer from gaseous to liquid 60– 70 gCO2 /kWh for the SCGT/CC.
phase. The better removal efficiency and the consequent specific ther-
Table 4 Main operating results for using ALSTOM cyclone GT in R-ATR cycle
CYCLE CO2 Fuel

ALSTOM mc T max m cool m sinj W GT W sp ␩ GT T exh m exh Emiss. LHV
Cyclone 关kg/s兴 ␤ 关K兴关kg/s兴关kg/s兴关kW兴关kJ/kg兴关%兴关K兴关kg/s兴关g/kWh兴关kJ/kg兴
Standard GT 38.66 16.23 1602 8.29 0 13124 339 35.9 827 39.4 695 50009
R-ATR 38.66 16.23 1602 4.86 3.42 22181 574 39.1 890 46.6 150 11201
共steam
cooling兲
mal request reduction are also due to the more favorable compo- the actual potential of the selected machine with very limited ad-
sition of the gas to be treated. In fact, a higher concentration of aptation to the proposed cycle. Even if the inquired GT model
CO2 enhances the gas-liquid CO2 transfer. In this type of cycle the works with air blade cooling, the configurations with steam cool-
CO2 mass concentration reaches the 42.5% in the case of the ing have been inquired too, in order to limit the change in design
R-ATR, while in the SCGT/CC case the CO2 mass concentration inlet turbine mass flow and maintaining the steam injection. The
is only 14.5%—already about three times that of a conventional blade cooling system should undergo some off design level by
GT—and the nitrogen is up to 77%, 关22兴. changing the coolant, which was not evaluated in this work. The
From the above considerations, it is possible to state that the main modifications 共common for example to biomass-fueled ap-
substantial reduction of the specific thermal request, with respect plications兲 are required for the combustion chamber, that should
to the SCGT/CC case, is due to the combined effects of increased manage a low heating value fuel 共highly diluted with nitrogen兲.
operating pressure and CO2 concentration. The main results are reported in Table 4. They show that a
From Fig. 9 it is also evident that when the temperature of the relatively limited efficiency improvement can be obtained,
gas at absorber inlet increases, the specific thermal request in- whereas a large increase in specific power is possible, especially
creases as well, due to the lower efficiency in the gas-liquid CO2 due to the steam injection. The relatively low heating value per
transfer. unit mass of reformed fuel implies the need of much higher
One problem that could affect the gas turbine of the R-ATR is amounts of fuel at combustion chamber inlet and a higher flow
that linked with the eventual entrainment of traces of amine 共DEA rate in the turbine with respect to design levels: this leads to a
and MDEA兲 in the clean fuel, that are likely to be corrosive for consistent degree of off design, that could bring to additional
blades, combustor and so on. From the calculation carried out losses in performance. The absence of intercooled compression
with ASPEN⫹, the possible presence of residual amine in the brings to consistently lower efficiency values with respect to those
cleaned fuel is marginal 共pollutant level兲, having estimated DEA reported in Fig. 2.
and MDEA mass fractions of 2.19•10⫺7 and 2.18•10⫺5 , respec-
tively. Literature data about the corrosive effects of amine solu- Conclusions
tions is lacking when such low concentration are encountered,
being mainly focused on the analysis of specific chemical plants A relatively innovative GT-based power cycle has been pro-
where the concentrations are much higher, 关27兴. Then, here it is to posed, with CO2 removal upstream of combustion chamber 共R-
be supposed that these effects are negligible, or, anyway, they ATR兲: The proposed cycle involves the use of auto-thermal re-
should be evaluated by dedicated test campaigns. forming 共ATR兲, with a recuperative GT. Downstream, further heat
Finally, the gas leaving CO2 absorption system is saturated in recovery from the exhaust proceeds into the fuel reforming—
DEA and MEA solutions, that could bring to the eventual carry- cleaning chain. The proposed solution makes extensive use of
over of amine droplets, having an high damaging potential for the compressor spray intercooling and steam injection into the com-
GT parts 共particularly blades and combustor兲. The prevention of bustion chamber. Moreover, the possibility of adopting steam
this phenomena requires an adequate design of the amine droplets blade cooling has been examined. The cycle does not involve any
separation system, also considering that the CO2 absorber is pres- steam bottoming cycle, ensuring thus a more limited complexity
surized and the effect of gravimetric contribution 共in cyclone sys- of the powerplant layout in comparison to reforming combined
tems, for example兲 is relatively reduced. cycles with CO2 removal and allowing the possibility to refer to a
wide range of power sizes.
Test Case of the R-ATR Cycle on a Current Production The performance potential in terms of efficiency and CO2 re-
moval have been inquired versus the typical parameters of chemi-
GT Model cally recuperated GTs. In particular, the influence of GT pressure
As stated in the previous sections of this work, the performance ratio has been investigated, since the gas turbine is the core of the
levels of the proposed R-ATR power cycle are interesting, espe- entire cycle layout. For the proposed configurations, the GT pres-
cially considering that there is no coupling with a steam bottom- sure ratio heavily influences the cycle performance, showing gen-
ing cycle. On the whole, the cycles efficiency levels under opti- erally low values for the efficiency optimization. The best perfor-
mized conditions are only 1–2% lower than other proposed mance has been obtained with the adoption of steam blade
chemically recuperated GT cycles with CO2 removal, involving a cooling, even if interesting values are shown also by air cooling.
steam bottomer cycle, 关11,13兴. The inquired R-ATR solutions provide interesting performance
In this section, the test on a current production GT model 共AL- levels 共around 45.5% with steam cooling and 44% with air cool-
STOM CYCLONE兲 has been performed to have a preliminary ing兲, together with low values of CO2 emissions 共70– 80 g/kWh兲
verification of the adaptability of the proposed cycle to the current if the highest possible K steam values are used.
production technology: in this way, a preliminary evaluation of The analysis of the CO2 removal section shows that the particu-
the short-term applicability chances of the R-ATR can be given. larly favorable fuel gas composition allows very low thermal re-
The basic data used for the simulation and tuning of the ALS- quests, due to the high concentration and pressure of CO2 . More-
TOM Cyclone gas turbine model with the manufacturer catalog over, the higher the operating pressure 共i.e., GT pressure ratio兲 the
data have been applied to the R-ATR cycle. Since the existing lower the CO2 removal heat demand.
machine does not include the possibility of compressor intercool- A test of the adaptability of the R-ATR with a current produc-
ing, this last has been canceled with respect to the previously tion gas turbine model has been performed, showing the possibil-
proposed configuration 共Fig. 1兲. This leads to a consistent loss in ity of increasing efficiency and specific power, as well as a con-
performance, but the aim of this section is just to give an idea of sistent reduction in specific CO2 emissions. The main problem is
surely linked to the higher turbine and combustion chamber mass 关4兴 Corti, A., Fiaschi, D., and Manfrida, G., 1999, ‘‘Thermo-Economic Evaluation
of the SCGT Cycle,’’ Energy Convers. Manage., 40, pp. 1917–1929.
flow rates with respect to the design values, due to the lower 关5兴 Facchini, B., Fiaschi, D., and Manfrida, G., 1996; ‘‘Semi-Closed Gas Turbine/
reformed fuel heating value and to the addiction of steam for Combined Cycle With Water Recovery,’’ ASME Paper No. 96-GT-317.
blade cooling and combustor. On the whole, the applicability of 关6兴 Facchini, B., Fiaschi, D., and Manfrida, G., 1997, ‘‘SCGT/CC: An Innovative
existing machines to the proposed cycle brings to a moderate in- Cycle With Advanced Environmental and Peakload Shaving Features,’’ Energy
crease in efficiency, but guarantees a consistent reduction potential Convers. Manage., 38„15–17…, pp. 1647–1653.
关7兴 Fiaschi, D., and Manfrida, G., 1999, ‘‘A New Semi-Closed Gas Turbine Cycle
of CO2 emissions. However, some dedicated studies for off design With CO2 Separation,’’ Energy Convers. Manage., 40, pp. 1669–1678.
operation are deserved in future works. 关8兴 Mathieu, Ph., Dechamps, P., and Distelmans, M., 1994, ‘‘Concepts and Appli-
cations of CO2 Gas Turbines,’’ Power-Gen Europe ’94, Cologne.
Nomenclature 关9兴 Mathieu, Ph., Chefneux, E., and Dechamps, P., 1995, ‘‘Energy and Exergy
Analysis of CO2 Based Combined Cycle Plants,’’ Proceedings of the Second
DT ⫽ temperature difference Law Analysis of Energy Systems: Towards the 21st Century, Roma.
K steam ⫽ steam-fuel mass ratio at reformer inlet 关10兴 Cau, G., and Cocco, D., 2000, ‘‘Performance Assessment of Semi-Closed
Lp ⫽ pressure loss Chemically Recuperated Gas Turbine Systems,’’ Proceedings of ASME Turbo
Expo, May 8 –11, Munich, Germany.
m ⫽ mass flow 关11兴 Lozza, G., and Chiesa, P., 2000, ‘‘Natural Gas Decarbonization to Reduce CO2
T ⫽ temperature Emission From Combined Cycles. Part B: Steam-Methane Reforming,’’
W ⫽ power ASME Paper No 2000-GT-0164.
关12兴 Andersen, T., Kvamsdal, M., and Bolland, O., 2000, ‘‘Gas Turbine Combined
Subscripts Cycle With CO2 -Capture Using Auto-Thermal Reforming of Natural Gas,’’
Proceedings of ASME Turbo Expo, May 8 –11, Munich, Germany.
ATR ⫽ related to the ATR 关13兴 Lozza, G., and Chiesa, P., 2000, ‘‘Natural Gas Decarbonization to Reduce CO2
CC ⫽ combustion chamber Emission From Combined Cycles. Part A: Partial Oxidation,’’ ASME Paper
CO2 ⫽ related to CO2 removal 2000-GT-0163.
comp ⫽ compressor inlet 关14兴 Carcasci, C., and Harvey, S., 1998, ‘‘Design Issues for the Methane-Steam
exh ⫽ related to gas turbine exhaust Reformer of a Chemically Recuperated Gas Turbine Cycle,’’ ASME IGTI In-
ternational Gas Turbine Congress, Stockholm.
GT ⫽ related to gas turbine cycle 关15兴 Tapinassi, L., 2001, ‘‘Studio di impianti di potenza innovativi con turbina a gas
inj ⫽ injected integrati con sistemi di rimozione della CO2 tramite conversione del metano in
max ⫽ maximum in cycle combustibile ad alto contenuto di idrogeno,’’ Graduation thesis, Environmental
pc ⫽ post combustion Engineering Faculty, Università degli Studi di Firenze, Italy 共in Italian兲.
关16兴 Chiesa, P., and Consonni, S., 1999, ‘‘Shift Reactors and Physical Absorption
pf ⫽ primary fuel for Low-CO2 Emission IGCCs,’’ ASME J. Eng. Gas Turbines and Power, 121,
RIG ⫽ recuperator pp 295–305.
s5 ⫽ steam added to the REF or ATR at point 5 关17兴 EFMA, 1995, ‘‘Best Available Techniques for Pollution Prevention and Con-
sat ⫽ water-steam saturation trol in the European Fertilizer Industry,’’ Booklet No. 1 of 8: ‘‘PRODUCTION
OF AMMONIA’’, copyright 1995—EFMA, European Fertilizer Manufactur-
scool ⫽ coolant steam ers’ Association.
sinj ⫽ injected steam 关18兴 IEA, 2000, ‘‘Carbon Dioxide Capture and Storage,’’ Report from Department
s tot ⫽ total steam of Trade and Industry, Sept. on www.dti.gov.uk/cct.
ws ⫽ spray water 关19兴 Adelman, S. T., Hoffman, M. A., and Baughn, J. W., 1995, ‘‘A Methane-Steam
Reformer for a Basic Chemically Recuperated Gas Turbine,’’ ASME J. Eng.
Greeks Gas Turbines and Power, 117, pp. 16 –23.
关20兴 Carcasci, C., Facchini, B., and Harvey, S., 1998, ‘‘Design Issues and Perfor-
␤ ⫽ GT cycle pressure ratio mance of a Chemically Recuperated Aeroderivative Gas Turbine,’’ Proc Instn
␩ ⫽ Efficiency Mech Engrs, 212, Part A.
关21兴 Fiaschi, D., and Lombardi, L., 2001, ‘‘Cellular Thermodynamic Model for Gas
Turbine Blade Cooling,’’ Proceedings of ECOS 2001 Conference, July 4 – 6.
References 关22兴 Lombardi, L., 2000, ‘‘LCA Comparison of Technical Solutions for CO2 Emis-
关1兴 Corti, A., Failli, L., Fiaschi, D., and Manfrida, G., 1998, ‘‘Exergy Analysis of sions Reduction in Power Generation,’’ Ph.D. thesis, Dipartimento di Ener-
Two Second-Generation SCGT Plant Proposals,’’ Proceedings of ASME IGTI getica ‘‘Sergio Stecco,’’ Università degli Studi di Firenze, Italy.
43rd Gas Turbine and Aeroengine Congress and Exhibition, Stockholm, Swe- 关23兴 Holland, M. L., and De Bruyn, H. J., 1996, ‘‘Metal Dusting Failures in Meth-
den, June. ane Reforming Plant,’’ Int. J. Pressure Vessels Piping, 66, pp. 125–133.
关2兴 Corti, A., Lombardi, L., and Manfrida, G., 1998, ‘‘Absorption of CO2 With 关24兴 Lai, G. Y., 1990, High Temperature Corrosion of Engineering Alloys, ASM
Amines in a Semi-Closed GT Cycle: Plant Performance and Operating Costs,’’ International, Materials Park, OH, pp. 47–72.
Proceedings of ASME IGTI 43rd Gas Turbine and Aeroengine Congress and 关25兴 Holland, M. L., 2000, ‘‘Practical Experience With Countering Metal Dusting
Exhibition, Stockholm, Sweden, June. in a Methane Reforming Unit,’’ Mossgas 共PTY兲 Ltd., Mossel Bay, South Af-
关3兴 Corti, A., and Manfrida, G., 1998, ‘‘Economic Analysis of a Semi-Closed Gas rica, on http://www.mossgas.com/research/nace2000a.pdf.
Turbine/Combined Cycle 共SCGT/CC兲 With CO2 Removal by Amines Absorp- 关26兴 Aspen⫹, 2000, v. 10 User Manual, Aspentech TM.
tion,’’ GHGT-4, 4th International Conference on Greenhouse Gas Control 关27兴 Kohl, A. L., and Riesenfeld, F. C., 1985, Gas Purification, 4th Ed., Gulf
Technologies, InterLaken, Switzerland, Aug. 30–Sept. 2. Publishing, Houston, TX.
Thermodynamic Performance
Analysis of New Gas Turbine
R. Gabbrielli Combined Cycles With No
Università di Pisa,
Dipartimento di Energetica,
via Diotisalvi 2,
Emissions of Carbon Dioxide
Pisa 56126, Italy
In the context of the reduction of the carbon dioxide 共 CO 2 兲 emissions as prescribed by the
Kyoto protocol, this paper describes a thermodynamic performance analysis of new gas
turbine combined cycles with no emissions of CO 2 and nitrogen oxides. Three new similar
cycles belonging to the same typology are proposed. These cycles use water/steam as
R. Singh working fluid, which is compressed in liquid and vapor phase, and the internal combus-
School of Mechanical Engineering,
tion process, which takes place between syngas and pure oxygen. The top Brayton cycle
Cranfield University,
and the bottom Rankine cycle are integrated together. The syngas is produced by steam-
Cranfield,
natural gas reforming with internal chemical heat recovery. The CO 2 produced in the
Bedfordshire MK43, 0AL, UK
combustion is captured simply by water condensation from the exhaust gas and liquefied
to be stored. A simulation analysis has been performed to evaluate the net efficiency and
the net specific work of the cycles. Varying the most important operative variables and
using the least-square regression and 2 k factorial design techniques, a very large sensi-
tivity analysis has permitted the highlighting of performance behavior of the cycles.
Including the energy penalty due to the liquefaction of CO 2 and to the oxygen production
and adopting standard operative conditions, the LHV-based net efficiency and the net
specific work may exceed 50% and 1000 kJ/kg, respectively. 关DOI: 10.1115/1.1610017兴
Introduction interest in the design and development of new power generation

technologies based on fossil fuels with lower CO2 emissions has
In the last hundreds of years, the content of some greenhouse
grown in the energy sector.
gases (CO2 , CH4 , N2 O, CFC, HFC, and PFC兲 in the atmosphere
In particular, many modifications to standard gas turbine power
has grown markedly. With a wide agreement in the scientific sec- generation technologies have been proposed in these last years
tor, this fact seems to have caused a continuous increase of the with the purpose of reducing or eliminating the CO2 emissions
Earth surface temperature with consequent alteration of the global from the fossil fuel combustion. The proposals may be catalogued
climate. essentially in the following well-known typologies:
In 1997, the Kyoto Protocol defined the goal of global green-
house gas emission reduction of about 5% within the years 2008 – • CO2 as working fluid and internal combustion with pure oxy-
2012 in comparison with the emission level of 1990. Although the gen, 关1– 6兴,
CO2 is the least aggressive greenhouse gas, the overall amount • water/steam as working fluid and internal combustion with
released into the atmosphere from the human activities is so high pure oxygen, 关7–10兴,
in comparison with all others gases that the CO2 is responsible for • air gas turbine combined cycles with post-combustion CO2
about 60% of the greenhouse effect due to the increase of green- capture, 关10–13兴,
house gas content in the atmosphere. The fossil fuel fired power • chemical looping combustion, 关14,15兴,
generation sector produces about 16% of the overall anthropo- • air gas turbine combined cycles with fossil fuel pre-
genic CO2 emissions. combustion decarbonization to produce a hydrogen enriched
A wide variety of techniques could be adopted to obtain the fuel without carbon content, 关16 –20兴, and
CO2 emission reduction, as the Kyoto Protocol commitments. The • water/steam working fluid gas turbine combined cycles with
more attractive possibilities are: reduction of the overall energy fossil fuel pre-combustion decarbonization to produce pure
consumption, by increasing the thermodynamic efficiency of en- hydrogen, 关21–23兴.
ergy conversion and utilization, for example converting the old In the present paper, three new similar gas turbine combined
steam power plants to new more efficient natural gas fired com- cycles with no CO2 emissions belonging to the second typology
bined cycles; use of lower carbon content fuels, for example natu- are proposed. The four main literature proposals of steam/CH4 gas
ral gas instead of coal; enhancement of the natural sinks for CO2 , turbine combined cycles cited above have the following common
such as forests and oceans, which naturally capture CO2 from the features: 共1兲 the working fluid is water/steam; 共2兲 the compression
atmosphere; use of power sources with intrinsically very low CO2 is executed by a pump in liquid phase with low energy consump-
emissions, as renewable 共for example, biomass兲 or nuclear energy; tion; 共3兲 the combustion takes place between natural gas and pure
deep modifications of standard fossil fuel fired power plants intro- oxygen in atmosphere of steam, which mitigates the combustion
ducing the capture and storage of CO2 . In this context, a large temperature 共recent interesting studies about the technological
feasibility of new combustors for the combustion of natural gas
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN with pure oxygen in steam atmosphere may be found in 关24兴兲; 共4兲
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF steam or saturated water is produced recovering internally the
exhaust gas heat; 共5兲 the exhaust gas, whose components are
erlands, June 3– 6, 2002; Paper No. 2002-GT-30117. Manuscript received by IGTI, steam and CO2 , is cooled down to nearly ambient temperature
Dec. 2001, final revision, Mar. 2002. Associate Editor: E. Benvenuti. with water condensation and consequent separation and capture of
the combustion CO2 without emissions into the atmosphere. Prac- • 18 –19: the exhaust gas, that has not been recirculated, ex-
tically, in these combined cycles, a top steam Brayton cycle is pands down to the condensation pressure 共0.055 bar兲;
integrated with a bottom steam Rankine cycle, where the heat is • 19–1: the water in the low pressure gas condenses so that the
provided by the top one. CO2 is separated in gas phase to be stored in liquid phase in
The cycles here proposed have the following peculiarities in a successive step without emissions into the atmosphere.
comparison with the other steam/CH4 gas turbine cycles: 共1兲 the
working fluid is compressed in both liquid and vapor phase, The cycles B and C differentiate from the cycle A only in the
adopting a pump and an intercooled multistage compressor, where high-pressure steam production process. In particular, the cycle B
the intercoolings are executed down to saturation conditions with has essentially the following differences in comparison with the
steam production; 共2兲 the exhaust gas is partially recirculated and cycle A: the pressure of the saturated steam is higher than the
mixed with the working fluid; 共3兲 the fuel is syngas (H2 , CO, exhaust gas pressure; the recirculated exhaust gas is compressed
CO2 , H2 O, and CH4 ), which is produced via steam–natural gas in two intercooled stages 共9–10, 10–11, and 11– 4兲, and it is
reforming with internal chemical heat recovery from the exhaust mixed 共3–11兲 with the saturated steam downstream the intercool-
gas, 关25兴. ing 共10–11兲; the vaporisation of the water is executed in the heat
In this paper, first the features of the proposed cycles are de- recovery steam generator and in the recirculated exhaust gas in-
scribed. Then, the results of thermodynamic simulations, in terms tercooling. It is important to note that in the intercooling of the
of thermodynamic efficiency 共␩兲 and net specific work 共sw兲, are compressed recirculated exhaust gas, the pinch point between the
reported. Finally, with the regression and 2 k factorial design tech- hot side and the cold side has been established equal to about
niques, the operative performance behavior of a possible power 10°C.
plant based on the thermodynamic processes here proposed are In the cycle C, the water is pumped directly up to the highest
highlighted. pressure of the cycle 共1–2兲; the exhaust gas, which is partially
recirculated, is compressed in two intercooled stages 共11–12, 12–
Cycle Configurations 13, and 13– 6兲; with the heat recovered in the intercooling 共12–
13兲, saturated intermediate pressure steam is produced 共3–5兲 and
The process flow diagram of each cycle here proposed, refer- then mixed with the intercooled compressed gas; downstream the
enced, respectively, as A, B, and C, is reported together with the second compression stage, the recirculated gas is mixed with the
qualitative temperature-specific entropy plane in Fig. 1. First of high pressure steam 共4 – 6兲 produced by the exhaust gas cooling
all, the thermodynamic transformations of the cycle A are de- 共2– 4 and 10–11兲.
scribed in the following using the number references of Fig. 1: The cycles here proposed may be considered as the integration
• 1–2: the water is compressed in a pump from the condensa- of a top steam intercooled regenerated Brayton cycle and a bottom
tion pressure 共0.055 bar兲 to a value lower than the exhaust gas steam Rankine cycle 共cycles A and B have one pressure level,
pressure, that is equal to 1 bar; whereas the cycle C has two pressure levels兲 where the energy
• 2–3: the compressed water is heated up to saturated condi- input is provided by the top cycle. For cycle A, B, and C the
tions in an economizer, where in the hot side the compressed Rankine cycle thermodynamic transformations are 1–2–3– 4 –5–
steam, downstream the first compression stage, is cooled 6 –18 –19, 1–2–3–10–9–12, and 1–2– 4 – 6 –5–1–3–5–12–11–
down to the saturation; 14, respectively, as in Fig. 1.
• 3– 4: the saturated water is first partially vaporized in two In the cycles proposed, once fixed the working fluid mass flow
heaters, corresponding to two intercoolings of the com- to mitigate the combustion temperature to a defined value, three
pressed steam, and then completely vaporized in a heat re- features have been introduced to reduce the heat to be provided by
covery steam generator where the exhaust gas is cooled down the exhaust gas in the saturated steam production: 共1兲 the com-
to 105°C; pression is subdivided in liquid and vapor phase so to decrease the
• 4 –5: the saturated steam is compressed up to the exhaust gas saturation pressure; 共2兲 the compression in vapor phase is inter-
pressure; cooled and the heat released is recovered in the steam production
• 5– 6/18 – 6: the exhaust gas is partially recirculated and mixed process; 共3兲 the exhaust gas is partially recirculated so to reduce
with the compressed steam; the steam mass flow necessary in the combustion temperature
• 6 –7, 7– 8, 8 –9, 9–10, 11–12, and 12–13: the mixed steam is mitigation.
compressed in four intercooled compression stages, where the The chemical heat recovery in the reformer makes it possible to
intercoolings are executed down to the saturation; increase the heat exchange efficiency because the average tem-
• 13: after heating without mixing in the high pressure turbine perature difference is lower than in standard heat exchangers 共see,
blade cooling, not reported in Fig. 1, the compressed steam is for example, 关28兴兲. The overall steam reforming process is endo-
split in two streams, one feeds the combustion chamber to thermic and so the heat exchange, that in standard heat exchangers
mitigate the combustion temperature, and the other is mixed is governed only by temperature difference between the hot and
with natural gas; cold side, is due also to the reaction heat absorption. For each
• 13–14: in a countercurrent plug-flow tubular reactor, 关26,27兴, mole of methane 206 kJ are required in the reforming reaction.
the steam–natural gas mixture reacts endothermically to pro- The water–shift reaction, which is always present in the reforming
duce a syngas stream, whose components are CO, H2 , CO2 , process, provides exothermically 41 kJ for each mole of reactant
and unreacted steam and CH4 , using the heat provided inter- and reduces the overall heat requirements. The methane conver-
nally by the exhaust gas 共chemical heat recovery兲; sion increases with low pressure, high equilibrium temperature,
• 14 –15: the syngas burns with pure compressed oxygen pro- and high steam-to-methane molar ratio.
vided in stoichiometric amount in atmosphere of steam which
mitigates the combustion temperature; this implies that NOx Thermodynamic Performance Analysis
are not produced in the combustion; The cycle operative behavior has been simulated with the well
• 15–16: the exhaust gas, whose components are steam 共about known process simulator ASPEN PLUS, 关29兴, evaluating the most
90% in volume兲 and CO2 , expands down to 1 bar in a important performance parameters, as ␩ and sw.
steam–CO2 turbine; the turbine blades are cooled by the First of all, for each cycle the most meaningful operative inde-
compressed steam upstream the splitting described above; pendent process variables have been selected. For cycle B and C
• 16 –17: the exhaust gas is cooled to provide the heat neces- the combustion pressure 共HP兲 and the firing temperature 共TIT兲,
sary to the endothermic reforming reaction; which is the temperature of the exhaust gas downstream the first
• 17–18: the exhaust gas is cooled down to 105°C to vaporize nozzle of the turbine, are the essential variables. For cycle A, the
the low pressure steam; low saturation pressure of the steam 共SP兲 has to be taken into
Fig. 1 Process flow diagram and relative qualitative temperature-specific entropy plane of the cycles
proposed
account as further independent variable. The simulations have becomes a simple heat exchange due to temperature difference.
been executed varying the operative variables selected. HP as- TIT assumes values from 1225°C to 1375°C in step of 50°C,
sumes values from 10 bar to 40 bar in step of 5 bar. When the which are standard firing temperatures in modern power genera-
reforming pressure is higher than about 40 bar, the methane con- tion equipment. SP assumes four values, from 0.55 bar to 0.85 bar
version is practically negligible and the chemical heat recovery in step of 0.10 bar. Hence, 7⫻4⫻4⫽112 and 7⫻4⫽28 simula-
Fig. 2 Thermodynamic efficiency and net specific work of the cycles proposed
tions have been executed for the cycle A and for cycles B and C, Once fixed the operative variables, ␩ and sw improve from
respectively. cycle A to cycle C. The bottom Rankine subcycle ␩ and sw grow
The operative assumptions to perform the simulations have markedly with the saturation pressure, as is well known. The best
been: 共1兲 the natural gas, whose volume composition is CH4 cycle is the cycle C whose ␩ reaches about 52.3%. In the opera-
⫽95.04%, CO2 ⫽0.20%, N2 ⫽2.65%, C2 H6 ⫽2.11%, is supplied tive ranges of the independent variables, ␩ grows monotonically:
at 1 bar and 15°C; 共2兲 the oxygen production plant has not been when TIT increases, the average temperature of heat introduction
taken into account; 共3兲 the oxygen stream is supplied at 1 bar and
increases; when HP grows, ␩ of both subcycles increases. The
15°C with a volume purity equal to 96% 共4% of argon兲; 共4兲 the
isoentropic efficiency of compressors, turbines and pumps is, re- Brayton subcycle ␩ starts to decrease with HP only for values
spectively, 0.90, 0.93, 0.80; 共5兲 the mechanical efficiency of the higher than 40 bar.
turbomachinery is 0.99; 共6兲 the difference between the combustion sw has always an increasing trend with both HP and TIT. In the
temperature at the outlet of the combustion chamber and the firing cycles B and C, the Rankine subcycle sw grows monotonically
temperature has been established equal to 70°C via blade cooling with HP. The Brayton subcycle sw assumes a decreasing trend
with the working fluid downstream the compression; this has been with HP only when HP is higher than 40 bar. When TIT increases,
simulated with a heat exchanger between the combustion chamber the power produced in the high pressure turbine grows. The high-
and the turbine 关30兴; 共7兲 further cooling of the turbine has been est value of sw 共for cycle C兲 is about 1150 kJ/kg.
simulated mixing some intermediate pressure working fluid with Among the cycles proposed, the best one is the cycle C: Once
the exhaust gas at the middle of the expansion path, 关30兴; 共8兲 in fixed, the operative variables, the compression, is executed at
the heat exchangers, including the reformer, the pinch point is
most in the liquid phase with the lowest energy consumption 共the
equal to 10°C; 共9兲 the working fluid mass flow 共recirculated flow
plus saturated steam兲 is fixed; 共10兲 the pressure ratios of the in- area inside the qualitative temperature-specific entropy diagram
tercooled steam compressor stages are equal; 共11兲 to calculate the assumes clearly the largest value兲.
thermodynamic properties, the Redlich-Kwong-Soave equation of
state and the steam NBS/NRC tables have been selected for
steam–CO2 mixtures and pure steam, respectively; 共12兲 the pres-
sure drop in the heat exchangers and reactors is 1% of the inlet
pressure; 共13兲 the heat loss in the heaters is equal to 1%; 共14兲 the Regression and 2 K Factorial Design Sensitivity Analysis
steam-to-methane molar ratio has been fixed equal to 3, which
may be considered a standard value, 关31兴; 共15兲 the liquefaction of Using the numerical values of ␩ and sw 共the available operative
CO2 , that is executed at 70 bar and 28.6°C, has been simulated conditions of cycles A, B, and C are 112, 28, and 28, respec-
with an intercooled three-stage compressor and a water-cooled tively兲, reported above, a deep sensitivity analysis based on the
condenser. regression technique has allowed us to define an explicit tool to
In the evaluation of ␩ and sw, the LHV of the fuel has been predict the thermodynamic performance of a power plant based on
supposed equal to 47.8 MJ/kg, adopting the combustion heat val- one of the cycles proposed. In particular, adopting a suitable con-
ues available in the software, and the overall net power has been solidated software, with the least-squares estimation method, all
decreased by the energy expense due to the oxygen production numerical results have been fitted to obtain approximated explicit
共0.24 kWh/kg, 关2兴, which is a standard value with consolidated functions that model the dependence of ␩ and sw on the operative
cryogenic technologies, 关32兴兲. In Fig. 2, ␩ and sw 共ratio between independent variables. With these functions, it is possible to pre-
the overall net power and the exhaust gas mass flow兲 are reported
dict approximately the process performances with a generic set of
for each version in function of the operative independent vari-
ables. For simplicity, the results of the cycle A are reported only operative variables within the ranges analyzed without running
when SP is 0.85 bar, because both ␩ and sw increase monotoni- further simulations.
cally with SP. Once established HP and TIT, when SP increases, The functions, used to approximate the dependence of ␩ and sw
the pressure ratio of the pump increases whereas that of the inter- on the operative variables and listed in the following, have proved
cooled compressor decreases. This fact reduces the power con- to be the best set after detailed statistical comparisons among
sumption due to the steam compression. many functions:
Table 1 Regression parameters Table 2 Performance data in the sensitivity analysis for
cycle A
Regression Average Coefficient of
Parameters Quadratic Error Determination EគO2 ,
HP, bar SP, bar TIT, °C kWh/kg ␩A , % swA, kJ/kg
␩A 1.038•10 ⫺3
9.969•10 ⫺1
swA 1.235•100 9.999•10⫺1 10 0.55 1225 0.10 40.13 647.7

␩B 9.209•10⫺4 9.981•10⫺1 40 0.55 1225 0.10 41.67 769.7
10 0.55 1375 0.10 42.63 757.5
swB 1.347•100 9.999•10⫺1 40 0.55 1375 0.10 44.25 935.1
␩C 6.994•10⫺4 9.991•10⫺1 10 0.85 1225 0.10 41.95 680.1
swC 1.569•100 9.999•10⫺1 40 0.85 1225 0.10 43.45 808.2
10 0.85 1375 0.10 44.37 790.8
40 0.85 1375 0.10 45.99 976.0
10 0.55 1225 0.32 33.49 540.5
40 0.55 1225 0.32 35.03 647.1
␩ A共HP, SP, TIT)⫽4.541•10⫺2 ⫹1.229•10⫺3 •HP⫹4.388•10⫺2 10 0.55 1375 0.32 35.99 639.6
40 0.55 1375 0.32 37.62 794.8
•SP⫹3.521•10⫺4 •TIT⫹1.667•10⫺5 •HP•SP 10 0.85 1225 0.32 35.31 572.5
40 0.85 1225 0.32 36.82 684.7
⫺7.404•10⫺7 •HP•TIT⫺2.374•10⫺5 •SP•TIT 10 0.85 1375 0.32 37.73 672.5
40 0.85 1375 0.32 39.35 835.1
⫺1.433•10⫺6 •HP2 ⫹4.242•10⫺3 •SP2 ⫺5.377
•10⫺8 •TIT2 ⫺8.779•103 /HP6 ⫺1.095•101 /TIT
⫺1.824•10⫺2 /SP; technology on the process performances. The extreme values of
the other operative variables are those used in the simulations. The
swA共HP, SP, TIT)⫽1.013•101 ⫺1.618•101 •HP⫺2.131•102 numerical values of ␩ and sw in the sensitivity analysis are re-
ported in Tables 2 and 3. Figure 3 reports the main effects of the
•SP⫹7.518•10⫺1 •TIT⫹7.242•10⫺1 •HP operative variables on ␩ and sw. In the operative ranges analyzed,
•SP⫹1.151•10⫺2 •HP•TIT⫹2.872•10⫺2 about ␩ and sw the most important variables are always EគO2 and
HP, respectively. The decrease of EគO2 from 0.32 kWh/kg to 0.10
•SP•TIT⫺6.870•10⫺3 •HP2 ⫺1.045•102 kWh/kg implies a large improvement of the performances of the
power generation process 共␩ grows 6.6% on average兲. About ␩,
•SP2 ⫺6.408•10⫺5 •TIT2 ⫺1.094 TIT is more important than HP only for cycle A. For each
•103 HP⫺0.5, 共 kJ/kg兲 ; cycle, the only interaction effect not negligible is that between
HP and TIT about sw. For cycle A, B, and C, the results equal
␩ B共HP, TIT)⫽1.206•10⫺1 ⫹2.130•10⫺3 •HP⫹3.180•10⫺4 ⫹2.644•101 kJ/kg, ⫹3.087•101 kJ/kg, and ⫹3.794•101 kJ/kg,
respectively.
•TIT⫺6.320•10⫺7 •HP•TIT⫺1.084•10⫺5
•HP2 ⫺5.392•10⫺8 •TIT2 ⫺1.268•102 /HP4 ; Conclusions
Three new gas turbine combined cycles with internal combus-
swB共HP, TIT)⫽1.210•102 ⫺1.737•101 •HP⫹8.585•10⫺1 •TIT
tion of natural gas with pure oxygen in steam atmosphere without
⫹1.406•10⫺2 •HP•TIT⫺3.202•10⫺2 •HP2 CO2 and NOx emissions have been proposed. They integrate many
features of several literature proposals: steam intercooled com-
⫺1.032•10⫺4 •TIT2 ⫺1.109•103 /HP0.5, 共 kJ/kg兲 ; pression, partial exhaust gas recirculation, internal chemical heat
recovery with syngas production in steam-natural gas tubular re-
␩ C共HP, TIT)⫽1.624•10⫺1 ⫹2.722•10⫺3 •HP⫹3.070•10⫺4 •TIT former, exhaust gas expansion down to subatmospheric pressure,
⫺8.795•10⫺7 •HP•TIT⫺1.270•10⫺5 •HP2 ⫺5.350 and CO2 separation from the exhaust gas without any capture
technology.
•10⫺8 •TIT2 ⫺1.565•101 /HP3 ; Cycle C, which is the integration of a top intercooled regenera-
tive Brayton cycle and a bottom two-pressure reheat Rankine
swC共HP, TIT)⫽1.018•102 ⫺1.969•101 •HP⫹1.187•100 •TIT
cycle, is the best one, because the power consumption due to
⫹1.746•10⫺2 •HP•TIT⫺4.915•10⫺2 •HP2 steam compression is lower than in the other cycles.
A wide thermodynamic analysis has permitted the discovery
⫺2.353•10⫺4 •TIT2 ⫺1.419•103 /HP0.5, 共 kJ/kg兲 ; that in the operative ranges of the selected process variables, the
where the unit measures of each coefficient are omitted for sim- thermodynamic performances improve monotonically. Thermody-
plicity and ␩ A 共HP, SP, TIT兲, swA 共HP, SP, TIT兲, ␩ B 共HP, TIT兲, namic efficiency and net specific work explicit functions, obtained
swB 共HP, TIT兲, ␩ C 共HP, TIT兲, swC 共HP, TIT兲 are the estimated with the regression analysis of the numerical results of the simu-
functions of ␩ and sw for the cycles A, B, and C, respectively. The
values of the regression parameters reported in Table 1 demon-
strate the goodness of the fitting executed. The fitted functions, Table 3 Performance data in the sensitivity analysis for cycle
B and C
reported above, make it possible to predict, with very good ap-
proximation, the performance behavior of the power generation EគO2 , swB, swC,
processes proposed. This may be useful in the automatic proce- HP, bar TIT, °C kWh/kg ␩B , % ␩C , % kJ/kg kJ/kg
dure of cost optimization of the new power generation plant.
10 1225 0.10 47.25 50.08 729.5 838.9
To highlight clearly the main effects and the interactions be- 40 1225 0.10 50.85 54.55 860.1 1052.3
tween the independent operative variables on ␩ and sw, a 2 k fac- 10 1375 0.10 49.43 52.16 844.3 956.1
torial design sensitivity analysis has been performed. In this kind 40 1375 0.10 53.12 56.54 1040.0 1249.5
of analysis, the energy expense to produce 1 kg of pure oxygen 10 1225 0.32 40.61 43.44 627.0 727.7
40 1225 0.32 44.21 47.92 747.8 924.3
(EគO2 ) has been added to the independent variable set. Its ex- 10 1375 0.32 42.79 45.52 730.8 834.3
treme values are 0.10 kWh/kg and 0.32 kWh/kg. This makes it 40 1375 0.32 46.49 49.90 910.0 1102.8
possible to analyze the effect of the available oxygen production
Fig. 3 Main effects of the operative variables on thermodynamic efficiency and net specific work with the 2 k
factorial design sensitivity analysis
lations, may be used as useful tools to predict the process perfor- tion,’’ Proceedings of ASME TURBOEXPO 2000, Munich, Germany, ASME
Paper No. 2000-GT-153.
mance behavior with a generic set of operative variables. A 2 k 关7兴 Pak, P. S., and Suzuki, Y., 1994, ‘‘A CO2 -Recovering Nonpolluting High-
factorial design analysis has highlighted the high importance of Efficiency Gas-Turbine Power-Generation System Utilizing Saturated Steam
the energy consumption due to the oxygen production. To define as Its Working Gas,’’ Electr. Eng. Jpn., 114, pp. 86 –97 共translated from Denki
the actual profitability of an eventual power plant based on the Gakkai Ronbunshi, 1993, 113-B, pp. 266 –272兲.
关8兴 Jericha, H., Sanz, W., Woisetschlager, J., and Fesharaki, M., 1995,
power generation processes proposed, it will be necessary to ‘‘CO2 -Retention Capability of CH4 /O2 -Fired Graz Cycle,’’ Proceedings of
evaluate the process technological feasibility and to perform de- 21th CIMAC World Congress on Combustion Engines, Interlaken, Switerland,
tailed economic and scenario analyses. These will be the topics of CIMAC, London.
following papers. 关9兴 Anderson, R., Brandt, H., Mueggenburg, H., Taylor, J., and Viteri, F., 1998, ‘‘A
Power Plant Concept Which Minimizes the Cost of Carbon Dioxide Seques-
tration and Eliminates the Emission of Atmospheric Pollutants,’’ Proceedings
Acknowledgments of 4th International Conference on Greenhouse Gas Control Technologies,
Interlaken, Switzerland, Elsevier Science Ltd., Oxford.
The authors want to thank Franco Donatini 共ENEL-Produzione 关10兴 Bolland, O., Undrum, H., and Nilsen, M., 2000, ‘‘Natural Gas Fired Power
Ricerca, Pisa兲 for his support provided during the research, a part Cycles With Integrated CO2 Capture,’’ Proceedings of 5th International Con-
ference on Greenhouse Gas Control Technologies, Cairns, Australia, CSIRO
of which has been described in this paper. Publishing, Collingwood.
关11兴 Freund, P., 1998, ‘‘Abatement and Mitigation of Carbon Dioxide Emissions
From Power Generation,’’ Proceedings of Power-Gen ’98, Milan, Italy,
Nomenclature Pennwell, London.
EគO2 ⫽ energy consumption to produce 1 kg of pure oxygen 关12兴 Bolland, O., and Undrum, H., 1999, ‘‘Removal of CO2 From Natural Gas
Fired Combined Cycle Plants,’’ Proceedings of Power-Gen ’99, Frankfurt,
共kWh/kg兲 Germany, Pennwell, London.
HP ⫽ combustion pressure 共bar兲关13兴 Fiaschi, D., and Manfrida, G., 1999, ‘‘A New Semi-Closed Gas Turbine Cycle
LHV ⫽ lower heating value 共kJ/kg兲 With CO2 Separation,’’ Energy Convers. Manage., 40, pp. 1669–1678.
SP ⫽ low saturation pressure in cycle A 共bar兲关14兴 Ishida, M., and Jin, H., 1994, ‘‘A New Advanced Power-Generation System
Using Chemical-Looping Combustion,’’ Energy, 19, pp. 415– 422.
sw ⫽ net specific work 共kJ/kg兲关15兴 Jin, H., and Ishida, M., 2001, ‘‘Reactivity Study on a Novel Hydrogen Fueled
TIT ⫽ firing temperature 共°C兲 Chemical-Looping Combustion,’’ Int. J. Hydrogen Energy, 26, pp. 889– 894.
关16兴 Hendriks, C. A., and Blok, K., 1992, ‘‘Carbon Dioxide Recovery Using a Dual
Greek Gas Turbine IGCC Plant,’’ Proceedings of First International Conference on
␩ ⫽ LHV-based thermodynamic efficiency Carbon Dioxide Removal, Amsterdam, Netherlands, Pergamon, Oxford.
关17兴 Audus, H., Kaarstad, O., and Skinner, G., 1998, ‘‘CO2 Capture by Pre-
Subscripts Combustion Decarbonization of Natural Gas,’’ Proceedings of 4th Interna-
tional Conference on Greenhouse Gas Control Technologies, Interlaken, Swit-
A ⫽ relative to cycle A zerland, Elsevier Science Ltd., Oxford.
B ⫽ relative to cycle B 关18兴 Andersen, T., Kvamsdal, M., and Bolland, O., 2000, ‘‘Gas Turbine Combined
C ⫽ relative to cycle C Cycle With CO2 -Capture Using Auto-Thermal Reforming of Natural Gas,’’
Proceedings of ASME TURBOEXPO 2000, Munich, Germany, ASME Paper
No. 2000-GT-162.
关19兴 Lozza, G., and Chiesa, P., 2000, ‘‘Natural Gas Decarbonization to Reduce CO2
References Emission From Combined Cycles. Part A: Partial Oxidation,’’ Proceedings of
关1兴 Mathieu, P., and De Ruyck, J., 1993, ‘‘CO2 Capture in CC and IGCC Power ASME TURBOEXPO 2000, Munich, Germany, ASME Paper No. 2000-GT-
Plants Using a CO2 Gas Turbine,’’ Proceedings of IGTI Cogen-Turbo Con- 0163.
gress, Bournemouth, UK, ASME, New York, pp. 77– 83. 关20兴 Lozza, G., and Chiesa, P., 2000, ‘‘Natural Gas Decarbonization to Reduce CO2
关2兴 Dechamps, P. J., Distelmans, M., Mathieu, P., and Pirard, N., 1994, ‘‘Perfor- Emission From Combined Cycles. Part B: Steam-Methane Reforming,’’ Pro-
mances of Combined Cycle Power Plants Using CO2 Gas Turbine,’’ Proceed- ceedings of ASME TURBOEXPO 2000, Munich, Germany, ASME Paper No.
ings of FLOWERS’94 Conference, Florence, Italy, SG Editoriali, Padova, pp. 2000-GT-0164.
671– 682. 关21兴 Cai, R., and Fang, G., 1991, ‘‘Analysis of a Novel Hydrogen and Oxygen
关3兴 Yantovski, E. I., 1996, ‘‘Zero Emission Fuel-Fired Power Plants Concept,’’ Combined Cycle,’’ Int. J. Hydrogen Energy, 16, pp. 249–254.
Energy Convers. Manage., 37, pp. 867– 877. 关22兴 Bannister, R. L., Newby, R. A., and Yang, W. C., 1999, ‘‘Final Report on the
关4兴 Ulizar, I., and Pilidis, P., 1997, ‘‘A Semiclosed-Cycle Gas Turbine With Car- Development of a Hydrogen-Fuelled Combustion Turbine Cycle for Power
bon Dioxide-Argon as Working Fluid,’’ J. Eng. Gas Turbines Power, 119, pp. Generation,’’ J. Eng. Gas Turbines Power, 121, pp. 38 – 45.
612– 616. 关23兴 Sugisita, H., Mori, H., and Uematsu, K., 1998, ‘‘A Study of Thermodynamic
关5兴 Mathieu, P., and Nihart, R., 1999, ‘‘Zero-Emission MATIANT Cycle,’’ J. Eng. Cycle and System Configurations of Hydrogen Combustion Turbines,’’ Int. J.
Gas Turbines Power, 116, pp. 116 –120. Hydrogen Energy, 23, pp. 705–712.
关6兴 Jackson, A., Neto, A. C., Whellens, M. W., and Audus, H., 2000, ‘‘Gas Turbine 关24兴 Inoue, H., Kobayashi, N., and Koganezawa, T., 2001, ‘‘Research and Devel-
Performance Using Carbon Dioxide as Working Fluid in Closed Cycle Opera- opment of Methane-Oxygen Combustor for Carbon Dioxide Recovery Closed-
Cycle Gas Turbine,’’ Proceedings of 23th CIMAC World Congress on Com- 关28兴 Abdallah, H., Facchini, B., Danes, F., and De Ruyck, J., 1999, ‘‘Exergetic
bustion Engine Technology for Ship Propulsion, Power Generation, Rail Optimization of Intercooled Reheat Chemically Recuperated Gas Turbine,’’
Traction, Hamburg, Germany, CIMAC, London. Energy Convers. Manage., 40, pp. 1679–1686.
关25兴 Kesser, K. F., Hoffman, M. A., and Baughn, J. W., 1994, ‘‘Analysis of a Basic 关29兴 ASPEN PLUS User Guide, release 10.2.1, 2000.
Chemically Recuperated Gas Turbine Power Plant,’’ J. Eng. Gas Turbines 关30兴 Carcasci, C., and Facchini, B., 2000, ‘‘Comparison Between Two Gas Turbine
Power, 116, pp. 277–284. Solutions to Increase Combined Power Plant Efficiency,’’ Energy Convers.
关26兴 Adelman, S. T., Hoffman, M. A., and Baughn, J. W., 1995, ‘‘A Methane-Steam Manage., 41, pp. 757–773.
Reformer for a Basic Chemically Recuperated Gas Turbine,’’ J. Eng. Gas 关31兴 Strait, M., Allum, G., and Gidwani, N., 1997, ‘‘Synthesis Gas Reformers,’’
Turbines Power 117, pp. 17–23. http://www.owlnet.rice.edu/⬃ceng403/nh3ref97.html.
关27兴 Dybkjær, I., 1995, ‘‘Tubular Reforming and Autothermal Reforming of Natural 关32兴 Smith, A. R., and Klosek, J., 2001, ‘‘A Review of Air Separation Technologies
Gas—An Overview of Available Processes,’’ Fuel Process. Technol., 42, pp. and Their Integration With Energy Conversion Processes,’’ Fuel Process. Tech-
85–107. nol., 70, pp. 115–134.
Thermoeconomic Analysis of Gas
Turbine Plants With Fuel
M. Bozzolo Decarbonization and Carbon
M. Brandani
Dioxide Sequestration
A. Traverso
In this paper the thermoeconomic analysis of gas turbine plants with fuel decarbonization
A. F. Massardo and carbon dioxide sequestration is presented. The study focuses on the amine (MEA)
Mem. ASME decarbonization plant layout and design, also providing economic data about the total
capital investment costs of the plant. The system is fuelled with methane that is chemically
Thermochemical Power Group, treated through a partial oxidation and a water-gas shift reactor. CO2 is captured from the
Dipartimento di Macchine, resulting gas mixture, using an absorbing solution of water and MEA that is continuously
Sistemi Energetici e Trasporti, recirculated through an absorption tower and a regeneration tower: the decarbonized fuel
Università di Genova, gas is afterwards burned in the gas turbine. The heat required by CO2 sequestration is
Genova, Italy mainly recovered from the gas turbine exhausts and partially from the fuel treatment
section. The reduction in efficiency and the increase in energy production costs due to fuel
amine decarbonization is evaluated and discussed for different gas turbine sizes and
technologies (microturbine, small size regenerated, aeroderivative, heavy duty). The nec-
essary level of carbon tax for a conventional plant without a fuel decarbonization section
is calculated and a comparison with the carbon exergy tax procedure is carried out,
showing the good agreement of the results. 关DOI: 10.1115/1.1587744兴
Introduction C Use of Carbon Dioxide as the Working Fluid. It is possible

to build a power plant where the expander flow is composed only
The Third Conference of the Parties to the Framework Conven-
of carbon dioxide and water 关4兴.
tion on climate change at Kyoto, Japan, required industry to seek
The proposed available technologies to separate carbon dioxide
new solutions to restrict the production and the emission of green-
from a syngas are based on the principles of:
house gases from human activities. For the power industry such
1 Chemical Absorption. The use of an aqueous solution of
restrictions mainly apply to carbon dioxide which is emitted into
blended amines represents the most mature technology that can be
the atmosphere when fossil fuels 共methane, coal, oil兲 are burnt to
transferred to power plants, 关1,3,5,6兴.
produce energy.
2 Physical Absorption. The technique is based on the variation
Fossil fuels are at present, and will be for the following de-
of CO2 solubility within a solvent at different pressures, 关5,6兴.
cades, the most important source for satisfying the world’s energy
3 Adsorption. Some materials can adsorb a particular gas: The
demands. Unfortunately this precludes the possibility of drasti-
flow is led to licking up the adsorbent bed, 关5,6兴.
cally reducing carbon dioxide emissions, despite the increases in
4 Cryogenic. Carbon dioxide can be removed from a syngas
power plant conversion efficiency and the exploitation of low C/H
using low temperature condensation, 关5,6兴.
content fuels, such as methane.
5 Separation Using Membranes. Separation is achieved by
Nuclear and renewable energy could be alternative solutions
compression of the feed stream, which forces the gaseous compo-
but the use of nuclear energy raises political and security issues
nents with the higher permeability values to permeate through the
and renewables present economic and infrastructural problems.
membrane, 关5,6兴.
Therefore, research efforts are being directed towards innova-
The TPG 共Thermomechanical Power Group兲 of the University
tive power plants capable of first concentrating and then removing
of Genoa has investigated models of simple-cycle gas turbine
carbon dioxide from fossil fuels. Three main strategies have been
plants with fuel decarbonisation and CO2 sequestration. The aim
proposed:
of this work is to explain what the use of a fuel treatment and CO2
A Exhaust Gas Treatment. With very few changes to existing
chemical absorption section involves in terms of performance,
gas turbines, it would be possible to install a chemical treatment
efficiency, capital costs, and cost of energy. The study focuses on
plant able to operate on the whole exhaust flow downstream from
four different gas turbines: a microturbine Turbec T-100, a Nuovo
the gas turbine, 关1,2兴.
Pignone PGT-2, a heavy-duty Siemens V64.3a, and an aeroderiva-
B Fuel Decarbonization. Hydrogen is not directly available in
tive GE-LM6000.
nature, but it can be produced by properly treating hydrocarbons:
To carry out thermodynamic and thermoeconomic analysis the
A partial oxidation or a steam-methane reforming reaction fol-
in-house TEMP code 共ThermoEconomic Modular Program兲, 关7,8兴,
lowed by a water-gas shift reaction could be used to achieve this
was used. Three new modules were added to the code to simulate
goal, 关3兴.
a partial oxidation reactor, a water-gas shift reactor and a chemical
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN absorption plant for CO2 sequestration. Thermodynamic results
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF and chemical compositions of streams were verified with the help
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- of the commercial code PRO/II, 关5,9兴.
erlands, June 3– 6, 2002; Paper No. 2002-GT-30120. Manuscript received by IGTI, Each TEMP module is provided with a specific ‘‘cost/costing
December 2001, final revision, March 2002. Associate Editor: E. Benvenuti. function’’ which evaluates the capital cost of the component on
Fig. 1 Layout of the complete plant model in PROII. 1. LP compressor, 2. HP compressor, 3. regenerator, 4. combustion chamber,
5. expanders, 6. partial oxidation reactor, 7. water gas shift reactor, 8. absorber column, 9. regenerator column.
the basis of thermodynamic and physical parameters. This ap- temperature that the incoming water flow can completely vaporize
proach allows capital costs and cost of energy to be estimated inside the mixer 共Component M5兲. Actually, to obtain a high car-
straightaway. bon monoxide efficiency conversion during the water-gas shift
reaction the water must be in a gaseous state. The syngas from the
Plant Layout water-gas shift reactor is split into two streams that preheat the
This work is based on a national test case, 关3兴, proposed for the fuel and the water and then it is carried to the carbon dioxide
study of very low CO2 emission power plants. The complete plant separation section where it enters the first column 共absorber兲 from
consists of three sections 共see Fig. 1兲: The first section is the the bottom. Clean gas, composed mainly of hydrogen and nitro-
power section, represented by a gas turbine with a heat recovery gen, leaves the tower from the top and is sent to the power sec-
system; the second is the fuel decarbonization section, represented tion, while the products of the reaction, basically monoethanola-
by partial oxidation and water-gas shift reactors; the third is the mine, water and carbon dioxide, leave the tower from the bottom
carbon dioxide separation unit represented by a chemical absorp- and are carried to the second column 共regenerator兲 in which the
tion plant using an aqueous solution of blended amines, 关5兴. chemical bond between the carbon dioxide and amines is broken
In the test case the regenerated 2.3 MWe Nuovo Pignone PGT-2 by heat provided by the reboiler. Since the heat required is at
gas turbine was employed to represent the power section. The relatively low temperature, about 180°C, it should be recovered
machine includes two centripetal compressors working at on- from the plant for economical operation. In this case 70% of the
design conditions with a polytropic efficiency of 0.845. A water heat comes from turbine exhaust gases, while 30% comes from
injection chamber 共Component M2 in Fig. 1兲 is placed between the coolers of the syngas. Carbon dioxide is available at the top of
them: the air flow coming from the low pressure compressor is the regenerator column and should be stored after being com-
cooled by water injection and the mixture is afterwards sent to the pressed and liquefied. However, this aspect of the process is be-
high pressure compressor. Pressurised air enters the regenerator at yond the scope of this work.
12.5 bar and is heated by hot gases coming from the expander. The flue gas leaving the top of the absorber column mainly
The turbine inlet temperature is set at 1080°C. A cooling flow consists of hydrogen and nitrogen; afterwards it is compressed
equal to 6.5% of the compressor inlet air mass flow is also con- and injected into the gas turbine combustion chamber, where it
sidered 共Stream A6兲. Hot gases are reduced to 1.055 bar and are reacts with the air coming from the regenerator. The nontradi-
then sent to the regenerator and then to the heat recovery system tional fuel molar composition is likely to need apt redesign of the
in which excess heat is used for cogeneration. Downstream from combustion chamber, 关10兴.
the gas regenerator a fraction of the air flow is split into two
streams in Component SP5 from which 10% of the total flow is
sent to the fuel decarbonization section. Operating conditions for Partial Oxidation for Fuel Decarbonization
the whole plant are set so that the turbine inlet temperature is the A partial oxidator is a chemical reactor where hydrocarbons
same as the simple gas turbine cycle: The inlet methane mass flow react with the incoming air flow without reaching complete com-
is chosen to produce the necessary flow rate of hydrogen to obtain bustion, thanks to the use of specific catalysts, 关11兴.
a 1080°C turbine inlet temperature. Methane is preheated by gases In a partial oxidation reactor fed by methane the main chemical
coming from the water-gas shift reactor, then sent to the partial reaction is
oxidation reactor where it reacts with the pressurised air coming
from the power section. The mixture leaves the reactor at such a CH4 ⫹ 21 O2 ⇔CO⫹2H2 ⫹35.67 kJ/mol at 298.15 K. (1)
Table 1 Molar composition „%… of the gas mixture at the outlet
of the partial oxidation reactor
T.E.M.P. PRO II
CH4 0.25 0.28
H2 32.70 32.38
CO 16.35 16.31
CO2 1.70 1.75
H2 O 5.30 5.33
N2 43.70 43.95
The reaction shows a favourable chemical equilibrium, 关12兴, for Fig. 2 Fuel decarbonization section
a wide range of temperatures: It is thus possible to choose the
operating temperature of the reactor on the basis of the require-
ments of the whole plant. The partial oxidation reactor is simu- methane and water flows 共see Fig. 2兲. It passes through a water
lated by the TEMP code with a new module called ‘‘Oxipar.’’ The knockout drum and then it enters the CO2 separation section.
program calculates the thermodynamic conditions and the molar The present paper mainly focuses on the CO2 separation plant;
composition of the mixture in a user-defined number of sections in that respect, partial oxidation and water-gas shift reactor design
inside the reactor. In each single section fuel reacts with the in- problems have not been analyzed in depth and their capital costs
coming air flow according to the reactions of partial oxidation and have been preliminarily estimated on the basis of the required
complete combustion (CH4 ⫹2O2 ⇔CO2 ⫹2H2 O). In this case catalyst surface: Operating temperature and pressure, mass flow
methane mass flow is chosen to obtain a 1080°C turbine inlet and catalyst type are the most influential parameters, 关5兴.
temperature. The program assumes a first attempt temperature and
solves the mass-action law in an iterative way in order to obtain
the molar composition at equilibrium. In that respect, it is neces- Carbon Dioxide Absorption With Amines
sary to evaluate the chemical equilibrium of the reactions through
Of all the techniques reported in open literature, chemical ab-
equilibrium constants, provided by the in-house code LNK, 关5兴.
sorption using an aqueous solution of blended amines represents
All data given by the new TEMP module were verified using the
the most mature technology for separating carbon dioxide from a
commercial code PRO/II 共see Table 1兲.
syngas, 关5兴. The separation plant 共see Fig. 3兲 consists of an ab-
As oxygen is poisonous for the absorber column catalysts, the
sorber column 共AC兲 and a regenerator column 共RC兲.
partial oxidation reactor is expected to completely consume the
Flue gas 共Stream no. 1兲 enters the first column from the bottom
incoming oxygen.
and comes into contact with the counter flow the water and mo-
The reactor outlet temperature is set so that incoming water for
noethanolamine absorbing solution 共Stream no. 5兲. Hydrogen
the water-gas shift reaction can completely vaporize: In this way
leaves the tower from the top 共fuel stream兲 while the CO2 rich
the reactor operating temperature depends on the incoming water
solution 共Stream no. 3兲 enters the regenerator column after being
mass flow. A unitary steam to carbon ratio is assumed, 关13兴. Car-
preheated. The chemical bond between the carbon dioxide and
bon monoxide coming from the partial oxidation reactor can be
amines is broken by heat provided by the reboiler 共component
converted into hydrogen by a water-gas shift reaction:
RE兲. CO2 leaves the column from the top while the aqueous
CO⫹H2 O⇔CO2 ⫹H2 ⫹41.15 kJ/mol at 298.15 K. (2) amine solution is carried back to the absorber column. The con-
denser at the top of the tower 共CN兲 is necessary to ensure a small
In order to obtain a high carbon monoxide efficiency conver- and gradual temperature gradient through the column.
sion, water must be in a gaseous state. Liquid water can be in- The most commonly used columns for chemical absorption sys-
jected directly into the partial oxidation reactor or downstream tems are plate towers or packed towers, 关5,12,13兴. Plate towers
from it to obtain complete vaporization. The water-gas shift reac- should be preferred when a chemical reaction occurs between the
tion is exothermic and it shows favorable chemical equilibrium at flue gas and the absorbing solution. In this model the two columns
low temperatures. This is the reason why the flow from the partial are plate towers as a chemical reaction takes place, 关6兴:
oxidator must be cooled before it enters the shift reactor.
The water-gas shift reactor is simulated in the TEMP code by a 2HOC2 H4 NH2 ⫹H2 O⫹CO2 ⇔ 共 HOC2 H4 NH2 兲 2* H2 CO3 . (3)
new module named ‘‘Shift Reactor.’’ Again, the program calcu-
lates the thermodynamic conditions and the molar composition of The right-hand side reaction is favored by low temperatures
the mixture in a user-defined number of sections inside the reactor. 共37–50°C兲, while the left-hand side one is favored by high tem-
In each section the gas mixture reacts with vapor according to the peratures 共120–180°C兲. The flue gas entering the absorber column
water-gas shift reaction. The results have been verified with the
help of PRO/II. The comparison between the two codes is re-
ported in Table 2.
Gas outlet temperature is 230°C. Downstream from the reactor
the mixture is split into two streams that preheat the incoming
Table 2 Molar composition „%… of the gas mixture at the water-

gas shift reactor outlet
T.E.M.P. PRO II
CH4 0.20 0.24
H2 39.20 38.91
CO 0.30 0.28
CO2 14.33 14.48
H2 O 10.40 10.89
N2 35.57 35.20
Fig. 3 Chemical absorption separation plant
Fig. 4 Reboiler duty versus pressure Fig. 5 Reboiler temperature versus pressure
must be kept at a low temperature; in the present model the flue

gas temperature is 43°C. A challenging problem is avoiding poi- the integrated plant. To reduce reboiler duty the aqueous solution
soning components: The most important impurity in the flue gas, of amines and carbon dioxide entering the regenerator column is
apart from SO2 and NO2 , is oxygen. Oxygen can react with the preheated by the lean solution leaving the column from the bottom
aqueous solution of amines and cause the formation of stable salts 共Fig. 3兲.
which become very difficult to regenerate. This is why the oxygen The second part of the ‘‘CO2 sequestration’’ module evaluates
must be completely burnt in the partial oxidation reactor. column sizes and system costs. In the model the two towers are
The chemical absorption system is simulated in the TEMP code plate columns, so that size calculations are the same as for the
by the new module called ‘‘CO2 Sequestration.’’ The first part of distillation columns, 关12兴. One of the most important parameters
the program calculates the thermodynamic and chemical compo- is rising gas velocity: If it is too low, very strong irregularities in
sition of outlet streams 共‘‘fuel’’ and ‘‘CO2 storage’’, see Fig. 3兲, the liquid-phase flow may occur, if too high, entrainment flooding
the second part evaluates the tower design and the costs of the may cause column malfunction. Attention must also be paid to the
entire separation plant. gas-liquid ratio in order to avoid downflow flooding. Once the gas
The program calculates the mass flow of the aqueous solution velocity and mass flow are chosen, column diameter and thickness
of amines required by the absorption process on the basis of the can be evaluated. Column costs are calculated on the basis of plate
mass fraction of carbon dioxide inside the incoming flue gas. For number, plate spacing, vessel diameter, and thickness, 关12兴.
every kg/s of CO2 entering the absorber column, 19.84 kg/s of
water and 4.86 kg/s of amines are needed for the process; for the
present plant 共Fig. 3兲, 0.52 kg/s of carbon dioxide from the bottom Test Case Thermodynamic Results
and 2.52 kg/s of monoethanolammine plus 10.31 kg/s of water The regenerated Nuovo Pignone PGT-2 is capable of producing
from the top. The solution strength, that is to say the ratio between 2.35 MWe and 3.27 MWth for cogeneration with a net electrical
MEA and water, cannot be greatly increased because of corrosion efficiency of 0.3. Inlet air mass flow is 10.4 kg/s, compression
problems inside the tower, 关6兴. ratio is 12.5. Exhaust gases leave the heat recovery system at
Flue gas inlet pressure is the most important parameter for the 117°C with a carbon dioxide specific emission of 634 g/kWh.
operating conditions of the plant: It influences not only the ther- For the complete plant equipped with fuel decarbonization and
modynamic and chemical operating conditions, but also the size CO2 sequestration sections, TEMP calculates an electric power of
of the towers. The model works properly in a large range of pres- 2.30 MWe, almost the same as the simple gas turbine cycle, but
sures, from 2 to 30 bar, and it is able to simulate the match be- net efficiency decreases dramatically to 0.227. Methane mass flow
tween the separation plant and a large number of gas turbines entering the plant 共Stream 16 in Fig. 6兲 is estimated at 0.202 kg/s,
共heavy-duty, aeroderivative, and microturbines兲. The program cal- 27.3% more than the simple gas turbine cycle 共turbine inlet tem-
culates the number of trays of the absorber column on the basis of perature is the same for the gas turbine simple cycle and the
the inlet pressure; the regenerator column has a fixed number of complete cycle兲. An output partial oxidation reactor temperature
trays for every operating pressure. Tray number is a very impor- of 933°C is reached using 9.4% of the inlet fuel flow for complete
tant parameter to evaluate pressure losses, and the size and cost of combustion, so that incoming water for the water-gas shift reactor
the columns. can completely vaporise. As the water-gas shift reaction has a
The regenerator column is meant to be very pressure sensitive: favorable chemical equilibrium at low temperatures, the gas mix-
Figure 4 shows the duty delivered by the reboiler to the column ture to the reactor is cooled to 151°C in component 41 共Fig. 6兲; an
with respect to the operating pressure and different solvent regen- additional water flow 共Stream 21兲 is injected into the shift reactor
eration percentages. in order to obtain a 97.8% carbon monoxide conversion effi-
The required duty has been adimensionalised by relating it to ciency. The syngas leaves the reactor at 231°C, preheats the in-
the amount of CO2 entering the system in kilograms per second. coming methane and water flows, enters the water knockout drum
According to the results, higher regeneration percentages mean 共component 47 in Fig. 6兲 and is sent to the carbon dioxide sepa-
greater energy requirements. A 95% regeneration efficiency means ration section 共component 16兲 at 43°C.
that 95% of the total CO2 flow rate entering the regenerator col- The sequestration plant operating pressure is related to the gas
umn is stored, while 5% is pumped back to the absorber column. turbine compression ratio: in the present plant, the reboiler works
In the present model, efficiency is fixed at 95% so that the heat at 10.1 bar and requires a 1.33 MW duty at 182.4°C. The required
recovery system of the whole plant can satisfy reboiler duty. An- heat is recovered from the recovery system downstream from the
other challenging problem is reboiler operating temperature; re- gas turbine 共component 11兲 and from the heat exchanger before
boiler temperature increases with operating pressure 共Fig. 5兲. the water-gas shift reactor 共component 41兲. This restricts the com-
It is important to note that reboiler temperature is not influenced plete plant to only 3 MWth for cogeneration. Fuel leaving the ab-
by regeneration efficiency. A remarkable fact is that an operating sorber column is compressed up to 14 bar and then is injected into
pressure increase causes reboiler duty to decrease and its tempera- the gas turbine combustion chamber. Fuel mass flow rate is 0.92
ture to rise: An agreement between these two parameters must be kg/s 共low heating value is equal to 8.96 MJ/kg兲; the molar com-
found in order to avoid a nonacceptable efficiency reduction for position is reported in Table 3.
Fig. 6 Plant layout with T.E.M.P. code
The presence of traces of carbon monoxide in the fuel compo- expanding fluid to have higher heat capacity, higher enthalpy and
sition is responsible for the carbon dioxide specific emission of finally allows the expander to produce higher work. Plant carbon
the plant which is lowered to 34.6 g/kWh. dioxide specific emission is 33.4 g/kWh 共it was 573 g/kWh兲.
Case „3… The simple cycle General Electric aeroderivative
Different Gas Turbine Plants LM-6000 gas turbine produces 43.7 MWe with a 0.41 net effi-
The present study focuses on three other gas turbine plants. ciency. Inlet air mass flow is 127 kg/s, compression ratio is 29.6.
In this case 9% of the outlet compressor air flow is sent to the
Case„2… The simple cycle Siemens heavy-duty V64.3a gas decarbonization section. Methane inlet mass flow is 11.2% more
turbine has a net power output of 68 MWe with an efficiency of than the simple gas turbine cycle; carbon monoxide conversion
0.347. Inlet air mass flow is 191 kg/s, compression ratio is 16.1.
efficiency is 97.2%. As a consequence of a 25.4 bar operating
Small plant layout differences have been adopted for the fuel de-
pressure, the reboiler works at a high temperature, 225°C, and
carbonization and CO2 sequestration sections. The heat recovery
requires a 25.4 MW duty, which is completely recovered from the
system downstream from the gas turbine is meant to be a steam
HRSG downstream from the gas turbine. The plant is capable of
generator and no cooling is adopted before the water-gas shift
44 MWe net power output with a net efficiency of 0.356 and a
reactor. In the present study 8.8% of the outlet compressor air
carbon dioxide specific emission of 20.7 g/kWh 共it was 479
flow is sent to the partial oxidation reactor: Outlet temperature is
g/kWh兲.
enough to vaporize the inlet water flow for the following water-
gas shift reaction. A water flow is injected into the shift reactor in Case „4… The regenerated micro-gas turbine Turbec T-100 is
order to reduce the reaction temperature so that a 95.5% carbon capable of producing 100 kWe with a net efficiency of 0.3. Inlet
monoxide conversion efficiency can be achieved. Methane inlet air mass flow, 0.78 kg/s, passes through the centrifugal compres-
mass flow is 5.8% more than the simple gas turbine cycle for the sor and is preheated in the regenerator by hot gases coming from
same turbine inlet temperature. The reboiler requires a 49.5 MW the centripetal expander. The compression ratio is 4.5, the turbine
duty at 196°C and works at 13 bar. The heat is recovered from the inlet temperature is 950°C as no blade cooling is adopted. As far
gas turbine exhaust gases in the heat recovery steam generator. as complete plant layout is concerned, an important change has
The complete plant produces 70.1 MWe, with a net efficiency of been made: A compressor operating on the air flow fraction des-
0.321. Electric power is 2.1 MWe more than the simple gas tur- tined for the fuel decarbonization section has been inserted before
bine cycle: This is due to the hydrogen combustion reaction (H2 the partial oxidation reactor. Such a device enables sequestration
⫹O2 ⇒H2 O) whose product is steam only. This fact causes the plant operating pressure to be chosen apart from the gas turbine
compression ratio. The complete plant produces 98 kWe with an
overall efficiency of 0.22 and a carbon dioxide specific emission
Table 3 Molar composition „%… of the fuel entering the com- of 44.3 g/kWh 共it was 640 g/kWh兲.
bustion chamber The installation of a fuel treatment-carbon dioxide absorption
plant reduces whole plant efficiency for many reasons. First of all
H2 N2 CO CH4 H2 O CO2 the reboiler at the bottom of the regeneration column needs to be
52.2 47 0.4 0.2 0.2 0 fed at a high temperature 共Fig. 5兲, and a plant with low compres-
sion ratio requires much more heat to be recovered from the sys-
Fig. 7 Net efficiency reduction and CO2 specific production
versus pressure ratio
Fig. 8 Capital cost allocation for Siemens V64.3a
tem. That is the main reason why efficiencies of smaller gas tur-
bines 共Cases 共1兲 and 共4兲兲 are more significantly reduced than
larger ones. Turbine exhausts and heat recovery exchangers in the crease varies considerably: the Nuovo Pignone PGT-2 capital cost
fuel treatment section do not provide enough heat to feed the increase is the highest, about 60% more than the simple cycle, the
chemical absorption section; so, an increase in the amount of fuel Turbec T-100 one is about 49%, while the LM6000 one is the
being introduced is needed, lowering the net system efficiency. lowest, about 32%. Such an enhanced initial expenditure directly
Referring to the PGT-2, the necessary heat is recovered after the affects the cost of electricity, which is further penalized by the net
partial oxidation reactor, where part of the introduced fuel is burnt efficiency decrease.
in order to raise outlet temperature. A different solution is adopted Figure 9 shows COE versus the overall efficiency for the simple
in the microturbine-based system, where regenerator size is recycle and complete plant configurations.
duced in order to recover more heat from the gas turbine exhausts; COE becomes considerably higher 共and net efficiency consid-
a larger methane flow is then converted in the fuel treatment sec- erably lower兲 when fuel decarbonization and CO2 sequestration
tion. A fired exhaust recovery system could be a third solution, sections are included in the plant: compared to simple cycle cases,
关5兴. the COE increase goes from 47% for the Siemens V64.3a to 164%
Figure 7 shows the percentage reduction of efficiency compared for the Turbec T100. Such a result clearly demonstrates that car-
to the simple cycle case for the considered plants together with bon dioxide sequestration becomes economically feasible and sus-
carbon dioxide specific storage of the complete plant configura- tainable in a deregulated energy market only if a heavy carbon-tax
tion and carbon dioxide specific emission of the simple cycle case. 共CT兲 is imposed on power plant carbon dioxide emissions. The
break-even point between the imposition a carbon tax and the use
Thermoeconomic Results of a sequestration plant has been evaluated for the gas turbines
considered: Figure 10 shows the break-even 共or economically
The installation of fuel decarbonisation and carbon dioxide se- equivalent兲 carbon tax versus net electrical efficiency. The result-
questration systems is responsible for a considerable increase in ing CT strongly depends on complete plant efficiency and size:
plant complexity which affects overall efficiency, capital costs, The Siemens V64.3a shows the lowest equivalent carbon tax
and cost of energy 共COE兲. The TEMP code is provided with value with $45 per ton of carbon dioxide emitted if in the simple
‘‘cost/costing equations,’’ 关8兴, that evaluate single component cycle configuration; the LM6000 and PGT-2 require, respectively,
capital costs on the basis of thermodynamic and physical param- $63/tonCO2 and $90/tonCO2 carbon taxes.
eters. In the framework of this work, the compressor, combustor, An alternative approach to deal with the problem of global
expander, and generator costing equations have been updated to warming is the carbon exergy tax 共CET兲 proposed by the authors,
obtain values conforming to current literature for the simple gas 关15–17兴. Such a complex procedure, based on the internal ther-
turbine cycle, 关14兴. New costing equations have been set up for moeconomic analysis of energy systems, allows a carbon tax to be
partial oxidation and water-gas shift reactor modules and for the evaluated for each power plant through properly pricing the inter-
CO2 separation section. As previously described, the latter capital nal irreversibilities of energy systems. The aim is to penalize the
cost is mainly affected by operating pressure, mass flow, column most inefficient cycles and to strongly push the energy market
size, plate type, plate number, and plate spacing. towards the most innovative and environmentally sustainable so-
The thermoeconomic analysis was firstly carried out on the
simple cycle: The results were verified with the data published in
关14兴. Then, such an analysis was extended to the complete plant,
in order to estimate both capital cost and variable cost increases
that lead to a significantly higher cost of electricity.
As far as investment cost is concerned, Fig. 8 shows, as an
example, the percent capital cost allocation of the Siemens V64.3a
simple cycle and complete plant configuration. The compressor is
the most expensive component requiring about 35.5% of the total
investment, followed by the expander 共21%兲 and the generator
共13%兲. The carbon dioxide absorption plant has almost the same
influence on total costs as the heat recovery steam generator
共HRSG兲 with 11%, while partial oxidation and water-gas shift
reactors 共REACT兲 make up less than 3%. The complete plant
capital cost is estimated to be 39.5% higher than the simple cycle
case.
Cost allocation for the other gas turbines examined is similar to
the Siemens machine. However, the total capital investment in- Fig. 9 Cost of electricity „8000 equivalent hours…
Acknowledgments
This work has been sponsored by MURST of Italy through a
Cofin.99 grant.
The authors wish to thank D. Watson 共Rolls-Royce兲 for his
invaluable help in developing the CO2 sequestration analysis.
References
关1兴 Corti, A., Lombardi, L., and Manfrida, G., ‘‘Absorption of CO2 With Amines
in a Semiclosed GT Cycle: Plant Performance and Operating Costs,’’ ASME
Paper No. 98-GT-450.
Fig. 10 CET curve and equivalent CT 关2兴 Cau, G., and Cocco, D., 2000, ‘‘Performance Assessment of a Semi-Closed
Chemically Recuperated Gas Turbine Systems,’’ ASME Paper No. 2000-GT-
161.
关3兴 Lozza, G., and Chiesa, P., 2000, ‘‘Natural Gas Decarbonization to Reduce CO2
lutions. In 关16兴, a new simplified approach for plants only produc- Emission From Combined Cycles. Part A: Partial Oxidation,’’ ASME Paper
ing electrical power that well approximates the results of the com- No. 2000-GT-163.
plete CET procedure is shown to be possible, and it is called the 关4兴 Jackson, A. J. B., Neto, A. C., Whellens, M. W., and Audus, H., 2001, ‘‘Gas
generalized CET. The generalized CET curve for the considered Turbine Performance Using Carbon Dioxide as Working Fluid in Closed Cycle
gas turbine cycles is reported in Figure 10. Except for the T-100 Operation,’’ ASME Paper No. 2000-GT-153.
关5兴 Bozzolo, M., and Brandani, M., 2001, ‘‘Study of Power Plants With CO2
microturbine, which is likely to be the less apt configuration for Sequestration Fed by Hydrogen From Fuel Decarbonization,’’ Degree thesis,
CO2 sequestration because of its very small size, the resulting University of Genoa.
Carbon Tax is in surprisingly good agreement with the evaluated 关6兴 Squeri, S., and Schiappacasse, R., 1999, ‘‘CO2 Capture and Sequestration
break-even CTs. This unexpected link demonstrates that the Gen- Study,’’ Degree thesis, University of Genoa.
eralized CET can be an effective tool in a deregulated energy 关7兴 Agazzani, A., and Massardo, A. F., 1997, ‘‘A Tool for Thermoeconomic Analy-
market to encourage the spread of low-CO2 -emitting plants by sis and Optimization of Gas, Steam, and Combined Plants,’’ ASME J. Eng.
Gas Turbines Power, 119, pp. 885– 892.
making them competitive with conventional configurations.
关8兴 Massardo, A. F., and Scialò, M., 2000, ‘‘Thermoeconomic Analysis of Gas
Turbine Based Cycle,’’ ASME J. Eng. Gas Turbines Power, 122, pp. 664 – 671.
Conclusions 关9兴 Simulation Sciences Inc., PRO/II 5.5 User’s guide.
The thermoeconomic models of four differently sized gas tur- 关10兴 Bozza, F., Cameretti, M. C., and Tuccillo, R., 2001, ‘‘Una metodologia di
bine plants with fuel decarbonization and carbon dioxide seques- analisi integrata di turbine a gas a bassa emissione di CO2 ,’’ X Convegno
TESEC Sergio Stecco, Genova, pp. 441– 456.
tration have been successfully built and verified. The CO2 sepa- 关11兴 Alpino, A., and Oliva, M., 2000, ‘‘Modellizzazione di un ossidatore parziale ed
ration section, based on the use of an aqueous solution of blended integrazione con impianti di potenza tradizionali,’’ Degree thesis, University of
amines, has been investigated in depth and a preliminary design Genoa.
study has been carried out in order to evaluate its capital cost. 关12兴 Perry, H., and Green, D., 1997, Perry’s Chemical Engineers’ Handbook, 7th
The results of the thermoeconomic analysis show that the con- ed., McGraw-Hill, New York.
siderable total investment cost increase is accompanied by a sig- 关13兴 Bosio, A., Bozzolo, M., Brandani, M., and Massardo, A. F., 2001, ‘‘Modelliz-
zazione di un impianto a gas da 2.3 MWe con emissioni nulle di biossido di
nificant net efficiency decrease, causing the COE to reach very
carbonio,’’ X Convegno TESEC Sergio Stecco, Genova, pp. 481– 496.
high values. The break-even carbon tax that can make CO2 se- 关14兴 Gas Turbine World Handbook 2000–2001, Pequot, CT.
questration economically feasible when applied to simple gas tur- 关15兴 Borchiellini, R., Massardo, A. F., and Santarelli, M., 2000, ‘‘An Analytical
bine cycles is shown to be in the range of $45– 90/tonCO2 for Procedure for the Carbon Tax Evaluation,’’ Energy Conversion and Manage-
conventional machines and very much higher for very small size ment Journal, 41, pp. 1509–1531.
gas turbines, which appear to be definitively excluded from CO2 关16兴 Traverso, A., Santarelli, M., Massardo, A. F., and Cali, M., 2002, ‘‘A New
sequestration. Generalized Carbon Energy Tax for CO2 Emission Internalisation: An Effec-
tive Rule to Control Global Warming,’’ ASME Turbo Expo 2002, Amsterdam,
The generalized carbon exergy tax approach, proposed by the submitted.
authors, seems to be an effective tool for encouraging the spread 关17兴 Traverso, A., 2000, ‘‘Thermoeconomic Analysis of STIG, RWI and HAT
of low-CO2 -emitting plants by making them competitive with Cycles With Carbon Dioxide (CO2 ) Emissions Penalty,’’ Degree thesis, Uni-
conventional configurations. versity of Genoa.
Comparative Analysis of
J. H. Kim
Senior Researcher,
Off-Design Performance
Aeropropulsion Department,
Korea Aerospace Research Institute,
Daejeon 305-600, Korea
Characteristics of Single and
Two-Shaft Industrial Gas Turbines
T. S. Kim
Assistant Professor, Off-design steady performance and operating characteristics of single and two shaft gas
Department of Mechanical Engineering, turbines for electric power generation have been investigated comparatively. A set of
Inha University, balance equations has been derived based on validated component models. A simulta-
Incheon 402-751, Korea neous calculation scheme has been employed, which is flexible to various engine configu-
rations. Part-load performance analyses of two commercial gas turbines have been car-
ried out to compare operating characteristics between single and two shaft engines. The
J. L. Sohn predicted performance characteristics of both engines coincide soundly with the manu-
BK 21 Assistant Professor facturer’s data and also correspond with the inherent characteristics of each configura-
tion. The adoption of the variable inlet guide vane (VIGV) modulation has been addressed
S. T. Ro in order to examine the possibility of leveling up the heat recovery capacity by maintain-
Professor, ing a high turbine exhaust temperature (TET) when those gas turbines are used for
Fellow ASME combined cycle plants. Maintaining TET at its design value as far as the VIGV modula-
tion allows has been simulated and it has been determined that the TET control is possible
School of Mechanical and Aerospace at up to 40% and 50% load in the single and two shaft engine, respectively. Combined
Engineering, cycle performances have also been investigated for two engine configurations in different
Seoul National University, operating modes. While the VIGV modulation produces a favorable influence over the
Seoul 151-742, Korea combined cycle performance of the single-shaft configuration, the two-shaft engine does
not appear to be effectively improved by the VIGV modulation since the degradation of
gas turbine performance counteracts the advantage of the higher performance of the
bottoming (steam turbine) cycle. 关DOI: 10.1115/1.1615252兴
Introduction corresponding to 20% to 50% in load reduction, 关2,3兴. In multi-

shaft engines, compressor shaft speed and thus airflow rate de-
Gas turbines for electric power generation usually operate at
crease as power reduces. Their part-load exhaust temperatures are
part-load conditions for a considerable amount of the time. Re-
higher than those of single shaft engines without VIGV control.
gardless of gas turbine configurations, performance generally de- Out of these descriptions, a question arises as to which of the
grades with a reduction in power. Accordingly, attention should be two alternatives 共single and two-shaft engines兲 are preferred, con-
paid to the task of enhancing the part-load performance of the gas sidering the part-load performance of both the gas turbine and the
turbine itself and furthermore, the performance of applied systems heat recovery system. Another factor for consideration is how
such as the combined cycle and cogeneration power plants. Most much the heat recovery performance would be affected when the
gas turbines developed for electric power generation 共heavy-duty兲 VIGV modulation is employed in the two shaft engine. This study
have a single-shaft configuration as shown in Fig. 1. The speed of aims at answering these questions through a comparative off-
the generator driven by the gas turbine should remain constant design performance analysis at part-load conditions. A program
during the load variation. Therefore, the single shaft engine is has been developed to analyze the part-load performance of gas
advantageous in the event of the load being shed because the turbines. In particular, the compressor model adopts an improved
compressor acts as a very efficient brake. For this reason, regula- stage-by-stage calculation enabling a practical estimation of
tion of output speed is easier to achieve than other configurations, VIGV and VSV control. The program has been validated by the
关1兴. Aeroderivative gas turbines have long been representatives of accurate prediction on the part-load performance of actual single
multishaft power generation engines. Recently, a couple of heavy- and two shaft engines. Subsequently, for single and two-shaft gas
duty engines with the two-shaft configuration shown in Fig. 2 turbines with similar design parameters 共pressure ratio, turbine
have also been introduced. Meanwhile, during the part-load op- inlet temperature, etc.兲 and performance 共efficiency and power兲,
eration of modern single shaft heavy-duty gas turbines, the airflow analysis has been carried out to compare the part-load perfor-
rate is actively controlled by modulating the setting angles of the mance. The adoption of VIGV modulation has been investigated
variable inlet guide vane 共VIGV兲 and the variable stator vanes to examine the possibility of leveling up the heat recovery capac-
共VSVs兲 in order to maintain the highest possible exhaust gas tem- ity when the gas turbines are used for combined cycle plants.
perature 共design temperature or higher兲. Thus, heat recovery ca- Variation of performance parameters and the range of VIGV con-
pacity is greater compared with the maximum airflow control trolled operations has been analyzed in detail. Combined cycle
共fuel only control兲. Practical mass reduction limit is 10% to 30% performances have likewise been investigated for two engine con-
figurations in different operating modes.
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- Off-Design Performance Analysis
erlands, June 3– 6, 2002; Paper No. 2002-GT-30132. Manuscript received by IGTI, The performance and operating characteristics of a gas turbine
December 2001, final revision, March 2002. Associate Editor: E. Benvenuti. in off-design conditions depend on individual components such as
These equations have been applied to each control volume cor-
responding to a component or a subcomponent 共for example, a
stage兲 with suitable F, Q̇, and Ẇ s , which must be specified at
every off-design condition. The force and shaft power in the bal-
ance equations have been determined by component characteris-
tics, which can be derived from either characteristic maps or off-
design models. In the above two equations, ‘‘*’’ represents the
value evaluated from component characteristics at a given inlet
Fig. 1 The schematic diagram of a single-shaft gas turbine condition.
Component Models
the compressor and the turbine. When they are linked together in Compressor. Multistage axial flow compressors of modern
an engine, the potential operating range of these components be- heavy-duty gas turbines usually incorporate VIGV and VSVs for
comes considerably reduced. It is, therefore, very important to stable startup and enhancement of exhaust heat recovery at part-
accurately estimate the performance characteristics of each com- load in combined cycle plants. In addition, some portion of the
ponent when attempting to predict the overall performance of the compressed air is extracted from intermediate stages of the com-
gas turbine. The performance characteristics of each component pressor for surge protection and turbine blade cooling. As such,
can be obtained using its mathematical model or actual test data. the operating characteristics of modern compressors are too com-
Once the operating characteristics of each component has been plicated to be handled by conventional methods, which generate a
specified, the performance and operating characteristics of the gas performance map first and then utilizes it in an engine simulation.
turbine during off-design conditions can be derived using ther- An efficient method, proposed by the authors of this paper, 关4,7兴,
mofluid dynamic and mechanical matching. is used in the present analysis to calculate the performance of
In the present study, component models have been adopted modern multistage axial flow compressors. The method requires
from previous studies, 关4,5兴. Special efforts have been made to stage performance maps but it differs from the conventional se-
estimate characteristics of multistage axial flow compressors and quential stage-stacking method in that all interstage state variables
turbines including the effect of intensive turbine blade cooling. are calculated simultaneously. Such a feature is very suitable for
Mass, momentum, and energy balances for each component 共each whole engine simulation, which is also based on simultaneous
stage in the cases of compressor and turbine in particular兲 have calculation. Generalized stage characteristics, 关8兴, have been used,
been considered for thermofluid dynamic matching. Work compat- which are represented by relations among flow coefficient, pres-
ibility has also been taken into account in order to describe me- sure coefficient, temperature coefficient, and efficiency. At
chanical matching, for example, between compressor and gas gen- changed vane setting angles, stage characteristics are modified by
erator turbine, or power turbine and load. These whole balances adopting an analytical approach, 关9兴. The feasibility of the method
are consequently represented by a set of balance equations. The can be found in previous papers, 关4,7兴, where comparisons be-
performance and operating characteristics of a gas turbine engine tween predicted characteristics and field data for both fixed and
is then obtained by solving these equations simultaneously. This variable geometry compressors have been presented.
simultaneous calculation scheme is more flexible to various en- Once the characteristics of the compressor have been specified,
gine configurations than other previous iterative sequential ‘‘*’’-values can be obtained at given inlet conditions. After that,
schemes, 关1,6兴, where the detailed calculation procedure must be the power required to drive the compressor can be determined by
modified as engine configurations change. In this study, without the following equation:
altering the calculation structure, a new component can be added nC
without difficulty and the balance equations will correspond to the
new components. This method can also be applied to a transient
Ẇ C ⫽ 兺 ṁ 共 H ⫺H *
i
i i i⫹1 兲 . (6)
analysis by merely adding unsteady terms to the balance equa-
tions. In practice, the balance equations for off-design perfor- Turbines. Modern multistage axial flow turbines also have
mance prediction in the present study are derived from the basic complicated operating characteristics mainly due to blade cooling.
governing equations in the previous studies, 关4,5兴, with unsteady The bleed air from the compressor is entrained into the turbine
terms neglected as follows: blade rows for blade cooling and thus influences stage character-
istics. Therefore, the effect of coolant flow must be considered in
mass: ṁ i⫹1 ⫺ṁ i ⫽0 (1) evaluating stage characteristics. For this reason, a row-by-row
momentum: ṁ i⫹1 u i⫹1 ⫺ṁ i u i ⫹p i⫹1 A i⫹1 ⫺p i A i ⫺F⫽0 (2) prediction method for the cooled turbine, 关10兴, has been employed
to obtain performance characteristics. An expansion pressure ratio
energy: ṁ i⫹1 H i⫹1 ⫺ṁ i H i ⫺Q̇⫹Ẇ s ⫽0 (3) has been obtained in off-design conditions by the following
Stodola equation:
冑冉冊
where
ṁ in 冑T T,in P out 2
* ⫺ṁ i u i ⫹p i⫹1
* u i⫹1
F⫽ṁ i⫹1 * A i⫹1 ⫺p i A i (4) ⫽K 1⫺ . (7)
P in P in
Ẇ s ⫽ṁ i 共 H i ⫺H i⫹1
* 兲. (5)
K has been deduced from design point values and kept constant
at off-design conditions. Total pressure loss due to the blade cool-
ing has been evaluated by using the model proposed by Kim et al.
关11兴. The amount of the coolant flow entrained to the turbine blade
rows at off-design conditions has been determined using the fol-
lowing equation:
ṁ c ⫽K c p c 冑 2 共 1⫺p T /p c 兲
RT c
. (8)
The coolant flow is driven by static pressure differences be-

tween points where the coolant is extracted and injected. K c is
Fig. 2 The schematic diagram of a two-shaft gas turbine also obtained from design conditions and p T is static pressure at
row inlet and exit for nozzle and rotor, respectively. As in the
compressor model, the turbine power can be obtained by an equa-
tion similar to Eq. 共6兲.
Combustor and Guide Ducts. Total pressure losses in the
combustor, inlet and exhaust ducts in off-design conditions have
been corrected using the following equation 关1兴:
共 ⌬ P/ P in 兲共 ṁ 冑T/ P 兲 in
2
R
⫽ . (9)
共 ⌬ P/ P in 兲 d 共 ṁ 冑T/ P 兲 in,d R d
2
The addition of thermal energy caused by combustion can be

treated as heat transfer to the combustor and introduced in the
energy balance equation of the combustor using the following
form:
Q̇⫽ ␩ combṁ f 共 LHV⫹⌬H T 0 ⫺T f 兲 . (10)
The subscript 0 indicates a reference point where the low heat-
ing value 共LHV兲 is defined. ␩ comb designates combustion effi-
ciency. Changes in gas composition and thermodynamic proper-
ties subsequent to combustion have also been taken into account.
Work Compatibility. Since steady-state off-design condi- Fig. 3 Calculation procedure for off-design performance
tions are being dealt with in this study, the net power produced by analysis at part-load conditions
the engine should be equal to the power or load required at any
speed. Otherwise, the engine will not be in equilibrium and will
either accelerate or decelerate depending on whether there is a
erator represented by Eq. 共12兲 should be included in the simulta-
surplus or deficiency of power. In the case of a single-shaft gas
neous calculation in order to solve the gas generator speed. With
turbine, a turbine drives both the compressor and the output load
the purpose of solving these state variables including the gas gen-
共electric generator兲 at a constant speed. The following equation
erator speed, the multivariable Newton-Raphson method has been
therefore represents the work compatibility:
employed.
Ẇ T ⫺Ẇ C ⫺Ẇ L ⫽0 共 at a constant speed兲 . (11) In actual situations, the net power and TET of an engine are
controlled by modulating fuel flow rate and VIGV angle. Numeri-
In the two-shaft configuration, the gas generator turbine drives cal simulation is carried out in the same manner, that is, the fuel
the compressor and the power turbine drives the output load. Un- flow rate and VIGV angle are modulated numerically so as to
like the single-shaft engine, the gas generator shaft is not linked to obtain the required power (Ẇ L ) and TET as shown in Fig. 3. As
the load; hence its speed is reduced as power diminishes. For a mentioned in the previous section, balance equations correspond-
given power or gas generator speed, the power turbine speed may ing to new components are simply added to the ‘‘simultaneous
vary over a wide range, depending on the load characteristics calculation’’ module shown in Fig. 3 when a new engine configu-
共mechanical drive, for example兲. In the present study, however, ration is simulated.
the power turbine speed has been fixed at the rated speed because
it must drive an electric generator. Two equations for work com-
patibility are required in the case of two shaft engine and are
Results and Discussion
presented as follows:
Gas Turbines. The performance and operating characteristics
Ẇ C ⫺Ẇ GT ⫽0 共 at variable speeds兲 (12)
of both a single and a two-shaft version of Titan™ 130 共Solar
Ẇ PT ⫺Ẇ L ⫽0 共 at a constant speed兲 . (13) Turbines Inc.兲 at steady-state part-load conditions have been in-
vestigated in the present study. Because they incorporate almost
Numerical Treatment the same components 共except for turbines兲 and have the same
power capacity level, an investigation of off-design performance
As mentioned in the previous section, the balance equations and of the two engines might indicate inherent differences between the
component characteristics in each component comprise a set of two configurations and could present reasonable comparative re-
algebraic equations. With a given fuel flow rate and VIGV angle, sults. The design point performance parameters and specifications
performance characteristics such as net power and TET can be of both engines are shown in Table 1, which have mostly been
obtained by solving this equation set. Static pressure, temperature, provided from technical brochures, 关12,13兴, and supplemented
and axial velocity at each station 共control surface兲 are state vari- with reference to Rocha and Etheridge 关14兴, Rocha and Kurz 关15兴,
ables to be solved. Therefore, there are 3(n⫹1) variables for n and Farmer 关16兴.
control volumes and three of them are given as known values, As seen in the component models, the majority of off-design
which are the inlet total temperature, pressure, and exhaust static characteristics have been evaluated by corrections based on design
pressure of the engine. Consequently, the remaining 3n variables point performances. Therefore, also required are detailed design
can be obtained as solutions of the equation set since there are 3n point performances and specifications such as stage efficiencies,
equations 共mass, momentum and energy balance equations for n stage pressure ratios, and the amount of bleed air, etc., which are
control volumes兲. The speed of the shaft linked to the load that hardly obtainable in most commercial engines. Accordingly, the
should be kept constant in electric power generation is a known design point analysis has been carried out based on the previous
variable both in a single and two shaft engine at steady-state part- work, 关17兴, so as to obtain such performance data. The predicted
load conditions. Hence, work compatibility Eqs. 共11兲 and 共13兲 are compressor and turbine stage efficiencies were 88%, 87%, and
not those to be solved but are rather to be used as criteria for 88% and 89% for the single and two-shaft engines, respectively.
determining the fuel flow rate and VIGV angle corresponding to As far as the total amount of coolant flow is concerned, it was
the required power (Ẇ L ) and TET. In the case of the two shaft estimated at 16.5% and 15.7% of the inlet airflow rate for the
engine, however, the work compatibility equation for the gas gen- single and two-shaft engines, respectively. These predicted data
Table 1 Design point performance parameters and specifica-
tions of Titan™ 130 engines
Gas Turbine Configuration Single Shaft Two Shaft

Electrical power, MW 13.5共13.5兲 14.5共14.5兲
Cycle efficiency, % 33.3共34.2兲 35.7共36.1兲
Compressor† 14 stages 14 stages
Speed, rpm 11,220 11,220
Inlet airflow rate, kg/s 48 47.7
Pressure ratio 17 16.4
Gas Generator Turbine 3 stages 2 stages
Firing temperature, °C 1,149 1,149
Exit temperature, °C 489共501兲
Power Turbine 2 stages
Speed 共rated兲, rpm 7,970
Exit temperature, °C 492共497兲
†
The compressor incorporates one VIGV and five VSVs. The values inside the pa-
rentheses are results of simulation. Fig. 5 Equilibrium running lines of the Titan™ 130 single and
two-shaft gas turbines
are thought to be reasonable compared with typical values of

modern engines. The simulated outputs are also shown in Table 1, Variations in parameters of the single and two shaft engine are
and are very similar to the actual data. shown in Fig. 6 and Fig. 7, respectively, compared with the data
To show the validity of the off-design analysis program, a com- provided by Solar Turbine Inc, 关12,13兴. The overall operating
parison between field data and simulation results for a single-shaft characteristics predicted in this study are in good agreement with
gas turbine, 关4兴, is exemplified in Fig. 4. Simulated results coin- the corresponding data from the manufacturer and these results
cide in accordance with manufacturer’s data. TET has been kept further confirm the validity of the component models. Clear dis-
constant for considerable power range through VIGV modulation. tinction between single and two shaft engines is the inlet airflow
Maximum airflow reduction is 30%, which corresponds to a 50% rates at identical load capacities, as shown in Fig. 5, Fig. 6, and
power reduction. This has been realized by the modulation of four Fig. 7. Compared with the single-shaft engine, the two-shaft en-
VSVs in addition to VIGV. The related VIGV and VSV modula- gine requires greater specific power 共power per unit airflow rate兲
tion has been adopted for the simulation of the present gas tur-
bines equipped with five VSVs as well as VIGV. Adopted from
the example calculation is the criterion regarding the limitation of
the airflow reduction during VIGV modulation, which will be ex-
plained further on.
At part-load conditions, the performance and operating charac-
teristics of the two engines were initially investigated without the
incorporation of VIGV modulation. Figure 5 displays operating
lines, known as ‘‘equilibrium running lines,’’ over the part-load on
the compressor characteristic map. The compressor characteristic
map 共constant speed lines in particular兲 is produced by an inde-
pendent compressor map generation routine that adopts the same
models as this work, 关7兴. While the single-shaft engine operates at
a constant speed, the gas generator speed of the two shaft engine
varies up to about 70% of the rated speed at no load, which results
in a drastic reduction in inlet airflow rate. Upon reference to per-
formance charts of the typical two shaft engine used in Walsh and
Fletcher 关6兴, about 75% of the rated speed at no load is observed.
This coincides soundly with the results of this study. Fig. 6 Predicted part-load characteristics compared with en-
gine data of the single-shaft version of Titan™ 130 „lines: pre-
diction, marks: data from †12‡…
Fig. 4 Predicted part-load characteristics compared with Fig. 7 Predicted part-load characteristics compared with en-
engine data of Siemens V64.3 „lines: prediction, marks: data gine data of the two shaft version of Titan™ 130 „lines: predic-
from †3‡… tion, marks: data from †13‡…
Fig. 8 Predicted part-load characteristics of the single shaft Fig. 10 Thermal efficiencies of the Titan™ 130 single and two
version of Titan™ 130 with VIGV modulation shaft gas turbines in different operating modes
in order to produce equivalent power due to the larger reduction in can operate at the design point TET up to 50% load by the VIGV
the inlet airflow rate. Accordingly, the two-shaft engine operates at modulation. However, a great difference in mass flow rate and
higher temperatures 共TRIT,TET兲 over the part-load conditions so pressure ratio cannot be found compared with the two shaft engine
as to acquire greater specific powers, as shown in Fig. 7. without VIGV modulation. From these results, it can be found that
At present, it does not seem that the VIGV and five VSVs are the VIGV modulation in two shaft engines does not change the
actively modulated during load changes except for the surge mar- equilibrium working line shown in Fig. 5. However, the closing of
gin control purpose taking place during startup and shutdown, VIGV changes operating characteristics of the compressor them-
关14兴. In this study, however, their active modulation to control selves such that constant speed lines as shown in Fig. 5, shift to
TET has been examined with the intention of enhancing the heat lower mass flow rate regions. For a given mass flow rate and
recovery capacity. It is assumed that the design point VIGV and pressure ratio, therefore, the gas generator operates at an increased
VSV angles are 15° and that they are connected by a common link speed as shown in Fig. 9. Thermal efficiencies of the single and
device driven by an actuator, 关7兴. The criterion for maximum pos- two-shaft engines in two different operating modes are shown in
sible airflow reduction by VIGV closing has been deduced from Fig. 10. Without VIGV modulation, the efficiency of the two shaft
the results of the previous study, 关4兴, where the part-load operating engine is higher than that of the single-shaft engine. Compressor
characteristics of Siemens V64.3 and GE 7F were evaluated. Ac- efficiency varies a great deal due to the change in VIGV and VSV
cording to the result analysis, it has been found that the first stage angles and may affect the engine efficiency. VIGV modulation
has the largest normalized flow coefficient at maximum VIGV reduces compressor efficiency as shown in Fig. 11. In particular,
closing and that its value is within 1.25–1.3, corresponding to a the degradation in engine efficiency during VIGV modulation for
nearly zero pressure rise in the generalized curve by Muir et al. the two shaft case may be explained by a drastic decrease in
关8兴. This can be an appropriate criterion for determining the load compressor efficiency. Once the VIGV is fully closed, compressor
up to which the VIGV closing can be continued. efficiency improves as power decreases further, which results in
The performance and operating characteristics of the two en- engine efficiency comparable to or rather higher than the case
gines with VIGV modulation are investigated and results are without VIGV modulation.
shown in Fig. 8 and Fig. 9. In the case of the single-shaft engine,
it is predicted that the inlet air-flow rate can be reduced up to Combined Cycle Consideration. In this section, the perfor-
about 60% of the design point value and TET can remain constant mance characteristics of combined cycles adopting the single and
up to 40% load reduction. This TET control range is even greater two-shaft gas turbines have been analyzed. Special attention has
than that of the Siemens V64.3 shown in Fig. 4 because the been given to the effect of VIGV modulation. A single pressure
present engine adopts a greater number of VSVs. An enhancement bottoming system 共heat recovery steam generator and steam tur-
of heat recovery, therefore, can be expected by the adoption of the bine兲 has been analyzed with an off-design calculation routine. A
VIGV and VSVs in a single-shaft engine. The two-shaft engine detailed analysis method can be found in Ref. 关18兴. The design
Fig. 9 Predicted part-load characteristics of the two shaft ver- Fig. 11 Compressor efficiencies of the Titan™ 130 single and
sion of Titan™ 130 with VIGV modulation two shaft gas turbines in different operating modes
Table 2 Design point performance parameters of the com- simple 共without VIGV modulation兲 two shaft engine in the higher
bined cycle power range 共over 50%兲. In the lower power range, the two shaft
engine is superior to the VIGV modulated single shaft engine,
Gas Turbine Configuration Single Shaft Two Shaft
which is mainly due to the higher gas turbine performance as
Steam pressure, kPa 3000 3000 indicated in Fig. 10. In the two shaft engine, VIGV modulation
Condensing pressure, kPa 4.2 4.2 results in higher steam turbine power for a given gas turbine as in
Steam turbine power, MW 6.17 6.00
Combined cycle power, MW 19.72 20.50 the single shaft case. However, the relatively lower gas turbine
Combined cycle efficiency, % 49.8 51.1 efficiency 共see Fig. 10兲 counteracts the bottoming cycle advan-
tage. Hence, the overall combined cycle efficiency is not effec-
tively influenced by the VIGV modulation as shown in Fig. 13
where both efficiencies fall on the same line. As explained previ-
calculation is performed for given gas turbine exhaust data 共TET ously, the lower gas turbine efficiency is mainly due to the de-
and mass flow兲 with both the specified heat transfer information creased compressor efficiency. Therefore, if the compressor per-
on the heat recovery sections and the steam turbine efficiency. The formance were prevented from degrading, improved combined
main features of the off-design calculation include sliding pres- cycle efficiency could be achieved.
sure operation, heat transfer data correction, and constant turbine
swallowing capacity. Major design parameters are summarized in Conclusion
Table 2. Figure 12 shows the variation of steam turbine power Part-load behavior of two gas turbines of single and two-shaft
with gas turbine power and Fig. 13 depicts the variation of com- configurations with similar design parameters 共pressure ratio, tur-
bined cycle efficiency. Without VIGV modulation for both single bine inlet temperature, etc.兲 and performance 共efficiency and
and two-shaft engines, the two-shaft case exhibits higher heat re- power兲 has been comparatively investigated. A practical estima-
covery capacity and results in far better combined cycle effi- tion of VIGV and VSV control is possible due to a novel com-
ciency, which is due to the much-increased TET. It is noticeable pressor model adopting an improved stage-by-stage calculation.
that the VIGV modulation for the single-shaft gas turbine brings For the simple mode, efficiency of the two-shaft engine is higher
about a very large increase in steam turbine power and thus than that of the single-shaft engine, especially in the lower power
achieves a considerable improvement in combined cycle effi- region. In the VIGV modulated single shaft case, the maximum
ciency, which also comes from the increased TET. This is the airflow reduction is estimated to be as much as 40%, correspond-
reason why VIGV modulation has been adopted in modern single- ing to a 60% reduction in power. In the two shaft engine, the
shaft gas turbine power plants. VIGV modulation reduces the thermal efficiency in the higher
It should also be noted that the VIGV modulated single-shaft power range due to a considerable reduction in the compressor
engine results in even higher combined cycle efficiency than the efficiency. In the combined cycle application, the VIGV modula-
tion increases the combined cycle efficiency considerably in the
case of the single shaft engine. The VIGV modulated single-shaft
engine exhibits even higher efficiency than the simple two shaft
engine in the higher power range but shows lower efficiency in
the lower power range. The combined cycle efficiency of the two-
shaft engine does not appear to be effectively improved by the
VIGV modulation since the degradation of gas turbine perfor-
mance counteracts the advantage of higher bottoming system
performance.
Acknowledgment
This work was supported by the Brain Korea 21 project.
Nomenclature
A ⫽ area
F ⫽ force
H ⫽ total enthalpy
Fig. 12 Variations in steam turbine power with gas turbine ṁ ⫽ mass flow rate
power in different operating modes
N ⫽ rotational speed
n ⫽ number of stages or control volumes
P ⫽ total pressure
p ⫽ static pressure
PR ⫽ total to total pressure ratio
Q̇ ⫽ heat transfer rate
R ⫽ gas constant
T ⫽ total temperature
TIT ⫽ turbine inlet temperature
TRIT ⫽ turbine rotor inlet temperature 共firing temperature兲
u ⫽ axial velocity
VIGV ⫽ variable inlet guide vane
VSV ⫽ variable stator vane
Ẇ ⫽ power
Subscripts
C ⫽ compressor
Fig. 13 Variations in combined cycle efficiency with combined c ⫽ coolant
cycle power in different operating modes d ⫽ design point
f ⫽ fuel taneous Calculation of Interstage Parameters,’’ Proc. Inst. Mech. Eng., Part A,
GG ⫽ gas generator shaft J. Power Energy, 215, pp. 89–98.
关8兴 Muir, D. E., Saravanamuttoo, H. I. H., and Marshall, D. J., 1989, ‘‘Health
GT ⫽ gas generator turbine Monitoring of Variable Geometry Gas Turbines for the Canadian Navy,’’
i ⫽ control volume index ASME J. Eng. Gas Turbines Power, 111, pp. 244 –250.
in ⫽ inlet 关9兴 Klapproth, J. F., 1958, ‘‘Effects of Stage Characteristics and Matching on
L ⫽ load Axial-Flow-Compressor Performance,’’ Trans. ASME, 80, pp. 1290–1291.
out ⫽ outlet 关10兴 Kim, T. S., and Ro, S. T., 1997, ‘‘The Effect of Gas Turbine Coolant Modu-
lation on the Part Load Performance of Combined Cycle Plants—Part 1: Gas
PT ⫽ power turbine Turbines,’’ Proc. Inst. Mech. Eng., Part A, J. Power Energy, 211, pp. 443– 451.
s ⫽ shaft 关11兴 Kim, J. H., Kim, T. S., Lee, J. S., and Ro, S. T., 1996, ‘‘Performance Analysis
T ⫽ turbine of a Turbine Stage Having Cooled Nozzle Blades With Trailing Edge Ejec-
tion,’’ ASME Paper 96-TA-12.
关12兴 Solar Turbines Inc., 1999, TITAN 130-Gas Turbine Generator Set, technical
References brochure, www.solarturbines.com.
关1兴 Cohen, H., Rogers, G. F. C., and Saravanamuttoo, H. I. H., 1996, Gas Turbine 关13兴 Solar Turbines Inc., 1999, TITAN 130-Gas Turbine Mechanical-Drive Pack-
Theory, 4th Ed., Longman Group Limited, London. age, technical brochure, www.solarturbines.com.
关2兴 Rowen, W. I., and Van Housen, R. L., 1983, ‘‘Gas Turbine Airflow Control for 关14兴 Rocha, G., and Etheridge, C. J., 1998, ‘‘Evolution of The Solar Turbines TI-
Optimum Heat Recovery,’’ ASME J. Eng. Power, 105, pp. 72–79. TAN 130 Industrial Gas Turbine,’’ ASME Paper 98-GT-590.
关3兴 Jansen, M., Schulenberg, T., and Waldinger, D., 1992, ‘‘Shop Test Result of the 关15兴 Rocha, G., and Kurz, R., 2001, ‘‘Field and Application Experience of the
V64.3 Gas Turbine,’’ ASME J. Eng. Gas Turbines Power, 114, pp. 676 – 681.
TITAN 130 Industrial Gas Turbine,’’ ASME Paper 2001-GT-0224.
关4兴 Kim, J. H., Song, T. W., Kim, T. S., and Ro, S. T., 2001, ‘‘Model Development
关16兴 Farmer, R., 2001, Gas Turbine World 2000–2001 Handbook, Pequot Publish-
and Simulation of Transient Behavior of Heavy Duty Gas Turbines,’’ ASME J.
Eng. Gas Turbines Power, 123, pp. 589–594. ing Inc., CT.
关5兴 Kim, J. H., Song, T. W., Kim, T. S., and Ro, S. T., 2002, ‘‘Dynamic Simulation 关17兴 Kim, T. S., and Ro, S. T., 1995, ‘‘Comparative Evaluation of the Effect of
of Full Start-up Procedure of Heavy Duty Gas Turbines,’’ ASME J. Eng. Gas Turbine Configuration on the Performance of Heavy-Duty Gas Turbines,’’
Turbines Power, 124, pp. 510–516. ASME Paper 95-GT-334.
关6兴 Walsh, P. P., and Fletcher, P., 1998, Gas Turbine Performance, 1st Ed., Black- 关18兴 Kim, T. S., and Ro, S. T., 1997, ‘‘The Effect of Gas Turbine Coolant Modu-
well Science Ltd., London. lation on the Part Load Performance of Combined Cycle Plants—Part 2: Com-
关7兴 Song, T. W., Kim, T. S., Kim, J. H., and Ro, S. T., 2001, ‘‘Performance bined Cycle Plant,’’ Proc. Inst. Mech. Eng., Part A, J. Power Energy, 211, pp.
Prediction of Axial Flow Compressors Using Stage Characteristics and Simul- 453– 459.
Steady and Dynamic Performance
and Emissions of a Variable
Geometry Combustor in a Gas
Turbine Engine
One of the remedies to reduce the major emissions production of nitric oxide 共 NO x 兲 ,
carbon monoxide (CO), and unburned hydrocarbon (UHC) from conventional gas turbine
engine combustors at both high and low operating conditions without losing performance
Y. G. Li and stability is to use variable geometry combustors. This type of combustor configuration
provides the possibility of dynamically controlling the airflow distribution of the combus-
R. L. Hales tor based on its operating conditions and therefore controlling the combustion in certain
lean burn conditions. Two control schemes are described and analyzed in this paper: Both
are based on airflow control with variable geometry, the second including fuel staging. A
School of Engineering, model two-spool turbofan engine is chosen in this study to test the effectiveness of the
Cranfield University, variable geometry combustor and control schemes. The steady and dynamic performance
Bedford MK43 0AL, England of the turbofan engine is simulated and analyzed using an engine transient performance
analysis code implemented with the variable geometry combustor. Empirical correlations
for NO x , CO, and UHC are used for the estimation of emissions. Some conclusions are
obtained from this study: (1) with variable geometry combustors significant reduction of
NO x emissions at high operating conditions and CO and UHC at low operating condition
is possible; (2) combustion efficiency and stability can be improved at low operating
conditions, which is symbolized by the higher flame temperature in the variable geometry
combustor; (3) the introduced correlation between nondimensional fuel flow rate and air
flow ratio to the primary zone is effective and simple in the control of flame temperature;
(4) circumferential fuel staging can reduce the range of air splitter movement in most of
the operating conditions from idle to maximum power and have the great potential to
reduce the inlet distortion to the combustor and improve the combustion efficiency; and
(5) during transient processes, the maximum moving rate of the hydraulic driven system
may delay the air splitter movement but this effect on engine combustor performance is
not significant. 关DOI: 10.1115/1.1615253兴
Introduction Gas turbines, especially aero gas turbines, work in a wide range
of environmental and operational conditions and their combustors
In recent years, more attention to aircraft emissions has been
experience dramatic variation of inlet pressure, temperature, air
paid with the emergence of environmental concerns of a global
flow rate and fuel flow rate. The control of the variable geometry
nature, such as climate change and depletion of the ozone layer. and therefore the air flow distribution in all working conditions is
Different combustion technologies have been developed in order one of the effective methods to keep the combustors working
to reduce gas turbine emissions. These can be classified into the stably, reliably, efficiently and with low emissions.
following combustor concepts, following the review by Wulff and The objective of this paper is to assess the effectiveness on the
Hourmouziadis 关1兴: variable geometry combustors, staged com- control of emissions production in gas turbine combustors with
bustors, lean premixed combustors, rich-burn, quick-quench, lean- two variable geometry combustor control schemes. The analysis is
burn 共RQL兲 combustors, and catalytic combustors. based on a typical two-spool large turbofan engine by using an
The variable geometry concept provides the flexibility to con- in-house gas turbine analysis computer code, Maccallum 关10兴,
trol the airflow distribution inside a combustor and therefore has implemented with a variable geometry combustor model and em-
the very attractive ability to reduce the emissions on both high and pirical emissions models 共Lefebvre 关11兴 and Reeves and Lefebvre
low operating conditions without losing combustor performance. 关12兴兲. The dynamic combustor performance and emissions char-
It also has great potential to improve combustion efficiency at low acteristics in engine environment with different control schemes
power operating conditions, Schultz 关2兴, and facilitate engine re- are analyzed and compared.
light at altitude, Schultz 关3兴. At the same time it certainly intro-
duces more complicated mechanical and control issues. Typical
low emissions variable geometry combustor configurations are Emissions Simulation
those proposed by Schultz 关2,3兴, Cupta et al. 关4兴, Fletcher 关5兴, There are three basic types of methods of predicting gas turbine
Saintsbury and Sampath 关6兴, Fletcher and Adkins 关7兴, Hayashi combustion emissions: empirical correlations such as that of Lefe-
et al. 关8兴, and Li et al. 关9兴. bvre 关11兴 based on combustor emissions measurement, analytical
methods using a stirred reactor assumption and simplified chemi-
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN cal reactions, and the computational combustion dynamics 共CCD兲.
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF CCD is able to provide a detailed description of the flow field and
emissions production mechanism but is numerically too expensive
erlands, June 3– 6, 2002; Paper No. 2002-GT-30135. Manuscript received by IGTI, for performance analysis. The analytical methods have more uni-
December 2001, final revision, March 2002. Associate Editor: E. Benvenuti. versal suitability but are more complicated than the empirical
models. In order to get fast emissions prediction within the engine
performance calculation environment with acceptable accuracy,
Lefebvre’s NOx and CO empirical correlation, 关11兴, and Reeves
and Lefebvre’s UHC empirical model, 关12兴, are used in this study.
These empirical correlations were developed based on large num-
ber of emissions measurements of gas turbine combustors and
reflect the influence of basic combustor fluid dynamic parameters.
It is assumed that the exhaust concentrations of NOx and CO are
proportional to the product of the mean residence time in the
combustion zone, the chemical reaction rates and mixing rates.
UHC is normally associated with poor atomization, inadequate
burning rates, chilling effects of film-cooling air, or any combina-
tion of these. Therefore, the parameters that determine the produc-
tion of NOx , CO, and UHC are the combustor inlet pressure, Fig. 1 Emissions characteristics
temperature, mass flow rate, flame temperature, and combustor
geometry. The empirical correletions of emissions used in this
study are as follows:
level. Because of the additional difficulties with rich burn control,
9⫻10⫺8 P 1.25
2 V c exp共 0.01T st 兲 the current analysis focuses on lean burn control.
共 NOx兲 ⫽ 共 g/kg•fuel兲 (1)
ṁ A T f Among all the emissions produced by gas turbines, CO and
UHC emissions can be reduced significantly by improving com-
86ṁ A T f exp共 ⫺0.00345T f 兲 bustion efficiency. The NOx production is mainly determined by
冉冊
共 CO兲 ⫽ 共 g/kg•fuel兲 (2)
⌬ P 0.5 1.5 the combustor inlet pressure, inlet temperature, flame temperature,
共 V c ⫺V e v 兲 P2 and residence time in the primary combustion zone and fuel type.
P2
According to Steele et al. 关14兴, the flame temperature has the most
11764ṁ A T f exp共 ⫺0.00345T f 兲 significant effect on the NOx production.
冉冊
共 UHC兲 ⫽ 共 g/kg•fuel兲 (3) According to Eqs. 共1兲 to 共3兲, there are several parameters, such
⌬ P 2.5
共 V c ⫺V e v 兲 P2 as combustor inlet pressure, temperature, mass flow rate, combus-
P2 tor geometry, and fuel flow rate, which influence the emissions
The NOx model suggests that in the combustion of a heteroge- production. In order to maintain a certain level of engine perfor-
neous fuel-air mixture it is the stoichiometric flame temperature mance for given gas turbine thermal cycles, the combustor inlet
T st that determines the formation of NOx . However, for lean burn pressure, temperature, mass flow rate, and fuel flow rate cannot be
combustors where the maximum attainable temperature is the changed. The only parameters that can be changed are the air flow
adiabatic flame temperature T f , the equation may still be used distribution and fuel flow distribution within the combustor, which
provided that T f is substituted for T st . The flame temperature of can be achieved with variable geometry and fuel staging respec-
the combustor primary zone is estimated with the formula of Gül- tively. Many fuel staging methods have been developed, but only
der 关13兴, where combustor inlet pressure, temperature, mass flow the circumferential staging is analyzed in this study.
rate, air flow ratio to the primary zone, and fuel flow rate are In conventional gas turbine combustors, the combustion system
required. is designed for a certain high-power operation condition, produc-
ing an unwanted effect on the requirements for combustion stabil-
ity and low limit of weak extinction at idle. Such a design requires
Model Engine and its Performance Calculation an equivalence ratio in the primary zone which is very close to
A representative civil aero turbofan engine is chosen for the unity at the maximum working point; the flame temperature and
analysis and has the following basic performance parameters: therefore the NOx emissions are very high in this situation. In
Total mass flow rate: 191.5 kg/s order to reduce the NOx emissions at high power level based on
Total pressure ratio: 16.7 the lean burn concept, the equivalence ratio in the primary zone
Bypass ratio: 3.03 must be within the lean burn low emissions window 共Fig. 1兲 so
Turbine entry temperature: 1300 K the air flow rate to the primary zone must be increased. Such a
In the combustor, the proportion of air flow to the primary change will definitely result in poor combustion stability at low
combustion zone is 0.34 for the conventional configuration, which operating conditions. One of the remedies to the problem is to
is typical for conventional gas turbine combustors. vary the air flow rate to the primary zone by changing the com-
The calculation of combustor performance is based on enthalpy bustor geometry according to the operating conditions. Due to
balance so whatever is the air flow ratio to the primary combus- difficulties in measuring and controlling the absolute air flow rate
tion zone the total fluid dynamic performance of the combustor is to the primary zone but being relatively easy to control the air
almost the same if small changes in pressure loss and combustion flow ratio to the primary zone because air flow ratio is approxi-
efficiency are ignored. The performance analysis of a variable mately proportional to cross-sectional area ratio, the air flow ratio
geometry combustor is carried out by replacing the combustor of is chosen to be the control parameter.
the engine with the variable geometry combustor. The engine is The variable geometry combustor suggested in this paper is
supposed to run in the same operating conditions and therefore the schematically shown in Fig. 2 where an air flow splitter driven by
combustor works in the same engine environment. a hydraulic system can move forward and backward to vary the
cross-sectional area ratio and therefore the air flow ratio to the
primary zone. The suggested position of the air splitter has the
Emissions Control Strategy advantages that it is located at the upstream of the combustor
The emissions characteristics are shown in Fig. 1, where the chamber and is relatively distant from the hottest gas. The control
emissions indices are plotted against equivalence ratio in the com- target is chosen to be the flame temperature, an indirect measure
bustor primary zone. Equivalence ratio close to the stoichiometric of combustion stability and the emissions production of the
condition enhances thermal NOx production, while lean and rich combustor.
equivalence ratios enhance CO and UHC production. There are The engine operates over a wide range of environmental and
two narrow low-emissions windows, one in lean burn and the operational conditions and therefore the air flow splitter may have
other in rich burn, where all the emissions are at a relatively low to move to an extreme position to satisfy the control target; this
Fig. 2 Variable geometry combustor
may cause severe distortion to the air flow field and result in large
pressure loss. Such a distortion may be crucial to the life of the
combustor chamber because the amount of cooling air may also
be far away from its design value. Therefore a circumferential fuel Fig. 4 Air flow ratio versus nondimensional fuel flow rate
staging is introduced in this study to reduce the range of move-
ment of the air flow splitter.
Control Scheme A. The starting point for the control system is chosen to be the indirect control parameter.
is to introduce a flame temperature predictor and to control the A throttle-back performance of the model turbofan engine at
flame temperature in the combustor primary zone to a set point standard day at sea level is calculated while adjusting the air flow
value. The flame temperature in the primary zone can be ad- ratio to the combustor primary zone with Control Scheme A to
equately predicted with combustor inlet pressure, temperature, keep the flame temperature at 1900 K. As a result of the calcula-
mass flow rate, fuel flow rate, and combustor geometry param- tion, the variation of the air flow ratio to the primary zone f pz with
eters. Generally, the best flame temperature range for lean-burn the nondimensional fuel flow rate (WF nd /WF nd,r ) is demon-
low-emissions combustion is between 1900 K and 2300 K. Out- strated in Fig. 4 and shows almost linear relationship, which pro-
side of the range, a higher flame temperature will enhance NOx vides the possibility of a flame temperature control scheme based
production and a lower flame temperature will carry the risk of on the information of nondimensional fuel flow rate. A linear re-
weak extinction and unstable combustion. For illustration, a con- lationship between the air flow ratio to the combustor primary
stant flame temperature in the primary combustion zone of 1900 K zone and the nondimensional fuel flow rate is introduced as
is chosen in this study. The introduced inferential control system follows:
冉冊
is schematically shown in Fig. 3 and is called Control Scheme A.
The accuracy of the control relies on the measurement accuracy of WF nd
f pz ⫽a• ⫹b (5)
those relevant combustor parameters and the prediction of flame WF nd,r
temperature. The drawback of such a control scheme is that the
control system may introduce more complexity to the combustor where a and b are case-dependent constants.
design and manufacture and require more attention to its reliabil- Fuel staging may be necessary when the total fuel/air ratio of
ity due to the existence of many sensors and the flame temperature the combustor is below a certain level and the air flow ratio to the
predictor. Accurate measurement of air flow rate to the combustor primary zone is significantly below its design value. The advan-
is even more difficult. tage of the fuel staging is that the air flow splitter does not have to
move far from its design position in most of the engine operation
Control Scheme B. An alternative simplified inferential con- conditions to meet the control requirement; accordingly the air
trol scheme may be used to prevent the above problem if there is flow field upstream of the combustor should not distort dramati-
a relationship between the flame temperature and another control cally. It would be better to control the fuel staging based on the
parameter. It is observed that the air flow rate required in the total air/fuel ratio of the combustor, but unfortunately accurate
primary zone to keep the flame temperature constant as the gas
turbine changes its power is approximately proportional to the
fuel flow rate into the combustor. Taking into account the wide
range of environmental and operating conditions, a nondimen- Table 1 Fuel staging scheme
sional fuel flow rate
No. CN/CN r Injectors fueled f inj
ṁ f 冑T 2 1 ⭓0.94 100%
WF nd ⫽ (4) 2 ⬍0.94 60%
P2
Fig. 3 Control Scheme A Fig. 5 Control Scheme B
air/fuel ratio is difficult to measure. One of the methods to control
the fuel staging is to use the HP spool rotational speed as an
indirect measure of the engine power level or the total air/fuel
ratio in the combustor. A two-step fuel staging scheme is chosen
in this study which is illustrated in Table 1 where all fuel injectors
are fueled when the HP spool nondimensional rotational speed is
greater than 0.94 and 60% injectors are fueled when the non-
dimensional rotational speed is below 0.94.
With fuel staging, Eq. 共5兲 is now modified to the following
form:
冋冉
f pz ⫽ a•
WF nd
WF nd,r冊册冉冊
⫹b •
1
f inj
. (6)
At the point of staging, the total fuel flow remains constant but the
number of injectors fueled changes, so the fuel flow per injector
changes. The control scheme based on the above relations is
called Control Scheme B and is schematically shown in Fig. 5.
The advantage of using the parameter WF nd for control is that
all the parameters in Eq. 共4兲 are measurable and the control sys-
tem can be simplified significantly. When the Eq. 共5兲 is used in the
Fig. 6 Flame temperature control system to control the air flow ratio to the primary zone, the
effectiveness of the flame temperature control can be observed in
Fig. 6 where the flame temperature is plotted against HP spool
speed. It can be seen that assuming such a linear air flow ratio
control is very effective compared to the original inferential Con-
trol Scheme A and the maximum control error is about 40 K at
about 82% nondimensional speed. Such a level of flame tempera-
ture variation would not adversely affect combustion
performance.
Transient Performance Calculation

Details of gas turbine transient performance modeling issues
can be found in Refs. 关15,16兴. The computer code for gas turbine
transient performance calculation used in this study was devel-
oped by Maccallum 关10兴 and the prediction accuracy has been
assessed during the calculation of Rolls-Royce RB183 engines.
The air splitter located at the upstream end of the combustor,
which is the salient part of the variable geometry combustor, is
Fig. 7 Hydraulic system capability controlled by the engine control system and driven by a hydraulic
Fig. 8 Index of NOx emissions
Fig. 9 Index of CO emission
system. The moving speed of the air splitter, if not high enough, Comparisons and Analysis
may have influence on the combustor transient performance. The
critical scenarios would be rapid acceleration and deceleration Steady State Behavior. When the turbofan engine is throttled
processes and fuel staging. back from its maximum operating condition at standard day at sea
A simple model is used to describe the moving speed of the air level, the performance and emissions of the engine combustor
splitter, Fig. 7, where a maximum moving speed of 0.4 m/s is produced by the conventional configuration and variable geometry
assumed. If the required moving speed of the air splitter is less combustors with different control schemes are compared as fol-
than 0.4 m/s, the hydraulic system would be able to drive the air lows. In the analysis, the indices of emissions produced by the
splitter to the required position, which is the case in most situa- combustor with conventional configuration at maximum power
tions. If the required moving speed of the air splitter is higher than level at sea level are chosen to be the reference values ((NOx) r ,
what the hydraulic system can drive, the movement of the air (CO) r and (UHC) r ) for the convenience of comparison.
splitter will be delayed. Details of the influence of the delay will In the combustors with conventional configuration where the air
be described in the next section. flow ratio to the primary combustion zone is chosen to have a
Fig. 10 Index of UHC emission
Fig. 11 Air flow ratio to primary zone
constant value of 34%, the flame temperature will reach around level, Fig. 11. As a consequence of the application of the control
2200 K at maximum operating condition at sea level, Fig. 6; as a schemes, the NOx index can be reduced by a factor of 14 at
consequence, a large amount of NOx will be produced, Fig. 8. maximum engine power level and kept almost constant with slight
When the engine throttles back the flame temperature decreases reduction at low engine power level. At about 72% of nondimen-
significantly and a large amount of CO and UHC may be pro- sional HP spool rotational speed the geometry control brings a
duced, Figs. 9 and 10. With the control schemes it can be seen that reduction of CO and UHC emissions indices by around a factor of
the flame temperature can be kept almost exactly at the set point 4.4 and reduces CO and UHC gradually as the high engine power
value in Control Scheme A and very close to the set point value in level is approached.
Control Scheme B 共within 40 K兲. Correspondingly, with both con- The benefit of the fuel staging can be seen in Fig. 11. The air
trol schemes the air flow ratio to the primary zone will be inflow ratio to the primary zone in Control Schemes A and B is at a
creased to around 45% at the maximum operating condition at sea level substantially different to the value at design point in most of
Fig. 12 Combustion efficiency
Fig. 13 Fuel flow rate and rotational speed
the engine operation conditions. With the introduction of fuel point for the same combustor is introduced in the calculation of
staging in Control Scheme B, the air flow ratio f pz to the primary combustion efficiency. In this way, the combustion efficiency bet-
zone can take values much closer to its design point in most of the ter reflects the combustion conditions for both the conventional
engine operation conditions, which is beneficial to the air flow and variable geometry combustors, and its effect can be seen in
field at the combustor inlet. Fig. 12. The improvement of the combustion efficiency in the
Compared with the conventional combustor configuration, the
variable geometry combustor has certain benefit to specific fuel
fuel/air mixture in the primary zone of the variable geometry com-
bustor 共whatever fuel staged or not兲 becomes richer at low oper- consumption 共sfc兲 but little influence on other engine performance
ating conditions. Correspondingly, the flame temperature in- parameters.
creases 共Fig. 6兲, which is a great benefit to the combustion It can be seen from Fig. 6 and Figs. 8 to 12 that the variable
stability and efficiency. A correction factor (T f /T f ,r ) s which is the geometry combustor has a great potential to reduce the emissions
ratio of flame temperature at off-design point to that at design and improve the combustor performance. Control Scheme B pro-
Fig. 14 Flame temperature
Fig. 15 Index of NOx emissions
vides equivalent control effect to that of Control Scheme A; the 2350 K at the end of the acceleration at about 3.8 second, Fig. 14,
difference in the performance and emissions production is very which causes a jump in NOx emissions accordingly, Fig. 15. With
small. the variable geometry combustor controlled by Schemes A and B,
Dynamic Behavior. An acceleration process of the engine the flame temperature varies around the set point value with con-
from steady idle to its maximum power at sea level was simulated trol error of less than 40 K, which results in much lower NOx
with the computer code, where the variations of the fuel flow rate emissions over operating conditions.
and corresponding HPC rotational speed are illustrated in Fig. 13. Much lower flame temperature 共Fig. 14兲 in the conventional
It shows a rapid increase in fuel flow rate in the first 0.5 second, combustor in the first 0.5 second, especially at idle, causes much
slow acceleration in the following 2.0 seconds and a further steep higher CO and UHC emissions, Figs. 16 and 17. These emissions
acceleration in the next 1.3 seconds. A steady state is attained after can be reduced significantly with the variable geometry combus-
around 7 seconds. tor. With the increase in engine power level, little difference in CO
The flame temperature in the conventional combustor increases and UHC emissions exist between the two control schemes, Figs.
rapidly during the acceleration and reaches its top value of about 16 and 17.
Fig. 16 Index of CO emission
Fig. 17 Index of UHC emission
The primary zone air flow ratios with Control Schemes A and B by the speed limit of the air splitter. This can be seen in Fig. 14
in the variable geometry combustor, Fig. 18, have larger values where the flame temperature in the variable geometry combustor
than those in the steady state situations, Fig. 11, during the accel- with Control Schemes A and B has sudden jumps at about 2.2
eration 共in the first 3.2 seconds兲. This is because the air flow ratio seconds when fuel staging is switching and at about 3.2 seconds
to the primary zone is controlled by the corrected fuel flow rate, when a steep acceleration occurs. The maximum flame tempera-
Eq. 共6兲, and the fuel flow rate at a specific rotational speed during ture jump is less than 20 K so the influence is not significant.
a transient is larger than that required to keep the same rotational The difference in combustion efficiencies between the conven-
speed at steady state. The two schemes exhibit a difference in air tional combustor and the variable geometry combustor during the
flow ratios until 2.2 seconds after the acceleration starts, at which acceleration is shown in Fig. 19. Maximum efficiency difference
time fueling is staged under Scheme B. occurs at idle due to maximum flame temperature difference but
The air splitter is constrained by the hydraulic system. There- the difference reduces when the engine power increases.
fore the combustion conditions during transient may be influenced In general, the combustor performance difference between the
Fig. 18 Air flow ratio to primary zone
Fig. 19 Combustion efficiency
conventional combustor and the variable geometry combustor dur- 共UHC兲 ⫽ unburned hydrocarbon index 共g/kg兲
ing the acceleration is obvious but the difference between the two Vc ⫽ volume of combustor chamber (m3 )
control schemes is very small. V ev ⫽ fuel evaporation volume (m3 )
WF nd ⫽ nondimensional fuel flow rate
Conclusions a, b ⫽ constants
The proposed variable geometry combustor combined with ap- f inj ⫽ proportion of injectors fueled
propriate control schemes can significantly reduce emissions pro- f pz ⫽ air flow ratio to primary zone
duction for engine operation at or near ground level, compared to ṁ A ⫽ air flow rate to combustor 共kg/s兲
a combustor with conventional configuration. Fuel staging can ṁ f ⫽ fuel flow rate 共kg/s兲
reduce the range of movement of the air flow splitter at most of Subscripts
the operational conditions and therefore is beneficial to the air
nd ⫽ nondimensional
flow field at combustor inlet.
pz ⫽ primary zone
The introduced linear relationship between the non-dimensional
r ⫽ reference value at engine maximum operation condi-
fuel flow rate and air flow ratio to the combustor primary zone for
tion at sea level
the control of flame temperature
0 ⫽ control set point
2 ⫽ combustor inlet
冋冉
f pz ⫽ a•
WF nd
WF nd,r冊册冉冊
⫹b •
1
f inj
References
is effective and can simplify the control system. 关1兴 Wulff, A., and Hourmouziadis, J., 1997, ‘‘Technology Review of Aeroengine
Emissions,’’ Aerospace Science and Technology, „8…, pp. 557–572.
The speed limit of the air flow splitter has little influence on 关2兴 Schultz, D. F., 1975, ‘‘Ground Idle Performance Improvement of a Double-
combustor performance. The variable geometry combustor has a Annular Combustor by Using Simulated Variable Combustor Geometry,’’
potential to improve combustion stability, efficiency and specific NASA TM X-3176.
fuel consumption compared to the conventional configuration. 关3兴 Schultz, D. F., 1975, ‘‘Variable Combustor Geometry for Improving the Alti-
tude Relight Capability of a Double Annular Combustor,’’ NASA TM X-3163.
The inferential Control Scheme A is able to provide good con- 关4兴 Cupta, A. K., Ramavajjala, M. S., and Chomiak, J., 1988, ‘‘Burner Geometry
trol accuracy of flame temperature but embodies significant com- Effects on Combustion and NOx Emission Characteristics of Variable Geom-
plexity. The simplified Control Scheme B provides equivalent etry Swirl Combustor,’’ Paper No. AIAA-89-0488.
control accuracy. 关5兴 Fletcher, R. S., 1973, ‘‘The Automotive Gas Turbine,’’ conference on Devel-
opments in Automotive Power Plants to Reduce Fuel Consumption, Air Pol-
lution and Noise, Queen Mary College, London, Apr.
Nomenclature 关6兴 Saintsbury, J. A., and Sampath, P., 1974, ‘‘Atmospheric Tests of a Variable
Combustor Geometry for Reducing Gas Turbine Emissions,’’ ASME/CSME
CN ⫽ engine high pressure 共HP兲 spool nondimensional rota- Conference, Montreal, May.
tional speed (N/ 冑T) 关7兴 Fletcher, R. S., and Adkins, R. C., 1976, ‘‘The Variable Geometry Combustor,’’
N ⫽ rotational speed 共rpm兲 AGARD Conference 共48 PEP Meeting兲, Paris, Sept.
关8兴 Hayashi, S., Yamada, H., and Shimodaira, K., 1996, ‘‘Engine Testing of a
共CO兲 ⫽ carbon monoxide index 共g/kg兲 Natural Gas-Fired, Low-NOx , Variable Geometry Gas Turbine Combustor for
(NOx) ⫽ nitric oxide index 共g/kg兲 a Small Gas Turbine,’’ ASME Paper No. 96-GT-455.
P 2 ⫽ combustor inlet pressure 共kPa兲关9兴 Li, Y. G., and Hales, R. L., 2002, ‘‘Gas Turbine Emissions Control Using
⌬ P ⫽ combustor liner pressure drop 共kPa兲 Variable Geometry Combustor and Fuel Staging,’’ Paper No. AIAA-2002-
0079.
T 2 ⫽ combustor inlet temperature 共K兲关10兴 Maccallum, N. R. L., 1984, ‘‘Computational Models for the Transient Perfor-
T f ⫽ flame temperature in primary combustion zone 共K兲 mance of RB183-02 共Spey兲 and RB183-03 共Tay兲 Engines,’’ Technical Report
T st ⫽ stoichiometric flame temperature 共K兲 RR/1, University of Glasgow, Aug. 13.
关11兴 Lefebvre, A. H., 1984, ‘‘Fuel Effects on Gas Turbine Combustion—Liner 1997, ‘‘Variables Affecting NOx Formation in Lean-Premixed Combustion,’’
Temperature, Pattern Factor and Pollutant Emissions,’’ Paper No. AIAA-84- ASME J. Eng. Gas Turbines Power, 119, pp. 102–107.
1491. 关15兴 Maccallum, N. R. L., and Qi, O. F., 1989, ‘‘The Transient Behavior of Aircraft
关12兴 Reeves, C. M., and Lefebvre, A. H., 1986, ‘‘Fuel Effects on Aircraft Combus- Gas Turbines,’’ IMechE Seminar 共Seminar S777兲 on Gas Turbines: Technology
tor Emissions,’’ ASME Paper No. 86-GT-212. and Development, Institute of Mechanical Engineering, London, Nov.
关13兴 Gülder, Ö. L., 1986, ‘‘Flame Temperature Estimation of Conventional and 关16兴 Pilidis, P., 1983, ‘‘Digital Simulation of Gas Turbine Performance,’’ Ph.D.
Future Jet Fuels,’’ ASME J. Eng. Gas Turbines Power, 108, pp. 376 –380. dissertation, Department of Mechanical Engineering, University of Glasgow,
关14兴 Steele, R. C., Jarrett, A. C., Malte, P. C., Tonouchi, J. H., and Nicol, D. G., Glasgow, Scotland.
A. Traverso
A New Generalized Carbon Exergy
Tax: An Effective Rule to Control
A. F. Massardo
e-mail: [email protected] Global Warming
Thermochemical Power Group,
Dipartimento di Macchine, An instrument for promoting CO 2 emission reductions, taking the Kyoto Protocol goal
Sistemi Energetici e Transporti, into account, could be the assignment to energy conversion plants of a monetary charge
Universita di Genova, Italy, linked to their specific emission intensity, usually called carbon tax. There are two main
Genova 16145, Italy problems closely connected with this approach: the estimation of the charge (that must be
related to the ‘‘external’’ cost associated with CO 2 emission) and the choice of the
strategy to determine the amount of the imposed charge. In this paper an analytical
M. Santarelli procedure proposed by the authors and called carbon exergy tax (CET) for the evaluation
e-mail: [email protected] of CO 2 emission externalities is presented. It is based on the thermoeconomic analysis of
energy systems, which allows second law losses to be quantified in monetary terms: the
M. Cali resulting cost represents the taxation that is to be applied to the energy system under
e-mail: [email protected] examination, calculated without any arbitrary assumption. Since the complete procedure
of the CET evaluation is too complex to become a feasible instrument of energy policy,
Dipartimento di Energetica, hereby, after applying the procedure to some conventional and advanced power plants,
Politecnico di Torino, gas, oil, and coal-fueled, a new generalized approach, based on the results of the com-
Torino, Italy plete CET procedure, is proposed. The generalized CET evaluation requires much less
information about the energy system and thus a simple and effective energy policy rule to
manage global warming is obtained and available. 关DOI: 10.1115/1.1610013兴
Introduction types and sizes, in order to discuss its effect on the economic
promotion of low CO2 emission configurations such as the HAT
The criteria that will influence the evolution of the energy mar-
or fuel-cell hybrid cycles. After that, a generalized approach that
ket this century will be based on the need to preserve the envi- maintains the same features of the rigorous CET method, but re-
ronment 共both locally and globally兲 through new technologies and quires much less information about the energy system, is shown to
sustainable use of existing resources. This need is also ratified by be possible. In this way a feasible proposal for CO2 emission
the guidelines of the various framework programs of the European taxation is issued and its ability to effectively contribute to the
community and by international agreements such as the Kyoto economic management of global warming is assessed.
Protocol. In particular, the global warming problem, linked to
CO2 emissions, requires an energy policy approach at an interna-
tional level devoted to CO2 regulation: The energy policy should Carbon Exergy Tax „CET…: Analytical Approach
aim at punishing the inefficient use of fossil energy sources that
The CO2 ‘‘external cost’’ should conceptually be equal to the
determines a larger emission of CO2 than a more efficient use.
cost of the CO2 emission damage caused to the environment:
Two main approaches are possible to deal with the global warm-
Once this cost is evaluated, it should be charged to the emitting
ing issue: the political approach and the economic approach, plant since it must repay the society for polluting the atmosphere.
关1–3兴, already discussed by the authors, 关4,5兴. The former When a procedure is being considered for the calculation of such
‘‘forces,’’ the latter ‘‘suggests’’ and makes low CO2 emission en- an ‘‘externality’’ of CO2 emissions, it should be taken into account
ergy systems more competitive than conventional solutions. The that carbon dioxide is not toxic and is even necessary for life. The
carbon exergy tax 共CET兲, proposed by the authors in the frame- CET method shifts attention away from the effects 共global warm-
work of the economic approach, 关6 –9兴, allows the CO2 emission ing兲 to the causes 共exploitation of fossil resources兲 of CO2 pro-
external cost to be evaluated on the basis of efficient utilization of duction: to punish the inefficient use of scarce energy resources
exergy resources. The idea is not to punish the energy production that determines a larger emission of CO2 than an efficient use. The
activity (CO2 emissions are not toxic and moreover they are in- proposed procedure is based on the concept of the inefficiency
evitably emitted in large quantities during hydrocarbon combus- penalty (⌸ ␧ ), and the index of CO 2 emission (I CO* ) and it high-
2
tion processes兲 but the inefficient use of fossil fuels.
lights system inefficiencies that cause larger amounts of CO2 to be
The carbon exergy tax should push the deregulated energy mar-
released into the environment. Both terms (⌸ ␧ and I CO * ) depend
ket towards the most efficient and advanced conversion technolo- 2
gies and towards renewable resources 共they should not be taxed on the technical characteristics of the system, contained in the
since they do not emit carbon dioxide兲; in addition, the CET is an vector x. The meaning and the expressions of ⌸ ␧ and I CO * are
2
effective regulation for making CO2 sequestration economically presented in the following.
feasible, 关10–14兴.
In the present work the carbon exergy tax evaluation procedure Inefficiency Penalty. The inefficiency penalty term has this
is briefly presented and is applied to power plants of different expression:
⌸ ␧ 共 x兲 ⫽3600•N• 兵 C irr 共 x兲 ⫹C w 共 x兲其 (1)
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF where the cost of destroyed exergy is
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- NC
兺c
erlands, June 3– 6, 2002; Paper No. 2002-GT-30139. Manuscript received by IGTI, C irr 共 x兲 ⫽ e i 关 ⌿ e 共 x 兲 ⫺⌿ o 共 x 兲兴 i (2)
December 2001, final revision, March 2002. Associate Editor: E. Benvenuti. i⫽1
and the cost of residual exergy is It is also possible to determine an equivalent tax imposed on
NC fuel cost, called the fuel carbon exergy tax 共FCET兲, whose expres-
C w 共 x兲 ⫽ 兺
w⫽1
c w ⌿ w 共 x兲 . (3)
sion is
C CO2 $/yr
In order to assign a weight to the irreversibilities and the FCET⫽ ⇒ 关 $/kJ兴 . (11)
residual exergy, corresponding to their relevance in terms of G F •LHV kJ/yr
resource waste, the unit inefficiency penalty is evaluated in this
manner: Finally, considering plants for electrical generation only, it is
c e : the cost of a unit of irreversibility produced by a compo- possible to evaluate the absolute increase of the electricity cost
nent i is calculated on the product of the economic fuel cost of the due to the imposition of the CET, defined as ⌬c el :
system 共e.g., cost of natural gas兲 and the unit exergetic cost (k * )
of the exergy input flow of the same component, which is CET•G CO2
⌬c el ⫽ (12)
W el
c e i ⫽c f •k i* 共 x兲 . (4)
c w : the cost of a unit of the residual exergy. Residual exergy is Software Tool and Case Studies
related to the exergy ejected into the biosphere through exhausts
and/or heat from the condenser, as it is considered to be a poten- All the results presented here are calculated using the in-house
tial product which is wasted; its unit cost is given by TEMP 共thermo-economic modular program兲 code, 关15–17兴, re-
cently upgraded to predict complex mixed air-steam cycle perfor-
c w ⫽c f •k *
p 共 x兲 . (5) mance, 关18,19兴. At present 64 modules are available, and they
allow the thermoeconomic and exergoeconomic, 关20兴, analysis of
The irreversibility cost is linked to the fuel cost because irrevers-
a large number of energy cycles such as the following to be ob-
ibility means waste of fuel.
tained: steam, gas turbine, combined, and advanced cycles 共mixed
* is
Index of CO2 Emission. The index of CO 2 emission I CO gas-steam cycles, fuel cells—SOFC and MCFC—and hybrid
2
cycles, partial oxidation cycles, chemical recovery cycles, 关21兴兲.
defined as
The system to be calculated is defined as an ensemble of intercon-
I CO2 1 G CO2 nected components. It is possible to calculate the internal irrevers-
* ⫽
I CO ⫽ (6) ibility of each component 共that determines an exergy expenditure兲
2 I CO2 ref I CO2 ref ⌿ p and its capital cost 共that determines a monetary expenditure兲.
where I CO2 ref is assumed to be equal to 1 kg/kWh. The assumption Thermoeconomic results are given at two different levels: at the
inner level as an estimate of the average unit cost 共c兲, the unit
is only made to define the I CO* as a nondimensional parameter.
2 exergetic cost (k * ) and the marginal cost 共␭兲 at each connection;
When plants for electrical generation only are considered, the at the outer level as an evaluation of the thermal efficiency, the
following formula can be derived: generated power, the generated heat, the electrical and thermal
1 ␣•关C兴 1 energy costs. In addition, it is possible to carry out plant through-
* ⫽
I CO • (7) life cost analysis, calculating financial parameters such as internal
2 I CO2 ref LHV ␩ rate of return, payback period, net present value, and others. In the
end plant thermodynamic and economic features are completely
* and the system thermal effi-
where the relationship between I CO 2 and fully described.
ciency ␩ is expressed. Parameter ␣, that is CO2 production per The TEMP code is also provided with an optimization tool that
mass of C, is equal to 3.664 for all fuels. allows energy system thermoeconomic optimization to be ob-
Charge for the CO2 Emitted by the Plant. tained with different objective functions: the most important are
thermal efficiency and the cost of electricity. The nonlinear opti-
* •⌸ ␧ .
C CO2 ⫽I CO (8) mization algorithm is described in 关15兴, and it has been recently
2
upgraded with the introduction of a genetic algorithm.
To better outline the influence of plant operating conditions the In the present paper, the complete CET calculation procedure
following relationship can be developed: has been applied to some conventional and advanced power plants
冋
differently fuelled 共coal, oil, and natural gas; see Table 1兲 and
1 兺 i⫽1
NC
c e i 关 ⌿ e 共 x兲 ⫺⌿ o 共 x兲兴 i
C CO2 ⫽ • ␣ • 关 C 兴 •G F •3600•N• sized 共0.5–350 MW; see Table 2兲. Of the conventional power
I CO2 ref ⌿ P 共 x兲 systems, a natural gas and oil-fueled 30 MW simple gas turbine
cycle 共GT兲, a natural gas-fueled 50 MW one-pressure-level steam
⫹
兺 i⫽1
NC
c w,i ⌿ w,i 共 x兲
⌿ P 共 x兲
册 . (9)
injected gas turbine 共STIG兲, a natural gas-fueled 200 MW three-
level-pressure combined cycle 共CC3LP兲, and the coal-fueled stan-
dard ENEL 共320 MW兲 have been considered in order to represent
CET. Finally the expression of CET is the state of the art of power plants.
Among advanced power systems, the following configurations
C CO2 ⌸␧ 1 ⌸␧ have been studied in order to comprehend a wide range of pos-
CET⫽ * •
⫽I CO ⫽ . (10) sible future plants.
G CO2 2 G CO2 I CO2 ref ⌿ p
An oil-fueled 354 MW externally fired combined cycle
It is important to note that the CET expression, as already dis- 共EFCC兲, 关22兴 and a coal-fueled 356 MW pressurised fluid bed
cussed in 关4,8兴, is only obtained through an objective thermoeco- combustor plant 共PFBC兲, 关23兴, have been considered as examples
nomic procedure, without subjective assumptions. of solid and liquid fuel exploitation. The EFCC cycle scheme is
When the CET, calculated according to 共10兲, is charged to the similar to the CC3LP cycle one considered, but with the addition
emitted CO2 , it causes a non-negligible increase in the energy of a high-temperature ceramic heat exchanger between the gas
cost that mainly depends on plant conversion efficiency and the turbine compressor and turbine: It is the most challenging compo-
type of fuel 共composition and cost兲: The least polluting plants will nent and at the present such a pressurised heat exchanger has
suffer the smallest increase. In that respect, the CET represents an never been built on such a large scale, because of problems re-
economic penalty linked to the inefficient utilisation of energy garding material high temperature corrosion, thermal stress, leak-
resources. age, and ceramic tube fouling, 关24兴.
Table 1 Fuel composition, LHV, and exergy
Coal Oil Natural Gas

Comp Mass Fraction Comp Mass Fraction Comp Molar Fraction
C 85 C 87.13 CH4 93.0
H 1.0 H 12.6 N2 7.0
O⫹N 3.0 O 0.044
S 1.0 N 0.006
H2 O 2.5 S 0.22
Ash 7.5 H2 O 0.0
LHV 29500 kJ/kg LHV 44720 kJ/kg LHV 44270 kJ/kg
Ex f 31060 kJ/kg Ex f 47650 kJ/kg Ex f 45800 kJ/kg
The pressurized fluid bed combustor plant scheme is a typical latter 共EFmGT兲 is similar to the EFGT without the bottoming
two-pressure-level combined cycle that has been modified in order cycle 共steam section兲. Present microturbine technology has been
to burn the coal in a PFBC. The PFBC replaces the original tur- considered.
bine natural gas combustor: The resulting scheme is an air-cooled
PFBC. This is the only alteration to the original combined cycle, Thermoeconomic Results
which remains the same for the bottoming steam cycle section.
Coal is burned in the presence of a sorbent at temperatures around To calculate electrical power cost, the assumptions reported in
1125 K: Above 1200 K there is the risk of a syntherization of the Table 3 are necessary. The cost of electricity without any penal-
fluidised bed; below 1025 K the combustion efficiency is too low; ization has been evaluated for all the power plants considered and
however, the most influential factor for bed temperature is the referred to the first construction year 共2002兲, except for microsys-
need for an in situ desulphuration, which operates well at a tem- tems, whose economic data in open literature are not yet detailed
perature around 1125 K, 关23兴. enough and, however, are expected to lower by much in next
A 50 MW regenerated water injected gas turbine cycle 共RWI兲 years. The presence of innovative components such as the humidi-
and a 50 MW humid air gas turbine cycle 共HAT兲, 关19兴, have been fication tower in the HAT cycle and the ceramic heat exchanger in
studied as examples of future applications of natural gas for power the EFCC cycle require some rough preliminary assumptions to
generation. The introduction of water into the compressed air after calculate capital costs: for the humidification tower, a cost equal
flue gas heat recovery is a very promising technique for increasing to the economizer cost has been considered; for the high-
gas turbine efficiency and specific work and, moreover, an effec- temperature heat exchanger, a cost equal to ten times the cost of a
tive technical device for drastically reducing NOx emissions. The metallic recuperator with the same exchanging surface has been
RWI cycle carries out compressed air intercooling and aftercool- supposed, in order to also comprehend external combustor cost
ing through water injection, which is a more irreversible process and the major operating and maintenance costs.
than the evaporation of water inside the humidification tower in It should be noted that capital costs do not affect CET values,
the HAT cycle. This fact justifies the lower efficiency of the RWI because they only depend on the process irreversibilities, which
to that of the HAT cycle. However, the HAT cycle does not re- are economically evaluated using the fuel cost c f 共see Eq. 共4兲 and
quire high-quality water: Conceptually it could work even with Eq. 共5兲兲.
sea-quality water. Figure 1 shows the COE of each plant before and after CET
On the other hand, the STIG and RWI cycles require very high- imposition. Considering the unpenalized COE, the HAT cycle has
quality water. At present water injection is implemented in a few the lowest electricity cost 共3.74 c$/kWh兲, despite the fact that it
commercial gas turbines for air intercooling, while no working does not achieve the high efficiency of the three-level-pressure
HAT cycle is operating in the world, except for the pilot plant combined cycle: This is due to the high specific work 共800 kJ/kg兲
installed at the University of Lund, 关25兴. and the absence of the bottoming cycle which make the HAT
Two innovative power plants have been considered as examples cycle a compact and cheap power plant. The oil-fueled EFGTCC
of sub-MW applications for distributed power generation: the has an electricity cost similar to the CC3LP 共3.84 c$/kWh兲: In
natural gas-fueled 500 kW micro HAT cycle 共mHAT兲 and the
natural gas and oil-fueled 500 kW externally fired micro gas
turbine 共EFmGT兲, 关26兴. Their layouts are simplified versions of Table 3 Main economic assumptions. When double values,
the large-size HAT and EFCC cycles: The former 共mHAT兲 does the first is for a simple gas turbine cycle.
not carry out compressed air intercooling and aftercooling, the O&M cost 共% of purchased equipment cost兲 4%
Inflation 2.5%
Nominal escalation rate of purchased equipment cost 2.5%
Nominal escalation rate of fuel cost 3.0%
Construction initial year 共1 Jan兲 2002
Table 2 Main plant characteristics Construction time 1–2 years
Plant economic life 共book life兲 15–20 years
Plant Fuel type Net Power 关MW兴 Efficiency Plant Life for tax purposes 10 years
Debts—financing fraction 50%
PFBC Coal 356 0.420 Preferred stocks—financing fraction 15%
EFCC3LP Oil 354 0.502 Common equities—financing fraction 35%
Standard ENEL Coal 320 0.423 Debts—required annual return 5.5%
CC3LP Natural Gas 200 0.537 Preferred stocks—required annual return 6%
HAT Natural Gas 50 0.528 Common equities—required annual return 6.5%
RWI Natural Gas 50 0.486 Average income tax rate 30%
STIG Natural Gas 50 0.459 Coal price 2.4e-6 $/kJ
GT Natural Gas 30 0.398 Oil price 3.2e-6 $/kJ
GToil Oil 30 0.393 Natural gas price 4.0e-6 $/kJ
mHAT Natural Gas 0.5 0.317 Demineralized water price 0.5 $/m3
EFmGT Natural Gas⫹Oil 0.5 0.292 Equivalent operating hours 4000– 8000
Fig. 1 Cost of electricity with or without CET Fig. 3 CO2 emission index versus LHV efficiency
fact, even if it burns cheaper fuel 共oil instead of natural gas兲, it is Values of 关 C 兴 and LHV for fuels are reported in Table 1: Regard-
heavily affected by the capital costs of the high-temperature heat ing fuel mixtures, LHV and 关 C 兴 are averaged on mass flow rates.
exchanger and NOx and SOx abatement equipment. The coal- The following relationship between the values of I CO* for fuel
2
fueled plants 共standard ENEL and PFBC combined cycle兲 show ‘‘x’’ and fuel ‘‘y’’ for the same plant efficiency ␩ can be derived
low conversion efficiencies 共less than 44%兲 that do not allow their from Eq. 共7兲:
COE to compete with the other systems, despite the low cost of
coal 共Table 3兲. Simple gas turbine COEs demonstrate that such a *
I CO 关 C 兴 x LHVy
2x
technology can stand alone only for peak power purposes: For that ⫽ . (13)
*
I CO 关 C 兴 y LHVx
reason 4000 equivalent operating hours have been set for a simple 2y
GT, while 8000 have been set for the other plants. The most efficient plants with low CO2 emissions, such as HAT
If the CET is applied to all power plants 共CET and FCET values and CC3LP, are the least penalized and so they become the most
are discussed in depth in the following paragraph兲, a considerable economic choice for investment; on the other hand, the oil-fueled
increase in the COE can be observed 共Fig. 2兲: up to 30 percent for and coal-fueled plants, due to the high carbon content of their
natural gas-fueled plants and over 35 percent for oil-fueled and fuel, cease to be competitive. In this way economic savings match
coal-fueled plants. environmental issues because the most efficient and less-emitting
Among the non-gas-fueled plants, only the EFGTCC shows a plants become the most attractive solutions for the energy market.
COE increase lower than 30 percent, because of the great impact
of capital costs on the total revenue requirement. Oil-fueled and
coal-fueled plants suffer a greater penalization due to the higher
Generalized Carbon Exergy Tax „CET…
mass content of carbon fuel, which is taken into account by the The complete procedure for CET evaluation, fully described in
index of CO 2 emission (I CO* ). the previous paragraphs, is too complex to become a feasible in-
2
strument of energy policy. In addition, many thermodynamic data
In Fig. 3 the hyperbolic relationship between I CO * and plant
2 are required to carry out the complete thermoeconomic analysis:
LHV efficiency is shown. Each different fuel 共or mixture of fuels兲 often the necessary information is not available or not even di-
has its own hyperbole as a function of plant efficiency 共Eq. 共7兲兲. rectly accessible. Here it is shown that the results of the complete
method allow a simplified approach to be developed: The simpli-
fied CET and FCET are named ‘‘generalized’’ because they can be
easily applied to all power plants. The generalized CET and FCET
evaluations require much less information about the energy sys-
tem and thus offer a simpler and more effective energy policy
regulation to manage global warming.
Figures 4 and 5 show the CET and FCET for the studied plants:
A hyperbolic-like curve can be traced to interpolate the calculated
values, but the analytical correlation is more complex. In the
present paper the following function is proposed:
y⫽a 冉冊
1
x
⫺1
b
(14)
where y represents the generalized CET 共GenCET兲 or the gener-

alized FCET 共GenFCET兲 and x stands for LHV efficiency. Such
an equation has two important characteristics that belong to the
complete CET procedure:
1. taxation equal to infinite for a zero-efficiency and
2. taxation equal to zero for a unit-efficiency.
It is important to state that in 共14兲 only two parameters are
Fig. 2 Electricity cost increase with CET used: ‘‘a’’ and ‘‘b.’’ The former defines the level of taxation
break-even carbon tax indicated in 关13兴 to economically sustain
CO2 sequestration. In addition, the detailed thermoeconomic
analysis of the CO2 sequestration facility integrated with a simple
gas turbine cycle with original efficiencies in the range 36 – 41
percent, carried out in 关27兴, shows that only a carbon tax at the
45– 65 $/tonCO2 level can make CO2 sequestration economically
feasible: such a value is very close to the generalized CET corre-
spondent for such an efficiency range. In that respect, it is very
interesting to note that an objective and nonarbitrary method like
the CET, for penalising the CO2 emissions through an internalisa-
tion of costs connected to plant inefficiencies, is capable of giving
similar results: It means that the calculated external ‘‘thermo-
economic’’ cost for CO2 emissions is equal to the ‘‘technical’’ cost
for preventing such emissions from happening. This unforegone
link demonstrates the reliability of thermoeconomics as an ana-
lytical and powerful tool for energy system analysis and the ca-
pabilities of the CET as an effective regulation for managing glo-
bal warming.
The FCET 共Fig. 5兲, that is the equivalent charge imposed on the
Fig. 4 CET and generalized CET versus LHV efficiency fuel price, depends not only on fuel cost but also on fuel carbon
content and LHV, as issued by 共11兲. So, coal is more heavily
charged than oil and oil more heavily charged than natural gas,
even if there is little difference between them.
共it is the GenCET or GenFCET applied to a plant achieving 50 However, fuel costs increase by up to 100 percent for small-size
percent efficiency兲, while the latter influences the curve trend. applications while by no more than 50 percent for midsize and
Values of ‘‘a’’ and ‘‘b’’ for generalized CET and FCET are shown large-size plants: However, such a considerable cost rise is justi-
in Table 4. fied only by an international energy policy aimed at globally re-
Considering the carbon exergy tax, GenCET curves 共dotted ducing CO2 emissions.
curves in Fig. 4兲 depend only on the fuel cost c f used in 共4,5兲 to The calculated FCET for the EFmGT cycle is an average
calculate the charge, and do not depend on fuel carbon content charge, because it has to be divided between the two fuels 共natural
and fuel LHV 共for fuel mixtures, c f is averaged on mass flow gas and oil兲: For instance, an equal percent cost increase criterion
rates兲: c f is the driving factor for GenCET and ensures market- can be used.
linked taxation. This is a very interesting result because a single When comparing CET 共and FCET兲 values with GenCET 共and
generalized CET law can be defined and then applied to all power GenFCET兲 curves for oil-fueled plants, the former seem to be too
systems according to their fuel cost, whatever their size and what- high: In fact, they lie above the corresponding generalized curve,
ever type of fossil fuel is used. which has been calculated on the basis of the results of gas-fueled
For midsize and large-size plants, with calculated efficiencies in plants. Such a problem would not be present if exergy efficiency
the range 43–53 percent, the resulting CET is always in the inter- replaced LHV efficiency in the abscissa. The relationship between
val 30– 15 $/tonCO2 : These values are in good agreement with the the two efficiencies is
␩ ex LHV
⫽ . (15)
␩ Ex f
Generalized curves have been based on gas-fueled plant CETs
共and FCETs兲. Generalized curves for the other fuels have been
obtained moving the curves based on gas-fueled plants propor-
tionally to fuel cost 共and, for FCETs , also to I CO
* ). Since the ratio
2
expressed by 共15兲 is equal to 0.975 for natural gas and 0.938 for
oil, if exergy efficiency replaced LHV efficiency in Figs. 4 and 5,
the oil-fueled plant CET and FCET would be ‘‘pushed back’’
more than those of gas-fueled plants in such graphics and would
‘‘ride’’ the generalized curve.
To obtain an easy-to-apply method for CO2 emission penaliza-
tion, LHV efficiency has been preferred to exergy efficiency be-
cause it is much more diffused in the energy system field: How-
ever, referring to the complete CET evaluation, the exergy
efficiency would be conceptually more correct.
Parameters ‘‘a’’ and ‘‘b’’ in 共14兲 have been calculated in the
Fig. 5 FCET and generalized FCET versus LHV efficiency efficiency range 30–55 percent, but such a correlation can be
extended to the whole range.
Table 4 Values of ‘‘a ’’ and ‘‘b ’’ for generalized CET and FCET
„Figs. 4 and 5… In fact, Fig. 6 demonstrates that the proposed correlation well
represents the carbon exergy tax results: The extrapolated value of
Coal Oil N.G.⫹Oil N.G. CET compared with the one calculated in 关21兴 for the hybrid gas
turbine-fuel cell plant called FCGT4 and achieving 73.2 percent
GenCET
a 关 $/tonCO2 兴 13.943 18.590 20.914 23.238 LHV efficiency, shows a quite satisfactory agreement: the pub-
b 1.325 1.325 1.325 1.325 lished CET is 8.21$/tonCO2 ; actually, considering the different
GenFCET assumptions regarding fuel equivalent cost and fuel composition,
a 关$/kJ兴 1.406e-6 1.330e-6 1.334e-6 1.284e-6
b 1.325 1.325 1.325 1.325 the reference value becomes 6.7$/tonCO2 ; the generalized CET
correlation gives the value of 6.14$/tonCO2 .
C CO2 ⫽ charge to emitted CO2 ($/yr)
C irr ⫽ destroyed exergy cost共$/s兲
Cw ⫽ residual exergy cost共$/s兲
关C兴 ⫽ fuel carbon mass fraction (kgC /kgfuel)
CET ⫽ carbon exergy tax ($/tonCO2 )
COE ⫽ cost of electricity 共$/kWh兲
Ex f ⫽ fuel exergy content 共kJ/kg兲
FCET ⫽ fuel carbon exergy tax 共$/kJ兲
GenCET ⫽ generalized CET 共$/kJ兲
GenFCET ⫽ generalized FCET 共$/kJ兲
*
I CO ⫽ index of CO2 emission 共kg/kWh兲
2
I CO2 ref ⫽ reference index of CO2 emission 共kg/KWh兲
G CO2 ⫽ emitted CO2 mass flow rate 共ton/yr兲
GF ⫽ fuel mass flow rate 共kg/s兲
k* ⫽ unit exergetic cost 共kJ/kJ兲
k i* ⫽ unit exergetic cost of input exergy flow of the ith
Fig. 6 Wide-range efficiency generalized CET with c f equal to
component 共kJ/kJ兲
4.0e-6 $ÕkJ
p ⫽
k* unit exergetic cost of the exergy product of the
plant 共kJ/kJ兲
LHV ⫽ low heating value 共kJ/kg兲
Since the CET rises dramatically at low efficiencies and be-
N ⫽ equivalent operating hours
comes quite flat at high efficiencies, an upper limit for the CET
NC ⫽ number of components of the plant
共e.g., 70$/tonCO2 ) and a target LHV efficiency 共e.g., 70 percent兲,
p ⫽ pressure 共bar兲
after which no carbon tax would be charged, could be imposed in N ⫽ number of components of the plant
order to strongly push the energy market beyond such a limit, T ⫽ temperature 共K兲
favoring the implementation of the most innovative solutions, x ⫽ vector of plant technical characteristics
such as hybrid fuel cell systems. W el ⫽ plant electrical power output 共kW兲
Greek Letters
Conclusions
␣ ⫽ CO2 production per mass of C (kgCO2 /kgc )
An effective and generalized method for CET and FCET evalu-
ation has been developed and it can easily be applied to all power ␩ ⫽ plant LHV efficiency
plants. It is based on the rigorous results of CET calculation and it ␩ ex ⫽ plant exergy efficiency
takes into account plant net efficiency and the type of fossil fuel. ⌬c el ⫽ electricity cost increase due to CET共$/kj兲
Such a method only requires the following information about the ⌸ ␧ ⫽ inefficiency penalty 共$/yr兲
energy system: ⌿ e ⫽ input exergy flow 共kW兲
⌿ irr ⫽ exergy destroyed 共kW兲
• net LHV efficiency, ⌿ o ⫽ output exergy flow 共kW兲
• average fuel cost, ⌿ p ⫽ plant product exergy content 共kWh/yr兲
• average carbon mass fraction 关 C 兴 for the employed fuels, and ⌿ w ⫽ residual exergy rejected by the plant 共kW兲
• average LHV for the employed fuels.
In a deregulated energy market, CO2 emission penalization References
through a generalized CET can 关1兴 OECD, 1994, ‘‘Managing the Environment: The Role of Economic Instru-
ments,’’ Organization for Economic Cooperation and Development, Paris.
• push the energy industry towards the most innovative, effi- 关2兴 Goodstein, E. S., 1999, Economics and the Environment, Prentice-Hall, Engle-
cient, and environmentally sustainable technical solutions, wood Cliffs, NJ.
• make CO2 sequestration economically feasible and competi- 关3兴 Parkin, M., Powell, M., and Matthews, K., 1997, Economics, Addison-Wesley,
Reading, MA.
tive, 关4兴 Borchiellini, R., Massardo, A. F., and Santarelli, M., 2000, ‘‘An Analytical
• increase the utilization of renewables, and Procedure for the Carbon Tax Evaluation,’’ Energy Convers. Manage., 41,
• help countries to achieve the Kyoto targets and, in that re- Sept.
spect, to improve the life of all the people in the world. 关5兴 Borchiellini, R., Massardo, A. F., and Santarelli, M., 2000, ‘‘A Carbon Tax
Evaluation Based on the Efficient Use of Energy Resources,’’ Int. J. Appl.
In this paper only plants for electrical generation have been Thermodyn., 3共3兲, pp. 129–145.
关6兴 Santarelli, M., 1998, ‘‘Considerazioni Termodinamiche, Economiche ed Am-
studied, without considering cogeneration systems. In that respect, bientali nella Analisi ed Ottimizzazione di Sistemi Energetici,’’ Ph.D. thesis,
while the complete thermoeconomic procedure for CET evalua- Dipartimento di Energetica, Politecnico di Torino, Torino, Italy.
tion can be immediately applied to cogenerative systems, the gen- 关7兴 Santarelli, M., Borchiellini, R., and Massardo, A. F., 1999, ‘‘Environomic Op-
eralized method should be further developed. timization for Combined Plants Including CO2 Influence,’’ IGTI Turbo Ekpo
’99, Indianapolis, IN, June, ASME, New York.
关8兴 Santarelli, M., Traverso, A., Cali, M., and Massardo, A. F., 2001, ‘‘La Metodo-
Acknowledgments logia dell’Efficiency Penalty Quale Criterio Termoeconomico di Penalizzazi-
one delle Emissioni di CO2 ,’’ X Convegno Tecnologie E Sistemi Energetici
The authors gratefully acknowledge the support of a mURST Complessi ‘‘Sergio Stecco,’’ Genova.
‘‘Cofinanziamento 1999’’ grant. 关9兴 Massardo, A. F., Santarelli, M., and Borchiellini, R., 2000, ‘‘Carbon Exergy
Tax 共CET兲: Impact on Conventional Energy Systems Design and Its Contribu-
tion to Advanced Systems Utilisation,’’ ECOS 2000, University of Twente, The
Nomenclature Netherlands, July, ASME, New York.
关10兴 Langeland, K., and Wilhelmsen, K., 1993, ‘‘A Study of the Costs and Energy
c f ⫽ economic fuel cost 共$/kJ兲 Requirement for Carbon Dioxide Disposal,’’ Energy Convers. Manage., 34共9–
c e ⫽ unit inefficiency penalty of input exergy flow 11兲.
关11兴 Summerfield, I. R., Goldthorpe, S. H., Williams, N., and Sheikh, A., 1993,
共$/kJ兲 ‘‘Costs of CO2 Disposal Options,’’ Energy Convers. Manage., 34共9–11兲.
c w ⫽ unit inefficiency penalty of residual exergy flow 关12兴 Schiappacasse, R., and Squeri, S., 1999, ‘‘CO2 Caption and Sequestration
共$/kJ兲 Study,’’ degree thesis, DIMSET, Università di Genova, Genova, Italy.
关13兴 Santarelli, M., Borchiellini, R., and Massardo, A. F., 1999, ‘‘Carbon Tax vs 关20兴 Lozano, M., and Valero, A., 1993, ‘‘Theory of the Exergetic Cost,’’ Energy–
CO2 Sequestration Effects on Environomic Analysis of Existing Power Int. J., 18, p. 939.
Plants,’’ ECOS ’99, Tokyo, Japan, June, ASME, New York. 关21兴 Massardo, A. F., and Lubelli, F., 2000, ‘‘Internal Reforming Solid Oxide Fuel
关14兴 Brandani, M., and Bozzolo, M., 2001, ‘‘Study of Power Plants With CO2 Cell-Gas Turbine Combined Cycles 共IRSOFC-GT兲: Part A—Cell Model and
Sequestration Fed by Hydrogen From Fuel Decarbonisation,’’ degree thesis, Cycle Thermodynamic Analysis,’’ ASME J. Eng. Gas Turbines Power, 122.
DIMSET, Università di Genova, Genova, Italy. 关22兴 Consonni, S., Macchi, E., and Farina, F., 1996, ‘‘Externally Fired Combined
关15兴 Agazzani, A., and Massardo, A. F., 1997, ‘‘A Tool for Thermoeconomic Analy- Cycles 共EFCC兲. Part A/B,’’ ASME Paper Nos. 96-GT-92/93.
sis and Optimization of Gas, Steam, and Combined Plants,’’ ASME J. Eng. 关23兴 Cohen, H., Rogers, G. F. C., and Saravanamuttoo H. I. H., 1996, Gas Turbine
Gas Turbines Power, 119, pp. 885– 892.
Theory, 4th Ed., Longman House, London.
关16兴 Agazzani, A., Massardo, A. F., and Frangopoulos, C. A., 1998, ‘‘Environmen-
关24兴 Ansaldo Ricerche 共ARI兲, 2001, private communication, Genova, Italy.
tal Influence on the Thermoeconomic Optimization of a Combined Plant With
关25兴 Rosén, Per M., 2000, ‘‘Evaporative Cycles—In Theory and in Practise,’’ doc-
NOx Abatement,’’ ASME J. Eng. Gas Turbines Power, 120, p. 557.
关17兴 Massardo, A. F., and Scialò, M., 2000, ‘‘Thermoeconomic Analysis of Gas toral thesis, Lund Institute of Technology, Sweden.
Turbine Based Cycle,’’ ASME J. Eng. Gas Turbines Power, 122. 关26兴 Magistri, L., Traverso, A., Scarpellini, R., and Zito, D., 2001, ‘‘Sviluppo di un
关18兴 Traverso, A., 2000, ‘‘Thermoeconomic Analysis of STIG, RWI and HAT Sistema a Combustione Esterna con Microturbina a Gas per la Generazione
cycles With Carbon Dioxide (CO2 ) Emissions Penalty,’’ degree thesis, Distribuita,’’ Tecnologie e Sistemi Energetici Complessi ‘‘Sergio Stecco,’’
DIMSET, Università di Genova, Genova, Italy. Genova.
关19兴 Traverso, A., and Massardo, A. F., 2001, ‘‘Thermoeconomic Analysis of Mixed 关27兴 Bozzolo, M., Brandani, M., Massardo, A. F., and Traverso, A., ‘‘Thermoeco-
Gas-Steam Cycles,’’ Applied Thermal Engineering, 22, Elsevier, New York, nomic Analysis of a Gas Turbine Plant With Fuel Decarbonisation and CO2
1–21. Sequestration,’’ ASME Paper No. GT-2002-30120.
A New Scaling Method for
Component Maps of Gas Turbine
C. Kong
Using System Identification
J. Ki A scaling method for characteristics of gas turbine components using experimental data
e-mail: [email protected] or partially given data from engine manufacturers was newly proposed. In case of cur-
rently used traditional scaling methods, the predicted performance around the on-design
M. Kang point may be well agreed with the real engine performance, but the simulated perfor-
e-mail: [email protected] mance at off-design points far away from the on-design point may not be well agreed with
the real engine performance generally. It would be caused that component scaling factors,
Department of Aerospace Engineering, which were obtained at on-design point, is also used at all other operating points and
Chosun University, components’ maps are derived from different known engine components. Therefore to
#375 Seosuk-dong, minimize the analyzed performance error in the this study, first components’ maps are
Dong-gu, constructed by identifying performances given by engine manufacturers at some operating
Kwangju 501-759, Korea conditions, then the simulated performance using the identified maps is compared with
performances using currently used scaling methods. In comparison, the analyzed perfor-
mance by the currently used traditional scaling method was well agreed with the real
engine performance at on-design point but had maximum 22% error at off design points
within the flight envelope of a study turboprop engine. However, the performance result by
the newly proposed scaling method in this study had maximum 6% reasonable error even
at all flight envelope. 关DOI: 10.1115/1.1610014兴
Introduction using experimental data or partially given data from engine manu-
facturers was proposed. This method can be realized using the
Performance simulation is one of the most important activities
following procedure: it is that first after obtaining the scaling
not only to minimize risk and cost in development phase but also maps with the scaling factors of components from the known
to monitor engine health in operation phase. performance data or the known experimental data by engine
In order to simulate gas turbine performance it is absolutely manufacturers at some operation points, and then constructing the
needed that a condition, which each engine performance must new component maps using the system identification technique,
follow its own components’ performance, should be satisfied, 关1兴. which build a mathematical model between input 共scaling factors
Generally components’ characteristics are obtained from compo- at some given conditions兲 and output parameters 共pressure ratio,
nent performance tests at various operating and environmental mass flow rate, and efficiency兲 with polynomial equations, and
conditions. However, it can be usually done in case of engine finally analyzing the engine performance with the identified maps.
development stage. Moreover, it is not generally disclosed to en- For this study, P&WC PT6A-62 turboprop engine was used for
gine users because of very expensive and important technical data. a calculation example, which has been a power plant of Korean
Therefore engine users or engineers, who are observing engine indigenous basic trainer KT-1. Basic data for identification of
performance, have been mostly utilizing the known engine com- component maps are uninstalled performance at sea level, 3000 m
ponents’ characteristics by scaling. 共10,000 ft兲, 6000 m 共20,000 ft兲 and 9000 m 共30,000 ft兲 altitudes,
Even though there are various scaling methods, current tradi- and standard atmosphere conditions.
tional methods are mostly using a technique, where first the scal- After analyzing performance at deferent altitude and flight
ing factors for the derived maps are derived from comparison M.N. conditions with uninstalled and installed conditions using
between a design point data of original known performance maps the newly obtained component maps, their results were compared
and a new design point then scaled maps can be obtained by with the results using component maps obtained from the cur-
multiplying the derived scaling factors to off-design point data of rently used scaling method.
the original performance maps, 关2,3兴. However, this technique
may be reasonable if the derived scaling factors are close to 1.0
because it means having similar performance between the original Engine Specification
engine and the scaled engine. Therefore, the performance analysis The engine selected for analysis is the PT6A-62 free-turbine
result using this technique nearby the on-design point may be turboprop engine. The engine’s shaft horsepower is 857.9 kW
usually agreed well with the measured performance result of the 共1150 hp兲 at sea level, but flat rated at around 708.7 kW 共950 hp兲.
scaled engine. However, if calculation at off-design points is per- The engine has a compressor with three axial stages and a single
formed far away from the on-design point, it has generally a draw- centrifugal stage, a reversed flow combustor, single-stage com-
back that the error of calculation by this technique may increase. pressor turbine with cooled NGV and blades, and a two-stage
Therefore, in order to reduce the analyzed performance error, a uncooled power turbine.
new scaling method for characteristics of gas turbine components The selected propeller for PT6A-62 of the Korean basic trainer
was Hartzell HC-E4V-3Exl/E9512CX model, which is four
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN blades, constant speed of 2000 rpm, featherable, and 95 inches
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF diameter, 关4兴. The station number and layout for performance
analysis of the study engine is shown in Fig. 1.
erlands, June 3– 6, 2002; Paper No. 2002-GT-30150. Manuscript received by IGTI, Table 1 shows the performance data at maximum takeoff con-
Dec. 2001, final revision, Mar. 2002. Associate Editor: E. Benvenuti. dition, which were provided by the engine manufacturer, 关5兴.
Table 3 The scaling factor using novel scaling method
Alt. 0m 3048 m 6096 m 9144 m

Compressor PR 0.5885 0.5716 0.5500 0.4664
MFP 0.5786 0.5538 0.5455 0.5461
␩ 0.9535 0.9535 0.9535 0.9535
Compressor PR 1.6636 1.4822 1.3675 1.3053
turbine MFP 0.3466 0.3498 0.3498 0.3498
␩ 1.0985 1.0985 1.0985 1.0985
Power PR 0.9808 1.1904 1.3153 1.4264
turbine MFP 0.7379 0.7054 0.7031 0.7023
␩ 1.0191 1.0191 1.0191 1.0191
Fig. 1 Station number and layout of the study engine
Newly Proposed Scaling Method

Traditional Scaling Method
Generally the design point performance, for instance perfor-
The currently used traditional scaling method is a way to obtain mance at maximum takeoff power condition, would be revealed to
scaled component maps by multiplying scaling factors derived at engine users. Therefore each component performance can be ap-
the on-design point to the original component maps at off-design proximately calculated using well-known thermodynamic equa-
points: The scaling equations used in this method are expressed as tions and the revealed performance. From these calculated pres-
follows, 关2兴: sure ratio, mass flow rate and efficiency of each component, the
PR D. ⫺1 scaling factors can be obtained using Eqs. 共1兲, 共2兲, and 共3兲.
PR⫽ 共 PR M . ⫺1 兲 ⫹1 (1) If performance data, such as power, specific fuel consumption
PR M.D.⫺1 ratio, etc., at some off-design points can be known from engine
ṁ D. manufacturers or experimental tests, scaling factors at some con-
ṁ⫽ ṁ (2) ditions can be calculated. In this case, if component maps, which
ṁ M.D. M . are obtained with scaling factors at some conditions, can be per-
␩ D. formed by the system identification, the closer component maps to
␩⫽ ␩ . (3) real engine may be obtained, where the system identification de-
␩ M.D. M . fines that it builds a mathematical model between input values and
As shown in the first column of Table 2, there are the design output values of a system.
point performance data taken from compressor, compressor tur- In this study, the mathematical model was built using polyno-
bine and power turbine maps, which are obtained from GAS- mial equations, where the input value was the scaling factor of
TURB program, 关6兴. And the second column has the design point each component map which was obtained from engine manufac-
data taken by PT6A-62 turboprop engine, and the third column turers’ performance data at the flight M.N. 0. and the altitude 0,
has the calculated scaling factors using above equations. 3048 m 共10,000 ft兲, 6096 m 共20,000 ft兲, and 9144 m 共30,000 ft兲,
This method may usually be effective if the scaling factors of and the output values were performance parameters such as pres-
scaled component maps are close to 1.0 range of 1.0. However, sure ratio, mass flow parameter, and efficiency of each component
if they are far away from 1.0, this method may generally be using the obtained scaling factors. Component maps used in the
inaccurate. present study were maps for compressor, compressor turbine, and
power turbine and each component has 20 beta-lines per a cor-
rected rotational speed, where the corrected rotational speed de-
fines as follows:
Table 1 Performance data by engine manufacturer
N
CN⫽ (4)
Variable Values 冑␪
Atmospheric condition Sea level static
standard condition where ␪ is T 1 /288.15. Therefore the corrected rotational speed is
Mass flow rate 共kg/s兲 4.18 100% at sea level, 103.63% at 3048 m 共10,000 ft兲, 107.68% at
Fuel flow rate 共kg/s兲 0.0818 6096 m 共20,000 ft兲, and 112.4% at 9144 m 共30,000 ft兲,
Shaft horse power 共hp兲 1150 respectively.
S.F.C. 共kg/kw•hr兲 0.3433
Nozzle throat area 共m2兲 0.058 Calculation Example. First, pressure ratio, mass flow param-
Gas generator rotational speed 37468
共100% rpm兲
eter, and efficiency were calculated to obtain the scaling factors
Propeller rotational speed 2000 for compressor, compressor turbine, and power turbine using air
共100% rpm兲 mass flow rate, fuel flow rate, shaft horsepower at the flight M.N.
0. and the altitude 0, 3048 m 共10,000 ft兲, 6096 m 共20,000 ft兲, and
9144 m 共30,000 ft兲, which were obtained from the PT6a-62 engine
Table 2 The scaling factor using traditional scaling method manufacturer’s performance data, 关3兴. Table 3 shows scaling fac-
M.D. Design data
tors at four conditions obtained using Eqs. 共1兲, 共2兲, and 共3兲. Sec-
Component 共GASTURB兲共PT6A-62兲 S.F. ondly, four sets of component maps were derived using each set of
scaling factors obtained at four conditions.
Compressor PR 15.275 9.4 0.5885 It can be assumed that pressure ratio, mass flow parameter, and
MFP 7.2245 4.18 0.5786
␩ 0.8076 0.77 0.9535 efficiency have the following linear function for the corrected ro-
Compressor PR 2.4252 3.34367 1.6445 tational speeds at each beta-line:
turbine MFP 46.499 16.5848 0.3498
␩ 0.8011 0.88 1.0985 y⫽ax⫹b (5)
Power PR 2.5844 2.5538 0.9807
turbine MFP 64.9699 48.9868 0.7540 where y is the pressure ratio, mass flow parameter, and efficiency,
␩ 0.8635 0.88 1.0191 respectively, and x is the corrected rotational speed. Polynomial
coefficients a and b were calculated using polyfit that is an m-file
of the MATLAB program. And then polynomial equations from
each beta-line were derived using polyval.m of the MATLAB pro-
gram. However, the scaling factors for component efficiencies are
fixed at the design point value because they are almost constant at
four conditions and exhibit simplification of calculation.
Finally, each component map was identified from 70% rpm to
110 rpm using the derived polynomial equations.
Figure 2 shows comparison between component maps obtained
by the newly proposed scaling method 共solid curves兲 and compo-
nent maps obtained by the traditional scaling method 共dotted
curves兲.
Steady-State Performance Analysis

In the steady-state performance analysis at off-design points,
the following assumptions and equations were considered.
The airflow passed through the intake, the compressor, the com-
pressor turbine, and the power turbine must be constant. There-
fore, the following flow compatibility equations can be applied to
this type of the engine, 关1兴.
ṁ a 冑T 03 ṁ a 冑T 01 P 01 P 02
P 03
⫽
P 01
⫻ ⫻
P 02 P 03
⫻ 冑 T 03
T 01
(6)
ṁ a 冑T 04 ṁ a 冑T 03 P 03
P 04
⫽
P 03
⫻
P 04
⫻ 冑 T 04
T 03
(7)
The work done by the compressor and the turbine connected

with the same shaft must be the same. Therefore, work compat-
ibility of the gas generator presents as the following equation:
␩ m C Pg ⌬T 034⫽C Pa ⌬T 012 . (8)
The steady-state performance analysis was carried out at the
uninstalled and installed conditions. Figure 3 shows the flow chart
of the steady state program, which satisfies the above conditions.
Performance Analysis at Uninstalled Condition. At the un-
installed condition, the bleed air for cooling and heating in a cock-
pit and the accessory power extraction were not considered. In
addition, the pressure losses at inlet duct and at nozzle caused by
engine installation to the aircraft were ignored, 关3兴.
When both gas generator and power turbines were running at
100% rpm, the uninstalled performance was analyzed in the
changes of the altitude using the developed program. Where the
altitude interval from sea level to 9144 m 共30,000 ft兲 was 1534 m
共5000 ft兲 and the flight Mach number was fixed at zero.
Table 4 and Fig. 4 show the result that the manufacturer’s per-
formance data 共M.D.兲 are compared with the results using the
newly proposed scaling method 共N.M.兲 and the traditional method
共T.M.兲, respectively.
In investigation of the analysis results, the performance using
the traditional scaling method, was well agreed upon with real
engine performance data at the on-design point, but the perfor-
mance at off-design points did not agreed well with real engine
performance data. However, in the case of the new scaling
method, it was noted that the analysis results at overall operation
range have a relatively reasonable error within 5%. Therefore, in
this study, it was confirmed that the newly proposed scaling
method would be more effective than the traditional scaling
method.
Fig. 2 Comparison between component maps obtained by the
Performance Analysis at Installed Condition. When an en- newly proposed scaling method „solid curves… and component
gine is installed on an aircraft, it happens that the temperature maps obtained by the traditional scaling method „dotted
rises and the pressure drops in the inlet duct, the bleed air loss is curves…
due to cooling and heating for the cockpit and the power loss is
due to accessory power extraction. Moreover, the effective ex-
haust area was influenced by the exhaust pressure loss, 关3,5兴. Table 6 and Fig. 5 show the result that the manufacturer’s per-
In this study, the performance was analyzed at both ECS 共En- formance data 共M.D.兲 are compared with the results using the
vironmental Control System兲 off condition and maximum opera- newly proposed scaling method 共N.M.兲 and the traditional method
tion of ECS. Table 5 shows loss values, which were provided by 共T.M.兲 in installed performance analysis at the ECS off condition,
the PT6A-62 engine manufacturer. flight M.N.⫽0.27, 0.38, 0.48, and sea level.
Fig. 3 Flow chart of steady state performance calculation program
Table 4 Uninstall performance with altitude variation
Table 7 and Fig. 6 show the result that the manufacturer’s per- Alt. Variable M.D. T.M. % Err N.M. % Err
formance data 共M.D.兲 are compared with the results using the
0 m a (kg/s) 4.18 4.1821 0.05 4.18 0.
newly proposed scaling method 共N.M.兲 and the traditional method m m f (kg/s) 0.0818 0.0818 0.00 0.0819 0.12
共T.M.兲 in installed performance analysis at ECS maximum condi- SHP 1150. 1150.2 0.02 1149.8 0.01
tion, flight M.N.⫽0.27, 0.38, 0.48, and sea level. 1524 m a (kg/s) 3.357 3.6729 9.41 3.5029 4.35
In this investigation, it was found that the analysis results using m m f (kg/s) 0.0724 0.0693 4.28 0.0703 2.90
SHP 1070. 977.7 8.63 1037.9 3.01
the newly proposed scaling method at overall operation range 3048 m a (kg/s) 3.0385 3.2560 7.16 3.0902 1.70
have a relatively reasonable error within 6%. Therefore, it could m m f (kg/s) 0.0642 0.0622 3.12 0.0608 5.30
be stated that the newly proposed scaling method would be more SHP 980. 866.2 11.6 934.3 4.67
effective than the traditional scaling method. 4572 m a (kg/s) 2.63 2.7744 5.49 2.6263 0.14
m m f (kg/s) 0.0562 0.0507 9.79 0.0540 3.92
SHP 868. 710.0 18.2 833.7 3.95
Conclusion 6096 m a (kg/s) 2.23 2.3658 6.09 2.2325 0.11
m m f (kg/s) 0.0466 0.0442 5.15 0.0447 4.08
In this study, in order to reduce the analysis performance error SHP 740. 608.7 17.7 705.0 4.73
of the traditional scaling method, a new scaling method for char- 7620 m a (kg/s) 1.88 1.9795 5.29 1.8802 0.01
acteristics of gas turbine components using experimental data or m m f (kg/s) 0.0393 0.0353 10.2 0.0376 4.33
partially given data from engine manufacturers was proposed. SHP 625. 488.5 21.8 591.3 5.40
9144 m a (kg/s) 1.57 1.6619 5.85 1.5680 0.13
Flow of the newly proposed scaling method is as follows: ob- m m f (kg/s) 0.0334 0.0303 9.28 0.0319 4.49
taining the scaling maps with the scaling factors of components SHP 520. 412.2 20.7 514.9 0.99
from the known performance data or the known experimental data
Table 6 Installed performance at the ECS off condition
M.N. Variable M.D. T.M. % Err. N.M. % Err.

0.27 m a (kg/s) 2.245 2.405 7.14 2.278 1.49
m f (kg/s) 0.047 0.040 14.68 0.047 0.43
SHP 712. 574.8 19.27 720.3 1.17
0.38 m a (kg/s) 2.381 2.493 4.72 2.373 0.34
m f (kg/s) 0.049 0.042 15.31 0.049 1.02
SHP 764. 596.8 21.89 749.6 1.88
0.48 m a (kg/s) 2.522 2.602 3.16 2.488 1.34
m f (kg/s) 0.049 0.043 11.84 0.047 4.08
SHP 792. 624.6 21.13 747.0 5.69
Fig. 4 Results of uninstall performance analysis with altitude

variation „flight Mach No.Ä0…
Table 5 Loss by install to an aircraft
Coefficient ECS off ECS max

Inlet efficiency 1.0 0.9922
Inlet temp. rise 3K 3K
Bleed air loss 0% 5.25%
Power extraction 3.73 kW 5.22 kW
Effective exhaust area 0.0477m2 0.0477m2 Fig. 5 Results of installed performance analysis at ECS off
condition „flight Mach No.Ä0.27, 0.38, and 0.48, altitudeÄ0 m…
Table 7 Installed performance at the ECS max condition by engine manufacturers at some operation points, and then con-
structing the new component maps using the system identification
M.N. Variable M.D. T.M. % Err. N.M. % Err.
technique, which build a mathematical model between input 共scal-
0.27 m a (kg/s) 2.223 2.404 8.2 2.300 3.45 ing factors at some given conditions兲 and output parameters 共pres-
m f (kg/s) 0.046 0.040 12.4 0.046 0.22 sure ratio, mass flow rate, and efficiency兲 with polynomial equa-
SHP 600. 546.0 9.0 612.1 2.01 tions, and finally analyzing the engine performance with the
0.38 m a (kg/s) 2.249 2.491 10.8 2.383 6.0
m f (kg/s) 0.048 0.042 13.3 0.048 0.42 identified maps.
SHP 653. 566.6 13.2 686.0 5.05 For this study, a P&WC PT6A-62 turboprop engine was used
0.48 m a (kg/s) 2.386 2.599 8.91 2.493 4.47 for a calculation example. After analyzing performance in unin-
m f (kg/s) 0.049 0.043 11.6 0.047 3.27 stalled and installed conditions at different altitude and variation
SHP 716. 592.5 17.2 712.5 0.49
of flight M.N. using component maps obtained from the newly
proposed scaling method, their results were compared with the
results using component maps obtained from the traditional scal-
ing method. In this investigation, it was found that the analysis
results using the newly proposed scaling method at overall opera-
tion range have a relatively reasonable error within 6%. Therefore
it could be stated that the newly proposed scaling method would
be more effective than the traditional scaling method.
Acknowledgment
This study was supported by research funds from Chosun
University, 2003.
Nomenclature
CN ⫽ corrected rotational speed
CP ⫽ specific heat at constant pressure
M.D. ⫽ provided data by manufacturer
MFP ⫽ mass flow parameter
M.N. ⫽ flight Mach number
ṁ ⫽ mass flow rate
ṁ a⫽ air mass flow rate
ṁ f⫽ fuel flow rate
N ⫽ true rotational speed
N.M. ⫽ new scaling method for component maps
PR ⫽ pressure ratio
rpm ⫽ revolution per minute
S.F. ⫽ scaling factor
SHP ⫽ shaft horse power
T.M. ⫽ traditional scaling method for component maps
␩ ⫽ isentropic efficiency
Subscripts
C ⫽ compressor
CT ⫽ compressor turbine
D. ⫽ design point map values of scaled components
M. ⫽ arbitrary map values
M .D. ⫽ design point map values of original components
a ⫽ air
g ⫽ combustion gas
m ⫽ mechanical
1, 2 ⫽ inlet and outlet station of compressor
3, 4 ⫽ inlet and outlet station of compressor turbine
5 ⫽ outlet station of power turbine
6 ⫽ nozzle throat
References
关1兴 Cohen, H., Rogers, G. F. C., and Saravanamuttoo, H. I. H., 1996, Gas Turbine
Theory, 4th Ed., Longman, London.
关2兴 Sellers, J. F., and Daniele, C. J., 1975, ‘‘DYNGEN-A Program for Calculating
Steady-State and Transient Performance of Turbojet and Turbofan Engines,’’
NASA-TN D-7901.
关3兴 Kong, C. D., and Ki, J. Y., 2001, ‘‘Performance Simulation of Turboprop
Engine for Basic Trainer,’’ ASME Paper No. 01-GT-391.
关4兴 Kong, C. D., 2000, ‘‘Propulsion System Integration of Turboprop Aircraft for
Basic Trainer,’’ ASME Paper No. 00-GT-10.
关5兴 Na, J. J., 1996, ‘‘A Study on PT6A-62 Engine Install Performance Analysis,’’
Agency of Defense Development, Technical Report.
关6兴 Kurzke, K., 1999, GASTURB 8.0 Manual.
Fig. 6 Results of installed performance analysis at ECS maxi- 关7兴 Douglas, I. E., 1986, ‘‘Development of a Generalized Computer Program for
mum condition „flight Mach No.Ä0.27, 0.38, and 0.48, Gas Turbine Performance Simulation,’’ Ph.D. thesis, Cranfield University,
altitudeÄ0 m… Cranfield, UK.
关8兴 Kong, C. D., and Chung, S. C., 1999, ‘‘Real Time Linear Simulation and Modularly, Structured Computer Code for Simulation of Dynamic Behavior of
Control for Small Aircraft Turbojet Engine,’’ KSME Int. J., 13, pp. 656 – 666. Aero-and Power Generation Gas Turbine Engines,’’ ASME J. Eng. Gas Tur-
关9兴 Palmer, J. R., and Yan, C.-Z., 1985, ‘‘TURBOTRANS—A Programming Lan- bines Power, 116, pp. 483– 494.
guage for the Performance Simulation of Arbitrary Gas Turbine Engines With 关11兴 Seldner, K. et al., 1972, ‘‘Generalized Simulation Technique for Turbojet En-
Arbitrary Control Systems,’’ Int. J. Turbo Jet Engines, 2, pp. 19–28.
gine System Analysis,’’ NASA-TN-D-6610.
关10兴 Schobeiri, M. T., Attia, M., and Lippke, C., 1994, ‘‘GETRAN: A Generic,
P. Bedont Off-Design Performance Analysis
e-mail [email protected]
Ansaldo Fuel Cells,
Genova, Italy
of a Hybrid System Based on an
O. Grillo
Existing Molten Fuel Cell Stack
This paper addresses the off-design analysis of a hybrid system (HS) based on the coup-
A. F. Massardo ling of an existing Ansaldo Fuel Cells (formerly Ansaldo Ricerche) molten carbonate fuel
e-mail: [email protected] cell (MCFC) stack (100 kW) and a micro gas turbine. The MCFC stack model at fixed
design conditions has previously been presented by the authors. The present work refers to
Thermochemical Power Group (TPG), an off-design stack model, taking into account the influence of the reactor layout, current
Dipartimento di Macchine Sistemi Energetici e density, air and fuel utilization factor, CO2 recycle loop, cell operating temperature, etc.
Trasporti, Finally, the design and off-design model of the whole hybrid system is presented. Effi-
Università di Genova, ciency at part load condition is presented and discussed, taking into account all the
Genova, Italy constraints for the stack and the micro gas turbine, with particular emphasis on CO2
recycle control. 关DOI: 10.1115/1.1587742兴
Introduction 1 Molten Carbonate Fuel Cell „MCFC… Stack and

Fuel cells are electrochemical reactors that allow an efficient Model
and ecological conversion of energy. High efficiency, close to The fuel cell system is made of a molten carbonate fuel cell,
50%, also at part-load condition, and low pollution, make fuel shown in Fig. 1, a sensible heat reformer 共SHR兲, and two catalytic
cells a very interesting system for distributed power generation. burners; the presence of blowers is due to pressure loss compen-
Molten carbonate fuel cells 共MCFCs兲 have been analyzed in this sation. The whole plant is pressurized at 0.35 MPa. Figure 2
paper. Their name comes from the electrolyte employed, a mixture shows the proof of concept power plant Ansaldo Fuel Cells
MCFC, while Fig. 3 shows a simplified process scheme of the
of alkaline carbonate 共K,Na兲, inserted into a ceramic matrix
whole hybrid system, obtained by the coupling of an MCFC group
(LiAlO2 ); the anode and cathode are made of Ni powder, via a 共FC-represented inside the dashes兲 and a gas turbine. As shown in
proper tape-casting process. MCFCs belong to the ‘‘high tempera- the scheme, the electrochemical reactions taking place in the
ture’’ fuel cell class: the operating range is about 600–700°C, in MCFC stack are reported below:
which the electrolyte has a good ionic conductivity.
In this work the MCFC stack has been considered with refer- H2 ⫹CO⫽
3 →H2 O⫹CO2 ⫹2e
⫺
anode (1)
ence to an existing Ansaldo fuel cells plant, named a fuel cell
CO2 ⫹1/2O2 ⫹2e⫺ →CO⫽ cathode (2)
system 共FC兲, 关1兴, based on a molten carbonate fuel cell stack 共100 3
kW size兲, a sensible heat reformer 共SHR兲, and two catalytic burn- H2 ⫹1/2O2 ⫹CO2,c→H2 O⫹CO2,a overall. (3)
ers. In a previous paper, the MCFC group performance was pre-
sented and analyzed by the authors, 关2兴. As already discussed in Besides the electrochemical one, the water-gas shift reaction,
the cited paper, the MCFC operating temperature 共600–700°C兲 whose equation is as follows, takes place in the anode:
well matches the gas turbine inlet one; therefore, in a pressurized CO⫹H2 O↔CO2 ⫹H2 shift. (4)
hybrid system configuration, an MCFC stack substitutes the tur-
bine combustor. Such a plant configuration is attracting wide in- The stack is fed with a mixture of H2 , CO, CO2 , H2 O 共anodic
terest as a solution to some of the key problems of small-size side兲 and with air and CO2 共cathodic side兲. The products of the
overall reaction are electrical power, heat, steam, and the transfer
turbines 共low efficiency and NOx emission, due to the presence of
of CO2 共in the form of CO⫺ 3 ) from the cathode to the anode
the combustor兲 and MCFC stacks 共high cost兲. The on-design effi-
through the electrolyte. A plant simulation model was previously
ciency of such an MCFC hybrid system has been evaluated as realized in Fortran language, 关1兴. Such a model, positively tested
close to 60%, as discussed by the authors for a small-size hybrid by experimental data, allows a thermal-economic analysis of the
system 共electrical power around 130 kW兲, 关3兴. plant. Nevertheless, being a variable geometry model, it only al-
The small size of the plant and its application in the distributed lows the analysis of the design point conditions, 关4兴. Thus, in
power field justify the study of its performance at part load con- order to study the off-design working conditions of the plant, a
dition. For this reason, the design and off-design performance of new model, written in Simulink, a graphic toolbox of Matlab lan-
the hybrid system 共HS兲, based on the Ansaldo fuel cells 共FC兲, is guage, was developed, 关5兴. The model structure is based upon the
presented and discussed in this paper, on the basis of a model energy balance equation. From Nernst’s equation, it is possible to
realized by coupling the design/off-design models of the micro evaluate the thermodynamic voltage, corresponding to the open
circuit condition. When electrical current is drawn from the stack,
gas turbine and the MCFC unit.
voltage losses arise due to irreversibility; these losses can be clas-
sified as ohmic losses, owing to the ohmic resistance of the stack
elements, and polarization losses, mainly due to an impoverish-
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN ment of the anodic and cathodic gases, 关6兴.
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF In the model realized, input data are stack area, chemical and
thermodynamic characteristics of inlet flow rates, temperature,
erlands, June 3– 6, 2002; Paper No. 2002-GT-30115. Manuscript received by IGTI, pressure and composition, current density. The starting assump-
Dec. 2001, final revision, Mar. 2002. Associate Editor: E. Benvenuti. tions are reported below:
Fig. 1 Molten carbonate fuel cell „MCFC… stack „Ansaldo Fuel Cells…
• cathodic flow rate composition: O2 , N2 , H2 O, CH4 , and 1. composition of outlet flow rates, since the shift equilibrium
CO2 ; constant depends on the operating temperature, and the elec-
• anodic flow rate composition: H2 , CH4 , CO, CO2 , and H2 O; trochemical reaction rate depends on current value 共input
• shift reaction at equilibrium; data兲;
• operating temperature considered as the average value 2. outlet flow rate specific heat;
between outlet gas temperatures. 3. heat flow rate associated with both electrochemical and shift
reactions;
Current value is known, as it depends on input data; the un- 4. thermodynamic voltage and voltage losses, from which it is
known term is the equilibrium temperature. The calculation is possible to calculate the electrical power value, since the
based on the stack energy balance equation, whose terms are cal- current is known.
culable thanks to a first attempt value of T eq . In particular, the
program evaluates: The calculation of heat transmission is based on the hypothesis
Fig. 2 Proof of concept power plant Ansaldo Fuel Cells MCFC
Fig. 3 Hybrid system simplified layout „fuel cell system is represented inside dashed boundary…
of sharing the electrochemical heat flow rate between the anode 2.1 Design Point Performance. The operating conditions
and cathode, on a fixed percentage; this assumption has been are reported in Table 1. The chemical and thermodynamic
adopted in a previous modelization, 关7兴. The model makes an characteristics of the inlet flow rate at design point are shown in
assumption of 3% of the electric power value for thermal dissipa- Table 2 since Table 3 shows the anode/cathode inlet and outlet
tion. From the energy balance equation, it is possible to calculate compositions.
outlet mass temperature and a new value of operating temperature The FC supplies a net power of 104.3 kW 共evaluated as the
for each electrode until data convergence. Moreover, the MCFC difference between the electrical power and the one absorbed by
model takes into account internal gas recycles; at design point, blowers兲 with an efficiency of 52.2%. Pressurization of the stack
their value is 89% anode to anode and 70% cathode to cathode, in has not been considered in the evaluation of the performance. The
order to recover effluent gas sensible heat, and 10.5% anode to
cathode, in order to recycle CO2 . A sensible heat reformer is a
catalytic reactor whose function is to convert methane into a hy-
drogen richer mixture. Reactant sensible heat, thanks to the anodic
exhausts recycle, provides the necessary energy so that the re- Table 1 Operating condition at design point
forming reaction can occur 共endothermic reaction兲. The model
structure is based upon the energy balance equation. Input data are Operating Condition
the chemical and thermodynamic characteristics 共temperature, m fuel in 0.00415 kg/s
pressure, and composition兲 of the inlet flow rates. The unknown m air in 0.139 kg/s
term is again equilibrium temperature. Thanks to a starting as- Current density 1580 A/m2
sumption of its value, it is possible to evaluate the composition of Recycle Anode to anode 89%
the outlet flow rates 共since both reforming and shifting equilib- Rates Anode to cathode 10.5%
Cathode to cathode 70%
rium constants depend on operating temperature兲, the specific heat
of the outlet flow rates, and the heat flow rate associated with both
the reforming and shifting reactions. From the balance equation it Table 2 Chemical and thermodynamic characteristics of the
is possible to calculate the outlet temperature and a new value for FC inlet flow rate
the operating temperature until data convergence. The model for
the burners, working at a fixed percentage of the reacted fuel, is T p Composition %
共°C兲共MPa兲共Molar Fraction Rate兲
also based upon the energy balance equation.
Air in 192 0.35 O2 20
2 Fuel Cell System Performance N2 80
H2 6.15
In order to understand the results of the overall plant simula- H2 O 4.17
tion, the performance of the fuel cell system alone 共not inserted in Fuel in 232 0.35 CH4 89.68
the hybrid system兲, is presented and discussed.
Table 3 Chemical composition in the stack
Cathode Anode
Molar Fraction
Rate % Inlet Outlet Inlet Outlet
O2 11.12 9.48 0 0
N2 71.64 76.92 0 0
CO2 8.14 3.83 59.77 63.02
H2 O 9.07 9.74 24.25 27.7
CH4 0.03 0.03 0.59 0.56
CO 0 0 5.80 4.52
H2 0 0 9.59 4.20
Fig. 5 Operating parameters at part-load condition

single-pass fuel utilization factor, defined as the ratio between
reacted fuel (H2 ⫹CO) and fuel at the stack inlet (H2 ⫹CO), is
0.4.
The model results have been verified, from a design point only, since the reactant content in the cell inlet streams is substantially
by comparison with constant. Under this condition, the reduction in current density
• the results of the above mentioned model, written in Fortran produces a consequent increase in cell voltage. The lowering of
language, 关4兴; the operating temperature has no particular influence on the ohmic
• the results of another simulation, 关8兴. losses and on the cell voltage, 共Fig. 8兲, 关5兴.
The parameters connected with the exploitation of the inlet flow
The calculation error for temperature and composition values, rates are reported in Fig. 9. At rated power, the percentage of
in each module of the plant, is always less than 2%, 关5兴. methane converted into hydrogen decreases, because of the lower
temperature. The fuel (U f ) and oxygen (U o ) utilization rates are
2.2 Off-Design Performance. To evaluate off-design FC defined as the ratio between the reacted flow rate and the one fed
performance, it is necessary to make some considerations on op- to the stack 共‘‘single pass’’兲. They both decrease, since the reacted
erating limits and working conditions: flow rates depend on the imposed current density value 共decreas-
• stack inlet temperature greater than 580°C, in order to guar- ing proportionally with rated power兲, while the percentage of re-
antee the molten condition of the carbonate; actant supplied to the stack is subjected to lower reductions thanks
• inlet cathode minimum percentage of CO2 : 9% 共mass frac- to the increase in the internal recycles. This definition of the fuel
tion rate兲. utilization rate does not take into account the recycle of the flow
rate at the anode outlet. Actually, also considering the recycled
At the first attempt, the FC off-design analysis was performed fraction as reacted flow rate, the fuel utilization rate (U f total) in-
by prescribing a proportional reduction of both inlet flow rates 共air creases by reducing the power, according to the plant efficiency
and fuel兲 and current density, at fixed recycle rates. Following this trend, 共Fig. 7兲, 关9兴.
operating procedure, the stack was fed with an excessive air flow It is necessary to underline that the described operating condi-
rate, making the anode inlet temperature decrease under the oper- tions cannot be adopted after the coupling of the fuel cell system
ating values, already at 90% of rated power, as shown in Fig. 4.
Therefore, to permanently keep the inlet temperatures over the
limit, different operating conditions were adopted; they are illus-
trated in Figs. 5 and 6. Part-load condition is obtained by reducing
fuel inlet flow rate and current density in proportion to the power
generated 共Fig. 5兲; the air inlet flow rate undergoes a greater re-
duction percentage, since a decrease in the air-to-fuel ratio is nec-
essary to avoid the results shown in Fig. 4. In order to recover the
outlet flow rate enthalpy, the plant works at increasing recycle
rates 共as for anode to anode and cathode to cathode ones兲. Despite
its reduction, the anode to cathode recycle rate has to guarantee a
sufficient percentage of CO2 to the cathode. Following these op-
erating conditions, FC thermodynamic efficiency increases by re-
ducing the power supplied by the plant, as shown in Fig. 7. This
fact can be explained on the basis of the cell characteristic curve:
Fig. 6 Internal recycle rates at part-load condition
Fig. 4 Stack inlet temperatures by reducing power at fixed re- Fig. 7 CU efficiency and plant fuel utilization rate at part-load
cycle rates condition
Fig. 8 Cell voltage and Nernst potential versus currently Fig. 10 Hybrid system „HS… operating parameters at part-load
density condition
with an MGT, without employing any dedicated solution, since In this study, the HS design point is the operating condition
the inlet air flow rate depends on the compressor and expander where turbine and compressor flow rates, rotational speed, and the
matching condition. pressure ratio are compatible with the existing fuel cell system
design point parameters. In particular, they are: m fuel
3 Molten Carbonate Fuel Cell „MCFC…—Hybrid ⫽0.00415 kg/s; m air⫽0.1394 kg/s; inlet temperature: 25°C; inlet
System pressure: 0.1013 MPa; pressure ratio: 3.4; current density: 1580
A/m2.
Hybrid cycles are derived from the integration of high- As shown in Fig. 3, the HS efficiency is close to 60%, which is
temperature fuel cell technology 共MCFC and SOFC兲 and small noteworthy, taking into account the small size of the plant. Net
size gas turbines, whose temperature and pressure working condi- power supplied by the hybrid system is 119 kW, 89% coming
tions well match a pressurized stack, 关3,10兴. Figure 3 shows a from the stack and 11% from the turbine. Power absorbed by the
schematic representation of a hybrid system FC-GT plant layout. air compressor is about 23 kW 共60% of the power supplied by the
The FC effluent gas enthalpy is exploited in a gas turbine, supply- turbine expander兲. All the data regarding internal recycle rates at
ing additional power and that necessary to move the inlet air com- design point are equal to the fuel cell system ones 共Fig. 3兲.
pressor. An auxiliary compressor performs the anode pressuriza-
tion. The hybrid system model has also been written in Simulink 3.2 Hybrid System Off-Design Performance at Constant
关5兴; simulation has been realized by coupling the FC model with Turbine Rotational Speed. When a variable speed control sys-
the gas turbine and compressor models. In order to determine the tem is not available, the only possible way of varying the power
matching of the air compressor and gas turbine, the unknown supplied by a system, including a gas turbine, is to vary the over-
terms are compressor flow rate; pressure ratio; and compressor all fuel flow rate. Air flow rate, in fact, depends on the compressor
and turbine isentropic efficiencies, depending on operating condi- working condition; this is a limit for hybrid system efficiency at
tions. Simulation requires an external loop for pressure ratio cal- part load operation, since the stack does not work at its optimal
culation and an internal one for flow rate and efficiency evaluation condition. In order to manage the heat produced into the stack,
共for both turbine and compressor models兲. The input data for the and to avoid an excessive and undesirable cooling of the stack, as
compressor module are air inlet pressure and temperature, rota- already discussed for the FC off-design behavior, two possibilities
tional speed and first-guess pressure ratio. The model is based on can be considered: increasing the cathodic recycle or supplying
the compressor, 关11兴, and expander maps, 关12兴共efficiency and more fuel to the anodic side. The first one is limited by pressure
pressure ratio versus nondimensional flow rate兲. losses, the second one causes a reduction in the whole efficiency,
3.1 Hybrid System Design Performance. There are differ- 关7兴.
ent ways to choose the design point of a hybrid system; for ex- Figure 10 shows the fuel flow rate and current density operating
ample, starting with the data of an existing stack and looking for conditions. The fuel reduction percentage is less than the power
a suitable turbine 共if it exists兲. Another way of defining the HS one, in order to keep the stack temperature over the operating
design point, when the gas turbine has been defined, is an MCFC limit (T min⬎580°C), to the detriment of stack efficiency. The dia-
group designed to be fed with the same air flow rate, to operate at gram also shows the inlet air flow rate, even if it is not an oper-
the turbine pressure, and to have an exhaust temperature equal to ating parameter 共it depends on the MGT matching point兲, in order
the turbine inlet one 共TIT兲. to better visualize the different working conditions between the
FC alone and the FC inserted into the hybrid system 共See also Fig.
5兲. The air inlet flow rate value remains almost constant, accord-
ing to the compressor and expander maps, 关5兴. Figure 11 shows
the internal recycle rates at part load condition; the cathode-to-
cathode rate increases, as well as the anode-to-anode one, in order
to avoid an excessive cooling of the electrodes, to the detriment of
the CO2 recycle.
Owing to the recycle rates values, the exhaust cathalitic burner
is almost inactive at design point; it is more used at part load
condition, since the increase of cathode to cathode recycle rate
causes a decrement of the air to fuel ratio. Figure 12 shows the
performance of both the hybrid system and the FC at part load
condition. The efficiency of the whole plant is calculated as the
ratio between the net power produced and the LHV of the feeding
fuel, while the FC efficiency is the ratio between the power sup-
plied by the stack alone and the LHV of the fuel supplied to the
Fig. 9 Fuel cell system utilization rates at part-load condition plant. At a fixed turbine rotational speed, the FC efficiency is
Fig. 14 Actual and Nernst potential of the cell against current
density
Fig. 11 Internal recycle rates at part-load condition
penalised by its insertion in the hybrid system, even if the plant undergoes the contrasting effects of Nernst potential increase, due
fuel utilization rate (U f total) increases at rated power 共since it to the greater average concentration of reactant, and electrical loss
takes into account the recycled flow rates, too兲. The impossibility increase, due to the reduction in the operating temperature.
of reducing the fresh air flow rate results in an excessive cooling The utilization rate trends are shown in Fig. 15. The fuel utili-
of the stack; this leads to increase the fuel supply to the anode 共in zation rate 共single pass兲, defined as the ratio between reacted
comparison to that supplied before the insertion into the hybrid (H2 ⫹CO) and supplied (H2 ⫹CO) fuel, decreases by reducing
system, at the same part load condition兲. Moreover, the operating the inlet flow rate. Reacted H2 , in fact, decreases by reducing the
temperature reduction causes an increase in electrical losses, be- current density value, while the anodic inlet flow rate is substan-
sides limiting the operating range at rated power. Due to these tially constant, thanks to the increase in the internal recycle. The
facts and the results of the modelling, it is impossible to reduce oxygen utilization rate decreases because the reduction in current
the power below 75% of the design point value. density is not followed by a fresh air reduction. The assessment of
Figure 13 shows the sharing of net power between the stack and operating limit feasibility 共inlet flow rate temperatures and per-
gas turbine, at part load condition: the FC net power is calculated centage of CO2 supplied to the cathodic side兲 is reported in Figs.
as the difference between the electrical power and the power ab- 16 and 17. The impossibility of reducing the air flow rate in the
sorbed by the blowers 共including the anodic compressor兲. The proper way causes a cooling of the stack; to 75% of the power
turbine net power is the difference between the power supplied by supplied by the plant; an increase of the cathode-to-cathode inter-
the turbine expander and that absorbed by the air compressor. nal recycle keeps the stack temperature over the operating limit.
At part-load condition, the contribution of the stack increases At 70% of nominal power, the cathodic inlet temperature prevents
since the turbine power is penalised by the reduction of the tur- a further reduction of the fuel inlet flow rate. With regard to the
bine inlet temperature 共Fig. 13兲. Figure 14 shows the cell voltage cathodic flow rate composition, the inlet percentage of CO2 is
and Nernst potential at part load condition. The first increases, but sufficient to allow the electrochemical reaction to occur until 75%
Fig. 12 Hybrid system „HS… and fuel cell system „FC… effi- Fig. 15 Fuel and oxygen utilization rates, and CH4 converted
ciency versus net nondimensional power supplied by the plant by sensible heat reformer, versus nondimensional power sup-
plied by the hybrid system
Fig. 13 Percentage of net power supplied by the stack and

inlet turbine temperature versus net nondimensional power
supplied by the whole hybrid system Fig. 16 Inlet temperature at part-load condition
Fig. 17 Inlet cathodic percentage of CO2 „mass fraction rate…
at the part-load condition
Fig. 19 Fuel cell efficiency at different turbine rotational speed

versus net nondimensional power supplied by the hybrid
of rated power is reached; its decrease is due to the reduction of system
the anode to cathode recycle rate and the excessive flow of fresh
air in the cathodic flow.
3.3 Hybrid System Off-Design Performance at Variable Conclusions

Turbine Rotational Speed: Preliminary Results. At a fixed
rotational speed, the impossibility of reducing the air flow rate, In this paper, the design and off-design performance of a hybrid
places the stack at nonoptimal operating conditions, and only a system 共HS兲, based on the coupling of a small-size gas turbine
25% reduction of power is possible. The typical operational mode with an existing molten carbonate fuel cell system 共MCFC兲 has
of large-size gas turbine plants does not usually involve the pos- been presented and discussed. The turbine and stack models have
sibility of changing the rotational speed of the turbine. This is been presented and discussed. The coupling of a small size non-
because typical plants do not include an inverter, and thus, the regenerated turbine with an MCFC has shown a potential for ef-
rotational speed of the gas turbine is chosen on the basis of the ficiency close to 60% at design point and always over 55% at
alternate current frequency required by the end user electrical net- part-load condition. With regard to the stack group, an existing
work. On the contrary, a hybrid system requires the presence of an plant, known as a fuel cell system 共100 kW兲 and realised by
inverter, which converts the electrical current produced by the fuel Ansaldo Fuel Cells, has been considered. The different behavior
cell and the alternator; thus, this configuration allows the opera- of the stack at part load condition, when both fuel and air flow
tion of the turbine at a variable rotational speed 共variable fre- rates and current density are reduced, has been presented for two
quency兲. This opportunity is of fundamental importance for oper- cases: at fixed or variable recycle rates:
ating the plant at very high efficiency and, above all, for enlarging • at fixed recycle rates, the stack temperature 共anodic side兲
the operating range at part-load condition. At variable speed, re- decreases under the operating limit already at 90% of rated
ducing the power, as shown in Fig. 18, according to the different power 共Fig. 4兲;
compressor working maps, 关11兴, can decrease the air inlet flow • at variable recycle rates, the stack works at optimal condition
rate. When a variable speed control is available, a reduction in and its efficiency increases by reducing power to 50% of
power supplied by the plant to 50% of design point value is pos- nominal value 共Fig. 7兲; feasibility of operating limits is
sible. Fuel cell system performance is shown in Fig. 19: It is shown in Fig. 7, too.
increased by reducing the turbine rotational speed, since this op-
erating condition allows working at rated power with stack tem- Hybrid system off-design performance has been analyzed for
perature and inlet flow rate ratio almost constant. Hybrid system two different gas turbine rotational speed control systems: fixed
performance is illustrated in Fig. 20; the overall efficiency of the and variable. In both cases, the analysis has required the modifi-
plant is very high, even at part-load condition. At a fixed power cation of the stack internal recycle rates, in order to verify the
supplied by the plant, decreasing the rotational speed increases operating limits 共minimum temperature and inlet cathodic per-
efficiency; for example, at 75% of rated power, the efficiency centage of CO2 ).
value is 58% at 85,000 rpm and 59.5% at 80,000 rpm. By reduc-
ing the power supplied to 50% of nominal value, the operating
pressure decrease is about 0.106 MPa.
Fig. 18 Air compressor working points at part load conditions, Fig. 20 Hybrid system efficiency at different rotational speed
when a variable turbine rotational speed is available versus net nondimensional power supplied by the plant
At a fixed rotational speed, owing to the impossibility of reduc- Subscripts
ing the air flow rate, as it depends on compressor and expander a ⫽ anode
matching condition: c ⫽ cathode
• part-load working is possible to 75% of rated power; a further f ⫽ fuel
fuel flow rate decrease does not allow the verification of the ref ⫽ reforming
operating limit on temperature and percentage of CO2 共Figs. D.P. ⫽ design point
16 and 17兲; HS ⫽ hybrid system
• stack efficiency is penalized by a temperature decrease 共51% FC ⫽ fuel cell system
at 70% of rated power against 52% at design point兲共Fig. 12兲; in ⫽ inlet
• owing to the stack performance, the overall plant efficiency out ⫽ outlet
decreases 共57.5% at 70% of rated power to 59.5% at design eq ⫽ equilibrium
point兲共Fig. 12兲. min ⫽ minimum
0 ⫽ reference condition
At variable turbine rotational speeds, the opportunity of varying
the air flow rate, according to the different compressor working
maps, allows stack feeding with the apt ratio between air and fuel; References
thus: 关1兴 Parodi, F., and Sanderson, R. A., 1994, ‘‘100 kW MCFC Power Plant: Ansaldo
System Analysis,’’ Proceedings of the 1994 Fuel Cell Seminar, FCS Organiz-
• part load working is possible to 50% of rated power, with ing Committee, San Diego, CA.
very high efficiency, at part load condition, too 共Fig. 20兲; 关2兴 Bosio, B., and Massardo, A. F., 2000, ‘‘Assessment of Molten Carbonate Fuel
• a rotational speed control system allows working at around Cell Models and Integration With Gas and Steam Cycles,’’ ASME J. Eng. Gas
Turbines Power, to be published.
constant stack operating temperature and FC efficiency in- 关3兴 Bedont, P., and Massardo, A. F., 2001, ‘‘Heat Recovering and Pressurisation
creases by reducing the power generated 共Fig. 19兲. for an Existing 100 kWe Molten Carbonate Fuel Cell,’’ TESEC Congress,
Genoa, June 共in Italian兲.
Acknowledgments 关4兴 Bedont, P., 1999, ‘‘Molten Carbonate Fuel Cells Modelisation and Integration
With Power Generation Plants,’’ Master thesis, University of Genoa 共in Ital-
This work has been partially sponsored by MIUR of Italy ian兲.
through the Contract FISR 2000, No. 10203– 836. 关5兴 Grillo, O., 2001, ‘‘Design and Part Load Performance of a Hybrid System
Based on a Molten Carbonate Fuel Cell and a Microgasturbine,’’ Master thesis,
University of Genoa 共in Italian兲.
Nomenclature 关6兴 Hirschenhofer, J. H., Stauffer, D. B., and Engleman, R. R., 1994, Fuel Cells
Handbook 共revision 3兲, US-DOE, Morgantown, WV.
T ⫽ temperature 共K兲关7兴 Bosio, B., 1999, ‘‘Development of Fuel Cell Technology,’’ Ph.D. thesis, Po-
TIT ⫽ inlet turbine temperature 共K兲 litecnico of Torino 共in Italian兲.
TFC ⫽ fuel cell temperature 共K兲关8兴 Bosio, B., Arato, E., and Parodi, F., 2001, ‘‘Fuel Cell Power Plant Process
␤ ⫽ pressure ratio Analysis Using Aspen Plus,’’ Proceedings of the 4th Conference on Process
Integration, Modelling and Optimization for Energy Saving and Pollution Re-
V ⫽ stack electrical potential 共V兲 duction, Florence, Italy.
i ⫽ current density 共A/m2兲关9兴 Bosio, B., Parodi, F., Arato, E., and Costamagna, P., 1999, ‘‘Process Analysis
p ⫽ pressure 共Pa兲 of a Molten Carbonate Fuel Cell Pilot Plant,’’ Proceedings of the Ichea P4
h ⫽ specific enthalpy 共J/kg兲 Conference, May, Florence, Italy, AIDIC-Servizi, Milano, p. 651.
关10兴 Costamagna, P., Magistri, L., and Massardo, A. F., 2001, ‘‘Design and Part
cp ⫽ specific heat 共J kg⫺1 K⫺1兲 Load Performance of a Hybrid System Based on a Solid Oxide Fuel Cell
m ⫽ flow rate 共kg/s兲 Reactor and a Micro Gas Turbine,’’ Journal of Power Resources, 96, pp. 352–
U ⫽ utilization rate 共%兲 368.
P ⫽ net power 共W兲关11兴 Huchida, H., Shiraki, M., Bessho, A., and Yagi, Y., 1994, ‘‘Development of a
Centrifugal Compressor for 100 kW Automotive Ceramic Gas Turbine,’’
Superscript ASME Paper No. 94-GT-73.
关12兴 Pullen, K. R., Baines, N. C., and Hill, S. H., 1992, ‘‘The Design and Evalua-
* ⫽ nondimensional value referred to design point one tion of a High Pressure Ratio Radial Turbine,’’ ASME Paper No. 92-GT-93.
J. C. Bailey
Experimental and Numerical
GE Corporate Research and Development,
Niskayuna, NY 12309 Study of Heat Transfer in a Gas
J. Intile
Turbine Combustor Liner
T. F. Fric Experiments and numerical simulations were conducted to understand the heat transfer
characteristics of a stationary gas turbine combustor liner cooled by impingement jets
GE Power Systems, and cross flow between the liner and sleeve. Heat transfer was also aided by trip-strip
Greenville, SC 29602 turbulators on the outside of the liner and in the flowsleeve downstream of the jets. The
study was aimed at enhancing heat transfer and prolonging the life of the combustor liner
components. The combustor liner and flow sleeve were simulated using a flat-plate rig.
A. K. Tolpadi The geometry has been scaled from actual combustion geometry except for the curvature.
GE Power Systems, The jet Reynolds number and the mass-velocity ratios between the jet and cross flow in the
Schenectady, NY 12345 rig were matched with the corresponding combustor conditions. A steady-state liquid
crystal technique was used to measure spatially resolved heat transfer coefficients for the
geometric and flow conditions mentioned above. The heat transfer was measured both in
N. V. Nirmalan the impingement region as well as over the turbulators. A numerical model of the com-
bustor test rig was created that included the impingement holes and the turbulators.
R. S. Bunker Using CFD, the flow distribution within the flow sleeve and the heat transfer coefficients
on the liner were both predicted. Calculations were made by varying the turbulence
GE Corporate Research and Development, models, numerical schemes, and the geometrical mesh. The results obtained were com-
Niskayuna, NY 12309 pared to the experimental data and recommendations have been made with regard to the
best modeling approach for such liner-flow sleeve configurations.
关DOI: 10.1115/1.1615256兴
Introduction cooling by film cooling methods, and combined film and convec-
tive cooling methods, is presented by Schulz 关10兴.
The combustor systems of modern high-efficiency, low-
In lieu of film cooling, combustor liners generally utilize im-
emission turbine engines have been under continual development
pingement, convection, and convective surface enhancement cool-
and improvement over the last decade. While firing temperatures
ing techniques. The use of impingement jets for the cooling of
have been increasing to gain efficiency, simultaneous require- various regions of modern gas turbine engines is widespread, most
ments have been enforced to reduce emissions levels, especially especially within the high pressure turbine. Since the cooling ef-
for power producing gas turbines. Today’s combustions systems fectiveness of impingement jets is very high, this method of cool-
seek to limit NOx emissions to 9 ppm or less on dry fuel, and 25 ing provides an efficient means of component heat load manage-
ppm or less prior to reduction measures on the turbine exhaust ment given sufficient available pressure head and geometrical
gases. To achieve such stringent emissions levels it is necessary to space for implementation. Regular arrays of impingement jets are
maintain combustion zone temperatures as low as possible, which used within turbine airfoils and endwalls to provide relatively uni-
in turn requires more air to be utilized in the premixing process form and controlled cooling of fairly open internal surface re-
and reaction zones. Such systems put a large burden on the com- gions. Such regular impingement arrays are generally directed
bustor liner, requiring that this structure be cooled without the use against the target surfaces by the use of sheet metal baffle plates,
of cooling air bleed as hot-side film cooling. While the geometry inserts, or covers which are fixed in position relative to the target
of many combustor liners is relatively simple, cooling must be surface. These arrangements allow for the design of a wide range
achieved through back-side convective methods alone. Due to the of impingement geometries, including in-line, staggered, or arbi-
extremely long operating intervals of power turbines, combustor trary patterns of jets. A summary of applicable impingement heat
liner cooling must be predictable and reliable to assure system transfer research may be found in Martin 关11兴.
durability and long life. Cases involving the use of a single line of impinging jets, or
Methods for the cooling of combustor liners are of many forms. arrays of circular jets, are of particular applicability to the cooling
Older power turbine engine product designs utilize combustion of combustor liners. Metzger and Korstad 关12兴 investigated the
dilution air and film cooling of various types, similar to their effects of a crossflow on a single line of circular jets within a
aircraft engine counterparts. One common cooling geometry used channel. The relative strengths of the jet flow and crossflow as
shingled combustor liners with continuous slots as investigated by determined by other system parameters such as target spacing and
Chin et al. 关1兴. Another typical film cooled combustor liner geom- Reynolds number are extremely important to the resulting heat
etry is that of full-coverage, multihole, or effusion cooling as de- transfer on the target wall. Several studies have also been per-
scribed in the various research of Crawford et al. 关2兴, Metzger formed to correlate the heat transfer under an array of impinging
et al. 关3兴, Andrews et al. 关4,5兴, Al Dabagh et al. 关6兴, Martiny et al. circular jets, primarily for normal impingement on flat surfaces.
关7,8兴, and Fric et al. 关9兴. An excellent summary of combustor liner Kercher and Tabakoff 关13兴 tested a matrix of square arrays of
in-line jets over ranges of target spacing and jet Reynolds number,
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN correlating streamwise heat transfer with geometry and flow pa-
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF rameters. The range of parameters in 关13兴 included jet average Re
up to 30,000, jet-to-jet spacing x/D of 3 to 12.5 jet diameters, and
erlands, June 3– 6, 2002; Paper No. 2002-GT-30183. Manuscript received by IGTI, target spacing z/D from 1 to 5. In a series of studies of Flor-
Dec. 2001, final revision, Mar. 2002. Associate Editor: E. Benvenuti. schuetz et al. 关14 –16兴 the heat transfer to a flat plate beneath
Fig. 1 Cross-section of typical ‘‘F’’ class combustor system per †23‡
nonsquare arrays of impingement jets was determined, including Experimental Apparatus and Method
in-line and staggered arrays and various effects of initial and de-
For the present study a test rig has been built to evaluate heat
veloping crossflow. Various jet array geometries were investigated
transfer coefficients and pressure drop in a large gas turbine re-
in these studies, providing a major portion of the current database
verse flow combustion system similar to Fig. 1. This system is
for heat transfer correlations of impingement arrays. In particular,
very similar to the low NOx design described in detail by Vand-
关14兴 presents correlations covering jet average Re up to 70,000,
ervort 关23兴. The combustion system model is composed of an
jet-to-jet spacing x/D and y/D of 5 and 4 to 15 and 8 jet diam-
inner liner to contain hot combustion gases and an outer flow
eters, target spacing z/D from 1 to 3, and crossflow ratios up to
sleeve to contain and control cooling flow. The experimental fa-
0.8.
cility used in this study is a cold flow parallel plate test section.
Many studies, both experimental and computational, exist in the
The test section geometry is modeled and scaled to match the
literature concerning fundamental means for enhancing internal
annulus geometry of the combustion liner and flow sleeve assem-
heat transfer within dominantly turbulent flow configurations.
bly. The test section is equivalent to a quarter sector of the com-
These include turbulation of passages by various repeated surface
bustor system. The annulus spacing is matched at each streamwise
features, impingement jets and arrays of jets, swirling flows, pin-
location. The test section is approximately 35.1 cm wide and
banks, roughness, and much more. One of the most common
113.4 cm long. The average passage height is 3.9 cm, varying
means for enhancing heat transfer coefficients within internal pas-
from 2.9 to 4.4 cm. The side walls and flow sleeve are aluminum.
sages, and especially the serpentine passages of many turbine
The test surface is 0.76-mm thick aluminum with 2.54 cm of
blade designs, is the use of turbulators, also known as rib rough-
acrylic backing for insulation and mechanical support. The test
eners. Basic turbulator research leading to widely used data and
section is contained in a three-piece ASME pressure vessel that is
correlations was performed by Webb et al. 关17兴, Burggraf 关18兴,
61 cm in diameter and 220 cm long when assembled. Each section
and Han et al. 关19,20兴. These and other researchers have concen-
of the vessel contains a pipe nozzle for air feed or air exhaust. The
trated on turbulators of relative height e/D less than 0.2, but
test section bolts to flanges inside the pressure vessel 共Fig. 2兲,
greater than that typically associated with uniform surface rough-
which provide sealing between the three sections creating separate
ness. The general findings of all such research has been that
plenums.
surface-averaged heat transfer coefficients within stationary, tur-
The test is operated using room temperature cooling air sup-
bulated passages may be enhanced by factors from 1.8 to 2.8,
plied from dedicated compressors. There are two controlled cool-
while the friction factors or required pumping power are increased
ing flows, each measured with a standard ASME square edge
by factors of 3 to 10. Many parameters are variable in turbulated
orifice station. The first cooling flow is brought in as initial cross
passages, and many more in serpentine circuits, but the range of
flow from one plenum supply. Secondary cooling flow is brought
effects has remained much the same over the years.
in through a second plenum feeding five rows of impingement
In integral structures of airfoils, endwalls, or liners, the total
jets. Both flows are metered and controlled independently. The
heat transfer distribution is important to the proper assessment of
combined cooling flows exhaust into the vessel top section, where
thermal-mechanical loading in the component. It is important also
a valve controls the back pressure. Figure 2 shows the flow circuit
to include the variations in geometry and flow conditions. Ferrara
of the test section. Passage pressures are monitored with static
et al. 关21兴 present a study of various regions of cooling for a
convective and film-cooled combustor liner. They developed an
analysis tool including impingement, turbulated, and film cooling
methods. Smith and Fahme 关22兴 specifically focus on a nonfilm
cooled combustor liner design for low emissions, which utilizes
only convective flow over turbulators.
The present study explores heat transfer distributions for a com-
bustor liner model utilizing both impingement jet cooling, high
Reynolds number turbulated flow between the liner and flow-
sleeve, and variable passage geometry. An experimental model is
used to obtain full-surface heat transfer distributions under engine
representative conditions. A CFD numerical model precisely rep-
resenting the experiment provides predictions of full-surface heat
transfer, as well as direct comparisons to data with both two-layer
and wall-function turbulence modeling methods. Fig. 2 Parallel plate test section module
Fig. 3 Flow cross-section and detailed surface construction
pressure taps in the flow sleeve at 13 axial positions. The inlet

pressure profile is measured with five static pressure taps distrib-
uted spanwise 共circumferentially兲 before the first row of cooling
jets. Air temperatures are measured at the orifice stations, each
section of the vessel, and axially at five locations in the test sec-
tion 共equally spaced from inlet to exit兲. There were also five ther-
mocouples spread spanwise at the channel exit to check unifor-
mity. Under all present test conditions, the inlet crossflow is quite
uniform in distribution, and all heat transfer tests show excellent
spanwise distribution uniformity for each impingement row and
the total downstream flow. Nominal rig flow conditions are listed
below:
Passage Re 共ave.兲 8.4⫻105 Fig. 4 Test rig pressure vessel „left… and internal test section
„right, jets inactive…
Jet Re j range 1.7⫻105 to 2.8⫻105
G j /G c range 0.26 to 0.6
Cross flow 0.98 kg/sec
Impingement flow 1.69 kg/sec the flow path surface. The impingement air supply temperature is
Impingement pressure 558 kPa used for T air inlet , and is the same as the initial crossflow supply
Air inlet temperature 22°C temperature. The heat up of the air over the heated test section
Passage Mach number 0.02–0.09 length was less than 1.1°C, while the minimum temperature po-
The impingement jet diameters are not uniform. Each row has tential between the surface and the air inlet was 11°C. Because the
different jet size, hence the range of jet Reynolds numbers and impingement region ‘‘h’’ is more appropriately based on the sup-
crossflow ratios. There are six spanwise rows of impingement jets. ply air temperature, this same basis was used for the entire test
The average target distance-to-diameter ratio of these jets is 2.3. region in order to allow full-surface comparisons. Experimental
The average jet array spacing streamwise and spanwise is 4. uncertainty in ‘‘h,’’ as calculated using the methods of Kline and
Sharp, square turbulators with full fillet radius are machined in the McClintock 关25兴, is between 8 and 15%. Higher uncertainty is
liner surface over the latter 50% of the flow path. The turbulators associated with higher heat transfer coefficients. The flow rate
are transverse to the flow, with height of 0.76 mm, a pitch-to- uncertainty is ⫾1%.
height ratio of 10, and an average height-to-channel height ratio of
0.022. Computational Technique and Model
Liner wall temperatures are measured utilizing the liquid crystal A fully three-dimensional, steady CFD analysis was performed
video thermography method of Farina et al. 关24兴. A wide band to validate the heat transfer coefficients within the impingement
liquid crystal pre-applied to a mylar sheet was calibrated over its zone of the liner/flow sleeve geometry. The purpose of this model
entire color band. The liquid crystal type was Hallcrest 40– 45°C. was to validate current methods and practices used to predict the
A curve fit of liquid crystal hue verse calibration temperature was local heat transfer coefficients on the liner. As CFD is becoming
then used to calculate liner wall temperatures. The liner heater the primary tool for providing three-dimensional heat transfer
system shown in Fig. 3 was a stack up consisting of 2.54 cm of boundary conditions for FEA lifing predictions, it is extremely
acrylic insulation, liquid crystal sheet, adhesive, foil heater, adhe- important to understand and validate the CFD methods and mod-
sive, and a 0.76-mm nominal aluminum plate. A thin aluminum els, detail the deficiencies, and provide a set of guidelines for
plate was used to allow for machining of turbulator trip strips on ‘‘best practices.’’
the liner cold side while minimizing thermal resistance. A uniform Symmetry conditions were used to reduce the size of the com-
heat flux boundary condition is created by applying high-current, putational model to that of 1/6th of the test rig geometry. Both the
low-voltage DC power to the foil heater. Liquid crystal images fluid and solid domain of the test rig were explicitly modeled as a
were taken with an RGB CCD camera. Site glasses in the rig conjugate heat transfer problem with temperature dependent prop-
provided viewing of the test section with the camera as well as erties. The model and results were created using StarCD-HPC
lighting access. Figure 4 shows the lighting and camera setup. Version 3.1 Patch B.
Each data set is comprised of four to eight images taken at differ- The computational domain and boundary conditions are shown
ent heat flux settings. Heat losses were measured to be less than in Fig. 5. The experimental test rig consisted of a series of heat
2% of the total power input. The definition of local heat transfer fluxes to allow the liquid crystals to respond within their tempera-
coefficient used in this study is ture limits over the liner surface. The CFD model surface heat flux
corresponds to the average experimental condition. The flow was
h⫽Q wall / 共 T surface⫺T air inlet 兲 assumed compressible using the ideal gas law relating pressure,
where Q wall is the power input to the heater divided by the heater density, and temperature. The heat fluxes were applied uniformly
area. The wall surface temperature is calculated using a one- across the liner. The cross flow rate into the flow sleeve was
dimensional temperature drop from the liquid crystal surface to assumed to be uniform. The impingement flow rates into the large
Fig. 5 Computational domain
plenum region of the CFD model closely resembled that of the

test rig. An assumed turbulence intensity magnitude of 10% was
used for both inlet flows. The turbulent viscosity was assumed to
Fig. 7 Two-layer discretization scheme
be 100 times the laminar value to evaluate the inlet turbulence
length scale. The definition of the heat transfer coefficient in the
CFD predictions was kept the same as that for the experiments.
The CFD models consisted of two volumes corresponding to ence of previous similar in-house CFD analyses shown to provide
the impingement zone and the turbulator zone of the liner. For good results, as well as maximum available computational re-
each zone the cell discretization was chosen such that either the sources for reasonable turn-around time.
wall-function approach of Launder and Spalding 关26兴 or the two- The convective terms in the governing equations were dis-
layer turbulence model/mesh approach of Norris and Reynolds cretized using the monotone advection and reconstruction scheme
关27兴 could be used. Thus two models were created for compari- 共MARS兲关28兴. Results shown in this paper include the standard
son. One model consisted of a wall-function mesh in the turbula- k-␧ model, 关26兴, and the two-layer model, 关27兴.
tor zone and the impingement zone. A second model used a two-
layer mesh for both the zones. The two models contained 1.1 and Results and Discussion
3.7 million cells, respectively, and required either one or three The distribution of the heat transfer coefficient for this sector
days to converge using available parallel resources. Generally, model is of value to design engineering in that the flow rate varia-
2500 iterations were required for convergence of the solutions. tion, both in the flow sleeve holes and crossflow, can be adjusted
The turbulator meshes used are shown in Figs. 6 and 7. The size to account for variations found in field operation.
of the first cell layer for the turbulators using the wall-function Before making a comparison of test data and CFD predictions,
mesh varied between y⫹ values of 30 and 50. For the two-layer some fundamental test data effects are of value. As noted in the
mesh the first cell layer varied between y⫹ values of 0.75 and experimental description above, the test surface was composed of
1.5. The grid density of these models was based upon the experi- a thin, heated aluminum plate, smooth under the impingement
Fig. 6 Wall-function discretization scheme
num surfaces above the heaters. The data are presented as span-
wise or circumferentially averaged heat transfer coefficients
across the middle 50% of the test plate, thereby eliminating any
endwall effects. Any sharp changes in the data are usually due to
the interface between adjacent foil heaters. The baseline configu-
ration for this study is that with both impingement and down-
stream turbulators; this is compared to CFD and literature data
later. The second configuration presents a convection only case,
without impingement, in which 100% of the flow is present at the
inlet to the channel 共2.67 kg/s兲. In this condition, the thermal entry
region effect is observed to decay, approaching a heat transfer
coefficient of about 320 W/m2/K, which is the value for turbulent,
fully developed smooth duct flow based on accepted literature. In
Fig. 8 Impingement heat transfer data with bare surface the present case, this shows a combined thermal and hydrody-
heater
namic entry length of about ten passage heights, which is some-
what less than typically observed under more ideal conditions.
The downstream turbulated wall is seen to enhance the heat trans-
region and turbulated downstream of impingement. Since the fer by a factor of about 2. The third configuration uses impinge-
heater is a thin foil of Inconel, it can also represent a smooth test ment in the upstream region, but eliminates the turbulators in the
surface without the presence of the aluminum plate. Figure 8 downstream region. Taking the turbulators away causes a redistri-
shows the impingement region heat transfer coefficients obtained bution of the pressure in the entire test section, due to the signifi-
by impingement directly onto the heater surface. This method cant fraction of total pressure loss contributed by the turbulators.
eliminates any surface conduction effects or corrections, and so As a result, the impingement jet Re numbers are altered, and the
provides the most detailed local heat transfer results. The interac- impingement heat transfer shows a more uniform jet-to-jet mag-
tion of the first jets with the initial crossflow is seen, both in the nitude. The average impingement region heat transfer is the same
shaping of the jet thermal footprint and in the upstream bow wave. for both impingement configurations. The downstream smooth
Subsequent jet impingement locations show less intensity due to surface heat transfer again decays to the turbulent, fully developed
increased crossflow 共remember that the jets do not have the same level from accepted literature, but with a small increase due to
diameter or Rejet). Regarding these jets as individuals for the mo- channel convergence. The most striking feature of these results is
ment, the peak impingement heat transfer is from 900 to 1000 the 40 to 50% enhancement in the turbulated region caused by the
W/m2/K. Using a very simple expression for peak impingement use of impingement upstream. This enhancement is essentially
Nusselt number, the present conditions result in Nu constant along the channel length, and felt to be due to the el-
⫽C * Rejet0.7Pr0.33, with C between 0.12 and 0.18. While no lit- evated bulk fluid turbulence levels created by the impingement
erature data exist for impingement jets at the present Re with jets. Recall that this configuration uses relatively large jets at very
crossflow, this approximate valuation is in the same range as data high Reynolds numbers 共up to 2.8e5兲, as well as relatively small
of Perry 关29兴 and also the evaluations provided by 关11兴 for jets turbulators of only 2.2% blockage. Under such conditions, an el-
without crossflow. As an additional point, a visual comparison of evated turbulence level generated in the bulk flow might be ex-
the spanwise-averaged data of Fig. 8 can be made with the pected to provide significant enhancement when aided by the tur-
impingement-smooth case in Fig. 9 for which the aluminum plate bulators. This has been shown to be a very repeatable effect in this
was employed. This comparison shows the effect of the thin plate and other geometries tested.
on spanwise-averaged results to be negligible. Figure 10共a兲 shows the predicted velocity vectors in a cross
The heat transfer data for three different liner cooling configu- section through the centerline of the model in the impingement
rations are shown in Fig. 9. Each case has the same total flow rate. region. The vectors show good penetration of the impingement
In this data, since some of the data include machined surface flow through all the holes. The crossflow under the holes is quite
turbulators, all of these tests were conducted using the thin alumi- strong but the impinging jets are able to penetrate and reach the
Fig. 9 Comparison of circumferentially averaged test data for varying geom-

etries
Fig. 10 „a… Computed velocities for impingement region „mÕs…; „b… Computed
velocities for flow past a turbulator „mÕs…
lower plate. A very small area of recirculation may be seen behind The experimental and predicted distributions of liner flow pres-
the jets. Figure 10共b兲 shows the velocity vectors around a turbu- sure drop are compared in Fig. 11. The inlet static pressure is that
lator approximately halfway down the turbulator section of the of the supply air entering as initial crossflow, upstream of the
plate. The boundary layer above the turbulator and also between heated test surface 共hence the negative location兲. The flow path
the turbulators can be discerned. The two-layer model is able to static pressure measurements are in the flow sleeve surface. The
capture the recirculation zone behind the turbulator. The zone liner cooling pressure drop is normalized to the inlet pressure,
length is about 1.5 turbulator heights. Reattachment of the flow is providing a common measure of combustor system performance
evident at about four step heights downstream. in the form of pressure loss fraction. Measurement uncertainty
Fig. 11 Comparison of liner pressure drop for experimental data and CFD
predictions
Fig. 12 Comparison of liner heat transfer for experimental data and CFD predictions
bars are shown on the experimental data indicating the ⫾0.05% as obtained using the wall function approach and the two-layer
accuracy of the pressure transducer used. The CFD static pres- approach, compared with the test data. The test data in this figure
sures are those predicted in the cells adjacent to the flow sleeve represent the center region of the test section. The first half of the
surface. The CFD predictions are seen to consistently underpredict surface represents the impingement zone and the second half rep-
the pressure loss throughout most of the liner channel; the two- resents the turbulators. In the impingement region, the heat trans-
layer model prediction is only slightly closer to the data overall. fer is somewhat underpredicted over the first hole and universally
This underprediction is not surprising in the region of the im- overpredicted over the last four holes with the wall function tur-
pingement jets, considering the significant shear layers involved bulence model. With the two layer approach, the heat transfer is
with crossflow mixing. The continued small divergence of experi- underpredicted over the first half of the impingement region and
mental and predicted loss in the turbulated region is less expected. overpredicted over the latter half but clearly shows better agree-
It is speculated that this is due in part to the continued jet mixing ment with the data. In the turbulator zone that follows, the predic-
which occurs for a considerable distance, as well as the additional tions with both approaches appear to be significantly under-
turbulation from the liner surface; i.e., these effects are not com- predicting data, although the two-layer approach predictions are
pletely captured by the turbulent shear stress terms in the transport higher and much closer to the data. For instance, the peak in the
models. The result is that pressure loss over the computational heat transfer in the vicinity of the first turbulator is predicted only
region is about 30% less than actual. by the two-layer approach.
Figure 12 shows the predictions of the heat transfer coefficient Figure 13 presents an error plot of the CFD predictions relative
Fig. 13 Relative error in heat transfer coefficients between test data and CFD predictions
Fig. 14 Comparison of circumferentially averaged heat transfer coefficient
across liner
to the data. The fractional difference two heat transfer coefficients are plotted as bounding values, both
assuming fully developed roughened flow as in 关17兴. It can be
兵共 hc f d⫺htest兲 /htest其 seen that the correlation of 关17兴 underpredicts the data almost
is calculated at all locations on the surface and plotted. Consistent universally. The correlation of 关17兴 is based upon turbulators in a
with the observations in Fig. 12, the heat transfer is always un- circular tube flow, not parallel plates, and furthermore is only
derpredicted over the turbulators. The wall function approach re- valid to Re of 105 . Moreover, it is to be expected that the up-
sults in much more deviation relative to the data, as much as stream conditions of the present test section will have an influence
⫺25%. In the impingement region, the deviation from the data is on the turbulated section heat transfer. It is interesting that the
less with the two-layer approach. two-layer prediction falls almost entirely midway between the
The local heat transfer coefficients were spanwise-averaged and predictions using 关17兴.
a one-dimensional heat transfer coefficient distribution along the Based on these averaged as well as the local heat transfer pre-
flow direction generated. Figure 14 presents both the test data and dictions, it is clear that the two-layer approach is recommended
predicted heat transfer coefficients. The bulk temperature used for for more accurate impingement and turbulator heat transfer pre-
definition of the heat transfer coefficient was the inlet temperature dictions. However, the use of a two-layer turbulence model/mesh
of the crossflow. The calculated CFD heat transfer coefficients for a real gas turbine combustor liner containing turbulators may
have the peaks and valleys within the impingement zone 共0–36 be practically infeasible because the number of cells required
cm兲 for both models, however, the levels predicted are not the could exceed available computer pre/post processing limits. In
same. Barring the first hole, the wall function approach over- this study, a two layer approach was used over 1/6th of the test rig
predicts the data over the remainder of the impingement region. geometry 共resulting in a 3.7 million cell mesh兲 to understand its
The two-layer approach underpredicts the data initially, perfectly impact on the heat transfer predictions and demonstrate its supe-
agrees with the data over the middle hole and then overpredicts riority over the wall function approach. In a fully three-
the data. Past the last impingement jet, the wall function model dimensional CFD simulation of a combustor liner, a judgment call
does not capture the large elevation of the average heat transfer therefore has to be made between the accuracy of predictions and
coefficient at the start of the turbulators. The two-layer approach the available computer resources.
is able to capture the large increase in heat transfer coefficients.
Overall, in the impingement region, the two-layer approach is the
better one to use. Using the jet Reynolds numbers and the cross
flow to jet momentum ratios, the heat transfer coefficient under
each of the five holes was calculated based on the correlation due Conclusions
to Metzger and Korstad 关12兴. The correlation of 关12兴 falls below Experiments and numerical simulations were conducted to un-
the data, as well as both the CFD predictions, universally at all the derstand the heat transfer characteristics of a stationary gas tur-
five jet locations. This is to be expected since the Re range of 关12兴 bine combustor cooled by impingement jets and also comprised of
is far below that of the present test, and the available correlations turbulators for heat transfer enhancement. A flat-plate test rig was
are all based on uniform jet sizing. Still, the comparison is rea- constructed corresponding to a one-quarter circumferential section
sonably good. of the combustor unwrapped. The Reynolds number through the
For the turbulator zone 共⬎36 cm兲, the wall-function mesh sig- impingement holes and the momentum ratios were matched to
nificantly underpredicts the average heat transfer coefficient rela- realistic combustor conditions. The turbulators in the test rig also
tive to the data. The two-layer calculation is able to predict the modeled the relative shape and size as those in a real combustor
initial spike of the heat transfer over the turbulators although fur- liner. A steady-state liquid crystal method was used to obtain the
ther downstream, it ends up underpredicting the data. However, local heat transfer coefficient distributions. A numerical model of
without any doubt, the two-layer method gives results that are far the exact test rig was created and CFD was used to calculate the
closer to the data as compared to wall functions. The correlation heat transfer coefficients. Two turbulence models were used: 共1兲
due to Webb et al. 关17兴 was used to estimate and compare the heat wall function, and 共2兲 two layer. Overall, the latter approach gave
transfer coefficients here. There are two reference locations in- predictions that were more accurate and much closer to the data.
volved in the passage, one corresponding to the inlet of the tur- However, the computational requirements with this approach
bulated section and the second corresponding to the exit of the could prove to be very prohibitive leaving the CFD practitioner
turbulated section. Based on the conditions at these two locations, with a difficult choice to make.
Nomenclature Pollutant Formation and Rising Thermal Efficiencies,’’ Heat Transfer in Gas
Turbine Systems, Ann. NY. Acad. Sci., 934, pp. 135–146.
Acf ⫽ passage crossflow area 关11兴 Martin, H., 1977, ‘‘Heat and Mass Transfer Between Impinging Gas Jets and
Aj ⫽ jet area Solid Surfaces,’’ Advances in Heat Transfer, 13, Academic Press, San Diego,
CA, pp. 1– 60.
Ah ⫽ heater area 关12兴 Metzger, D., and Korstad, R., 1972, ‘‘Effects of Crossflow on Impingement
D ⫽ jet diameter 共mm兲 Heat Transfer,’’ ASME J. Eng. Power, 94, pp. 35– 41.
h ⫽ heat transfer coefficient 共W/m2/K兲关13兴 Kercher, D., and Tabakoff, W., 1970, ‘‘Heat Transfer by a Square Array of
Gc ⫽ crossflow mass velocity⫽m c f /A c f Round Air Jets Impinging Perpendicular to a Flat Surface Including the Effect
of Spent Air,’’ ASME J. Eng. Power, 92, pp. 73– 82.
Gj ⫽ jet mass velocity⫽m/A j 关14兴 Florschuetz, L., Truman, C., and Metzger, D., 1981, ‘‘Streamwise Flow and
m ⫽ jet mass flow rate 共kg/s兲 Heat Transfer Distributions for Jet Array Impingement With Crossflow,’’
mcf ⫽ passage cross flow rate 共kg/s兲 ASME J. Heat Transfer, 103, pp. 337–342.
Q total ⫽ total heater power 共W兲关15兴 Florschuetz, L., Berry, R., and Metzger, D., 1980, ‘‘Periodic Streamwise Varia-
Pr ⫽ Prandtl number tions of Heat Transfer Coefficients for Inline and Staggered Arrays of Circular
Jets With Crossflow of Spent Air,’’ ASME J. Heat Transfer, 102, pp. 132–137.
Re ⫽ channel Reynolds number based on 2⫻ height 关16兴 Florschuetz, L., Metzger, D., and Su, C., 1984, ‘‘Heat Transfer Characteristics
Re j ⫽ jet Reynolds number⫽(4m)/(D/ ␮ ) for Jet Array Impingement With Initial Crossflow,’’ ASME J. Heat Transfer,
T air ⫽ plenum supply temperature 共°C兲 106, pp. 34 – 41.
T surface ⫽ liner wall temperature 关17兴 Webb, R. L., Eckert, E. R. G., and Goldstein, R. J., 1971, ‘‘Heat Transfer and
␮ ⫽ viscosity Friction in Tubes With Repeated-Rib Roughness,’’ Int. J. Heat Mass Transfer,
14, pp. 601– 617.
关18兴 Burggraf, F., 1970, ‘‘Experimental Heat Transfer and Pressure Drop With Two-
References Dimensional Turbulence Promoter Applied to Two Opposite Walls of a Square
Tube,’’ Augmentation of Convective Heat and Mass Transfer, Bergles and
关1兴 Chin, J., Skirvin, S., Hayes, L., and Burggraf, F., 1961, ‘‘Film Cooling With Webb, eds., ASME, New York, pp. 70–79.
Multiple Slots and Louvers—Part 1: Multiple Continuous Slots,’’ ASME J. 关19兴 Han, J. C., Glicksman, L. R., and Rohsenow, W. M., 1978, ‘‘An Investigation
Heat Transfer, 83, pp. 281–286. of Heat Transfer and Friction for Rib-Roughened Surfaces,’’ Int. J. Heat Mass
关2兴 Crawford, M., Kays, W., and Moffat, R., 1980, ‘‘Full-Coverage Film Cooling
Transfer, 21, pp. 1143–1156.
Part 1: Comparison of Heat Transfer Data for Three Injection Angles,’’ ASME
关20兴 Han, J. C., Park, J. S., and Lei, C. K., 1984, ‘‘Heat Transfer Enhancement in
J. Eng. Power, 102, pp. 1000–1005.
Channels With Turbulence Promoters,’’ ASME Paper 84-WA/HT-72.
关3兴 Metzger, D. E., Takeuchi, D., and Kuenstler, P., 1973, ‘‘Effectiveness and Heat
关21兴 Ferrara, G., Innocenti, L., Migliorini, G., Facchini, B., and Dean, A. J., 2000,
Transfer With Full-Coverage Film Cooling,’’ ASME J. Eng. Power, 95, pp.
‘‘Heat Transfer Analysis in a Modern DLN Combustor,’’ ASME Paper 2000-
180–184.
关4兴 Andrews, G. E., Al Dabagh, A. M., Asere, A. A., Bazdidi-Tehrani, F., Mkpadi, GT-254.
M. C., and Nazari, A., 1992, ‘‘Impingement/Effusion Cooling,’’ AGARD CP- 关22兴 Smith, K., and Fahme, A., 1999, ‘‘Backside Cooled Combustor Liner for Lean-
527, Heat Transfer and Cooling in Gas Turbines, Paper Number 30. Premixed Combustion,’’ ASME Paper 99-GT-239.
关5兴 Andrews, G. E., Khalifa, I. M., Asere, A. A., and Bazdidi-Tehrani, F., 1995, 关23兴 Vandervort, C. L., 2001, ‘‘9 ppm NOx /CO Combustion System for ‘‘F’’ Class
‘‘Full Coverage Effusion Film Cooling With Inclined Holes,’’ ASME Paper Industrial Gas Turbines,’’ ASME J. Eng. Gas Turbines Power, 123, pp. 317–
95-GT-274. 323.
关6兴 Al Dabagh, A. M., Andrews, G. E., Abdul Husain, R. A. A., Husain, C. I., 关24兴 Farina, D. J., Hacker, J. M., Moffat, R. J., and Eaton, J. K., 1994, ‘‘Illuminant
Nazari, A., and Wu, J., 1990, ‘‘Impingement/Effusion Cooling: The Influence Invariant Calibration of Thermochromic Liquid Crystals,’’ Exp. Therm. Fluid
of the Number of Impingement Holes and Pressure Loss on the Heat Transfer Sci., 9, pp. 1–9.
Coefficient,’’ ASME J. Turbomach., 112, pp. 467– 476. 关25兴 Kline, S. J., and McClintock, F. A., ‘‘Describing Uncertainties in Single-
关7兴 Martiny, M., Schulz, A., and Wittig, S., 1995, ‘‘Full-Coverage Film Cooling Sample Experiments,’’ Mech. Eng. 共Am. Soc. Mech. Eng.兲, Jan.
Investigations: Adiabatic Wall Temperatures and Flow Visualization,’’ ASME 关26兴 Launder, B. E., and Spalding, D. B., 1974, ‘‘The Numerical Computation of
Paper 95-WA/HT-4. Turbulent Flows,’’ Comput. Methods Appl. Mech. Eng., 3, pp. 269–289.
关8兴 Martiny, M., Schulz, A., and Wittig, S., 1997, ‘‘Mathematical Model Describ- 关27兴 Norris, L. H., and Reynolds, W. C., 1975, ‘‘Turbulent Channel Flow With a
ing the Coupled Heat Transfer in Effusion Cooled Combustor Walls,’’ ASME Moving Wavy Boundary,’’ Report No. FM-10, Department of Mechanical En-
Paper 97-GT-329. gineering, Stanford University, Stanford, CA.
关9兴 Fric, T. F., Campbell, R. P., and Rettig, M. G., 1997, ‘‘Quantitative Visualiza- 关28兴 STARCD, 2001, Version 3.1B manual.
tion of Full-Coverage Discrete-Hole Film Cooling,’’ ASME Paper 97-GT-328. 关29兴 Perry, K. P., 1954, ‘‘Heat Transfer by Convection From a Hot Gas Jet to a
关10兴 Schulz, A., 2001, ‘‘Combustor Liner Cooling Technology in Scope of Reduced Plane Surface,’’ Proc. Inst. Mech. Eng., 168, pp. 775–780.
Disintegration of Oil Jets
A. Glahn
M. F. Blair
Emerging From Axial Passages at
United Technologies Research Center,
411 Silver Lane, M/S 129-19,
the Face of a Rotating Cylinder
East Hartford, CT 06108 A fundamental study has been performed to examine the disintegration of oil films emerg-
ing from axial passages at the face of a rotating cylinder. The investigation has been
conducted in parallel to a similar study on atomization processes at rotating radial holes
K. L. Allard (Glahn, A. et al., 2001, ‘‘Disintegration of Oil Films Emerging From Radial Holes Inside
Pratt & Whitney, a Cylinder,’’ ASME Paper No. 2001-GT-0202) and has used the same approaches in
400 Main Street, simulating one of the droplet generation sources in aeroengine lubrication systems. Both
East Hartford, CT 06108 papers aim to contribute to the establishment of a database that can be used for the
development of droplet generation models directly applicable to engine conditions. As
with the parallel investigation, the near-term objectives of fundamental oil film disinte-
S. Busam gration studies are (i) to determine droplet sizes under relevant aeroengine bearing com-
partment operating conditions, and (ii) to measure individual droplet diameter/velocity
O. Schäfer relationships. The long-term objective is to incorporate this information into advanced
design systems such as CFD-based tools. In the present study, flow visualization has been
S. Wittig used to identify the dominant disintegration processes. Droplet diameters and velocities
have been obtained for relevant engine operating conditions. Data are presented in terms
Institut für Thermische Strömungsmaschinen, of both characteristic diameters and size-class resolved droplet velocities and flow angles.
Universität Karlsruhe, A comparison of droplet sprays measured in the present study with those generated by
Kaiserstrasse 12, disintegration of oil films at the rim of a rotating disk (Glahn, A. et al., 2000, ‘‘Droplet
Karlsruhe 76128, Germany Generation by Disintegration of Oil Films at the Rim of a Rotating Disk,’’ ASME Paper
No. 2000-GT-0279) has been enabled by introducing nondimensional parameters for
atomization products and operating conditions. 关DOI: 10.1115/1.1586310兴
Introduction Glahn et al. 关1兴 report on the first phase of a multiyear, funda-
mental investigation into generic oil system atomization processes
Some of the consequences of current aeroengine development
by examining the disintegration of oil films at the rim of a rotating
trends for lubrication system designs have been discussed in sev- disk. The present paper is a continuation of this effort and deals
eral prior publications on bearing compartment two-phase flow with the disintegration of oil films emerging from axial passages
and internal heat transfer. Most recently, Glahn et al. 关1兴 pointed at the face of a rotating cylinder. This study has been conducted in
out that high pressure, high temperature, and high rotational parallel with an investigation of oil film breakup processes at ro-
speeds of the turbine rotor significantly reduce the design margin tating radial holes, 关10兴. Both studies aim at simulating droplet
for lubrication system components and especially bearing com- generation sources in aeroengine bearing compartments and gear-
partments. Improved design tools are required to meet the chal- boxes. The present paper focuses on the oil film disintegration
lenges of current and future engine developments. downstream of rotating axial holes and passages that exist in in-
In recent years, progress has been made on several subtasks tershaft oil flow configurations and gear support systems.
involved in the overall design process, 关2– 8兴. All these studies The objectives of the study were as follows: 共i兲 to identify
were helpful in getting an overall picture of the flow phenomena disintegration modes relevant to aeroengine bearing compartment
involved in bearing compartments. They establish a database for operating conditions, 共ii兲 to determine droplet sizes under those
validation purposes and some of them appear to be directly appli- operating conditions, and 共iii兲 to measure individual droplet
cable for certain aspects of the engine design. However, the de- diameter/velocity relationships.
velopment of a design system and its adaptation to different geo- Following a brief description of the experimental setup, results
metrical and operational boundary conditions requires a platform are presented in terms of samples from the visualization work,
that is able to integrate these submodels. typical spatial droplet diameter distributions, characteristic diam-
A viable approach would be to use CFD methods to predict the eters, and droplet velocities as a function of flow rate and rota-
two-phase air/oil flow in lubrication system components. A study tional speed. Nondimensional parameters for atomization products
presented by Glahn et al. 关9兴 suggested the feasibility of such an and operating conditions are introduced that are useful in compar-
approach. However, a prerequisite for a successful implementa- ing droplet sprays measured in the present study with those gen-
tion of these tools into a standard design environment is the avail- erated by disintegration of oil films at the rim of a rotating disk,
ability of boundary conditions and the development of submodels 关1兴. Conclusions drawn from this comparison will summarize the
which describe the complex phase interaction mechanism and at- paper.
omization effects prevailing in these parts. Therefore, further stud-
ies of film behavior and atomization processes are essential. Experimental Setup and Techniques
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN Test Rig. The test facility used for the fundamental oil film
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF disintegration studies reported in this paper is basically the same
tional Gas Turbine and Aeroengine Congress and Exhibition, New Orleans, LA, June
used by Glahn et al. 关1兴. Minor modifications similar to those
4 –7, 2001; Paper 2001-GT-0201. Manuscript received by IGTI, December 2000, undertaken in a parallel study by Glahn et al. 关10兴 are highlighted
final revision, March 2001. Associate Editor: R. Natole. in the following section. The rotating disk used in the prior inves-
Fig. 1 Rotating Cylinder test rig
tigation has been replaced by a rotating outer cylinder that sup- transparent housing walls. Gravitation forces the oil to flow down-
ports an insert which forms the axial passages 共Fig. 1兲. Rotational ward into the oil sump from where it is scavenged back into the
speeds and physical dimensions of this arrangement had to be reservoir.
chosen such that two goals were met: 共i兲 to provide relevant bear-
ing compartment operating conditions and 共ii兲 to guarantee full Measurement Techniques. As in the previous investigation
optical access to the rotating holes. focussing on the disintegration of oil films at the rim of a rotating
The first goal was met based on a simulation of the relevant disk, 关1兴, the experimental objectives of this study required apply-
Weber number which is the parameter dominating the atomization ing techniques for flow visualization as well as laser Doppler tech-
process. By using a test fluid 共Dow Corning 200 Fluid 5cS兲 with niques for the droplet sizing and velocimetry.
the same fluid properties at room temperature that turbine oil The first task, aimed at identifying the oil film breakup mecha-
would have at 373 K, engine Weber numbers are obtained for nisms, used two types of flow visualization arrangements. In one
冉冊冉冊
setup, stroboscopic illumination along with a simple video camera
␻T 2
DC 3
was employed to obtain a detailed full-access documentation of
1⬇ . (1)
␻E DS oil flow patterns. At high rotational speeds, however, the open
housing of the test rig had to be covered in order to avoid oil
The conclusion to be drawn from Eq. 共1兲 is that it would be droplet penetration into the laboratory room. Utilization of a
favorable to simulate the Weber number by aiming for a larger simple transparent cover in combination with standard video
cylinder diameter and keeping the rotational speeds at a moderate equipment had limitations due to droplet coalescence and film
level. For this investigation, a maximum rotational speed of N T flow formation at the inner wall. Therefore, another recording
⫽10,000 rpm and a cylinder diameter of D C ⫽0.13 m 共5.12 in.兲
system was adapted with an endoscope optic (6.5•10⫺3 m diam-
have been chosen. The exit diameter of the rotating axial passages
eter, 110° apex angle兲 in front of the video camera. For both
is D P ⫽0.11 m 共4.33 in.兲. These values translate into an engine
arrangements all flow phenomena were monitored on a screen,
condition of N E ⫽16,000 rpm and a shaft diameter of D S
recorded on videotape, and examined in detail using digital image
⫽0.10 m 共3.94 in.兲 which are expected to be typical values for
processing.
current aeroengine designs.
For quantitative measurements of spray characteristics, a single
The cylinder diameter of D C ⫽0.13 m in combination with a
particle counter, Aerometrics’ PDPA, was used. In recent years,
relatively short cylinder length of about L C ⫽0.07 m 共2.76 in.兲
the PDPA has become a standard technique for atomization stud-
also supports the second design requirement by enabling a canti-
ies. Thus, a lengthy description of the measuring principle is
lever arrangement and, therefore, full optical access from the non-
avoided here and the interested reader is referred to the specific
drive side. The cylinder is located in a stationary housing with an
literature 共e.g., Bachalo and Houser 关11兴兲. However, the adapta-
inner diameter of 0.301 m 共11.85 in.兲. The housing consists of an
tion of this measuring technique to the current test is special
aluminum frame supporting a transparent thermoplastic that al-
because it used a backscatter arrangement at an off-axis angle of
lows optical access for flow visualization studies throughout the
130 deg, which became available for measurements due to a rela-
whole circumference. The shaft is positioned by spindle bearings
tively high index of refraction for oil that is greater than 1.4. A
and driven by an electromotor in combination with a flexible cou-
comprehensive analysis of this type of measurement configuration
pling and a flat belt.
is given by Willmann et al. 关12兴.
Oil is pumped out of a reservoir through an electrical oil heater
(T max⫽473 K for ṁ⫽0.1 kg/s) into the rig. It is fed along the Test Matrix for Droplet Sizing and Velocimetry. Measure-
centerline onto the spinning disk face using a tube with an inner ments for the characterization of sprays generated by the disinte-
diameter of 6•10⫺3 m 共0.236 in.兲. The oil is driven by centrifugal gration of oil films emerging from rotating axial passages were
forces along the inner cylinder wall until it reaches an annulus carried out for some 152 different combinations of operating con-
from where it is evenly distributed to 12 axial passages that are dition and measurement location. As indicated in Fig. 2共a,b兲 varia-
equally spaced along the circumference. As shown in the next tions of operating conditions included three different rotational
section on flow visualization results, oil is discharged through the speeds and two different flow rates. Measurements of spatial spray
axial passages as a thin film that eventually separates and disinte- characteristics were configured to obtain axial and radial droplet
grates into droplets. The atomized oil is collected at the stationary, size and velocity distributions and assumed a 共time-averaged兲 axi-
Fig. 2 Test matrix for droplet size and velocity measurements. „a… Configuration 3:
D eq Ä2.6"10À3 m, L Õ D eq Ä10, q Ä1.67"10À5 m3 Õs. „b… Configuration 3: D eq Ä2.6
"10À3 m, L Õ D eq Ä10, q Ä2.50"10À5 m3 Õs. „c… Nomenclature.
symmetric flow pattern. The circumferential measurement loca- accumulates in the corners of the noncircular rotating passages
tion and the nomenclature used in the subsequent discussion of and 共ii兲 surface tension tries to minimize the overall surface area.
test results are introduced in Fig. 2共c兲. Note the use of a right-hand The thick edges of the oil film eventually disintegrate into liga-
side cylindrical coordinate system that is aligned with the center- ments and subsequently droplets. At higher rotational speeds 共Fig.
line of the rotor (Z-axis兲 and the face of the rotating cylinder at 3共b兲兲 the same mechanisms occur, but the oil film breaks up more
the exit side of the axial passages (R-axis兲, respectively. rapidly. The oil film disintegration appears to be completed at
about ⌬R/D P ⫽0.10, about twice the radial distance between oil
Experimental Results film exit location and the rim of the cylinder (5•10⫺3 m,
0.2 in.).
Flow Visualization. Flow visualization studies were per-
formed in order to identify the fundamental oil film break-up
mechanisms that have to be considered in the modeling and analy-
sis of quantitative test results. Figure 3 shows typical results
gained from the full-access flow visualization.
A close examination of the picture visualizing oil atomization at
very low rotational speeds 共Fig. 3共a兲兲 shows that a thin continuous
oil film separates from the surface at the outer pitch diameter (D p )
of the rotating axial passage. The thin film extends radially out-
ward from the OD side of the passage and is bent in the circum-
ferential direction. At low speeds, equilibrium between the surface
tension force at the free edge and the kinetic energy of the liquid
sheet is maintained over almost the entire radial gap between cyl-
inder and housing. From further video and photo analysis it was Fig. 3 Full-access visualization of axial jet disintegration. „a…
concluded that the oil film thickens at the edges of the sheet. The ␻ Ä105 radÕs, q Ä1.67"10À5 m3 Õs, ␯ L Ä5"10À6 m2 Õs. „b… ␻
following two effects are expected to cause this behavior: 共i兲 oil Ä524 radÕs, q Ä1.67"10À5 m3 Õs, ␯ L Ä5"10À6 m2 Õs.
Fig. 4 Endoscope visualization of axial jet disintegration. „a…
␻Ä314 radÕs, q Ä2.50"10À5 m3 Õs, ␯ L Ä5"10À6 m2 Õs. „b…
À5
␻Ä890 radÕs, q Ä2.50"10 m Õs,
3
␯ L Ä5"10À6 m2 Õs. „c…
␻Ä1047 radÕs, q Ä2.50"10À5 m3 Õs, ␯ L Ä5"10À6 m2 Õs.
In Fig. 4, samples from the endoscope visualization are shown

for rotational speeds and flow rates that reflect test conditions as
defined in 共Fig. 2共b兲兲. Though the viewport is much smaller than
before and the endoscope optic distorts the image, the previous
observations are confirmed. In addition, the impact of rotational
speed on the droplet spray is visualized. A jet, which is character-
ized best at relatively low speed 共Fig. 4共a兲兲 as oil mist, transitions
into a cloud or fog flow 共Fig. 4共b,c兲兲.
The qualitative flow investigation strongly suggests that the
processes driving the atomization of oil films at rotating axial
passages and rotating disks are somewhat similar. However, it is
recognized that 共i兲 the disintegration of oil films emerging from
axial passages starts at discrete locations as opposed to a continu-
ous rim and 共ii兲 the turning from an axial to a radial flow direction
with possible reattachment at the face of the cylinder will have an
impact on the spray characteristics.
Typical results from the quantitative flow measurements are
discussed in the following sections in terms of actual dimensional
quantities. In order to assess differences in the atomization quality
relative to rotating disk oil film disintegration, nondimensional
quantities are introduced and compared in the subsequent
sections. Fig. 5 N Ä3000 rpm, q Ä1.67"10À5 m3 Õs, L Õ D eq Ä10, ⌬ R Õ D P
Ä0.145. Axial profiles of „a… Mean diameters, „b… Cumulative
Typical Droplet Sizes and Velocities. Figure 5 summarizes diameters.
results obtained at a radial distance of ⌬R⫽16•10⫺3 m from the
rim of the rotating cylinder (⌬R/D P ⫽0.145). Measurements
were acquired along an axial traverse ranging from about 0.10
⬍⌬Z/D P ⬍0.10. In Fig. 5共a兲, results are presented in terms of 5共a兲兲 is that the average diameter (D 10) of the spray contributes
arithmetic mean droplet diameter (D 10), area-averaged droplet di- less than 10% to the overall volume. A much better overview on
ameter (D 20), volume-averaged droplet diameter (D 30), and a the volume distribution can be obtained from a combination of all
combination of the latter two, the so-called Sauter mean diameter three cumulative diameters given in 共Fig. 5共b兲兲, e.g. by tracking
(D 32). The shape of the curves shows a distinct asymmetry rela- the mass mean diameter (D V,0.5 ) and recognizing significant con-
tive to the face of the disk (⌬Z/D P ⫽0) with maximums in the tributions to the volume distributions up to D V,0.9 ⫽200 ␮ m
diameter distribution located above the cylinder surface (⌬Z/D P (⌬Z/D P ⫽0). However, it should be pointed out that cumulative
⫽⫺0.02 to ⫺0.04). As will be seen in the discussion of droplet diameters are sensitive to the correct and statistically relevant ac-
flow velocities 共Fig. 6兲, differences in the relative velocities be- quisition of a few large droplets at the upper limit of the measure-
tween oil fragments and surrounding airflow are expected to be ment range. In the present study, a minimum of 10,000 valid
responsible for the shape of the size distribution curve. In the samples was acquired for each of the 152 measurements. In addi-
annular gap between rotating cylinder and stationary housing, the tion to that, a huge number of exploratory measurements were
air velocity is higher than in the rather quiescent environment in conducted in order to find the best suitable combination of instru-
front of the cylinder. Thus the relative velocity between the drop- ment settings and to avoid conditioned sampling.
lets that are spun off the cylinder at high speed and the air is The overall trend of the velocity distributions was briefly dis-
smaller at negative ⌬Z/D P -values and the droplet size distribution cussed above. Figure 6共a兲 summarizes size-class resolved droplet
is skewed towards larger diameters in this region. The decline in velocities for three droplet classes and compares this data with
droplet sizes at both ends of the measured traverse is due to the ensemble-averaged velocities. The positive leg of the velocity dis-
fact that only small particles are able to diffuse into these regions. tribution (⌬Z/D P ⬎0) does not show any size dependency of the
Arithmetic mean diameters in the center of the distribution droplet velocity indicating a uniform flow behavior and a narrow
(⌬Z/D P ⫽0) are in the range of D 10⫽55 ␮ m, whereas Sauter size distribution. The negative leg, however, shows the effects of
mean diameters are about twice as large. larger droplets by moving the ensemble-averaged values
Similar conclusions relative to the droplet propagation can be 共‘‘spray’’兲 above those of rather small size classes.
drawn from another set of characteristic diameters that is shown in In Fig. 6共b兲, droplet flow angles corresponding to the velocity
共Fig. 5共b兲.兲 Diameters D v ,x that define the diameter at which the distributions discussed above are shown. The distribution displays
fraction x of the total droplet volume has been accumulated are a fairly homogeneous behavior for the whole spray and is almost
plotted against the axial coordinate. In the center of the droplet symmetrical relative to ⌬Z/D P ⫽0.
spray (⌬Z/D P ⫽0), a value of D V,0.1 ⫽67 ␮ m was obtained. Analogous to the presentation of axial profiles of diameters and
Thus, a conclusion that can be drawn in comparison with 共Fig. velocities, Fig. 7 presents data for a radial traverse measured at
Fig. 6 N Ä3000 rpm, q Ä1.67"10À5 m3 Õs, L Õ D eq Ä10, ⌬ R Õ D P
Ä0.145. Axial profiles of „a… Droplet and spray velocities, „b… Fig. 7 N Ä8500 rpm, q Ä1.67"10À5 m3 Õs, L Õ D eq Ä10, ⌬ Z Õ D P
Droplet and spray flow angles. Ä0.0. Radial profiles of „a… Mean diameters, „b… Cumulative
diameters.
the location ⌬Z/D P ⫽0, i.e., aligned with the exit plane of the
axial passages at the face of the cylinder. It is interesting to note
that values for D 10 , D 20 , and D V,0.1 remain approximately con- quired for 6 different operating conditions. An overview and com-
stant for radial distances greater than ⌬R/D P ⫽0.10. Characteris- parison of the atomization quality gained for these conditions is
tic diameters that are more dependent on the upper limit of the presented in Fig. 9. Arithmetic mean diameters (D 10) are plotted
measured size distribution display the ongoing atomization pro- against rotational speeds with the flow rate as a curve parameter.
cess of larger oil fragments by showing variations up to a distance Apparently, the latter does not have an identifiable impact on the
of about ⌬R/D C ⫽0.20 where the atomization process is com- droplet generation though it has to be mentioned that the variation
pleted. For the subsequent analysis and correlation of test results, examined here was relatively small. However, small and even
only those data at locations beyond where the mass mean diameter negligible effects of flow rates are also known from other oil film
(D V,0.5 ) has converged into a constant level were considered. In disintegration studies, 关1,10兴. Opposite to that, the rotor speed has
most cases this requirement was met at about ⌬R/D P ⫽0.145. a significant impact on the droplet size. Figure 9 suggests a rela-
Droplet velocities decrease as the particles move radially out- tionship D 10⬃N ⫺0.33.
ward 共Fig. 8共a兲兲. Again, ensemble-averaged velocity information
has been combined with size-class resolved data. For the rela-
tively high speed of N⫽8500 rpm considered here, the ensemble- Comparison With Oil Film Disintegration at the Rim of a
average velocity is close to that of the 32 ␮m class which reflects Rotating Disk. Similar observations to those reported here were
the finer atomization at high rotational speeds. As expected, larger made by Glahn et al. 关1兴 for oil film atomization studies at the rim
droplets maintain a higher velocity at a given radius which reflects of a rotating disk. Therefore, it is believed to be instructive to
the dependency of the droplet/gas interaction on the droplet Rey- compare the spray characteristics generated under these different
nolds number. circumstances.
In comparison with the lower speed case 共Fig. 6共a兲兲 the droplet In order to compare the disintegration process in different con-
velocity relative to the rim speed of the rotating cylinder seems to figurations, nondimensional parameters are required. The previous
decrease with increasing rotational speed. Opposite to that, the study by Glahn et al. 关1兴 adopted Hinze and Milborn’s 关13兴 sug-
droplet flow angle does not depend on either rotational speed or gestion of using a modified Weber number
droplet size. 1 ␳ L␻ 2D 3
We* ⫽ (2)
8 ␴
Impact of Rotational Speed and Flow Rate on the Spray
Characteristic. In the present study, data sets have been ac- and a nondimensional flow rate
Fig. 10 Average spray diameters compared with those gener-
ated at the rim of a rotating disk
diameter, D D , in the case of Glahn et al.’s 关1兴 data while the pitch
diameter of the rotating axial passages, D P , is the proper dimen-
sion in the current study. The same diameters have been used
to normalize measured mean diameters, D 10 /D D and D 10 /D P ,
respectively.
In Fig. 10, normalized arithmetic mean diameters of sprays
generated by rotating axial passages and rotating disks are com-
pared for various Weber numbers. The slopes of the trendlines that
combine data point from different geometries have been included
in the figure and indicate a similar though more pronounced im-
pact of aerodynamic to surface tension impact on the rotating disk
oil film atomization process. Furthermore, it is interesting to note
that in the most relevant range of 106 ⬍We* ⬍107 , nondimen-
Fig. 8 N Ä8500 rpm, q Ä1.67"10À5 m3 Õs, L Õ D eq Ä10, ⌬ Z Õ D P sional mean diameters differ merely by some 30%. This is re-
Ä0.0. Radial profiles of „a… Droplet and spray velocities, „b… markable because oil film atomization processes at rotating disks
Droplet and spray flow angles. offer the opportunity of predicting the diameter distribution based
on an assessment of the oil film thickness at the face of the disk
共Glahn et al. 关10兴兲. A comparison with the latter case, therefore,
␳ Lq 2 enables at least a rough estimate of droplet sizes generated by
V ⫹⫽ (3) rotating axial passages at the face of a rotating cylinder.
␴ •D D3
As seen before in Figs. 5 and 7, the sprays considered here are
for the characterization of disintegration processes and atomiza- non-homogeneous, i.e., the spray has to be regarded as a spectrum
tion products. However, an invariance relative to the flow rate has of drop sizes distributed about a mean value (D 10). Thus, in ad-
been observed in the present study and, therefore, only the Weber dition to this mean diameter, another parameter is needed in order
number as defined in Eq. 共2兲 will be used subsequently to scale to be able to calculate this distribution. The data obtained in the
operating conditions. Note that the diameter to be used is the disk present study has been approximated best by use of a log-normal
distribution of droplet sizes, 关14兴:
dn d
⫽
1
dD d 冑2 ␲ D d s g
exp⫺
1
冋
2s 2g 册
共 ln D d ⫺ln D 10兲 2 , (3)
where s g is the geometric standard deviation and n d is the number

of droplets per size class D d . Measured standard deviations are
shown in Fig. 11 for both geometries at a comparable flow rate.
Though the data show some scatter, no distinct dependence on the
Weber number has been identified. Oil film disintegration at the
rim of a rotating disk apparently produces a narrower droplet size
distribution than rotating axial passages. The latter case is reason-
ably well characterized by a constant value of
s g ⫽0.335. (4)
From Fig. 10 and 11, the mean droplet diameter and the spec-
trum of the size distribution can be estimated from known dimen-
sions and operating conditions. Besides that, additional informa-
tion relative to the droplet motion, namely flow velocity and flow
angle, are required. Again, an attempt has been made to relate the
Fig. 9 Arithmetic mean droplet diameter versus operating current oil film atomization process to that prevailing at the rim of
conditions a rotating disk. Such a comparison is shown in Fig. 12 for the
V̄ d
⫽1.5⫺0.1327•ln共 We* 0.5兲 . (5)
V rim
As mentioned before, the flow angle is invariant to both oper-
ating conditions and remains constant at
␣ ⫽34°. (6)
Conclusions
The disintegration of oil films emerging from axial passages at
the face of a rotating cylinder is one of the droplet generation
sources inside aeroengine lubrication system components. In order
to establish a database for the development of atomization models,
which are required in advanced design tools, this situation was
studied experimentally and the following conclusions were drawn:
All measurements were carried out for a 5cS liquid, thus in
terms of the fluid properties at relevant engine conditions.
Fig. 11 Standard deviation of droplet diameter distribution
The breakup process shows similarities to those observed at the
rim of a rotating disk. It consists of the separation of a thin oil film
from the outer most part of the inner surface of the rotating pas-
sage and the subsequent disintegration of this film into ligaments
normalized mean velocity, V tot /Vrim , and the flow angle, ␣, re- and droplets.
spectively. Note that the dimensions used in the definition of a The droplet size distribution is driven by the rotational speed.
‘‘rim velocity’’ have to be consistent with those used before, i.e., No impact of oil flow rates has been detected which suggests an
for rotating axial passages, the pitch radius has to replace the rim indifference relative to a variation of oil Reynolds numbers 共based
radius to calculate the reference velocity. Using this scaling on the hydraulic diameter of the axial passage兲. The effect of the
parameter yields identical relationships between droplet velocity passage length was not investigated though it can be expected that
and operating conditions in terms of Weber numbers for both the disintegration results are different for very short passages.
configurations: More data are needed to establish these characteristics.
By introducing nondimensional parameters for the description
of atomization products, a quantitative comparison of the disinte-
gration processes at rotating axial passages with those at the rim
of rotating disks has been enabled. This comparison confirms phe-
nomenological interpretations of the oil film breakup and offers
the opportunity to scale droplet size distributions and flow char-
acteristics to other operating conditions.
Acknowledgments
The authors are grateful to Lou Dobek, Pratt & Whitney, East
Hartford, for his encouragement and support of the work pre-
sented here.
Nomenclature
D ⫽ diameter 共m兲
D 10,20,30 ⫽ averaged droplet diameters based on number, sur-
face area, and volume 共m兲
D V,x ⫽ characteristic droplet diameter, representing the
fraction x of the total droplet volume accumulated
at the specified diameter 共m兲
N ⫽ rotational speed 共rpm兲
n ⫽ number
q ⫽ volumetric flow rate (m3 /s)
R, ␸, Z ⫽ cylindrical coordinates 共m, deg, m兲
s ⫽ standard deviation
T ⫽ temperature 共K兲
V ⫽ velocity 共m/s兲
V ⫹ ⫽ nondim. volumetric flow rate
We* ⫽ modified Weber number
␣ ⫽ flow angle, tan ␣⫽Vrad /V tan 共deg兲
␳ ⫽ density (kg/m3 )
␮ ⫽ dynamic viscosity 共kg/共m•s兲兲
␴ ⫽ surface tension 共N/m兲
␻ ⫽ angular velocity 共rad/s兲
Subscripts
Fig. 12 Spray characteristics compared with those of droplet C ⫽ cylinder
flows emerging from the rim of a rotating disk. „a… Initial droplet D ⫽ disk
velocity. „b… Initial droplet flow angle. d(rop) ⫽ droplet
E ⫽ engine conditions 关6兴 Glahn, A., Busam, S., and Wittig, S., 1997, ‘‘Local and Mean Heat Transfer
L ⫽ liquid Coefficients Along the Internal Housing Walls of Aero Engine Bearing Cham-
bers,’’ ASME Paper No. 97-GT-261.
P ⫽ pitch 关7兴 Zaidi, S. H., Ishaq G., Aroussi A., Azzopardi, B. J., 1998, ‘‘Two-Phase Flow
rad ⫽ radial Study Around a Rotating Liquid Film Using Laser Techniques,’’ Proceedings
T ⫽ test conditions of the VSJ-SPIE98, Dec. 6 –9, Yokohama, Japan.
tan ⫽ tangential 关8兴 Busam, S., Glahn, A., and Wittig, S., 1999, ‘‘Internal Bearing Chamber Wall
tot ⫽ total Heat Transfer as a Function of Operating Conditions and Chamber Geometry,’’
ASME J. Eng. Gas Turbines Power, accepted for publication.
关9兴 Glahn, A., Kurreck, M., Willmann, M., and Wittig, S., 1996, ‘‘Feasibility
References Study on Oil Droplet Flow Investigations Inside Aero Engine Bearing
Chambers—PDPA Techniques in Combination with Numerical Approaches,’’
关1兴 Glahn, A., Busam, S., Blair, M. F., Allard, K. L., and Wittig, S., 2000, ‘‘Drop-
ASME J. Eng. Gas Turbines Power, 118, pp. 749–755.
let Generation by Disintegration of Oil Films at the Rim of a Rotating Disk,’’
关10兴 Glahn, A., Blair, M. F., Allard, K. L., Busam, S., Schäfer, O., and Wittig, S.,
ASME Paper No. 2000-GT-279.
关2兴 Wittig, S., Glahn, A., and Himmelsbach, J., 1994, ‘‘Influence of High Rota- 2001, ‘‘Disintegration of Oil Films Emerging From Radial Holes Inside a
tional Speeds on Heat Transfer and Oil Film Thickness in Aero Engine Bearing Rotating Cylinder,’’ ASME Paper No. 2001-GT-0202.
Chambers,’’ ASME J. Eng. Gas Turbines Power, 116共2兲, pp. 395– 401. 关11兴 Bachalo, W. D., and Houser, M. J., 1984, ‘‘Phase Doppler Spray Analyzer for
关3兴 Chew, J., 1996, ‘‘Analysis of the Oil Film on the Inside Surface of an Aero- Simultaneous Measurements of Drop Size and Velocity Distributions,’’ Opt.
Engine Bearing Chamber Housing,’’ ASME Paper No. 96-GT-300. Eng., 23, pp. 583–590.
关4兴 Glahn, A., and Wittig, S., 1996, ‘‘Two-Phase Air Oil Flow in Aero Engine 关12兴 Willmann, M., Glahn, A., and Wittig, S., 1997, ‘‘Phase-Doppler Particle Sizing
Bearing Chambers—Characterization of Oil Film Flows,’’ ASME J. Eng. Gas with Off-Axis Angles in Alexander’s Darkband,’’ Part. Part. Syst. Charact.,
Turbines Power, 118共3兲, pp. 578 –583. 14共3兲, pp. 122–128.
关5兴 Glahn, A., and Wittig, S., 1999, ‘‘Two-Phase Air Oil Flow in Aero Engine 关13兴 Hinze, J. O., and Milborn, H., 1950, ‘‘Atomization of Liquids by Means of a
Bearing Chambers—Assessment of an Analytical Prediction Method for the Rotating Cup,’’ ASME J. Appl. Mech., 17, pp. 145–153.
Internal Wall Heat Transfer,’’ Int. J. Rotating Mach., 5共3兲, pp. 155–165. 关14兴 Lefebvre, A. H., 1989, Atomization and Sprays, Hemisphere, Washington, DC.
A. Glahn Disintegration of Oil Films
M. F. Blair Emerging From Radial Holes in a
United Technologies Research Center,
411 Silver Lane, M/S 129-19,
East Hartford, CT 06108
Rotating Cylinder
A fundamental study has been performed to examine the disintegration of oil films emerg-
ing from radial holes in a rotating hollow cylinder. The configuration investigated is an
K. L. Allard abstraction of one of the droplet generation sources in an aeroengine bearing compart-
Pratt & Whitney, ment; similar configurations may also occur inside gearboxes. The paper aims to con-
400 Main Street, tribute to both the determination of directly applicable droplet characteristics and the
East Hartford, CT 06108 establishment of a database that can be used for the development of droplet generation
models. Similar to a prior paper on droplet generation processes at the rim of a rotating
disk (Glahn, A. et al., 2000, ‘‘Droplet Generation by Disintegration of Oil Films at the
S. Busam Rim of a Rotating Disk,’’ ASME Paper No. 2000-GT-0279.) the near-term objectives of the
study are (i) to determine droplet sizes under relevant aeroengine bearing compartment
O. Schäfer operating conditions, and (ii) to measure individual droplet diameter/velocity relation-
ships. The long-term objective is to incorporate this information into advanced CFD-
S. Wittig based design tools. Therefore, special emphasis has been directed towards a correlation
of test results that enables determination of boundary conditions for a two-phase (oil
Institut für Thermische Strömungsmaschinen, droplets/air) simulation of lubrication system components. Based on the results of the
Universität Karlsruhe, present paper, droplet flow boundary conditions in terms of mean diameter, standard
Kaiserstr. 12, deviation of the diameter distribution, starting velocity, and flow angle are available for
Karlsruhe 76128, Germany oil droplets generated by disintegration of oil films emerging from rotating radial holes
and rotating disks. 关DOI: 10.1115/1.1586311兴
Introduction Glahn et al. 关9兴 suggested the feasibility of such an approach.

However, a prerequisite for a successful implementation of these
The major development driver for new engines is still the desire
to increase the gas turbine’s efficiency. This goal is accomplished tools into a standard design environment is the availability of
by increases of both the overall pressure ratio and the turbine inlet boundary conditions and the development of submodels which
temperature. Although beneficial for the cycle performance of the describe the complex phase interaction mechanism and atomiza-
engine, the combination of high pressure, high temperature, and tion effects prevailing in these parts. Therefore, further studies of
high rotational speeds of the turbine rotor leads to specific prob- film behavior and atomization processes are required.
lems within the heat management of the engine’s lubrication sys- Glahn et al. 关10兴 report on the first phase of a multiyear, fun-
tem, 关1兴. As a consequence, the design margin for the lubrication damental investigation into generic oil system atomization pro-
system and especially the bearing compartments becomes smaller cesses: specifically the disintegration of oil films at the rim of a
and design tool improvements are required to meet the challenges rotating disk. The present paper is a continuation of this effort and
of current and future engine developments. deals with the disintegration of oil films emerging from radial
In recent years, progress has been made on several subtasks holes in a rotating cylinder. This is a generic representation of one
involved in the overall design process. Wittig et al. 关2兴 report on of the droplet generation sources in an aeroengine bearing com-
the identification of bearing compartment flow patterns. The char- partment, e.g., under-race lubrication oil supply arrangements or
acterization of oil film flows was subject of studies by Chew 关3兴 intershaft oil flow configurations; similar configurations may also
and Glahn and Wittig 关4兴. Glahn et al. 关5兴, Busam et al. 关6兴, and occur inside gearboxes.
Glahn and Wittig 关7兴 investigated the heat transfer along the in- As in the previous study, the objectives were as follows: 共i兲 to
ternal bearing chamber housing walls. Droplet flows in a rotating, identify disintegration modes relevant with respect to aeroengine
annular two-phase flow environment were studied for air/water bearing compartment operating conditions, 共ii兲 to determine drop-
mixtures at low speeds by Zaidi et al. 关8兴. All these studies were let sizes under those operating conditions, and 共iii兲 to measure
helpful in getting an overall picture of the flow phenomena in- individual droplet diameter/velocity relationships.
volved in bearing compartments. They help to establish a database The current experimental paper aims to contribute to both the
for validation purposes and some of them appear to be directly definition of directly usable droplet characteristics and the estab-
applicable for certain aspects of the engine design. However, the
lishment of a database that can be used for model development.
development of a design system and its adaptation to different
Following a brief description of the experimental setup, results are
geometrical and operational boundary conditions requires a plat-
presented in terms of samples from the visualization work, typical
form that is able to integrate these submodels. A viable approach
spatial droplet diameter distributions, characteristic diameters, and
would be to use CFD methods to predict the two-phase air/oil
flow in lubrication system components. A study presented by droplet velocities as a function of flow rate and rotational speed. A
theory is developed that enables comparison of oil film atomiza-
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN tion by rotating radial holes with that of a rotating disk. From this
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF theory, correlations are developed for all atomization products,
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- namely the mean droplet diameter, the standard deviation of the
4 –7, 2001; Paper 2001-GT-0202. Manuscript received by IGTI, December 2000, diameter distribution, the mean droplet velocity, and the droplet
final revision, March 2001. Associate Editor: R. Natole. flow angle.
Fig. 1 Rotating cylinder test rig
Experimental Setup and Techniques diameter of 6•10⫺3 m 共0.236 in.兲. The oil is driven by centrifugal
forces along the inner cylinder wall until it reaches a groove from
Test Rig. The test facility used for the fundamental oil film where it is evenly distributed to six radial holes that are equally
disintegration studies reported in this paper is the same as was spaced around the circumference. As shown in the next section on
used by Glahn et al. 关10兴. However, the rotating disk used in the flow visualization results, oil is discharged through the radial
prior investigation has been replaced by a rotating cylinder. The holes as a thin film that separates from the hole wall and disinte-
rotational speeds and the physical dimensions of the cylinder had grates into droplets. The atomized oil is collected at the stationary,
to be chosen such that two goals were met: 共i兲 to achieve an transparent housing walls. Gravitation forces the oil to flow down-
investigation relevant to bearing compartment operating condi- ward into the oil sump from where it is scavenged back into the
tions and 共ii兲 to aim for a design that guarantees full optical access reservoir.
to the rotating holes. Measurement Techniques. As in the previous investigation
The first goal was met based on a simulation of the relevant focussing on the disintegration of oil films at the rim of a rotating
Weber number which is the parameter dominating the atomization disk 共Glahn et al., 关10兴兲, the experimental objectives of this study
process. By using a test fluid 共Dow Corning 200 Fluid 5cS兲 with required applying techniques for flow visualization as well as la-
the same fluid properties at room temperature as turbine oil at 373 ser Doppler techniques for the droplet sizing and velocimetry.
K, engine Weber numbers are obtained for For the first task, aimed at identifying the oil film breakup
1⬇ 冉冊冉冊
␻T
␻E
2
DC
DS
3
. (1)
mechanisms, stroboscopic illumination along with a simple video
camera was used to obtain a detailed documentation of oil flow
patterns. All flow phenomena were monitored on a screen, re-
Equation 共1兲 suggests that it would be favorable to simulate the corded on videotape, and examined in detail using digital image
Weber number by aiming for a higher cylinder diameter and keep- processing.
ing the rotational speeds at a moderate level. For this investiga- For quantitative measurements of spray characteristics, a single
tion, a maximum rotational speed of N T ⫽10,000 rpm and a cyl- particle counter, Aerometrics’ PDPA, was used. In recent years,
inder diameter of D C ⫽0.13 m 共5.12 in.兲 have been chosen. These the PDPA has become a standard technique for atomization stud-
values translate into an engine condition of N E ⫽16,000 rpm and a ies. Thus, a lengthy description of the measuring principle is
shaft diameter of D S ⫽0.10 m 共3.94 in.兲 which are expected to be avoided here and the interested reader is referred to the specific
typical values for current aeroengine designs. literature 共e.g., Bachalo and Houser 关11兴兲. However, the adapta-
The cylinder diameter of D C ⫽0.13 m in combination with a tion of this measuring technique to the current test is special be-
relatively short cylinder length of about L C ⫽0.096 m 共2.76 in.兲 cause it used a backscatter arrangement at an off-axis angle of 130
also supports the second design requirement by enabling a canti- deg, which became available for measurements due to a relatively
lever arrangement and, therefore, full optical access from the non- high index of refraction for oil that is greater than 1.4. A compre-
drive side. As shown in Fig. 1, the cylinder is located in a station- hensive analysis of this type of measurement configuration is
ary housing with an inner diameter of 0.301 m 共11.85 in.兲. The given by Willmann et al. 关12兴.
housing consists of an aluminum frame supporting a transparent
thermoplastic that allows optical access for flow visualization Test Matrix for Droplet Sizing and Velocimetry. Measure-
studies throughout the whole circumference. The shaft is posi- ments for the characterization of sprays generated by the disinte-
tioned by spindle bearings and driven by an electro-motor in com- gration of oil films emerging from rotating radial holes were car-
bination with a flexible coupling and a flat belt. ried out for two configurations and 224 different combinations of
Oil is pumped out of a reservoir through an electrical oil heater operating condition and measurement location. As indicated in
(T max⫽473 K for ṁ⫽0.1 kg/s) into the rig. It is fed along the Fig. 2共a,b兲, variations of operating conditions included three dif-
centerline onto the spinning disk face using a tube with an inner ferent rotational speeds and four different flow rates. Measure-
Fig. 2 Test matrix for droplet size and velocity measurement. „a… Configuration
1A: D H Ä3"10À3 m, „ L Õ D … H Ä1.67. „b… Configuration 1B: D H Ä9"10À3 m, „ L Õ D … H
Ä0.56. „c… Nomenclature.
ments of spatial spray characteristics were designed to obtain The qualitative flow investigation strongly suggests that the
axial and radial droplet size and velocity distributions and as- processes driving the atomization of oil films at rotating radial
sumed a 共time-averaged兲 axisymmetric flow pattern. The circum- holes and rotating disks are basically the same. This information
ferential measurement location and the nomenclature used in the was used to derive a nondimensional description of disintegration
subsequent discussion of test results are introduced in Fig. 2共c兲. products and their subsequent correlation with operating condi-
Note the use of a right-hand side cylindrical coordinate system tions. However, some typical results will be discussed first in
that is aligned with the centerline of the rotor (Z-axis兲 and the
centerline of the radial hole (R-axis兲, respectively.
Experimental Results
Flow Visualization. Flow visualization studies were per-
formed in order to identify the fundamental oil film break-up
mechanisms to be considered in the modeling and analysis of
quantitative test results. Figure 3 shows typical results gained
from the flow visualization. The picture in the left-hand side 共Fig.
3共a兲兲 shows in some detail that a thin continuous oil film separates
from the inner surface of the rotating radial hole (D H ⫽9
•10⫺3 m). This separation process occurs along the whole cir-
cumference of the hole. The thin film extends radially outward
from the rim of the hole 共initially forming a hollow fluid cylinder兲
until equilibrium is reached between the surface tension force at
the free edge and the kinetic energy of the liquid sheet. The oil
film thickens at the edges of the sheet and then disintegrates into
ligaments and droplets 共Fig. 3共b兲兲. This process is very similar to
the disintegration of oil films at the rim of a rotating disk, Fig. 3 Oil atomization at a rotating radial hole „ D H Ä9
关10,13,14兴. "10À3 m…; „a… sheet separation, „b… droplet trajectory
⫽10•10⫺3 m from the rim of the rotating cylinder. Measurements
were acquired along an axial traverse ranging from about ⫺0.07
⬍Z/D C ⬍0.05. The axial location of the radial holes is indicated
by vertical lines at R H /D⫽⫾0.0115. Results are presented in
terms of characteristic diameters, mean velocities, and velocity
distributions.
Figure 4共a兲 shows the arithmetic mean droplet diameter (D 10),
the area-averaged droplet diameter (D 20), the volume-averaged
droplet diameter (D 30), and a combination of the latter two, the
so-called Sauter mean diameter (D 32), which represents a droplet
with the same volume-to-surface ratio as the whole spray. The
shape of the curves reflects the jet-like flow of oil into the annulus
between rotating cylinder and stationary housing. Larger droplets
are concentrated in a region more or less aligned with the center-
line of the radial holes. Only smaller droplets separate from this
flow path and penetrate into regions beyond ⫾Z/D C ⫽0.02. Arith-
metic mean diameters in the center of the distribution are approxi-
mately D 10⫽75 ␮ m, whereas Sauter mean diameters are about
twice as large.
Similar conclusions relative to the droplet propagation can be
drawn from another set of characteristic diameters that is shown in
Fig. 4共b兲. Diameters D v ,x that define the diameter at which the
fraction x of the total droplet volume has been accumulated are
plotted against the axial coordinate. In the center of the droplet
spray (Z/D C ⫽0), a value of D V,0.1 ⫽75 ␮ m was obtained. Thus,
a conclusion that can be drawn from the combined results of Fig.
4共a兲 and 4共b兲 is that the average diameter (D 10) of the spray
contributes only 10% to the overall volume. A much better over-
view on the volume distribution can be obtained from a combina-
tion of three cumulative diameters. However, cumulative diam-
eters are sensitive to the correct and statistically relevant
acquisition of a few large droplets at the upper limit of the mea-
surement range. In the present study, a minimum of 10,000 valid
samples was acquired for each of the 224 measurements. In addi-
tion to that, a huge number of exploratory measurements were
conducted in order to find the best suitable combination of instru-
ment settings and to avoid conditioned sampling.
Mean radial and tangential velocities are shown in Fig. 4共c兲.
The velocity distributions along the axial traverse support the phe-
nomenological explanation of droplet distributions given above.
Only low inertia droplets outside of the core of the droplet spray
can be decelerated significantly by the surrounding air. Note that
the ⫾1␴ velocities for both components are about ⫾40% – 50%
of the mean velocities. This rather broad distribution of velocities
reflects the fact that they represent the whole spray, i.e., large
variations in droplet size and, therefore, droplet Reynolds number.
Analogous to the presentation of axial profiles of diameters and
velocities, Fig. 5 shows data for a radial traverse measured at the
location Z/D C ⫽0, i.e., aligned with the radial holes. It is interest-
ing to note that values for D 10 and D V,0.1 remain approximately
constant for radial distances greater than ⌬R/D C ⫽0.05. All other
characteristic diameters display the ongoing atomization process
of larger oil fragments by showing continuously decreasing drop-
let sizes up to a distance of about ⌬R/D C ⫽0.18 at which the
atomization process is completed. For the subsequent analysis and
correlation of test results, only those data were considered for
which this ‘‘completed atomization’’ condition was reached. The
velocity distributions 共Fig. 5共c兲兲 show the expected transition
from a tangentially orientated flow 共in the stationary frame of
Fig. 4 N Ä3450 rpm, q Ä1.39"10À5 m3 Õs, L Õ D H Ä1.67, ⌬ R Õ D C reference兲 to a more and more radially oriented flow vector. The
Ä0.077. Axial profiles of „a… mean diameters, „b… cumulative
spread of measured velocities (⫾1 ␴ ) was approximately the
diameters, „c… mean velocities and velocity range.
same as for the axial distribution and does not change significantly
with increasing radii.
terms of actual dimensional quantities, which are often more suit- Size Class Resolved Droplet Velocity Distributions. The
able than nondimensional quantities to develop an initial under- previous discussion of typical traverse data included mean and
standing of the droplet flow phenomena. velocity range information useful for an overall understanding of
the flow situation. However, these velocities represent the whole
Typical Droplet Sizes and Velocities. Figure 4 summarizes spray and are affected not only by operating conditions such as
results obtained for configuration 1A at a radial distance of ⌬R rotational speed and flow rate but also by the atomization quality,
Fig. 6 N Ä3450 rpm, L Õ D H Ä1.67, Z Õ D C Ä0.0. Effect of flow rate
on „a… droplet velocities based on rim speed, „b… flow angles.
figures contain droplet velocities based on the rim speed of the

rotating cylinder and flow angles relative to the tangential direc-
tion 共Fig. 2共c兲兲.
In Fig. 6, the effect of the oil flow rate on the droplet velocity
and the flow angle is shown for a droplet diameter class of
D drop⫽63 ␮ m along a radial traverse. The overall impact of the
flow rate on both parameters is small. The droplet velocity de-
creases with increasing distance from the cylinder until it reaches
a constant level of about V tot /Vrim⫽0.5. For the lowest flow rate,
however, the velocity distribution seems to have a slightly steeper
slope that those measured for higher flow rates. Similar observa-
tions were made for other size classes.
A possible reason for this difference in the flow behavior is the
slower disintegration process that has been observed for these
lower flow rates. The droplet flow angle increases from an almost
tangential orientation to values slightly above 30 deg 共at ⌬R/D C
⫽0.18). No dependence on the oil flow rate has been observed.
In contrast to the relatively weak dependence on the oil flow
rate displayed in Fig. 6, rotational speed has a substantial impact
Fig. 5 N Ä3450 rpm, q Ä1.39"10À5 m3 Õs, L Õ D H Ä1.67, Z Õ D C on both droplet velocity and flow angle. Figures 7共a–c兲 show the
Ä0.0. Radial profiles of „a… mean diameters, „b… cumulative di- results for droplets with D drop⫽40 ␮ m for three speed conditions
ameters, „c… Mean velocities and velocity range. and a constant flow rate of q⫽5.44•10⫺5 m3 /s. Note that the
distances (⌬R/D C ) from the cylinder to the first data point for
Fig. 7 are inversely dependent on the rotational speed. These ‘‘in-
nermost’’ data points indicate the location where the first statisti-
i.e., the diameter distribution. An analysis of cause/effect relation- cally meaningful 40 ␮m droplet data could be obtained. This re-
ships on the droplet flow has to be based on individual size-class sult indicates that the film breakup/atomization process depends
resolved velocities that enable a better separation of droplet gen- on the rotor speed. While Fig. 7共a兲 demonstrates that the absolute
eration mechanisms from droplet propagation phenomena. velocity of the droplets increases directly with the rotational
Samples from such an analysis are given in Figs. 6 and 7. Both speed, Figs. 7共a兲 and 7共b兲 show that the radial decrease of the
Fig. 8 Deceleration of droplets at various rotational speeds
mines the initial droplet velocity兲 but depends on the absolute

velocity magnitude. The stronger deceleration at higher cylinder
speeds has been visualized in Fig. 8 by transferring the measured
spatial velocity distributions into the time domain. The time in-
crement has been calculated from the radial location and the mea-
sured radial velocity component.
As noted above, the data indicate that the droplet generation
due to oil film disintegration occurs much more rapidly at higher
rotor speeds. The film breakup process is driven by the ratio of
destabilizing aerodynamic forces to surface tension forces, thus
depending on some function of the Weber number which is pro-
portional to the square of the relative velocity. Therefore, the ve-
locity distributions shown in Figs. 7共a,b兲 are a result of superim-
posed effects of stronger deceleration and earlier onset of
deceleration. The latter is a result of an earlier completion of the
atomization process for higher speeds which enables the acquisi-
tion of relatively small droplets 共40 ␮m diameter兲 at smaller radii.
The relative magnitude of these effects could be determined based
on further studies with independent simulations of droplet Weber
number and Reynolds numbers, respectively.
Another consequence of different time scales involved in the
disintegration process is that of significantly different flow angles
共Fig. 7共c兲兲. Close to the rotating surface, droplet flow angles show
a steeper angle for higher rotational speeds but also a more rapid
transition into a constant flow angle characteristic. That is consis-
tent with a faster breakup process as discussed above.
Impact of Hole Diameter, Rotational Speed, and Flow Rate

on the Spray Characteristic. Data sets have been acquired
from two rotating hole geometries in the present study. Further-
more, similarities to the disintegration processes at the rim of a
Fig. 7 q Ä5.44"10À5 m3 Õs, L Õ D H Ä0.56, Z Õ D C Ä0.0. Effect of ro- rotating disk have been identified that suggest an incorporation of
tational speed on „a… absolute droplet velocities, „b… droplet ve- this kind of data into the subsequent data analysis. In order to
locities based on rim speed, „c… flow angles compare the three different configurations, nondimensional pa-
rameters are required. The previous study by Glahn et al. 关10兴
adopted Hinze and Milborn’s 关13兴 suggestion of using a modified
Weber number
droplet velocity is also highly speed dependent. For the analysis
of these flow conditions it is helpful to examine the equation of 1 ␳ L␻ 2D D
3
motion for a single liquid particle in air: We* ⫽ (3)

8 ␴
dVd 18• ␮ G and a nondimensional flow rate
⫽ 共 V ⫺Vd 兲 . (2)
dt ␳ d •D 2d G
␳ Lq 2
Note that Eq. 共2兲 is valid for small droplet Reynolds numbers, V ⫹⫽ (4)
␴ •D D3
where the coefficient of drag can be approximated by C W
⫽24/ReD . This condition is met for the droplet size and velocity for the characterization of disintegration processes and atomiza-
range considered in Fig. 7. It can readily be concluded from the tion products. Though it was not introduced as such, it is interest-
equation of motion that the deceleration of a liquid droplet of ing to note that the nondimensional flow rate as defined in Eq. 共4兲
given size does not scale with the characteristic velocity of its is basically the ratio of radial inertia to surface tension forces, thus
generation mechanism 共Fig. 7共a兲 shows that the rotor speed deter- a radial Weber number. Recognizing this physical interpretation of
Fig. 9 Arithmetic mean droplet diameter versus nondimen-
sional volumetric flow rate Fig. 10 Mean diameter based on oil film thickness at the rim of
the holeÕdisk
the nondimensional flow rate, it is useful to rewrite Eq. 共4兲 to

account for the specifics of the rotating hole configurations: h̄ F ⫽ 冉 6•q• ␯ L
␻ •D D,C •n• ␲ •D D,H
2 冊 1/3
. (6)
␳ Lq 2 Note that different diameters have to be used in Eq. 共6兲 if it is

V ⫹⫽ 3 , (5) desired to compute the oil film thickness at the rim of a rotating
n 2 • ␴ •D H disk (D D ) or the oil film thickness at the exit of a rotating hole
(D C and D H ).
with n as the number of holes with diameter D H . Note that Eqs. In Fig. 10, measured mean droplet diameters D 10 have been
共4兲 and 共5兲 are identical if the rim of the rotating disk is treated in related to film thicknesses calculated from Eq. 共6兲 and plotted for
terms of the flow rate impact as a single hole (n⫽1) with a various rotational speeds as a function of non-dimensional flow
diameter D H ⫽D D . rates V ⫹ . It is most interesting to note that all data acquired for
In Fig. 9, arithmetic mean diameters of sprays generated by the the test case of a rotating cylinder with radial holes seem to group
three different configurations are compared for various flow rates around a value of unity, i.e., droplet and film dimensions are ap-
and two rotational speeds. V ⫹ as defined in Eq. 共5兲 appears to be proximately the same. More accurate information may be ob-
a proper scaling parameter for this comparison though the impact tained by use of the following equation:
of the flow rate on the diameter seems to be relatively small. In
contrast to the weak dependence on flow rate, rotational speeds D 10
⫽1.25• 共 V ⫹ 兲 ⫺0.15. (7)
significantly influence the atomization quality. This is highlighted h̄ F
by lines of constant speed 共note that We* 0.5⬃ ␻ ) which demon-
strate an almost linear relationship between speed increase and Equation 共7兲 represents data for engine relevant variations of
diameter reduction. Both observations relative to the impact of flow rate and speed and predicts mean diameters within a rela-
operating parameters are consistent with conclusions drawn from tively narrow scatterband of ⫾15%. The same equation also gives
the rotating disk studies by Glahn et al. 关10兴. a reasonable estimate for the size of droplets generated for low
speed conditions at the rim of a rotating disk (We* ⫽1073). For
Correlation of Test Results. Employing the type of data re- higher disk speeds, which are more applicable to engine situa-
duction shown in Fig. 9, the test results could have been expressed tions, ratios of droplet diameter to film thickness also approach a
as a function of some power law representation of V ⫹ and We* . value of 1.
Such an approach would lead to correlations that could predict the As seen before in Figs. 4 and 5 and in the previous study by
mean diameter, D 10 , and the Sauter mean diameter, D 32 , within a Glahn et al. 关10兴, the sprays considered here are nonhomoge-
confidence range of about 30% and 50%, respectively. Consider- neous, i.e., the spray has to be regarded as a spectrum of drop
ing the complexity of the flow problem, this would already be a sizes distributed about a mean value (D 10). Thus, in addition to
satisfactory result. However, in the current study, an attempt has the mean diameter, another parameter is needed in order to be able
been made to relate the atomization product more closely to its to calculate the droplet size distribution.
generation mechanisms, i.e., to the oil film flow conditions at the The data obtained in the present study can be approximated best
atomizer edge. It is hoped that this approach will further enhance by use of a log-normal distribution of droplet sizes, 关14兴:
冋册
the understanding of the disintegration process, will lead to a
broader applicability of subsequent model formulations, and will dn d 1 1
⫽ exp⫺ 共 ln D d ⫺ln D 10兲 2 , (8)
result in a higher confidence level towards these models. dD d 冑2 ␲ D d s g 2s 2g
It has been shown by flow visualization that the droplet spray in
rotating disk and rotating cylinder configurations is generated by where s g is the geometric standard deviation and n d is the number
disintegration of thin oil films and, therefore, it is instructive to of droplets per size class D d . It has also been recognized that the
compare the size of the droplets with the thickness of the oil sheet standard deviation is not only affected by the flow rate but also by
prior to its break up. the hole diameter. The latter has an impact because a broader oil
For the calculation of film conditions at the rim of a rotating column at the exit of the rotating hole 共Fig. 3兲 has the tendency to
disk or at the exit plane of a rotating radial hole, a simple model shield oil film fragments at its leeward side from high aerody-
has been derived that is explained in some detail in the Appendix. namic forces. Although the majority of droplets eventually reach
According to this model, the oil film thickness can be calculated comparable size 共Fig. 10兲, the larger hole and especially the rotat-
from ing disk with its continuous oil film maintain a small but recog-
Fig. 11 Standard deviation of droplet diameter distribution Fig. 13 Droplet flow angle after completed disintegration
nizable number of larger droplets which influence the standard V̄ d

deviation. This circumstance was considered in Fig. 11 by refer- ⫽1.5⫺0.1327•ln共 We* 0.5兲 . (10)
V rim
ring the measured standard deviation to the nondimensional hole
diameter, D H /D C . Note that for the case of the rotating disk D H No significant impact of the flow rate on the average droplet ve-
and D C become equal to the disk diameter D D . locity has been observed and Eq. 共10兲 appears to be applicable for
For the relationship between standard deviation, hole diameter, all flow rates relevant to engine designs.
and flow rate, the following simple equation has been found to As the final quantity required for a characterization of oil film
represent measured data within about ⫾25%: disintegration processes in rotating disk and rotating cylinder con-
figurations, the droplet flow angle has to be defined in terms of
DH operating conditions. This has been done in Fig. 13 for same test
s g ⫽2.91• 共 V ⫹ 兲 0.5• . (9)
DC conditions that have been considered in Fig. 12 for the droplet
By use of Eqs. 共6兲 to 共9兲, the mean droplet diameter and the velocity. The flow angle, however, appears to be only weakly
spectrum of the size distribution can be calculated from known dependent on variations in either flow rate or rotational speed. In
dimensions and operating conditions. Besides that, additional in- fact, for intermediate and high rotational speeds the flow angle is
formation relative to the droplet dynamics, namely flow velocity represented best by a constant value of
and flow angle, are required. Again, an attempt has been made to ␣ ⫽30 deg, We* ⬎2•106 . (11)
relate droplet to oil film flow characteristics. However, a compari-
son of both oil film velocity components shows that the tangential For the rotating cylinder with radial holes at low angular veloci-
velocity, i.e., the rim speed of the disk or the cylinder, is about one ties,
to two orders of magnitude larger than the radial oil film velocity. ␣ ⫽34•log共 We* 0.5兲 ⫺78, We* ⬍2•106 , (12)
Therefore, the droplet’s average total velocity has been related
simply to the rim speed and the result of this comparison is shown describes the relationship between flow angle and shaft speeds.
in Fig. 12 as a function of Weber numbers, We* . Almost identical
characteristics for the three different configurations have been ob- Conclusions
served though the high-speed datapoint for the smaller of the two The disintegration of oil films emerging from radial holes inside
radial holes deviates from the overall trend by roughly 20%. The a rotating cylinder is one of the droplet generation sources inside
relationship between droplet velocity and operating conditions in aeroengine lubrication system components. In order to establish a
terms of Weber numbers is represented by database for the development of atomization models, which are
required in advanced design tools, this situation was studied ex-
perimentally and the following conclusions were drawn:
All measurements were carried out for a 5cS liquid, thus in
terms of the fluid properties at relevant engine conditions.
The breakup process is basically the same as observed at the
rim of a rotating disk and consists of the separation of a thin oil
film from the inner surface of the rotating hole and the subsequent
disintegration of this film into ligaments and droplets.
The atomization quality, i.e., the droplet size distribution, is
mainly driven by the rotational speed. The impact of oil flow rates
is much weaker.
A dimensional analysis confirmed similarities between film dis-
integration processes at rotating cylinders and rotating disks. A
nondimensional flow rate and a modified Weber number were
identified as suitable scaling parameters for a comparison of dif-
ferent geometries.
A comparison of mean droplet sizes and oil film flow conditions
prior to separation and disintegration shows that averaged droplet
diameters and film thicknesses are approximately the same, thus
Fig. 12 Droplet velocity after completed disintegration D 10 /h̄ F ⬇1 appears to be a reasonable estimate for the average
diameter. A simple model for the calculation of oil film properties the oil film, the radial velocity profile for the oil film along the
was developed and included as an Appendix to this paper. rotating surface can be written in wall coordinates as
Within the limited range of geometrical variations investigated,
results obtained from dimensional analysis and comparisons with d v F⫹ 1⫺ ␴ CF •y F⫹
⫽ , (A1)
the simple film model suggest that the ratio of hole diameter to dy F⫹ 1⫹ ␯ t / ␯ F
hole length does not have a significant effect on the disintegration
with ␴ CF as a parameter that describes the impact of centrifugal
products.
forces versus frictional forces on the film profile
Based on the results of the present paper, droplet flow boundary
conditions in terms of mean diameter, standard deviation of the ␻ 2 r• ␯ F
diameter distribution, starting velocity, and flow angle are avail- ␴ CF ⫽ . (A2)
v 3␶
able for oil droplets generated by disintegration of oil films
emerging from rotating radial holes and rotating disks. This approach is an analogy to film flow calculations that have
been verified by Glahn and Wittig 关4,7兴 for oil film flows in the
Acknowledgments stationary frame of reference. In those situations the gravitational
acceleration, g, has been used in Eq. 共A2兲 instead of the centrifu-
The authors are grateful to Lou Dobek, Pratt & Whitney, East gal acceleration, ␻ 2 r.
Hartford, for his encouragement and support of the work pre- Assuming that the oil film is thin in comparison with either the
sented here. hole diameter or the disk diameter, a film Reynolds number can be
calculated from
Nomenclature
q
D ⫽ diameter 共m兲 ReF ⫽ ⫽ v̄ F⫹ •h̄ F⫹ . (A3)
D 10,20,30 ⫽ averaged droplet diameters based on number, sur- n• ␲ •D H • ␯ F
face area, and volume 共m兲 Integration of Eq. 共A1兲 and substitution into Eq. 共A3兲 requires
D V,x ⫽ characteristic droplet diameter, representing the an assumption for the eddy viscosity term, ␯ t / ␯ F . In the present
fraction x of the total droplet volume accumulated study, Glahn and Wittig’s 关4,7兴 proposal of a turbulent velocity
at the specified diameter 共m兲 profile as well as a laminar approach ( ␯ t / ␯ F ⫽0) have been used.
h F ⫽ oil film thickness 共m兲 However, differences in calculated film thicknesses and velocities
N ⫽ rotational speed 共rpm兲 were negligible because the h̄ F⫹ values of the film flows consid-
n ⫽ number ered here were usually less than 12–14. Thus, for the sake of
q ⫽ volumetric flow rate (m3 /s) simplicity, a laminar oil film profile has been assumed in the
R, ␸ ,Z ⫽ cylindrical coordinates 共m, deg, m兲 following.
⌬R ⫽ radial distance from rim 共m兲 The assumption of no interfacial shear translates into
T ⫽ temperature 共K兲
V ⫽ velocity 共m/s兲 1
V ⫹ ⫽ nondim. volumetric flow rate h̄ F⫹ ⫽ . (A4)
␴ CF
v ⫽ radial oil film velocity 共m/s兲
We* ⫽ modified Weber number With Eq. 共A4兲, the integration of 共A1兲 and substitution of v̄ F⫹
␣ ⫽ flow angle, tan ␣⫽Vrad /V tan 共deg兲 into Eq. 共A3兲 readily leads to
␳ ⫽ density (kg/m3 ) q 1
␮ ⫽ dynamic viscosity (kg/(m•s)) ⫽ (A5)
n• ␲ •D H • ␯ F 3• ␴ CF
2
␯ ⫽ kinematic viscosity (m2 /s)
␴ ⫽ surface tension 共N/m兲 which can be rearranged to
冉冊
␻ ⫽ angular velocity 共rad/s兲
6•q• ␯ L 1/3
Subscripts h̄ F ⫽ (A6)
␻ •D D,C •n• ␲ •D D,H
2
C ⫽ Cylinder
D ⫽ Disk and
d(rop)
E
F
⫽
⫽
⫽
Droplet
Engine conditions
Film
v̄ F ⫽ 冉 q 2 • ␻ 2 •D D,C
6•n • ␲2 2
•D D,H
2
•␯F 冊 1/3
. (A7)
G ⫽ Gas Note that for the rotating disk, Eq. 共A7兲 can further be simpli-
H ⫽ Hole fied to
冉冊
L ⫽ liquid
max ⫽ maximum value ␳ F •q 2 • ␻ 2 1/3
v̄ F ⫽ (A8)
rad ⫽ radial 6• ␲ 2 •D D • ␮ F
S ⫽ shaft
which is identical to Hinze and Milborn’s 关13兴 result in the event
T ⫽ test conditions
of ␸ ⫽90 deg, i.e., when the equation derived for a rotating cup
tan ⫽ tangential
with opening half-angle ␸ is applied to the test case of a rotating
tot ⫽ total
disk.
Appendix
References
The following assumptions and models have been used to cal-
culate oil film flow properties, i.e., film thickness and film veloc- 关1兴 Zimmermann, H., Kammerer, A., Fischer, R., and Rebhahn, D., 1991. ‘‘Two-
Phase Flow Correlations in Air/Oil Systems of Aero Engines,’’ ASME Paper
ity, at the rim of a rotating disk and a radial hole in a rotating No. 91-GT-54.
cylinder, respectively. 关2兴 Wittig, S., Glahn, A., and Himmelsbach, J., 1994, ‘‘Influence of High Rota-
Considering a rotating frame of reference, axisymmetric flow tional Speeds on Heat Transfer and Oil Film Thickness in Aero Engine Bearing
Chambers,’’ ASME J. Eng. Gas Turbines Power, 116共2兲, pp. 395– 401.
conditions ( ␻ 2 rⰇg), assuming no interfacial shear stress be- 关3兴 Chew, J., 1996, ‘‘Analysis of the Oil Film on the Inside Surface of an Aero-
tween the co-current gas stream and the film surface, and neglect- Engine Bearing Chamber Housing,’’ ASME Paper No. 96-GT-300.
ing tangential relative motion between the hole/disk surface and 关4兴 Glahn, A., and Wittig, S., 1996, ‘‘Two-Phase Air Oil Flow in Aero Engine
Bearing Chambers—Characterization of Oil Film Flows,’’ ASME J. Eng. Gas Study on Oil Droplet Flow Investigations Inside Aero Engine Bearing
Turbines Power, 118共3兲, pp. 578 –583. Chambers—PDPA Techniques in Combination with Numerical Approaches,’’
关5兴 Glahn, A., Busam, S., and Wittig, S., 1997, ‘‘Local and Mean Heat Transfer ASME J. Eng. Gas Turbines Power, 118共4兲, pp. 749–755.
Coefficients along the Internal Housing Walls of Aero Engine Bearing Cham- 关10兴 Glahn, A., Busam, S., Blair, M. F., Allard, K. L., and Wittig, S., 2000, ‘‘Drop-
bers,’’ ASME Paper No. 97-GT-261. let Generation by Disintegration of Oil Films at the Rim of a Rotating Disk,’’
关6兴 Busam, S., Glahn, A., and Wittig, S., 1999, ‘‘Internal Bearing Chamber Wall ASME Paper No. 2000-GT-279.
Heat Transfer as a Function of Operating Conditions and Chamber Geometry,’’
关11兴 Bachalo, W. D., and Houser, M. J., 1984, ‘‘Phase Doppler Spray Analyzer for
ASME Journal of Engineering for Gas Turbines and Power, accepted for pub-
Simultaneous Measurements of Drop Size and Velocity Distributions,’’ Opt.
lication.
关7兴 Glahn, A., and Wittig, S., 1999, ‘‘Two-Phase Air Oil Flow in Aero Engine Eng., 23, pp. 583–590.
Bearing Chambers—Assessment of an Analytical Prediction Method for the 关12兴 Willmann, M., Glahn, A., and Wittig, S., 1997, ‘‘Phase-Doppler Particle Sizing
Internal Wall Heat Transfer,’’ Int. J. Rotating Mach., 5共3兲, pp. 155–165. with Off-Axis Angles in Alexander’s Darkband,’’ Part. Part. Syst. Charact.,
关8兴 Zaidi, S. H., Ishaq G., Aroussi A., and Azzopardi, B. J. 1998. ‘‘Two-Phase 14共3兲, pp. 122–128.
Flow Study Around a Rotating Liquid Film Using Laser Techniques,’’ Pro- 关13兴 Hinze, J. O., and Milborn, H., 1950, ‘‘Atomization of Liquids by Means of a
ceedings of the VSJ-SPIE98, Dec. 6 –9, Yokohama, Japan. Rotating Cup,’’ Journal of Applied Mechanics, pp. 145–153.
关9兴 Glahn, A., Kurreck, M., Willmann, M., and Wittig, S., 1996, ‘‘Feasibility 关14兴 Lefebvre, A. H., 1989, Atomization and Sprays, Hemisphere, Washington, DC.
Mechanical Surface Treatment
Technologies for Gas Turbine
W. Zhuang
Engine Components
B. Wicks Various mechanical surface treatment technologies have been invented and developed to
induce a protective layer of compressive residual stress at the surface of gas turbine
Air Vehicle Division, engine components where the operating loads are tensile dominated. The benefits of these
Aeronautical and Maritime Research Laboratory, surface treatments are to prevent crack initiation, to retard propagation of small cracks,
Defence Science and Technology Organization, and even to resist corrosion and wear damage. In this paper the literature on these
Melbourne, VIC 3207, Australia technologies and their effects on component reliability and durability is reviewed, with an
emphasis on shot peening (SP), laser shock peening (LSP), and low plasticity burnishing
(LPB). The relative advantages of these three surface treatment technologies, as well as
their limitations, are identified and evaluated. The most important issues for these three
technologies, and the relative merits of the resultant residual stress fields, are presented
and discussed with a view to enhancing gas turbine engine component operating
lives. 关DOI: 10.1115/1.1610011兴
Introduction cold work at the surface that not only can increase fatigue initia-
tion life but also can decrease fatigue crack growth rates, although
For gas turbine engines, fatigue and corrosion damage, often
they have very different features and capabilities that will be pre-
initiated from the surfaces of components, represents a threat to
sented and discussed in this paper.
the safety and availability of aircraft 关1,2兴. The inspection require- It is worth noting that there are other kinds of surface treatment
ments for fatigue and corrosion damage, the maintenance to re- technologies under research and development. It is impossible for
place critical components, and the rework required to remove fa- this paper to exhaustively review and cover all of these technolo-
tigue and corrosion damage considerably increase operating costs. gies, and this paper highlights only the mechanical type of surface
The down-time required to perform these inspections and repairs treatment technologies and their applications to gas turbine en-
significantly impacts on aircraft airworthiness and readiness. gines. The technical features and induced residual stress fields will
Various surface treatment and modification technologies have be presented and discussed and the paper will conclude with a
been developed to induce a protective layer at the surface of gas statement of the future research requirements for this area, includ-
turbine engine components to enhance wear, corrosion and fatigue ing the generation and relaxation of residual stresses, and the
properties. Some of these technologies, such as plating, coating, likely implications for gas turbine engine life assessment, life ex-
ion implantation, heat treatment for surface hardening, and shot tension, and repair.
peening are well established and routinely applied to engine com-
ponents in manufacturing and maintenance, 关3–7兴. However, Shot Peening „SP…
these technologies have a number of limitations. For instance,
shot peening 共SP兲 induces only a shallow compressive residual The Basic Shot Peening Process. The shot peening process
stress layer 共about 0.2 mm deep兲 at the surface, with a large de- is a traditional and inexpensive mechanical surface treatment
gree of cold work 共strain hardening兲 that can cause damage to the method, and it is currently applied to many aircraft structures and
microstructure of materials and promote accelerated relaxation of engine components. The process is shown schematically in Fig. 1.
the beneficial compressive residual stresses, 关8兴. In this regard, Hard shots 共metallic shots or glass or ceramic beads兲 in an air-
although shot peening is routinely applied to engine components, blast repeatedly bombard the surface of the workpiece resulting in
engine manufacturers are unable to quantify the benefits and local plastic surface deformation. Since the plastically deformed
hence cannot allow the operators to derive any benefit in the form surface layer tends to stretch/expand and the adjacent layer of
of extended component replacement intervals. In particular there elastic material beneath restrains the expansion there is a com-
is a lack of confidence in the development and sustainment of the pressive residual stress field generated in the surface layer bal-
residual stresses over long periods of service, 关7兴. anced by subsurface tensile residual stresses.
These shortcomings have lead to the development of advanced A simple Heyn spring model can be used to describe the bal-
mechanical surface treatment technologies, such as laser shock anced residual stress field in the depth direction of a component
peening 共LSP兲 and low plasticity burnishing 共LPB兲. LSP was first produced by the shot peening process, 关12兴, as shown in Fig. 2,
developed in early 1970s as a fatigue enhancement surface treat- based on the principle of local plastic deformation. Two compres-
ment for metallic materials, 关9,10兴. Recently, based on the prin- sive surface springs are balanced by five tensile springs in the
ciple of roller and ball burnishing used originally for refinement of core. The model shows that there are basically two features for the
surface finishes, a new concept—low plasticity burnishing—was residual stresses induced by shot peening. The first feature is a
patented by Lambda Research in Cincinnati, Ohio, 关11兴. The pub- high level of compressive residual stresses at the surface. The
lished data have demonstrated that both LSP and LPB are able to second is a low level of tensile residual stresses over a large
introduce large and deep compressive residual stresses with low section in the subsurface region to maintain the component in
equilibrium.
Contributed by the Manufacturing Materials and Metallurgy Division of THE The Development of Shot Peening. The development of the
AMERICAN SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME
JOURNAL OF ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received
shot peening process has taken place in three stages, 关13兴. The first
by the Manufacturing Materials and Metallurgy Division Nov. 2001; final revision stage involved manual control with the quality of the SP process
received Feb. 2003. IGTI Chair: S. Gollahalli. being heavily dependent on the skill of the operator. The second
men intensity can be easily measured using an Almen gauge
which is commercially available. From the Almen intensity the
approximate depth of compressive residual stress at the surface
may be calculated. However, there is a possibility that shot peen-
ing with different shot sizes may generate the same Almen inten-
sity, but induce different residual stress distributions at the sur-
face, 关8兴. Further the steel selected for Almen strip is heat treated
so that it neither work hardens nor softens. Differences in yield
strength and work hardening behavior of actual components will
result in quite different stress distributions that are determined by
destructive residual stress measurements by either mechanical or
X-ray diffraction techniques.
The coverage is related to the amount of surface area treated by
shot peening. Complete coverage means that the surface area is
impacted at least once and totally by shot particles. The com-
monly used specification of 200% coverage implies that the area
is impacted for twice the time required for 100% coverage. The
definition of the coverage is obviously arbitrary since there is no
such way to control the shots impacting on the area exactly once
or twice without overlapping, 关15兴. The coverage may be esti-
mated using the Peenscan method which uses the color change of
a fluorescent tracer liquid under a UV light to estimate different
Fig. 1 Schematic of air-blast shot peening coverage. Similarly, fluorescent indicator or fluorescent marker is
also applicable.
Applications of Shot Peening to Gas Turbine Engines. One
stage involved hydraulic control, which was developed to ensure of the main merits of the SP process is that it can be applied to
the accuracy and repeatability of the SP process. The third stage of large and irregular surfaces of components at relative low cost.
development involved the use of Computerized Numerical Con- Because of the enhancement in fatigue life produced by the pro-
trol 共CNC兲, which not only maintained control of the SP process cess, shot peening was applied initially to a variety of gas turbine
but also allowed monitoring of the process to achieve the required engine components, such as crankshafts, connecting rods, ring
residual stress field and surface finish. gears and so on in the 1940s, 关16兴. A wide range of materials has
A further stage of development of the SP process is expected since been SP treated, including steel, titanium, and aluminum
which will quantify the benefits with concomitant nondestructive alloys relevant to the aircraft industry.
monitoring of compressive residual stresses and the elimination of To ensure the quality of shot peening, a military specification,
surface damage. MIL-S-13165 has been established to guide the SP process for gas
The Effect of Shot Peening Control Parameters. Although turbine blades and metallic components, 关17兴. Subsequently, a
shot peening is simple to apply, it is much more difficult to de- modified SP process with a shot deflector was developed by Gen-
velop a model of the mechanism and the subsequent induced de- eral Electric Company to enhance the fatigue strength of bolt
formation and residual stress relaxation. It is equally difficult to holes in engine turbine disks, 关6兴. The diameter of bolt hole is 6.5
quantify the SP benefits and the effect on the fatigue behavior of mm. The depth of compressive residual stresses which can be
shot peened components. In general, of the various important shot achieved is about 0.15 mm.
peening parameters 共shot size, shape, mass, density, hardness, ve-
locity, coverage, etc.兲, the Almen intensity and the coverage are
the important parameters that can be controlled once the media Laser Shock Peening „LSP…
has been selected for shot peening for controlling the quality of
the shot peening process. The Basic Laser Shock Peening Process. Laser shock peen-
An Almen strip intensity indirectly represents the compressive ing was first developed for enhancing the fatigue life of metal
residual stress profile of a shot peened component. The Almen parts in the early 1970s, but the LSP process has only recently
strip is a piece of steel 共the 1070 steel兲 which is measured for been applied to gas turbine engine components such as fan blades
flatness with an indicator, and then peened. The change in flatness and compressor blades. The main reason for the delay in applying
is called the Almen arc height, or Almen intensity, 关14兴. The Al- the process has been the low repetition rate of the original pulsed
Nd-glass laser system—about one cycle every eight minutes, 关18兴.
A new laser system developed by LSP Technologies, Inc. in Dub-
lin, OH is a high-energy, pulsed neodymium-glass laser which
reduces pulse duration to 15 to 30 nanoseconds and produces
output energy per pulse of 200 joules or more, 关19兴.
Figure 3 shows the basic process of laser shock peening. A coat
of black paint is first applied to the area of the component to be
peened. A high energy pulsed laser is then used to vaporize the
opaque coating into a plasma. As the plasma explosively expands,
a powerful shock wave is generated that impacts and propagates
into the components due to the presence of a transparent layer
confining the energy outwards. The intensity of the shock wave
travelling within the component can exceed two times the dy-
namic yield strength of the material at the surface and the shock
wave then gradually disperses through the thickness, 关20兴. The
shock wave penetrates a relatively deep surface layer to plastically
Fig. 2 The Heyn spring model for the generation of residual deform the material to produce a relatively deep compressive re-
stresses sidual stress layer.
Fig. 3 Schematic of the laser shock peening process on a flat
surface
The Control Parameters for Laser Shock Peening. The re-

sidual stress distribution induced by the LSP process is strongly
dependent on the type and intensity of the process and the mate-
rial properties. In principle, by increasing the intensity and expo-
sure, the depth of the compressive residual stresses can be in- Fig. 4 Scheme of the LPB process on a flat surface
creased from 0.9 mm to 1.8 mm depending on the material, 关20兴.
Since the compressive residual stresses have to be balanced by
equivalent tensile residual stresses, the increasing intensity of LSP the resultant residual stresses at the surfaces. The most important
will not necessarily increase the depth of the compressive stresses parameters are the LPB burnishing force, the burnishing ball di-
for thin-wall structures. Further, based on the test results from ameter, the feed and speed.
Smith in 2000, it is expected that the percentage of cold work The effect of the LPB burnishing loads on residual stresses has
increases with the increase of pulse number from one to three. been investigated, 关24兴. Two levels of burnishing loading were
chosen to represent the low and high LPB conditions. It was dem-
Applications of Laser Shock Peening in Gas Turbine En- onstrated that the compressive residual stresses in IN718 increase
gines. The LSP process has been successfully applied to en- with the LPB burnishing load. Like increasing laser intensity in
hance the fatigue resistance of engine materials such as A17075- the LSP process, increasing the burnishing load will increase the
T7351, 关21兴, IN718, 关8兴, Ti-6-4, 关22兴, and Ti-6-2-4-6, 关20兴. percentage of cold work, not necessarily increasing the depth of
However, one of the significant barriers to the application of LSP the compressive stresses for thin-wall structures.
is the slow processing rate, which is mainly due to the inability to Based on studies performed on a lathe, high-speed coverage can
remove the black coating rapidly for multiple shots. be achieved without significant difference in the residual stress
The first production application of LSP was in 1997 on a United distributions. The use of a 0.030 in. 共0.76 mm兲 feed appears to be
States military aircraft gas turbine engine blade after a Rapid- excessive. Any feed below that is probably a reasonable maximum
Coater™ Processing System was developed. Following this suc- for both residual compression at the surface and finish.
cessful application, the LSP has been used to enhance fatigue- High chromium steel, beta-silicon nitride, and tungsten carbide
prone engine components, such as blades, disks, and gears on the burnishing balls have been tested. The results, 关8兴, show that the
F-14 Tomcat, the F-16 Falcon, and the B-1B Lancer. Most re- effect of burnishing ball modulus on residual stresses is almost
cently, a F119 integrally bladed rotor has been attached to a robot negligible. So far no results have been published on the effect of
for the LSP processing, 关23兴. the size of the LPB burnishing ball.
Applications of Low Plasticity Burnishing in Gas Turbine
Low Plasticity Burnishing Engines. LPB has been demonstrated as a cost-effective surface
enhancement technology with process speeds comparable to those
The Basic Low Plasticity Burnishing Process. Low plastic- of conventional machining operations such as surface milling. It
ity burnishing 共LPB兲 has been developed as an improved surface has been demonstrated that LPB can 共i兲 improve fatigue strength
enhancement technology that can induce deep compressive re- of the engine materials, IN718, 关8兴, Ti-6-4, 关24兴, and Al 7075-T6,
sidual stress layers reaching to 1.0 mm with a low degree of cold 关25兴; 共ii兲 increase resistance to foreign object damage, 关8兴; and
work 共about 5%兲, and a proportionate increase in the stability of 共iii兲 eliminate corrosion damage, 关25兴.
the residual stress layers. The basic LPB process is shown in Fig. The advantages of LPB have attracted attention from USAF
4. Rather than using shots to bombard the component surface, a and US Navy. It is currently being investigated for application to
hydraulically floated ball is used to press and roll freely along the both F404 共F/A-18 aircraft兲 and T56 共C-130 and P-3C aircraft兲.
surface so that near-surface layers are deformed plastically. A se- Figure 5 is an example of the LPB application to an engine com-
ries of overlapping passes are applied until the intended surface pressor blade using four-axis mode in CNC milling machine, 关26兴.
area has been treated. By using X-ray diffraction stress measure-
ment system, it was found that great depths of compressive re- Evaluation and Comparison
sidual stress can be induced by the LPB process. Because the
rolling ball minimizes plasticity 共no shearing due to sliding兲, less Residual Stress Profiles. In general, shot peening with small
cold work is generated at the surface, 关8兴. shots generates a low Almen intensity and a shallow depth of
plastic deformation, resulting in shallower compressive residual
The Effect of Low Plasticity Burnishing Processing Param- stresses. A comparison of residual stress profiles induced by SP,
eters. There are several LPB processing parameters that affect LSP, and LPB for engine material IN718 is shown in Fig. 6, 关27兴.
Table 1 Comparison of surface treatment technologies
Surface
Treatment Depth
Technology Speed Coldwork 共mm兲 Relative Cost
SP Fast High ⬃0.2–0.5 Low
LPB Moderate Low ⬃1.0 Moderate
LSP Slow Low ⬃1.5 High
Figure 6共a兲 shows that, after the SP process, high compressive

residual stresses are confined to a shallow surface layer 共0.2 mm兲
while low tensile residual stresses spread deeper through the cross
section. Unfortunately, the degree of cold work can be exception-
ally high 共up to 30– 40%兲 at the surface, Fig. 6共b兲.
Residual Stress Relaxation. Although the mechanism of re-
sidual stress relaxation needs to be investigated, it has been found
Fig. 5 An engine compressor blade under PPB processing us- that the degree of cold work induced during residual stress gen-
ing a four-axis mode in CNC mill eration will influence the rate of thermal relaxation of the residual
stresses. A comparison study of thermal relaxation for a variety of
surface treatment technologies has been conducted, 关8兴, and the
results indicated that compressive residual stresses at heavily cold
worked SP’d surfaces relaxed very rapidly, compared with that of
a low cold worked surface produced by LSP and LPB.
Figure 6 indicates that, using laser shock peening, the depth of
compressive residual stress is significantly greater than that pro-
duced by shot peening for the same peak compressive residual
stress. The degree of cold working is beneficially less 共⬇9% or
less兲.
Figure 6 also indicates that LPB can generate great depths of
compressive residual stress. Because the rolling ball minimizes
plasticity 共no shearing due to sliding兲, less cold work 共⬇4%兲 is
generated at the surface. Deep compressive residual stresses de-
crease fatigue crack growth rates, in addition to prolonging initia-
tion life, and hence further increase the life of treated components,
关28兴.
Comparison of The Features of Each Process

It is known that a deep and high-magnitude compressive layer
at the surface 共⬃1 mm⫹ by LPB and LSP兲 can significantly resist
both crack initiation and short crack propagation. Very low cold
work 共2–5%兲 can improve both thermal stability and resistance to
thermal and cyclic load relaxation of these residual stresses.
Moreover, application of the process to industrial components re-
quires a process which is adaptable to existing CNC machine
tools, at speeds comparable to conventional machining operations
such as surface milling. A comparison of the applicability of the
three surface treatment technologies is summarized in Table 1.
In summary Table 1 shows that
• LPB can generate deep and large compressive residual
stresses, comparable to LSP;
• High cold work by SP is achieved by repeated applications of
shot peening; and
• LSP is the most expensive and slow process and the most
difficult to incorporate into an overhaul shop environment.
Conclusions
A review has been made of three mechanical surface treatment
technologies and their application to aircraft engine component
life extension and repair. The two most advanced technologies,
laser shock peening and low plasticity burnishing are described
and compared with shot peening, a well-established process. The
Fig. 6 Residual stress and cold work distributions in IN 718 advantages of the technologies are highlighted as well as the
after SP, LSP, and LPB limitations.
The review has indicated that neither LSP nor LPB has been 关13兴 Bailey, P. G., Lombardo, D. R., Popp, H. G., and Thompson, R. A., 1996,
‘‘Full Assurance Shot Peening of Aircraft Gas Turbine Engine Components,’’
fully understood and effectively applied to engine components. Proceedings of The Six International Conference on Shot Peening, San Fran-
Further research work is needed for the better understanding in the cisco, CA, pp. 320–327.
following areas: 关14兴 Kirk, D., 1999, ‘‘Shot Peening,’’ International Journal of Aircraft Engineering
and Aerospace Technology, 71, pp. 349–361.
• the mechanism of residual stress generation and relaxation by 关15兴 Lombardo D., and Bailey, P., 1996, ‘‘The Reality of Shot Peening Coverage,’’
all the mechanical surface treatment technologies and Proceedings of The Six International Conference on Shot Peening, San Fran-
• the effectiveness of these surface treatment technologies for cisco, CA, pp. 493–504.
关16兴 Brookman, J. G., and Kiddle, L., 1946, ‘‘The Prevention of Fatigue Failures in
engine component fatigue life extension and damage repair in Metal Parts by Shot Peening,’’ The Failure of Metals by Fatigue, Melbourne
an overhaul shop environment. University Press, Melbourne, Australia.
关17兴 Eckersley, J. S., 1990, ‘‘Effect of Shot Peening Process Controls on Consistent
Treatment,’’ Surface Engineering, S. A. Meguid, ed., Elsevier, New York, pp.
References 417– 427.
关1兴 James, D., 1991, ‘‘Ageing aircraft–The AATF Present Programme,’’ Proceed- 关18兴 Vaccari, J. A., 1992, ‘‘Laser Shocking Extends Fatigue Life,’’ American Ma-
ings of an International Conference on Ageing Aircraft and Structural Airwor- chinist, July, pp. 62– 64.
thiness, Washington, DC, Nov. 19–21, pp. 15–26. 关19兴 Clauer, A. H., Walters, C. T., and Ford, S. C., 1983, ‘‘The Effects of Laser
关2兴 Ratwani, M. M., Koul, A. K., Immarigeon, J-P., and Wallace, W., 1997, ‘‘Ag- Shock Processing on the Fatigue Properties of 2024-T3 Aluminum,’’ Lasers in
ing Airframes and Engines,’’ AGARD-CP-600 Vol. 1. Materials Processing, American Society for Metals, Metals Park, OH, pp.
关3兴 Lessells, J. M., and Brodrick, R. F., 1956, ‘‘Shot-Peening as Protection of 7–22.
Surface-Damaged Propeller-Blade Materials,’’ Proceedings of The Interna- 关20兴 Clauer, A. H., 1996, ‘‘Laser Shock Peening for Fatigue Resistance,’’ Surface
tional Conference on Fatigue of Metals, London, Sept. 10–14, pp. 617– 627. Performance of Titanium, J. K. Gregory, H. J. Rack, and D. Eylon, eds., TMS,
关4兴 Levine, S. R., 1978, ‘‘High Temperature Surface Protection,’’ Surface Treat- Warrendale, PA, pp. 217–230.
ments for Protection, The Chameleon Press Ltd., London. 关21兴 Peyre, P., Fabbro, R., Berthe, L., and Dubouchet, C., 1996, ‘‘Laser Shock
关5兴 Meetham, C. W., 1978, ‘‘The Protection of Compressor and Turbine Compo- Processing of Materials, Physical Processes Involved and Examples of Appli-
nents of the Aero Gas Turbine Engine,’’ Surface Treatments for Protection, cations,’’ J. Laser Appl., 8, pp. 135–141.
The Chameleon Press, Ltd., London. 关22兴 Ruschau, J., John, R., Thompson, S., and Nicholas, T., 1999, ‘‘Fatigue Crack
关6兴 Happ, M. B., 1984, ‘‘Shot Peening Bolt Holes in Aircraft Engine Hardware,’’ Growth Rate Characteristics of Laser Shock Peened Ti-6al-4V,’’ ASME J. Eng.
Proceedings of The Second International Conference on Shot Peening, Chi- Mater. Technol., 121, pp. 321–329.
cago, IL, pp. 43– 49. 关23兴 LSPT 2001, 2001, website publications of LSP Technologies, Inc.
关7兴 Bailey, P. G., and Whalen, J. M., 1990, ‘‘Shot Peen Controls and Aircraft 关24兴 Prevey, P. S., and Shepard, M. J., 2000, ‘‘Surface Enhancement of Ti-6A1-4V
Engine Requirements,’’ Surface Engineering, S. A. Meguid, ed., Elsevier, New Using Low Plasticity Burnishing,’’ Proceedings of 11th AeroMat Conference,
York, pp. 379–385. Bellevue, WA, June.
关8兴 Prevey, P. S., 2000, ‘‘The Effect of Cold Work on the Thermal Stability of 关25兴 Prevey, P. S., and Cammett, J. T., 2000, ‘‘Low Cost Corrosion Damage Miti-
Residual Compression in Surface Enhanced IN718,’’ Proceedings of the 20th gation and Improved Fatigue Performance of Low Plasticity Burnished 7075-
ASM Materials Solutions Conference & Exposition, St. Louis, MO, Oct. 10– T6,’’ Proceedings of 4th International Conference Aircraft Corrosion Work-
12. shop, Solomons, MD, Aug.
关9兴 Metz, S. A., and Smidt, F. A., 1971, ‘‘Production of Vacancies by Laser Bom- 关26兴 Cammett, J. T., and Prevey, P. S., 2001, Fatigue Strength Restoration in Cor-
bardment,’’ Appl. Phys. Lett., 19, pp. 207–208. rosion Pitted 4340 Alloy Steel via Low Plasticity Burnishing, Lambda-
关10兴 Fairand, B. P., Wilcox, B. A., Gallagher, W. J., and Williams, D. N., 1972, Research, Cincinnati, OH.
‘‘Laser Shock-Induced Microstructural and Mechanical Property Changes in 关27兴 Zhuang, W., and Halford, G., 2001, ‘‘Investigation of Residual Stress Relax-
7075 Aluminum,’’ J. Appl. Phys., 43, pp. 3893–3895. ation Under Cyclic Load,’’ Int. J. Fatigue, 23, pp. S31–S37.
关11兴 U.S. Patent 5,826,453, Oct. 1998, other patents pending. 关28兴 Telesman, J., Kantzos, P., Gabb, T., and Prevey, P., 1999, ‘‘Effect of Residual
关12兴 Sigwart, H., 1956, ‘‘Influence of Residual Stresses on the Fatigue Limit,’’ Stress Surface Enhancement Treatments on Fatigue Crack Growth in Turbine
Proceedings, International Conference on Fatigue of Metals, Institution of Engine Alloys,’’ The Third Joint FAA/DoD/NASA Conference on Aging Air-
Mechanical Engineers, London. craft, Albuquerque, NM, Sept.
Imbalance Response of a Rotor
Supported on Flexure Pivot Tilting
Pad Journal Bearings in Series
With Integral Squeeze Film
L. San Andrés
Dampers
Professor Measurements of the imbalance responses of a massive 45 kg rotor supported on series
(flexure pivot) tilting pad bearings and integral squeeze film dampers (SFDs) are pre-
O. De Santiago sented. The rotor-bearing configuration is of interest in compressor applications where
Research Assistant often oil lubricated dampers are introduced in series with fluid film bearings to relocate
critical speeds, enhance the overall system damping, and reduce the risks of rotordynamic
Mechanical Engineering Department, instabilities due to seals and impellers, for example. Coast-down experiments from 9000
Texas A&M University, rpm are conducted for increasing levels of rotor imbalance, and equivalent system damp-
College Station, TX 77843-3123 ing coefficients identified from the peak amplitude of rotor response while traversing
cylindrical mode critical speeds. The tests performed with locked (inactive) and active
SFDs demonstrate the effectiveness of the flexible damped support in reducing the system
critical speed and improving the overall rotor response with reduced transmitted forces to
ground. The SFDs allow safe rotor operation with values of imbalance twice as large as
the maximum sustained by the rotor supported on tilting pad bearings alone. The experi-
ments reveal a linear relationship between the peak amplitude of vibration at the critical
speeds and the imbalance displacement, even for rotor motions larger than 50% of the
tilting pad bearing and damper clearances. The tests also show little cross-coupling
effects with the shaft centerline moving along a nearly vertical path. The rotor-bearing
system remained stable in the entire range of operation and without the appearance of
subsynchronous vibration or nonlinear damper jump response.
[DOI: 10.1115/1.1492831]
Introduction cage spring centered dampers are the most commonly employed
design. Their most distinctive feature is the relatively large axial
Current turbomachinery design demands higher power-to-
length required by the elastic support in comparison to the damper
weight ratios. These systems require flexible supports in order to
hydrodynamic length, often requiring three to four times as much
locate and traverse critical speeds and to reduce midspan rotor length as the damper itself. The integral squeeze film damper
excursions. Furthermore, there exists a demand for increased tol- 共ISFD兲 recently advanced by Zeidan 关2兴 provides advantages such
erance to imbalance and sudden loads for satisfactory engine per- as short axial length, light weight, and easiness of installation. In
formance, which accrues additional requirements to the rotor sup- this case, the cylindrical journal is replaced by a set of arcuate
ports for vibrational energy dissipation. Tilting pad journal segments 共pads兲, their number depending on the particular appli-
bearings 共TPJBs兲 are extensively used in land turbines and com- cation. Thin structural EDM webs 共springs兲 provide the support
pressors due to their inherent stability, long life, and low power flexibility and connect the inner and outer rings. The thin gap
consumption. Recently, flexure pivot tilting pad journal bearings between the sector pads and the outer ring forms the squeeze film
共FPJBs兲 have been implemented due to their easiness of installa- lands. Each pad can be manufactured with different clearances to
tion and reduced number of parts 共De Choudhury et al. 关1兴兲. How- counter the static deflection due to the rotor weight. Additionally,
ever, tilting pad bearings may possess large stiffness coefficients the ISFD can be split for ease of installation and retrofit.
that inhibit damping to properly dissipate vibrational energy at the Integral squeeze film dampers provide satisfactory damping
support locations. Introducing a flexible support in series with the forces to rotor-bearing systems to safely traverse critical speeds.
tilting pad bearing allows to relocate critical speeds in nearly rigid De Santiago et al. 关3兴 and De Santiago and San Andrés 关4兴 present
rotors and provides a means to increase viscous damping by using experimental results on integral squeeze film dampers 共ISFDs兲
squeeze film dampers 共SFD兲. supporting a three-disk rigid rotor subjected to rotating loads in-
Squeeze film dampers derive their behavior from a lubricant duced by calibrated imbalance masses. In the tests, a pair of stiff
being squeezed in the annular space between a nonrotating cylin- ball bearings connected the shaft to the damper journal. Measure-
drical journal and its housing. A rolling bearing usually connects ments of the rotor synchronous response demonstrate the ability
the rotating shaft with the damper journal, and an elastic element of the ISFDs to attenuate rotor vibrations in the range of operation
共spring兲 centers the journal within the damper housing. Squirrel from zero to 9000 rpm, both for the first 共cylindrical兲 and second
共conical兲 modes of vibration. De Santiago and San Andrés 关5兴
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN incorporated end plate seals on the faces of the ISFDs and ob-
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF served enhanced damping capability. The particular construction
of the centering springs enables larger damping coefficients for
4 –7, 2001; Paper 2001-GT-0257. Manuscript received by IGTI, December 2000, tighter end seals without a severe reduction in the through lubri-
final revision, March 2001. Associate Editor: R. Natole. cant flow, which could cause excessive lubricant temperature rise.
Fig. 1 Test rig for measurements of imbalance response of rotor supported on tilting pad
bearings and squeeze film dampers
Several applications of the FPJB-SFD bearing combination have ings and squeeze film dampers through separate lines and allows
already been implemented in industrial turbomachinery 共Edney for independent control of the supply pressure to the bearing and
and Nicholas 关6兴 and Zeidan 关2兴兲. damper elements.
Experiments on the dynamic response of a rigid rotor mounted Eddy current proximity probes located at three axial planes and
on a pair of flexure pivot TPJBs in series with Integral SFDs in two orthogonal directions measure the rotor vibration during
follow. The aim of the tests is to determine the combined effect of coast down. Piezoelectric accelerometers with magnetic bases
these elements on the location of critical speeds and estimation of record the pedestals’ vibrations. A data acquisition system records
the rotor-bearing system damping ratio, and to demonstrate their the rotor response and acceleration at both supports as functions
effectiveness on reducing amplitudes of rotor vibration. The ex- of the rotor speed and stores the data in a personal computer. Type
periments extend prior work in a comprehensive research program K thermocouples and meters display reservoir, inlet, and exit lu-
dedicated to study and to demonstrate the effectiveness of novel bricant temperatures from the test bearings as well as ambient
squeeze film dampers and bearings for applications in turboma- temperature for reference, and the motor temperature for safe op-
chinery. eration. A light beam optical tachometer displays the rotor speed
and provides a reference signal to the data acquisition system for
phase angle measurements. Turbine-type flow meters digitally dis-
Test Rig Description
play the overall and individual lubricant consumption of the test
Figure 1 depicts the test rig with a 7.5 kW 共10 HP兲 DC motor bearings and dampers, and gauges display the lubricant supply
driving the rotor up to a top speed of 9000 rpm through a flexible pressure into each bearing and damper. The lubricant employed in
coupling and a drawn cup-type roller clutch. The rotor mass is all tests is an ISO VG 10 turbine oil with a viscosity 共␮兲 of 21.6
43.5 kg 共95.9 lb兲 and consists of a 76.2 mm 共3 in.兲 diameter shaft, cP at 17°C (63°F) and 8.40 cP at 40°C (104°F).
686 mm 共27 in.兲 long. The total vibrating mass is 45.3 kg includ-
ing one half of the coupling weight. The rotor has three equally
spaced, press-fit disks 25.4 mm 共1 in.兲 thick. The drive end and Design of the Flexure Pivot-Tilting Pad Bearing
middle disks are 279 mm 共11 in.兲 in diameter whereas the free end The rotor support assembly consists of three elements in series:
disk has a diameter of 229 mm 共9 in.兲. Each disk has 12 threaded the flexure pivot tilting pad journal bearings 共FPJBs兲, the integral
holes, equally spaced on its circumference, for installation of cali- squeeze film dampers 共ISFDs兲, and the bearing pedestals. Figure 2
brated imbalance masses. The span distance between the support depicts the FPJB-ISFD bearing combination. The tilting pad bear-
bearing centerlines is 406.4 mm 共16 in.兲. ing is incorporated into the existing ISFD previously tested by De
Hardened steel rings mounted on the shaft provide the sliding Santiago et al. 关3兴, whose main dimensions are given in Table 1.
surfaces at the fluid film bearing locations. Two identical pedestals The objective of the bearing design is to furnish a rotor support
bolted to a metal base plate hold the test bearings and provide with engineered location of critical speeds through the appropriate
grooves for lubricant feeding into the damper lands, as well as choice of the combined stiffness as well as optimum damping
drainage ducts for oil return. The pedestals also have transparent coefficients. The two fluid films in series posses combined stiff-
side covers allowing visualization of the bearing assembly. ness and damping force coefficients that can be optimized as de-
Custom-made plastic hose seals prevent leakage of lubricant to scribed by Barrett et al. 关7兴. Given a rotor-bearing system, there
the pedestal’s exterior while allowing free rotor lateral motions. A exists a specific value of the bearing to damper damping ratio 共␤兲
plastic pin facing the rotor free-end constrains axial motions since that minimizes the rotor response for given values of bearing to
no thrust bearings are installed in the test apparatus. The rotor, damper stiffness ratio 共␣兲 and damper damping ratio 共␨兲 at any
bearing pedestals, and support steel plate rest on a steel-reinforced rotor speed 共␻兲共frequency ratio f ⫽ ␻ / ␻ n 兲. In the design of the
concrete table. flexure pivot journal bearing, the optimum bearing to support
The facility includes a 151 L 共40-gallon兲 lubricant reservoir damping ratio ( ␤ opt) is selected at the undamped natural fre-
with one main pump and a circulation system with a heater and a quency of the original system 共i.e., f ⫽1兲.
cooler to maintain the lubricant temperature at a desired preset Prior measurements show the bearing supports 共pedestals兲 to be
value. The lubrication system provides oil to the tilting pad bear- flexible with asymmetric stiffness coefficients in the vertical and
series relationship K E ⫽K s K b /(K s ⫹K b ) . The condition of sup-
port isotropy is needed to keep the vertical and horizontal first
critical speeds as close as possible. The desired values of bearing
stiffness obtained are K bV ⫽27.6 MN/m, and K bH ⫽4.7 K bV . Note
that these predicted stiffness values define a strongly anisotropic
fluid film bearing. A computational analysis 共San Andrés 关8兴兲 gen-
erates the synchronous stiffness and damping coefficients for a
four pad, load between pads, flexure pivot TPJB. The fluid film
model is laminar with adiabatic thermal boundaries.
Table 3 gives the geometry of the FPJB, lubricant properties
and operating conditions. With the current hardware, the predic-
tions show that the maximum attainable bearing stiffness asym-
metry (K xx ⫽K y y ) is on the order of K bV /K bH ⫽1/1.008 at 3000
rpm. However, these bearing stiffness coefficients yield an equiva-
Fig. 2 Assembled ISFD and tilting pad bearing lent support stiffness asymmetry (K EV /K EH ) of only 1.12 at this
speed, which is considered satisfactory.
The predicted FPJB force coefficients at 3000 rpm, ( f ⬇1)
horizontal directions as shown in Table 2. On the other hand, the combined with the pedestal-ISFD effective stiffness and damping
integral dampers have 共both from design and static tests兲 a radial
ratios 共␨兲 for the open-ended ISFDs 共see Table 2兲, yield optimum
stiffness very close to 3.5 MN/m. The effective stiffness of the
combined bearing pedestals and integral dampers found from im- bearing to support damping ratios of ␤ opt V ⫽21.2, and ␤ opt H
pact tests on the rotor supported on ball bearings and ISFDs ⫽26.7. Table 2 includes the measured values of support damping,
equals K sV ⫽5.6 MN/m and K sH ⫽4.83 MN/m in the vertical and i.e., overall system damping for the open ends ISFD and bearing
horizontal directions, respectively. pedestals. The selected tilting pad bearing is expected to provide
Desired bearing stiffness coefficients (K bV ,K bH ) that render an damping coefficients that are smaller than those needed for opti-
isotropic support 共i.e., K EV ⫽K EH 兲 are estimated from the simple mum damping conditions 共slightly over 50% of the optimum
damping兲. Figure 3 depicts predictions of the rotordynamic force
coefficients representing the combined pedestals, ISFDs and
Table 1 Integral squeeze film damper main characteristics FPJBs for the drive end bearing as a function of the rotor
speed. The force coefficients for the free end bearing support are
very similar to the coefficients in the drive end, the only differ-
ence being the carrying load and the structural stiffness of the
pedestals.
Table 2 Measured stiffness of pedestal supports and integral dampers
Table 3 Flexure pivot tilting pad bearing main dimensions and operating conditions
Fig. 3 Predicted equivalent stiffness and damping synchronous force coefficients for tilting pad bearings and integral dampers,
including supports flexibility. Drive and support.
Experimental Results due to known imbalance masses inserted at a radius of 114.3 mm

共4.5 in.兲 on the rotor middle disk and exciting the first 共cylindri-
Coast-down imbalance response tests serve to evaluate the ef-
fectiveness of the series bearing combination in limiting orbit am- cal兲 mode of vibration.
plitudes when the rotor traverses the first critical speed. In these The current bearing support configuration allows conducting
experiments, the rotor speed is brought up to 9000 rpm and then measurements with locked squeeze film dampers. These experi-
the motor power is shut off. The rotor disengages from the motor ments are of importance to isolate the effect of the tilting pad
through the clutch action, continues spinning after the motor stops bearings on the dynamic response of the rotor. Bronze shims in-
and decelerates until it reaches rest. The experimental results evi- serted within the integral damper film lands and lateral plates
dence the very low friction in the FPJBs since it takes more than tightly pressed against the damper journal and housing ring effec-
ten minutes for the decelerating rotor to come to a full stop. In all tively eliminated 共locked兲 any damper journal motions relative to
the synchronous response measurements, the baseline response to its housing. Table 4 sums the imbalance conditions and recorded
remnant imbalance is subtracted 共vector form兲 from the responses feed lubricant temperatures. The largest imbalance displacement
Table 4 Measured responses at the first critical speed of rotor supported on FPJBs with locked dampers
Fig. 4 Summary of imbalance response of rotor supported on FPJBs. Measure-

ment at rotor midspan „vertical direction…. Locked ISFDs.
about 5750 rpm. Table 4 also gives the peak-peak amplitudes of
response at the first critical speed in the vertical direction as well
as the location of the critical speeds. The largest imbalance (u
⫽22.4 microns) excites rotor motions up to ⬃80% of the as-
sembled bearing clearance at the drive end location. The measure-
ments show the linearity of the response, i.e., amplitudes of mo-
tion proportional to the imbalance displacement, even though
these motions are rather large relative to the bearing clearance.
Predictions of rotor response from a transfer matrix model of the
rotor supported on the tilting pad bearings 共XLTRC 关9兴兲 show
good correlation with the location of the experimental critical
speeds and response amplitudes in the vertical direction.
A test with no imbalance in the rotor evidences the rotor lift
within the bearing clearance with the dampers active 共unlocked兲.
Figure 5 shows in dimensionless form the centerline travel of the
rotor at the drive end for increasing rotor speeds. The FPJB mini-
mum diametrical clearance 共0.152 mm兲 is used to set dimension-
less rotor displacements. The main direction of the shaft centerline
motion is along a vertical line thus denoting small cross-coupled
forces. The coordinate 共0,0兲 denotes the beginning of the test with
the rotor at its lowest position and in contact with the pads. At the
Fig. 5 Shaft centerline motion for test with no imbalance. Ro- highest speed of ⬃9000 rpm, the rotor has lifted to about 20% of
tor supported on FPJBs in series with ISFDs. Drive end, static the bearing diametrical clearance. Note that the rotor center does
load: 247.3 N. not return to the same start position, thus revealing some static sag
at the end of the experiment. This may be attributed to the bearing
construction and some thermal-induced bow on the rotor.
(u) corresponds to 22.4 microns, i.e., approximately 30% of the Imbalance response experiments were conducted next to assess
FPJB assembled radial clearance. Figure 4 shows the rotor syn- the dynamic behavior of the rotor supported on tilting pad bear-
chronous response at midspan in the vertical direction. The am- ings in series with the ISFDs. The squeeze film lands are open to
plitudes of rotor vibration are consistently larger in this direction ambient in the current tests, i.e., without end seals installed. Table
than in the horizontal direction, the critical speeds occurring at 5 sums the imbalance conditions and inlet lubricant temperatures.
Table 5 Measured responses at the first critical speed of rotor supported on series FPJB-ISFDs with locked dampers
Fig. 6 Summary of imbalance response of rotor supported on series FPJB-ISFDs. Measurements at rotor midspan in the vertical
and horizontal directions. ISFDs active.
Fig. 7 Measured first critical speed versus imbalance distance for tests on
rotor supported on series FPJB-ISFDs
The largest imbalance displacement (u) corresponds to 49 mi- with the locked dampers. However, operation with the dampers
crons 共18.7 grams兲, i.e., larger than twice the largest value tested results in an increased tolerance to imbalance due to the larger
with the FPJBs alone. Figure 6 depicts the rotor synchronous re- clearances in the ISFDs. Note that the maximum available clear-
sponse, vertical and horizontal, at midspan for increasing imbalance with the two elements is 0.610 mm 关0.152 (TPJB)⫹0.458
ances. Similar measurements were recorded at the free and drive 共ISFD兲兴.
ends and are not shown for brevity. Note that the amplitudes of Table 5 includes the peak-peak amplitudes of rotor response at
rotor vibration in the two directions are nearly identical while the first critical speed as well as estimates of the average ampli-
crossing the 共new兲 lower critical speed at approximately 2700 tude and standard deviation in the three planes of measurement.
rpm. The peak response in the vertical direction at about 4500 rpm Figure 7 shows the average value of critical speeds 共vertical and
is due to a resonance of the bearing pedestals. No subsynchronous horizontal兲 versus the imbalance displacement. Note the remark-
vibrations were ever observed in any of the experiments. For iden- able reduction in critical speed of almost 10% for the largest im-
tical imbalance conditions, the peak amplitudes of vibration for balance condition tested.
the rotor with the dampers active are larger than those recorded Figure 8 depicts the average peak amplitudes of motion at the
critical speed versus the imbalance distance. Note that the rotor
amplitudes of motion are proportional to the imbalance displace-
ment (u), in spite of the fact that the largest rotor motions are
about 70% of the damper radial clearance. The equivalent system
damping ratios 共␨兲 are estimated from the peak responses using
the amplification factor (Q) at the first critical speed as, i.e., from
2 ␨ 2 ⫽⫺1⫾ 冑1⫹1/Q 2 . The simple analysis implies rigidity of the
rotor, as the experiments demonstrate from the measurements of
displacements at three planes. The amplification factors Q from
the linear regressions given in Fig. 8 are equal to 3.5 and 2.85 in
the vertical and horizontal directions, respectively. Thus, the cor-
responding system damping ratios are ␨ V ⫽0.141 and ␨ H ⫽0.173
for the configuration of rotor mounted on tilting pad bearings and
active integral dampers. These values of system damping ratio are
compared with results obtained in prior experiments for the rotor
Fig. 8 Average peak-peak rotor response at first critical speed supported on integral dampers and ball bearings, ␨ sV ⫽0.137 and
versus imbalance distance for rotor supported on FPJB-ISFDs. ␨ sH ⫽0.106 共De Santiago et al. 关3兴兲. Thus, the present test con-
Vertical direction and horizontal planes. figuration shows a marked increase in system damping of about
Table 6 Calculated rotordynamic properties of test rotor supported on FPJBs only „locked dampers… and FPJBs in series with
active ISFDs
overall rotor response with reduced transmitted forces to ground.
The ISFDs allow the rotor to operate safely with values of imbal-
ance twice as large as the maximum sustained by the rotor sup-
ported on tilting pad bearings only.
Predicted system damping ratios compare well with those esti-
mated from the peak amplitude responses. Critical speeds are also
favorably predicted. Further experimental effort is needed to de-
velop parameter identification techniques that will allow the accu-
rate estimation of the bearing and damper force coefficients over
frequency ranges of interest.
Acknowledgments
The support of the TAMU Turbomachinery Research
Consortium and National Science Foundation are gratefully
acknowledged.
Nomenclature
C ⫽ system damping coefficient 共N.s/m, lbf.s/in.兲
Fig. 9 Comparison of predicted and measured horizontal syn- f ⫽ frequency ratio, ␻ / ␻ n
chronous response for 18.6 gram imbalance „u Ä49 microns…. K ⫽ bearing and support stiffness 共N/m, lbf/in.兲
Measurements at rotor midspan „integral dampers active…. m ⫽ imbalance mass attached to rotor middle disk 共kg兲
M ⫽ rotor mass for the first mode of vibration 共45.3 kg,
99.8 lbm兲
60% for motions in the horizontal direction. The increase in Q ⫽ rotor-bearing system amplification factor at first criti-
damping ratios for motions in the vertical direction is more mod- cal speed
est 共only 2.9%兲 since the rotor displacements in this direction now r ⫽ rotor middle disk radius for location of imbalance
are similar to the displacements in the horizontal direction. masses 共114.3 mm, 4.5 in.兲
A transfer matrix model of the test rotor on the bearing sup- R2 ⫽ linear regression correlation factor
ports, i.e., series FPJBs and ISFDs, predicts the system damped u ⫽ m"r/M , imbalance distance 共m, in.兲
eigenvalues 共natural frequencies and damping ratios兲 as well as X,Y ⫽ coordinates of motion, vertical and horizontal
the rotor imbalance response. Table 6 presents the calculated sys- ␣ ⫽ K b /K s , bearing-to-support stiffness ratio
tem critical speeds and damping ratios for rotor speeds below ␤ ⫽ C b /C s , bearing-to-support damping ratio
9000 rpm. All predicted modes of vibration are well damped ␮ ⫽ lubricant viscosity 共Pa-s兲
within the range of test speeds. The predicted damping ratios for ␻ ⫽ frequency of imbalance excitation, synchronous with
the cylindrical mode of vibration 共␨ V ⫽0.127, ␨ H ⫽0.134兲 com- rotor speed 共rad/s兲
pare relatively well with the values derived from the measure- ␻n ⫽ system natural frequency of rotor supported on ISFDs
ments 共␨ V ⫽0.141, ␨ H ⫽0.173兲, the differences being 5% and 26% and rigid bearings 共rad/s兲
in the vertical and horizontal directions, respectively. Figure 9 ␨ ⫽ C/2M ␻ n , system damping ratio
shows a comparison of the experimental and predicted synchro- Subscripts
nous rotor response for the largest imbalance 共18.6 gram兲, u
⫽0.049 mm. Note that the test response in the vertical direction V,H ⫽ vertical, horizontal (X,Y )
shows the resonance of the support not well accounted for in the s ⫽ refers to support 共pedestal–ISFD兲 properties
predictive model. In general, there is good agreement between the b ⫽ refers to tilting pad bearing properties
predicted and experimental responses, the horizontal response be- E ⫽ equivalent ‘‘static’’ property
ing slightly closer to the experimental measurements than the ver- opt ⫽ optimum value
tical amplitudes.
References
Conclusions 关1兴 De Choudhury, P., Hill, M. R., and Paquette, D. J., 1992, ‘‘A Flexible Pad
Bearing System for a High Speed Centrifugal Compressor,’’ Proceedings of
Measurements of the imbalance response of a three disk mas- the 21st Turbomachinery Symposium, Dallas, TX, pp. 57– 64.
sive rotor supported on flexure pivot tilting pad bearings in series 关2兴 Zeidan, F., 1995, ‘‘Application of Squeeze Film Dampers,’’ Turbomach. Int.,
with integral squeeze film dampers are detailed. This support con- 11, 共Sept./Oct.兲, pp. 50–53.
关3兴 De Santiago, O., Oliveras, J., and San Andrés, L., 1999, ‘‘Imbalance Response
figuration shows many advantages over the common ball bearing- of a Rotor Supported on Open-Ends Integral Squeeze Film Dampers,’’ ASME
SFD arrangement. These are namely, ease of installation and ret- J. Eng. Gas Turbines Power, 121, pp. 718 –724.
rofit 共split elements兲, reduced number of parts and weight, and as 关4兴 De Santiago, O., and San Andrés, L., 2000, ‘‘Dynamic Response of a Rotor-
demonstrated by the measurements, much larger viscous damping Integral Squeeze Film Damper to Couple Imbalances,’’ ASME Paper No.
2000-GT-388.
action with increased tolerance to disk imbalances. 关5兴 De Santiago, O., and San Andrés, L., 1999, ‘‘Imbalance Response and Damp-
Coast-down experiments from 9 krpm were conducted for in- ing Force Coefficients of a Rotor Supported on End Sealed Integral Squeeze
creasing levels of disk imbalance. The measurements show the Film Dampers,’’ ASME Paper No. 99-GT-203.
rotor center line moves in a nearly vertical direction, thus evidenc- 关6兴 Edney, S. L., and Nicholas, J. C., 1999, ‘‘Retrofitting a Large Steam Turbine
With a Mechanically Centered Squeeze Film Damper,’’ Proceedings of the
ing little cross coupled effects in the flexure pivot bearings. Sys- 28th Turbomachinery Symposium, Houston, TX, pp. 29– 40.
tem damping coefficients are identified from the peak amplitude 关7兴 Barrett, L. E., Gunter, E. J., and Allaire, P. E., 1978, ‘‘Optimum Bearing and
of rotor response 共Q-factor兲 while traversing cylindrical mode Support Damping for Unbalance Response and Stability of Rotating Machin-
critical speeds 共⬃5700 rpm for locked SFDs and 3200 rpm for ery,’’ ASME J. Eng. Gas Turbines Power, 100, pp. 89–94.
关8兴 San Andrés, L., 1996, ‘‘Turbulent Flow, Flexure-Pivot Hybrid Bearing for
active SFDs兲. The tests performed with locked 共inactive兲 and ac- Cryogenic Applications,’’ ASME J. Tribol., 118, pp. 190–200.
tive ISFDs demonstrate the effectiveness of the flexible damped 关9兴 XLTRC, 1998, Rotordynamics Analysis Software, TAMU Turbomachinery
support in relocating the system critical speed and improving the Laboratory.
Contact Stresses in Dovetail
Attachments: Alleviation via
G. B. Sinclair
Department of Mechanical Engineering,
Louisiana State University,
Precision Crowning
Baton Rouge, LA 70803-0100
Previous studies of contact stresses in dovetail attachments provide verified finite element
stresses as well as some simple physical models of such stresses. Generally there is good
N. G. Cormier agreement between the finite element analysis and the simple physical models. Together
General Electric Aircraft Engines, the two identify a pinching mechanism as leading to large fluctuations in hoop stresses at
Evendale, the edges of contact. These fluctuating hoop stresses can be expected to be a major
Cincinnati, OH 45215 contributor to the fatigue of dovetail attachments. The present study investigates a method
of alleviating the fluctuating hoop stresses. This method entails crowning of the contacting
flat on a blade. Such crowning is in two orthogonal directions on the blade flat. Two-
dimensional finite element analyses are made in both directions. These analyses indicate
that significant reductions in fluctuating hoop stresses can be achieved by the method,
provided the crowning is controlled sufficiently precisely. 关DOI: 10.1115/1.1584477兴
1 Introduction A highly simplified model in 关2兴 indicates that the blade slips
relative to the disk during loading up if
1.1 Motivation and Background. Single tooth attachments
or ‘‘dovetails’’ are used to secure fan and compressor blades to tan ␣ ⬎ ␮ , (1)
disks in gas turbines. A section through a typical dovetail is shown
in Fig. 1共a兲. Therein the base of the blade is pulled as a result of where ␮ is the coefficient of friction. Typically in gas turbine
the centripetal acceleration of its remainder, while it is restrained engines ␣⬇␲/4 whereas ␮ⱗ0.4 共Hamdy and Waterhouse, 关3兴兲, so
by contact with the disk on two flats 共e.g., C⫺C ⬘ in Fig. 1共a兲兲. At that 共1兲 predicts slip during loading up. The finite element analysis
the edges of these contact regions, fretting can occur when loads of 关1兴 confirms this is the case.
vary. This fretting can lead to fatigue crack initiation and ulti- Given slipping during loading up,
mately to failures 共e.g., in the disk at C, the blade at C ⬘ ). The
primary intent of this work is to try to devise means of reducing T⫽ ␮ N. (2)
the incidence of such failures.
This research builds on an earlier stress analysis of a dovetail Then resolving vertically leads to the following expressions for
attachment with a section as in Fig. 1共a兲共Sinclair et al., 关1兴兲. This the nominal contact and shear stress:
stress analysis uses finite elements. It also builds on an earlier
development of simple companion physical models 共Sinclair and F ␻ ⫹F b
¯␴ c ⫽ , ¯␶ c ⫽ ␮ ¯␴ c . (3)
Cormier 关2兴兲. These models are largely analytical. Preliminary to 2a 共 cos ␣ ⫹ ␮ sin ␣ 兲
devising an approach for reducing fatigue failures in dovetail at-
tachments, we summarize the findings of these two papers. We In 共3兲, ¯␴ c is taken to be positive when compressive 共as in Fig.
begin with some results for nominal contact stresses during load- 1共b兲兲, and bars atop stresses denote nominal values. The finite
ing up, then we discuss results for peak contact stresses during element analysis in 关1兴 tracks the results in 共3兲 during loading up
loading up, then stresses during unloading. to within 3%. This analysis also furnishes nominal bending
stresses due to M: There is no corresponding formula in terms of
1.2 Nominal Contact Stresses During Loading Up. A F ␻ , F b for these stresses because they are statically indetermi-
free-body diagram for half of the blade section of Fig. 1共a兲 is nate.
given in Fig. 2. The loading of this half-blade comes from the
centripetal acceleration of its self mass, F ␻ , and that of the por- 1.3 Peak Contact Stresses During Loading Up. While it is
fairly easy to obtain accurate nominal stresses with finite ele-
tion of the blade not shown, F b . These ‘‘forces’’ are balanced by
ments, it is not so easy to obtain accurate edge-of-contact stresses
normal and shear forces, N and T, acting on a contact region of
extent 2a inclined at an angle ␣. By virtue of symmetry, there are 共i.e., stresses at C, C ⬘ in Fig. 1共a兲,共b兲兲. There are challenges to
obtaining such stresses even within two-dimensional elasticity.
no counterbalancing forces on the centerline and only a horizontal
force there, H. All these forces are per unit thickness in the out- While contact is conforming at C, C ⬘ and stresses therefore non-
singular provided contact inequalities are policed,1 the policing of
of-plane direction in Fig. 2.
these inequalities requires elements that are small relative to local
In general there can be a moment reaction on the contact re-
radii of curvature and here these local radii are small themselves.
gion, M 共per unit thickness兲. This is balanced by an equal and
In addition there are locally high stresses, and even infinite stress
opposite moment on the centerline, which also has a further mo-
gradients, at the edges of contact. An illustration of this for the
ment M F to offset the moment produced by F ␻ and F b about the normal contact stress ␴ c is given in Fig. 3. This is for maximum
head of N. load 共rpm兲 and when there is no friction, the situation producing
the highest ␴ c 共cf., 共3兲兲. Normalization is by ␴ 0 , a stress repre-
sentative of the loading and defined by
tional Gas Turbine and Aeroengine Congress and Exhibition, New Orleans, LA, June 1
See, e.g., 关1兴, Section 2, for a discussion of contact inequalities and their roles in
4 –7, 2001; Paper 2001-GT-0550. Manuscript received by IGTI, Dec. 2000, final conforming contact problems.
revision, Mar. 2001. Associate Editor: R. Natole.
Fig. 3 Contact stress distributions „␮Ä0…
␴ max
c ⫽6.3␴ c .
¯ (5)
Hence, at maximum load, a stress concentration factor of 6.3 in
Fig. 1 Dovetail attachment configuration: „a… section of overall effect ( ¯␴ c ⫽1.9␴ 0 , thus the lower factor in 共5兲 than in Fig. 3兲.
attachment, „b… closeup of disk near lower contact point with Crucial to determining the peak contact stress of Fig. 3 with
stresses acting, „c… in-plane crown on blade flat, „d… out-of- finite elements is to try and ensure convergence: Only then can
plane section BB ⬘ one be reasonably sure that peak stresses are being accurately
calculated. Furthermore, convergence can be nonuniform in con-
forming contact problems because of changing contact extents.
Under these circumstances, it is relatively easy to be mislead as to
F ␻ ⫹F b convergence with just a two-grid check. Consequently in 关1兴 and
␴ 0⫽ , (4)
b here we adopt the following convergence checks. We employ a
sequence of at least three grids: a coarse, a medium, and a fine.
where b is the half-blade width 共Fig. 2兲. The sharp peaks in Fig. 3
The last two of these grids are formed from their respective pre-
are not equal because of the presence of bending contributions
decessors by successively halving element sides. Then if the
共i.e., the presence of M兲. For the higher of the two peaks, the value
coarse grid has n elements, the medium and fine have 4n and 16n,
of ␴ max
c can be expressed by respectively. On this sequence, we judge a contact stress to be
converging if
兩 ␴ Cc ⫺ ␴ cM 兩 ⬎ 兩 ␴ cM ⫺ ␴ Fc 兩 , (6)
wherein superscripts indicate the grid used in the evaluation. Pro-
vided 共6兲 holds, we judge this contact stress to have converged if
兩 ␴ cM ⫺ ␴ Fc 兩 ⬍␧ 兩 ␴ Fc 兩 , (7)
wherein ␧ is the level of accuracy sought. We view ␧⫽0.10 共i.e.,
10%兲 as a satisfactory level of accuracy for present purposes,
␧⫽0.05 共5%兲 as good, and ␧⫽0.01 共1%兲 as excellent.2
In order to meet the convergence criteria of 共6兲, 共7兲 for the
dovetail attachment of Fig. 1共a兲, 关1兴 uses a submodeling proce-
dure 共Cormier et al. 关5兴兲. This choice results in three, successively
refined, global grids, followed by three, successively refined, sub-
model grids. Convergence is only achieved on the latter three-grid
sequence. Thereon 共6兲 is complied with as illustrated in the clo-
seup in Fig. 3. Further, 共7兲 is complied with to within a good level
of accuracy 共the actual ␧⫽0.017 or 1 2/3%; 关1兴, Section 4兲. These
are the only edge-of-contact stresses we are aware of in the litera-
ture that comply with 共6兲, 共7兲共see 关1兴, Section 1, for a review兲.
The sides of the elements in the fine submodel grid are 0.5% of
the local radii of curvature: This is 30 times smaller than elements
used in any analysis reported in the literature and referenced in
关1兴. The submodeling procedure enables this level of resolution to
be achieved without undue computational effort.
2
For stress analysis in general, some justification for convergence checks of this
Fig. 2 Free-body diagram of half of the blade section genre is given in Sinclair 关4兴.
With friction present, slipping during loading up leads to local
shear stresses, ␶ c of Fig. 1共b兲, with
␶ c⫽ ␮ ␴ c . (8)
The shear contact stresses in 关1兴 comply with 共8兲 to within 3.5%.
These shears pull the material within the disk just before C, and
push the material just ahead of C ⬘ . Accordingly the hoop stress
␴ h 共Fig. 1共b兲兲 is tensile at C and compressive at C ⬘ in the disk.
The converse holds true for the blade. Thus the peak tensile hoop
stresses occur at C in the disk, C ⬘ in the blade.
For Hertzian contact in the presence of slip, Poritsky 关6兴 has
that the hoop stress at the edge of contact, ␴ max
h , is given by
h ⫽2 ␶ c ⫽2 ␮ ␴ c
␴ max max max
. (9)
For the maximum coefficient of friction considered here, 0.4, this Fig. 4 Hoop stress variation near the edge of contact for 20%
means hoop stresses that are quite comparable in magnitude to the unloading „␮Ä0.4…
normal contact stresses. This is indeed found to be the case in 关1兴.
There the peak value of the converged hoop stress at C at maxi-
mum load is 88% of the peak normal contact stress when ␮⫽0.4.3
A further consequence of this pinching is that it produces large
1.4 Contact Stresses During Unloading. Unloading can fluctuations in the peak hoop stress, ␴ max
h . In the first instance,
occur with reductions in engine rpm. It can also occur with varia- this is because ␴ max
h is largely due to ␶ max
c 共recall 共9兲兲, and ␶ max
c
tions in stiff-wise bending 共i.e., bending out-of-plane in Fig. 1共a兲兲. drops significantly with unloading and pinching. In the second
Irrespective of the source, two different responses can be identi- instance, this is because ␴ h can go from being tensile just out-
max
fied. side of the contact region to compressive and within the contact
Without friction, N and ␴ max
c during unloading take on the same region as N and the contact region grow with unloading and
values as during loading up for corresponding loads 共to within pinching. These effects in combination are illustrated in Fig. 4.
0.5%, 关1兴, Section 4.2兲. With friction, N and ␴ maxc during unload- Therein a 20% reduction in load changes ␴ max h /␴0 from 8.0 to
ing increase, at least initially. This result is contrary to any one- ⫺5.2 when ␮⫽0.4.
dimensional physical reasoning. However, it can be explained The foregoing large fluctuations in hoop stresses are from an
with two-dimensional reasoning. elastic analysis. With yielding and plastic flow, these fluctuations
Consider what happens at the section through the disk at AA ⬘ in can be expected to be lessened. Even so, large fluctuations in
Fig. 1共a兲. With unloading, F ␻ ⫹F b in Fig. 2 is reduced. Therefore elastic stresses are the harbingers of fatigue failure 共cf., S-N
so is the radially outward pull on the disk at AA ⬘ . Consequently curves兲. It would seem reasonable that finding a means for low-
the material above AA ⬘ in the disk retracts radially inward. Be- ering these fluctuations would offer a potential way for reducing
cause the periodic symmetry line through A ⬘ is not parallel to the the incidence of fatigue failures in dovetail attachments.
central symmetry line through B 共Fig. 1共a兲兲, this retraction is ac-
1.5 Objective and Scope. The primary objective of the
companied by a tendency for the disk contact region to move
present work, then, is to reduce the large fluctuations in hoop
laterally towards the centerline. Without friction, the blade can
stresses that attend unloading of dovetail attachments. One clear
slip radially inwards and thereby accommodate lateral motion of
means of doing this is to lower friction. Our understanding is that
the disk contact region. With friction, the blade can stick and get
the industry has introduced surface treatments to this end. Any
pinched by lateral motion of the disk contact region. It is this
further improvements in this regard would be all to the good.
pinching mechanism that increases N and ␴ max c during unloading. Here, instead, we adopt a different tactic.
A simple analysis in 关2兴, Section 2.5, shows that indeed sticking
The tensile ␴ max
h prior to unloading largely stems from ␶ maxc 共as
and pinching during unloading is consistent with increases in N,
in the first of 共9兲兲: ␶ max in turn is directly proportional to ␴ max 共as
hence ␴ max . Results from the analysis for 关1兴 and reported in 关2兴 c c
c
confirm sticking with pinching and quantify the increases in N and in the second of 共9兲兲. It follows that reducing ␴ max c reduces the
␴ max 共e.g., a 20% unload leads to a 13% increase in ␴ max when ␴ max
h that can fluctuate at the outset. Here therefore we seek to
c c
␮⫽0.4兲. reduce ␴ max
c at maximum load.
One consequence of this pinching is that T and ␶ max drop sig- The method chosen for implementing this tactic is precision
c
nificantly on unloading. This is because there are two sources of crowning. This method places a crown in the shape of an arc of a
reduction in the shear force T. The first is the expected reduction circle with a small height ␦ on the blade contact flat 共Fig. 1共c兲兲.
occurring because there is less load to be balanced 共i.e., because The intent is to promote Hertzian-like normal contact stresses 共in-
F ␻ ⫹F b in Fig. 2 is less兲. The second is the less-expected reduc- dicated with a broken line in Fig. 3兲. Frictionless Hertzian contact
tion occurring because the normal reaction is increased and con- of a cylinder on a flat has a stress concentration factor of 4/␲
sequently is balancing a greater share of the load 共i.e., because the 共Hertz 关7兴兲. The dovetail attachment of Fig. 1共a兲 with its blend
vertical component of N in Fig. 2 is increased兲. Results from the radii has a stress concentration factor of 6.3 when ␮⫽0 共see 共5兲兲.
analysis for 关1兴 reported in 关2兴, Section 2.5, confirm and quantify Hence provided we can approach the same contact extent with
the significant drops that can occur in T and ␶ max 共e.g., a 20% crowning as without, a marked reduction in ␴ max c may be possible.
c
In order for contact to spread throughout most of the available
unload leads to a 68% decrease in ␶ max when ␮ ⫽0.4兲.
c
contact extent (CC ⬘ of Fig. 1共a兲兲, the height of ␦ of the crown
3
There are other contributions to the ␴ h in the dovetail attachment that are not
must be comparable to the elastic approach 共i.e., the elastic de-
present in the configuration in Poritsky 关6兴, so that precise compliance with 共9兲 is not flection of the crown center兲. Typically elastic approaches under
to be expected. Hertzian contact are too small to be manufactured. In the jet en-
Fig. 5 Crown profiles
gine industry, however, loads are high enough relative to elastic

moduli that this need not be the case, albeit some precision may
be required in the manufacture.4
If precision crowning in the plane of Fig. 1共a兲 is successful in
reducing ␴ max
c and thereby fluctuations in hoop stresses, then to be
consistent a similar crowning needs to be employed in the out-of-
plane direction. That is, for section BB ⬘ of Fig. 1共a兲 shown in Fig. Fig. 6 Coarse finite element grid for in-plane analysis
1共d兲. Absent crowning, this cross section can be expected to pro-
duce a double-peak, normal, contact stress similar to that of Fig.
3. Accompanying shear stresses out of the plane of Fig. 1共d兲 where E is the Young’s modulus and ␯ Poisson’s ratio. These
would have a similar scaled distribution and lead to large hoop moduli are taken to be the same for both the blade and the disk.
stresses. With crowning, this cross section can be expected to Further,
冉冊
produce a more Hertzian-like normal contact stress with corre-
spondingly reduced shear and hoop stresses. ␦ a 2
⫽ . (11)
In what follows, therefore, we first assess the effects of in-plane ␦a aa
precision crowning in Section 2. Thereafter, in Section 3, we as- From 共10兲, 共11兲, for the specifications given in 关1兴, we obtain a
sess the effects of out-of-plane precision crowning. This latter minimum ␦ of 25.4 ␮m 共1/1000 in兲 for 87% contact when ␮⫽0
activity requires a special development in order to keep the analy- and N is a maximum. That is, a a ⫽0.87a when ␦⫽25.4 ␮m. This
sis two-dimensional and thus be able to check convergence fairly crown height we denote by ␦ 0 .
readily.5 The paper closes in Section 4 with some remarks in light To see if precision crowning is effective within some range
of the results found for precision crowning in both planes.6 above the minimum ␦ 0 , we consider higher crown heights of 2 ␦ 0
and 4 ␦ 0 for ␮⫽0.4. If it is effective, this would enable precision
crowning to be performed at a higher ␦ than ␦ 0 with some toler-
2 In-Plane Precision Crowning ance. To check our expectation with respect to edge-of-contact
2.1 Crown Heights. The major concern of this analysis is peaks in ␴ c , we also consider a lower crown height of ␦ 0 /2 for
when the fluctuations in hoop stresses are worst. That is, when ␮⫽0. After all, the estimate of the minimum ␦ 0 came from Hert-
friction is at its likely maximum 共␮⫽0.4, Hamdy and Waterhouse zian formulas, and even with crowning the dovetail attachment of
关3兴兲. However, we need precision crowning to work for any level Fig. 1共a兲 is not strictly a Hertzian configuration.
of friction if it is to be really effective. Consequently we also Absent the crown, a precise formulation for the in-plane contact
evaluate precision crowning for ␮⫽0 as a lower bound check. problem attending Fig. 1共a兲, including contact inequalities, is
Although the hoop stresses unload without any dramatic fluc- given in 关1兴, Section 3.1. The only change needed for this formu-
tuations for the frictionless case, it is this case which sets the lation to be applicable here is the minor realignment of the s
minimum crown height. To see this, consider Fig. 5. This depicts coordinate on the blade in 关1兴 so that is now follows the crowning.
an exaggerated crown in the form of a circular arc of central 2.2 Finite Element Analysis. To police the contact in-
height ␦ spanning the maximum available contact extent 2a be- equalities in the foregoing formulation within a finite element
tween CC ⬘ of Fig. 1共a兲. At the ends of this arc, the crown surface analysis, we use the point-to-surface contact elements of ANSYS
is smoothly blended into the blade 共this blending is not shown in 关11兴, CONTAC 48. There are some input parameters required to
Fig. 5兲. Even with such blending, we want to avoid contact with run CONTAC 48. Here we take these to be the same as in 关1兴,
this part of the blade. Otherwise we can expect a normal contact Section 3.2.
stress with large peaks at the edges of contact 共cf., Fig. 3兲. Under Following ANSYS 关11兴 recommendations for the analysis of
load, therefore, we want the actual contact extent, 2a a , to be contact problems, we use the CONTAC 48 elements in conjunc-
distinctly less than 2a, say about 90% of 2a. And we want this tion with ANSYS four-node quadrilateral elements, PLANE 42.
for the maximum normal load, and this occurs for the frictionless The latter are arranged in a sequence of three grids to enable
case. Hence this case sets ␦. convergence to be checked in accordance with 共6兲, 共7兲. The coarse
More specifically, the elastic approach ␦ a under load 共Fig. 5兲 grid has relatively small elements arranged uniformly in the criti-
can be estimated from Hertzian formulas 共see, e.g., Johnson, 关10兴, cal areas bordering the contact elements 共Fig. 6; no crown is in-
Appendix 3兲. Thus cluded in this figure, but for ␦ 0 it would be barely discernible
anyway, even in the closeup兲. The remainder of this grid is
4 共 1⫺ ␯ 2 兲 N
␦ a⫽ , (10) meshed with elements of increasing size as they get further from
␲E the contact elements. This is done with the convenient mesh-
generation command AMESH 共ANSYS, 关11兴兲. Grid refinement
4
For an in-plane geometry like that of Fig. 1共a兲, Evans 关8兴 suggests barreling a within the uniform element region is completely systematic with
contact flat to position the line of action of N. While such barreling profiles consist of
curved arcs as do precision crowns, the heights involved are considerably larger.
element sides being successively halved to form successive grids.
Pape and Neu 关9兴 also employ a similar curved geometry, with a far larger height Grid refinement with the AMESH command follows a similar
than that for precision crowning, in fretting test pieces. sizing pattern but is not completely systematic. Thus the number
5
In 3D, the exacting convergence checks of 共6兲, 共7兲 can be expected to involve of elements only roughly quadruples with grid refinement. Table 1
two-three orders of magnitude more degrees-of-freedom than in 2D.
6
As an aside here, we report that we did consider a third tactic for reducing
gives the actual numbers of four-node quadrilateral elements
fluctuations in ␴ h , namely varying ␣ of Fig. 2. We could not find an ␣ which used. The corresponding numbers of contact elements used are
resulted in any real reduction. comparable.
Table 1 Summary of finite element grids Table 3 Reduction in hoop stress fluctuations with in-plane
presicion crowning „␮Ä0.4…
Number of Elements
⌬␴h /␴0 , ⌬␴h /␴0 ,
Grid In-Plane Analysis Out-of-Plane Analysis ␦/␦0 20% Unload 40% Unload
Coarse 915 800 0 13.2 20.7
Medium 2,604 3,200 1 3.0 4.4
Fine 8,314 12,800 2 3.7 5.9
4 4.4 7.4
2.3 Results. For maximum load and ␮⫽0, ␦ ⫽ ␦ 0 , the peak,

normal, contact stress is converging in the sense of 共6兲, and con-
verged in the sense of 共7兲 to within ␧⫽0.002 共0.2%兲. The resulting included for comparison are corresponding values without crown-
␴ c is actually the stress shown in Fig. 3 with a broken line. It acts ing 共from 关1兴兲.
over 88% of 2a, in close agreement with the prediction from Confirmation of the Hertzian nature of the normal contact stress
Hertz theory 共87兲. Moreover, it is practically indistinguishable can be obtained by comparing peak values. From Hertz theory
from a Hertzian contact stress distribution over 2a a . That is, in 共see, e.g., Johnson, 关10兴, Appendix 3兲,
terms of the coordinates in Fig. 1共a兲, this stress is essentially
given by ␴ max 兩 / ␴ max 兩 ⫽ 冑␦ 1 / ␦ 2 , (14)
冑冉冊
c c
2N z 2 ␦1 ␦2
␴ c⫽ 1⫺ . (12)
␲aa aa where ␦ 1 is one crown height, ␦ 2 another. With ␦ doubling, there-
With 共12兲, the central maximum value in terms of the original, fore, the ratio of the second ␴ max
c to the third, as well as the ratio
nominal, normal, contact stress of 共3兲 can be expressed by of the third to the fourth, should be 1/& in Table 2. In fact, both
ratios are 0.7, in good agreement with this Hertzian prediction.
␴ max
c ⫽1.4␴ c .
¯ (13) With crowning, the peak shear stresses of Table 2 satisfy 共8兲 to
two figures. This is better agreement with this shear check than
This is more than a factor of four lower than ␴ max
without crown-
c obtained in 关1兴 without crowning. Presumably this is because
ing 共cf., 共5兲兲. Also with 共12兲, bending contributions have been stresses are less concentrated with crowning.
completely removed.
Both ␴ max
c and ␴ max
h values are significantly lowered in Table 2
Given the preceding close accord with Hertz theory, it is not
surprising that for ␦ ⫽ ␦ 0 /2 we get edge-of-contact peaks in ␴ c . by crowning. Even for a crown height of 4 ␦ 0 , both are lowered by
Hertz theory has that this happens whenever ␦ ⭐0.75 ␦ 0 共see a factor of two.
共11兲兲, and indeed it does. The values of the peak stresses so pro- While promising, the results of Table 2 are not the critical ones.
duced are uneven, revealing the presence of a bending contribu- Rather it is the fluctuations in hoop stresses that attend unloading,
tion. While stresses for the larger of the peaks have yet to con- ⌬ ␴ h , that are key. Table 3 presents values for ⌬ ␴ h for 20 and
verge on the sequence of grids used here, the fine grid value is 40% unloading for the varying crown heights. Also included for
comparison are corresponding values without crowning 共from the
already a factor of two greater than ␴ maxc of 共13兲. Further grid analysis for 关1兴兲. Clearly these fluctuations are markedly reduced.
refinement can only be expected to increase this factor. As antici- Even for a crown height of 4 ␦ 0 , reductions approach a factor of
pated, then, crowning which leads to contact regions extending off three. These results suggest that, here, in-plane precision crown-
the circular crown arc should be avoided. ing would be effective with crown heights in the range
For maximum load and ␮⫽0.4, ␦ ⫽ ␦ 0 , 2 ␦ 0 , 4 ␦ 0 , the peak
normal and shear contact stresses are converging in the sense of ␦ ⫽2 ␦ 0 ⫾ ␦ 0 . (15)
共6兲 and converged in the sense of 共7兲 to within ␧⫽0.002 共0.2%兲.
The peak hoop stress is converging in the sense of 共6兲 and con- Hence for the present dovetail attachment, precision crowning
verged in the sense of 共7兲, but only to within ␧⫽0.04 共4%兲. There could be implemented with a 50% tolerance in the crown height.
is a step discontinuity in the derivative of the hoop stress at its
peak which slows convergence.7
Even with friction present, ␴ c for all crown heights at maxi- 3 Out-of-Plane Precision Crowning
mum load continue to closely follow Hertzian distributions. Thus
no bending is present. Further, ␶ c also follow Hertzian distribu- 3.1 Loading Mechanism. While the means of loading for
tions 共cf., 共8兲兲. the in-plane section of Fig. 1共a兲 is clear, the means of loading for
Peak values of the contact stresses at maximum load when the out-of-plane section of Fig. 1共d兲 is not so clear. Though a
␮⫽0.4 are given in Table 2 for the varying crown heights. Also pressure is to be expected between the upper disk section and the
lower blade section in Fig. 1共d兲, it is not obvious what agency
produces this pressure, at least from Fig. 1共d兲. In Fig. 1共a兲, how-
ever, this agency can be discerned. Therein the dotted segment of
the disk can be thought of as acting like a cantilever beam with a
Table 2 Stresses at maximum load with in-plane precision
crowning „␮Ä0.4… uniform distributed load on its under side. From beam theory,
then, we have the standard result
␦/␦0 ␴ cmax/␴0 ␶ cmax/␴0 ␴ max
h /␴0 dV
0 9.0 3.5 8.0 p⫽⫺ , (16)
dz
1 2.3 0.9 3.0
2 3.2 1.3 3.7 wherein p is the pressure due to varying shear force, V, and z is as
4 4.5 1.8 4.0
in Fig. 1共a兲. Thus the basic loading mechanism for the section of
Fig. 1共d兲 is shear transfer out of the plane of this section.
To explain further, Fig. 7 shows a uniform nominal pressure p
with the stress resultants it induces acting on a geometry corre-
sponding to the dotted segment of Fig. 1共a兲. With these resultants
7
A similar discontinuity in ␴ h is present in the solution in Poritsky 关6兴. as Saint-Venant end conditions, it is straightforward to construct a
Fig. 7 Nominal pressure and stress resultants acting on a
segment of the disk above contact flat
two-dimensional, elastic, stress field which reflects shear load

transfer. The approach for this construction is described in Ti-
moshenko and Goodier 关12兴, Chap. 3. Thus
Fig. 8 Comparison of normal stress with shear load transfer
h⫺2y and simulating body force
␴ z ⫽p 关 30共 z⫹a 兲 2 ⫹3h 2 ⫺5 共 h⫺2y 兲 2 兴 ,
10h 3
(17)
h⫹2y h⫺y
␴ y ⫽⫺p 共 h⫺y 兲 2 , ␶ yz ⫽⫺6p y 共 z⫹a 兲 , where ⵜ 2 is the Laplacian operator in x, y-coordinates.
h3 h3 To proceed with a plane-strain analysis of the section in Fig.
1共d兲, we need to provide loading. This loading comes from shear
with all other stress components being zero, where h is the height transfer and fields like that of 共17兲. Clearly these are not plane-
of the segment 共Fig. 7兲, and y and z are as in Fig. 1共a兲. This stress strain fields in the xy-plane. Actually they are plane stress fields
field complies with the stress equations of equilibrium and com- in the yz-plane.
patibility and the boundary conditions indicated in Fig. 7 on y To introduce the effects of 共17兲 in a way which is compatible
⫽0, h, as well as reproducing the stress resultants indicated in with plane strain in the xy-plane, we simulate the shear transfer
Fig. 7 on z⫽⫾a. That this is so may be verified directly by with a body force, F y . We take F y to be as simple as possible
substitution. Away from the ends z⫽⫾a, as at the central section which yet produces a distribution similar to ␴ y of 共17兲. We also
BB ⬘ , we can expect the stresses in 共17兲 to be physically represen- set Poisson’s ratio ␯⫽0 so that the only displacement attending
tative of the stresses attending shear transfer. this F y and ␴ y is v , and this v is a function of y alone. Then these
Associated displacements for the stress field of 共17兲 are those in three elements in the simulation are admissible contributors to a
the y and z directions, v and w, respectively. These are both func- plane-strain state in the xy-plane.
tions of y and z alone 共i.e., independent of x of Fig. 1共d兲兲. Specifically the fields we take for simulating shear transfer are
Complementary fields are active within the blade section of Fig. thus as follows, distinguished by a superscript s. For 0⭐y/h
1共a兲. Together these fields produce a nominal contact pressure p ⭐1/10,
for the section in Fig. 1共d兲. py
␴ sy ⫽⫺p, F y ⫽0, v s ⫽⫺ ; (20)
3.2 Plane Strain With Simulated Shear Transfer. The ex- E
pectation for the central section BB ⬘ of Fig. 1共a兲 shown in Fig.
for 1/10⭐y/h⭐9/10,
1共d兲 is that fields do not vary appreciably in the out-of-plane di-
rection of Fig. 1共d兲. That is, in terms of the coordinates in Fig. 1, p 5p
⳵ / ⳵ z is a null operator in Fig. 1共d兲. This is the essential condition ␴ sy ⫽ 共 10y⫺9h 兲 , F y ⫽⫺ ,
8h 4h
for a state of plane strain in the xy-plane. Accordingly we look to (21)
use a plane strain analysis for the section in Fig. 1共d兲. p
In the xy-plane, a plane-strain state has that the only displace- v s⫽ 共 100y 2 ⫺180yh⫹h 2 兲 ;
160Eh
ments are those in the x and y directions, u and v , respectively.
Furthermore, these displacements are functions of x and y alone. for 9/10⭐y/h⭐1,
Then the existing companion stress components are ␴ x , ␴ y , ␶ xy ,
ph
and ␴ z . These stresses satisfy the following field equations: the ␴ sy ⫽0, F y ⫽0, v s ⫽⫺ ; (22)
stress equations of equilibrium, 2E
␴ x,x ⫹ ␶ xy,y ⫹F x ⫽0, with all other stresses and displacements being zero. Correspond-
(18) ing fields act in the blade section (⫺1⭐y/h⭐0) with F y replaced
␴ y,y ⫹ ␶ xy,x ⫹F y ⫽0, by ⫺F y . In 共20兲–共22兲 and these corresponding fields, h is now
taken so that the cross section these fields apply to spans BB ⬘ of
where coordinates preceded by a comma denote partial differen- Fig. 1.
tiation with respect to that coordinate and F x , F y are body-force Figure 8 presents a comparison of ␴ sy with ␴ y of 共17兲. Typically
components; and the stress equations of compatibility, the two agree to within 2% of p: The maximum discrepancy be-
tween the two is 3.5%. This sort of agreement is judged to be
ⵜ 2 共 ␴ x ⫹ ␴ y 兲 ⫽0, ␴ z⫽ ␯共 ␴ x⫹ ␴ y 兲, (19) sufficient for the proof-of-concept analyses being undertaken here.
In combination then we have the plane-strain stresses for the
cross section in Fig. 1共d兲 determined via finite element analysis
together with the simulated shear transfer stress. That is, on dis-
tinguishing the former by the superscript f,
␴ x ⫽ ␴ xf , ␴ y ⫽ ␴ yf ⫹ ␴ sy , ␶ xy ⫽ ␶ xy
f
, (23)
with F x ⫽0, F y as in 共20兲–共22兲 times signum 共y兲, and ␯⫽0. These

stresses satisfy 共18兲, 共19兲 with ␴ z ⫽0 because ␯⫽0 rather than its
actual value. We do not expect, though, that setting ␯⫽0 effects
the in-plane stresses significantly: This is something we check
with finite element analysis. In fact, with F y prescribed to be as in
共20兲–共22兲 times signum 共y兲, the finite element analysis also cal-
culates ␴ sy . This is also something we check and see how well
finite elements do this.
3.3 Crown Heights. For the analysis of the cross section in

Fig. 1共d兲 with and without crowning, we restrict attention to the
frictionless case ␮⫽0. This is because when ␮⫽0 a shear stress in
induced which acts out of the plane of Fig. 1共d兲, and therefore
cannot be captured with a plane-strain analysis. However, as in
Section 2.1, ␮⫽0 is the critical case when sizing crown heights if
one is to prevent contact expanding out to the blending at the ends Fig. 9 Coarse finite element grids for out-of-plane analysis: „a…
of crowns. without crown, „b… with circular crown
The basic crown profile continues to be as show schematically
in Fig. 5. That is, a circular arc blended into the blade at its ends
共this blending not being shown in Fig. 5兲. Now, though, crown
heights need to be considerably larger if contact is not to expand They are, though, perhaps more appropriate than stress-free con-
off the circular arc. This is because of bending contributions to the ditions because they would be correct if the section BB ⬘ inter-
deformation. These contributions are not accounted for in 共10兲, sected the centerline at ␲/2 instead of ⬃␲/4.
共11兲, and consequently these equations cannot be used to size The foregoing formulation can be perturbed to accommodate
crown heights. Instead an actual effective value of crown height either of the crown profiles considered simply by altering the
can be found using finite element analysis. Such analysis leads to blade surface at y⫽0 to conform to these profiles. With these
perturbations, the blade occupies a region wherein ⫺h⭐y⭐ ␦
a baseline, out-of-plane, crown height, ␦ 0⬘ , of 5 mm 共1/5 in兲. This
while the disk now occupies a region wherein ␦ ⭐y⭐ ␦ ⫹h, where
results in contact over 81% of C ⬙ C ⵮ at maximum load.
This baseline crown height is sizable within the confines of
␦ is either ␦ ⬘0 or ␦ ⬘0 /4 depending on the profile.
dovetail attachments in jet engines. To provide a smaller, out-of-
plane, crown height, we introduce a flat on the central portion of 3.4 Finite Element Analysis. As previously, we use the
the crown profile which then smoothly blends into two outer cir- point-to-surface contact elements of ANSYS, 关11兴, CONTAC 48,
cular arcs with a common radius 共indicated by the broken line in in conjunction with ANSYS four-node quadrilateral elements,
Fig. 5兲. In a first analysis, we take the extent of the flat to be a, or PLANE 42. The latter are arranged in sequences of grids to enable
half of the available contact region. This greatly reduces bending convergence to be checked via 共6兲, 共7兲.
contributions so that the crown height is reduced to ␦ 0⬘ /4 共again, For the original cross section, high stresses are expected near
though, this is something that needs to be checked with finite the outer blend radii 共cf., the original in-plane section兲. To try and
element analysis兲. capture these stresses, the coarse grid for this cross section has
With or without either crown profile, the symmetry of the sec- relatively small elements near its blend radius 共Fig. 9共a兲兲. The
tion in Fig. 1共d兲 enables analysis to be confined to x⭓0. Without coarse grids with crowning are not expected to need this type of
crowning, the stress analysis problem for the section in Fig. 1共d兲 refinement and thus are more uniform in their element sizes 共e.g.,
can then be stated as follows. Throughout this half section we Fig. 9共b兲 for ␦ ⬘0 ). Both types of grid have a completely uniform
seek the plane-strain stresses ␴ x , ␴ y , ␶ xy satisfying: the stress arrangement of elements throughout most of their vertical extents
equations of equilibrium 共18兲 with F x ⫽0, F y as in 共20兲–共22兲 away from the contact surfaces to facilitate the introduction of the
when y⭓0, and an equal and opposing F y when y⭐0; the stress body force simulating shear transfer. As a result, grid refinement is
equation of compatibility, 共19兲 with ␯⫽0; frictionless contact con- systematic 共to the nearest 100 of elements兲. The actual numbers of
ditions, which match normal stresses and displacements while set- four-node quadrilateral elements used are included in Table 1 for
ting shears to zero, on y⫽0 between x⫽0 and C ⵮ ; symmetry the original cross section 共i.e., for a grid sequence with coarse grid
conditions on x⫽0 for ⫺h⭐y⭐h; stress-free conditions on the as in Fig. 9共a兲兲. Fewer quadrilateral elements are used for the
remainder of the boundary of this half-section; and contact in- crowned profiles 共about 40% less兲, but grid refinement remains
equalities, which prohibit tensile contact stresses within the con- systematic. The numbers of contact elements used are comparable
tact region and overlapping of displacements without. to the numbers of quadrilateral elements for respective grids.
The stress-free conditions in the preceding are not really appro- In addition to the foregoing grids, we run the three grids for the
priate on the bottom of the blade 共i.e., at y⫽⫺h). This is because original cross section with the blend radius removed. Then, with
this ‘‘edge’’ is actually still inside the blade, being at B in Fig. perfect alignment of the disk and blade edges, the stress field
1共a兲. We do not expect the choice of boundary conditions at this produced reduces to simply ␴ sy of 共20兲–共22兲. This checks the abil-
location to be that significant, however. This is because the body ity of our finite element analysis to calculate ␴ sy .
force in the blade is yet to start acting when y⫽⫺h. As a check For the coarse grids of Fig. 9, we run our finite element analysis
on this expectation, we also enforce symmetry conditions on this with Poisson’s ratio equal to 0.0 and 0.3. This checks the effects
edge. Symmetry conditions are also not really appropriate here. on stresses of making the simplification ␯⫽0. We also run these
The reductions in ␴ c attending either crown profile can be ex-
pected to reduce the ␶ c out of the plane of Fig. 1共d兲 induced
during loading up. Hence these profiles can be expected to reduce
␴ h of Fig. 1共b兲, and thus fluctuations in the hoop stress at the
edges of contact.
4 Concluding Remarks
In dovetail attachments, fluctuations in hoop stresses at the
edges of contact can occur. A pinching mechanism can amplify
these changes in hoop stresses over those in loading. As a result,
fluctuations in hoop stresses can be more than an order of magni-
tude greater than variations in loading. Moreover, these fluctua-
tions include hoop stresses which are tensile at maximum load and
compressive at reduced load. Such fluctuating hoop stresses can
be expected to be a major contributor to fatigue failure in dovetail
attachments. The principal objective of the work described here is
to offer a method of alleviating these fluctuations.
The peak hoop stresses at maximum load largely result from the
shear contact stresses. These stresses in turn are directly propor-
tional to the normal contact stresses. Thus reducing the normal
contact stresses at maximum load should reduce the hoop stresses
Fig. 10 Out-of-plane contact stress distributions „␮Ä0… that can fluctuate. This is the tactic adopted here, with crowning
being chosen as the method for reducing contact stresses.
Crowns in both in-plane 共Fig. 1共a兲兲 and out-of-plane 共Fig. 1共d兲兲
directions are analyzed via finite elements. The finite element
analysis employs point-to-surface contact elements to ensure con-
grids with stress-free and symmetry conditions on their lower vergent edge-of-contact stresses. For the in-plane configuration, a
edges. This checks the effects of boundary condition choice. state of plane strain applies keeping analysis two-dimensional. For
the out-of-plane configuration, plane strain is not strictly appli-
3.5 Results. For our first check on the finite element calcu- cable. However, by simulating shear load transfer with body
lation of ␴ sy of 共20兲–共22兲, we find results from all three grids lie forces, it can be approximately treated as such. While it seems
on top of the analytical result except near the two discontinuities likely that this approach yields physically reasonable stresses, ul-
at y/h⫽1/10, 9/10. There results are converging in the sense of timately this is something that needs to be checked with a full
共6兲 as indicated in the closeup in Fig. 8. They have converged in three-dimensional analysis. Given confirmation, plane strain with
the sense of 共7兲 to within ␧⫽0.008 provided 兩 ␴ Fc 兩 on the right- simulated shear transfer then furnishes stress analysts with a
hand side of 共7兲 is replaced by p 共to avoid dividing by zero when means of keeping analysis two-dimensional in dovetail attach-
estimating ␧ for y/h⫽9/10). That is, finite element results for ␴ sy ments, with consequent reductions in computational effort.
have converged to within 0.8% of p, an excellent level of accuracy Results from crowning in both directions indicate that signifi-
for present purposes. cant reductions in hoop stress fluctuations can be achieved via this
For our check on the effects of Poisson’s ratio, we find results means. For the in-plane direction, crown heights are small. Hence
for ␴ c with ␯⫽0, 0.3 to be close 共within 5%兲. This is not unex- some precision is expected to be required in their manufacture.
pected. Peak contact stresses in Hertzian contact differ by only 5% Nonetheless, even with crown heights varying by ⫾50%, hoop
when ␯ varies from 0 to 0.3 共see, e.g., Johnson, 关10兴, Appendix 3兲. stress fluctuations can be reduced by a factor of three. For the
For our check on boundary condition choice, we find no differ- out-of-plane direction, quite a range of crown profiles is available
ences in the contact stress with stress-free or symmetry conditions which can reduce hoop stresses by a further factor of two. All told,
on the lower edge of the original cross section of Fig. 1共d兲. With then, crowning shows some promise of alleviating hoop stress
crowning, though, stresses are about 8% higher with the symme- fluctuations in dovetail attachments.
try condition. This is because it restrains bending deformation.
Henceforth, therefore, we only report crowning results with sym- Acknowledgments
metry conditions, thereby minimizing any reductions in ␴ c af-
forded by crowning. We are pleased to acknowledge the benefit of technical discus-
At maximum load, results for the contact stress for the original sions with engineers at General Electric Aircraft Engines 共GEAE兲.
cross section are shown in Fig. 10. These stresses are for the fine We are also grateful for the financial support of this work by
grid. The peak stresses for this cross section are not converging in GEAE.
the sense of 共6兲 on the grid sequence used. Further grid refinement
can only be expected to increase the peak contact stress for this References
cross section. 关1兴 Sinclair, G. B., Cormier, N. G., Griffin, J. H., and Meda, G., 2002, ‘‘Contact
At maximum load, results for the contact stress for the circular Stresses in Dovetail Attachments: Finite Element Modeling,’’ ASME J. Eng.
crown of height ␦ 0⬘ are included in Fig. 10. These are for the fine Gas Turbines Power, 124, pp. 182–189.
关2兴 Sinclair, G. B., and Cormier, N. G., 2002, ‘‘Contact Stresses in Dovetail At-
grid. The peak stresses for this crown are converging in the sense tachments: Physical Modeling,’’ ASME J. Eng. Gas Turbines Power, 124, pp.
of 共6兲, and have converged in the sense of 共7兲 to within ␧⫽0.005. 325–331.
Peak values of ␴ c with this crown are a factor of 2.7 less than the 关3兴 Hamdy, M. M., and Waterhouse, R. B., 1981, ‘‘The Fretting Wear of Ti-
low estimate of peak ␴ c for the original cross section. 6A1-4V and Aged Inconel 718 at Elevated Temperatures,’’ Wear, 71, pp. 237–
248.
At maximum load, results for the contact stress for the crown 关4兴 Sinclair, G. B., 1998, ‘‘FEA of Singular Elasticity Problems,’’ Proceedings of
with a flat and a height of ␦ ⬘0 /2 are also included in Fig. 10. These the Eighth International ANSYS Conference, ANSYS Inc., Canonsburg, PA, 1,
are from only a medium grid because of its close agreement with pp. 225–236.
关5兴 Cormier, N. G., Smallwood, B. S., Sinclair, G. B., and Meda, G., 1999, ‘‘Ag-
coarse grid results 共e.g., peak stresses differ by less than 1.5%兲. gressive Submodelling of Stress Concentrations,’’ Int. J. Numer. Methods
Peak values of ␴ c with this crown are still a factor of two less than Eng., 46, pp. 889–909.
the low estimate of peak ␴ c for the original cross section. 关6兴 Poritsky, H., 1950, ‘‘Stresses and Deflections of Cylindrical Bodies in Contact
With Application to Contact of Gears and Locomotive Wheels,’’ ASME J. Fretting Fatigue Testing,’’ Wear, 225–229, pp. 1205–1214.
Appl. Mech., 17, pp. 191–201. 关10兴 Johnson, K. L., 1985, Contact Mechanics, Cambridge University Press, Cam-
关7兴 Hertz, H., 1882, ‘‘On the Contact of Elastic Solids,’’ Journal für die reine und bridge, UK.
angewandte Mathematik, 92, pp. 156 –171 共in German; for an account in En- 关11兴 ANSYS personnel, 1995, ANSYS User’s Manual, Revision 5.2, I, ANSYS Inc.,
glish, see Johnson, 关10兴, Chap. 4兲.
Canonsburg, PA.
关8兴 Evans, N. M., 1992, ‘‘Attachment of a Gas Turbine Engine Blade to a Turbine
Rotor Disc,’’ Patent No. 5, 110, 262, U.S. Patent Office. 关12兴 Timoshenko, S. P., and Goodier, J. N., 1970, Theory of Elasticity, 3rd Ed.,
关9兴 Pape, J. A., and Neu, R. W., 1999, ‘‘Influence of Contact Configuration in McGraw-Hill, New York.
Computation of the Optimal
Normal Load of a Friction Damper
D. Cha Under Different Types of
A. Sinha Excitation
Department of Mechanical Engineering,
The Pennsylvania State University,
In this paper, forced responses of a frictionally damped turbine blade are investigated for
University Park, PA 16802
three different types of excitation: white noise excitation, narrowband random excitation,
and deterministic sinusoidal excitation. To determine the steady-state nonlinear response,
the harmonic balance method is used for sinusoidal excitation, and the equivalent linear-
ization method is used for white noise and narrowband random excitations. Using a new
set of nondimensionalized variables, the optimal value of normal load of a friction
damper is found to be almost independent of the nature of excitation. The effectiveness of
the damper in reducing the vibration level is also examined for the aforementioned three
different types of excitation. 关DOI: 10.1115/1.1584474兴
1 Introduction integration because the accuracy of HBM has been established in

the literature. Lastly, the nondimensionalized variables are used to
The fatigue failure of turbine blades can be caused by excessive
obtain the optimal value of normal load.
resonant stresses. In order to reduce the resonant amplitude of a
turbine blade, friction dampers are used in practice. In many situ-
ations, a friction damper can be designed as a blade-to-ground
共B-G兲 damper which is composed of a linear spring and a Cou- 2 Formulation
lomb friction element, as shown in Fig. 1, 关1兴. The B-G damper Let m t , k, and c be the mass, stiffness, and damping coefficient,
provides slip between points experiencing relative motion and dis- respectively. The governing differential equation of motion for the
sipate the system energy. The single-degree-of-freedom model system shown in Fig. 2 can be written as follows:
used for the investigation, shown in Fig. 2, represents the funda-
mental mode of a blade with B-G friction damper. m t ẍ⫹cẋ⫹kx⫹ f G ⫽F 共 t 兲 (1)
Many researchers have worked on the calculation of the opti-
where f G is the hysteretic force of the friction damper and can be
mal normal load of a friction damper. While the harmonic balance
defined as follows:
method 共HBM兲 is widely used for sinusoidal excitation, 关1–5兴, the
共i兲 when friction damper is stuck: K G 兩 x⫺d G 兩 ⭐ ␮ f F N
equivalent linearization method 共ELM兲 is commonly used for the
white noise excitation to investigate the steady-state nonlinear re- f G ⫽K G 共 x⫺d G 兲 (2a)
sponse, 关6 – 8兴. It has been reported that turbine blades can expe-
rience random loads, 关9–11兴. However, the system responses to 共ii兲 when friction damper is slipping: K G 兩 x⫺d G 兩 ⫽ ␮ f F N
random and deterministic excitations are different in nature.
Hence the optimal normal load obtained for one type of external f G ⫽ ␮ f F N sign共 ḋ G 兲 (2b)
force may not work effectively for other types of excitation. where d G , K G , ␮ f , and F N represent the displacement, the stiff-
In this paper, three different types of excitation are considered: ness, the friction coefficient and normal load of friction damper,
white noise excitation, narrowband random excitation, and sinu- respectively. For such systems, analytical calculation of the re-
soidal excitation. Using a new set of nondimensionalized vari- sponse depends on the characteristic of external force, F(t).
ables, the performance of the friction damper is plotted under
different loading conditions. The resulting curves of responses Case I: White Noise Excitation. If the external force, F(t),
with these variables are independent of the intensity of external is assumed as an ergodic and Gaussian white noise, w(t), with
force and are not sensitive to the change in the frequency band of zero mean, the equivalent linearization method 共ELM兲 can be
excitation. Therefore, the optimal value of normal load can be used to obtain the statistics of response. The autocorrelation of the
easily obtained irrespective of the excitation characteristics. It is external force is given as follows:
noted that this approach can be easily applicable in the actual R w 共 ␶ 兲 ⫽E 关 w 共 t 兲 w 共 t⫹ ␶ 兲兴 ⫽Q 0 ␦ 共 ␶ 兲 (3)
situation, because instead of the measurements of external forces,
only the experimental data of response when the friction damper where Q 0 is the intensity, ␦共␶兲 is the dirac-delta function, and
is fully stuck are required. E 关 • 兴 is the expected value.
First, the ELM and HBM are briefly described. Then, the ana- The nonlinear force of the friction damper can also be de-
lytical results from ELM only are compared with those from time scribed, 关6兴, as
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN f G 共 ẋ,y G ;x c ,K G 兲 ⫽K G y G 关 U 共 y G ⫹x c 兲 ⫺U 共 y G ⫺x c 兲兴
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- ⫹K G x c 关 U 共 y G ⫺x c 兲 U 共 ẋ 兲
tional Gas Turbine and Aeroengine Congress and Exhibition, Indianapolis, IN, June
7–10, 1999; ASME Paper 99-GT-413. Manuscript received by IGTI, Oct. 1998; final ⫺U 共 ⫺y G ⫺x c 兲 U 共 ⫺ẋ 兲兴 (4)
revision received by the ASME Headquarters, Mar. 1999. Associate Editor: D.
Wisler. where
c Gy ⫽E 冋册
⳵hG
⳵ ẋ
, (7c)
k Gy ⫽E 冋册
⳵hG
⳵yG
. (7d)
These coefficients are evaluated by assuming a joint Gaussian

probability distribution of the multidimensional response process
and presented in the Appendix A.
Introducing the state variables, z(t), the Eq. 共6a兲 and 共6b兲 can
be written as
ż⫽Az⫹Bw 共 t 兲 (8)
where
冋册
z T ⫽ 关 x ẋ y G 兴 T (9a)
0 1 0
⫺k ⫺ 共 c⫹c G
x
兲 ⫺k G
x
A⫽ , (9b)
Fig. 1 A blade-to-ground damper mt mt mt
c Gy k Gy
冋册
0
ẏ G ⫽h G 共 ẋ,y G ;x c 兲 0
1
⫽ẋ 关 U 共 y G ⫹x c 兲 ⫺U 共 y G ⫺x c 兲 ⫹U 共 y G ⫺x c 兲 U 共 ⫺ẋ 兲 B⫽ . (9c)
mt
⫹U 共 ⫺y G ⫺x c 兲 U 共 ẋ 兲兴 (5a) 0
y G ⫽x⫺d G (5b) Since E 关 w(t) 兴 ⫽0, it has been shown that E 关 z(t) 兴 ⫽0. Let P be
the correlation matrix of state variables: i.e.,
␮ fFN
x c⫽ . (5c) P⫽E b z 共 t 兲 z 共 t 兲 T c . (10)
KG
Note that U(•) is the unit step function. Using ELM, the equa- Then the correlation matrix, P, satisfies the following algebraic
tions of motion, Eq. 共1兲 and Eq. 共5a兲, can be expressed as equation under steady-state conditions.
A P⫹ PA T ⫽B R (11)
x
m t ẍ⫹cẋ⫹kx⫹c G x
ẋ⫹k G y G ⫽w 共 t 兲 (6a)
where
ẏ G ⫽c Gy ẋ⫹k Gy y G (6b)
B R ⫽⫺BQ 0 B T . (12)
where
冋册
Note that the variance of the blade’s displacement, R x , is ob-
⳵fG tained from the correlation matrix of the system response P: i.e.,
x
cG ⫽E , (7a)
⳵ ẋ
R x ⫽ ␴ 2x ⫽ 共 P 兲 1,1 .
冋册
(13)
⳵fG x x
x
kG ⫽E (7b) Because the linearized coefficients, cG , kG
, c Gy
, and k Gy
, are
⳵yG 2
functions of E b ẋ 2 c , E b y G c , and E 关 ẋy G 兴 , which are the elements of
correlation matrix P, Eq. 共11兲 is solved numerically to evaluate
the correlation matrix P, 关7兴.
Case II: Narrowband Excitation. If the external excitation,
F(t), is assumed as a narrow band excitation, g(t), which can be
generated by passing the white noise excitation, w(t), through the
bandpass filter: i.e.,
g̈⫹2 ␰ F ␻ F ġ⫹ ␻ F2 g⫽ ␻ F2 w 共 t 兲 (14)
where ␰ F and ␻ F are damping ratio and natural frequency of the
bandpass filter, respectively. Using ELM, the governing differen-
tial equation, Eq. 共1兲 and Eq. 共5a兲, are given by
x
m t ẍ⫹cẋ⫹kx⫹c G x
ẋ⫹k G y G ⫽g 共 t 兲 (15a)
ẏ G ⫽c Gy ẋ⫹k Gy y G . (15b)
If the state vector is selected as
z T ⫽ 关 g ġ x ẋ y G 兴 T , (16)
then Eqs. 共14兲, 共15a兲, and 共15b兲 can be represented in the state-
space form as
ż⫽Az⫹Bw 共 t 兲 (17)
Fig. 2 Equivalent single-degree-of-freedom system where
Fig. 3 Hysteretic characteristic of a friction damper „sinusoid- Fig. 4 A p as a function of x c „sinusoidal excitation…
al response…
A p , and the resonant frequency, ␻ p , can be obtained, 关1兴. The

numerical procedure is presented in Appendix B. The HBM is
冤冥
0 1 0 0 0 frequently used for the analysis of response to sinusoidal excita-
⫺ ␻ F2 ⫺2 ␰ F ␻ F 0 0 0 tion and its accuracy has been proved in previous papers, 关1,4,12兴.
The peak response of the blade, A p , is plotted in Fig. 4 as a
0 0 0 1 0 function of x c when the amplitude of the sinusoidal excitation, f 0 ,
A⫽ (18a) is chosen to be 1 N.
1 ⫺k ⫺ 共 c⫹c G
x
兲 ⫺k G
x
0
mt mt mt mt 3 Accuracy of Equivalent Linearization Method
0 0 0 c Gy k Gy „ELM…
B⫽ 关 0 ␻ F2 0 0 0 兴 T (18b) The computational efficiency of ELM makes it an attractive
approach to find the response of a nonlinear system. However, its
Therefore, Eq. 共11兲 can be used again to calculate the correla- accuracy depends on the system characteristics because the ELM
tion matrix P. is an approximation to replace nonlinear terms by linear ones
which minimize the error between them, 关13兴. Therefore, the ana-
Case III: Deterministic Sinusoidal Excitation. When the
lytical results by ELM are compared with those from the numeri-
external force is assumed to correspond to a particular engine
cal integration of the differential Eq. 共1兲. The nominal values of
order excitation,
modal parameters are given in Table 1 and are representative of a
F 共 t 兲 ⫽ f 0 cos共 ␻ t 兲 , (19) high pressure turbine stage, 关14兴. The analytical results are ob-
tained by solving Eq. 共11兲 iteratively with the MATLAB routine
and the harmonic balance method 共HBM兲 is used for the analysis ‘‘lyap,’’ 关15兴. For the time integration, the MATLAB routine
of system response. It is assumed that the steady-state response is ‘‘randn’’ is used to generate the Gaussian white noise excitation. It
almost sinusoidal: i.e., yields random numbers which have Gaussian distribution with
x 共 t 兲 ⫽A x cos共 ␻ t⫹ ␾ 兲 . (20) zero mean and specified standard deviation. The time response is
generated at 30,000 instants with the time interval equal to 5
Then, in steady state, the damper force will be periodic with the ⫻10⫺5 sec. The external excitation, w(t), is assumed to be con-
fundamental frequency same as ␻. The resulting nonlinear force- stant during the time interval to facilitate the numerical integration
displacement relationship of a friction damper is shown in Fig. 3. of differential equation.
The displacement of the friction damper, d G , is expanded in a For the white noise excitation, the intensity of external force,
Fourier series and only the terms with the fundamental frequency
Q 0 , is taken to be 1 N 2 and the variance of displacement of blade,
are considered, 关1,12兴. Therefore the nonlinear force of the fric-
R x , is shown in Fig. 5 as a function of the slip distance, x c . The
tion damper can be represented as
optimal value of x c is close to 0.46⫻10⫺3 m. These results ob-
f G ⫽K G 共 x⫺d G 兲 ⫽K G 共 a d cos共 ␻ t⫹ ␾ 兲 ⫹b d sin共 ␻ t⫹ ␾ 兲兲 tained analytically from Eq. 共11兲 are compared with results from
(21) the numerical integration of differential equation. Three different
where values of slip distance, x c ⫽0.1⫻10⫺3 , 0.46⫻10⫺3 , and 2.98
⫻10⫺3 m, are chosen and the results are presented in Table 2. The
a d ⫽A x 共 ␪ c ⫺0.5 sin共 2 ␪ c 兲兲 / ␲ (22a) responses at 30,000 discrete times are divided into six batches of
5000 successive data where each batch is considered as a result of
b d ⫽⫺A x sin 共 ␪ c 兲 / ␲
2
(22b)
⫺1
␪ c ⫽cos 共 1⫺2x c /A x 兲 . (22c)
Table 1 System parameters
It should be noted that a d ⫽A x and b d ⫽0 when there is no slip
at the friction damper: i.e., A x ⭐x c where the slip distance x c is m t ⫽0.0114 kg, c⫽1.45 Ns/m, k⫽430,000 N/m, K G ⫽43,000 N/m
defined by Eq. 共5c兲. Analytically, the peak response of the blade,
Fig. 5 R x as a function of x c „white noise excitation…
an independent simulation, 关16兴. Therefore, the value of R x has

been computed for each of six batches separately. It is found that
R x of each batch fluctuates around the analytical result. However,
the average value of six batches compares well with the analytical
result.
The bandpass filter is used to generate narrow band excitation
with ␰ F equal to 0.01, Eq. 共14兲. The bandpass filter amplifies the
white noise excitation at frequencies which are close to the reso-
nant frequency of the filter ␻ F . Therefore, the exciting force to
the bladed disk system has the dominant frequency band near the
resonant frequency of the filter ␻ F . In the Fig. 6共a兲, R x for the
narrow band excitation is plotted as a function of x c and ␻ F .
Because the bandpass filter amplifies the white noise excitation, Fig. 6 „a… R x as a function of x c and ␻ F „narrowband excita-
w(t), near the resonant frequency of the filter, ␻ F , the actual tion…; „b… peak value of R x as a function of x c „narrowband
force on blade, g(t), can be increased by as much as the quality excitation…
factor of the bandpass filter, Q f , which can be defined as
1 five different points are selected to compare with the results from
Qf⫽ . (23) time integration. The results are presented in Table 3. The average
2␰F
values obtained using all 30,000 data are reasonably close to ana-
Therefore, the intensity of external white noise excitation, Q 0 , is lytical results, however, the values of six batches show large fluc-
chosen to be 10⫺4 N2 . From analytical results, shown in Fig. 6共a兲, tuation and the errors are increased compared to the results for
Table 2 The response correlation, R x , for white noise excitation „units: Ã10À7 m2 … „optimal
x c Ä0.00046 m…
Batch number 1 2 3 4 5 6 Average ELM

x c ⫽0.0001 m 7.154 5.951 5.273 6.474 4.001 4.802 5.609 5.894
x c ⫽0.00046 m 4.195 3.526 3.286 4.203 2.857 2.855 3.487 3.837
x c ⫽0.00298 m 6.628 7.177 8.373 8.484 8.246 7.387 7.716 7.247
Table 3 The response correlation, R x , for narrow band excitation „units: Ã10À8 m2 … „optimal
x c Ä0.00015 m…
Batch number 1 2 3 4 5 6 Average ELM

x c ⫽0.00005 m 5.815 4.279 3.793 6.080 3.011 3.036 4.336 4.332
␻ F ⫽6170 rad/s
x c ⫽0.00015 m 2.919 2.740 2.783 3.210 1.792 1.548 2.499 3.131
␻ F ⫽6270 rad/s
x c ⫽0.0003 m 3.501 4.119 4.330 4.645 5.041 4.225 4.310 4.802
␻ F ⫽6450 rad/s
x c ⫽0.0006 m 6.327 7.622 9.550 8.700 10.02 8.565 8.465 7.030
␻ F ⫽6420 rad/s
x c ⫽0.000975 m 6.945 8.158 13.78 9.658 13.31 9.286 10.99 8.988
␻ F ⫽6440 rad/s
sponding ␻ F is 6270 rad/s. It should be noted again that the nor-
mal load, F N , is proportional to the slip distance x c , Eq. 共5c兲.
In comparison with the results from time integration, Tables 2
and 3, the results from ELM are greater than those from numerical
simulation when x c is around that for the minimum value of R x .
It implies that the ELM seems to be underestimating the effect of
damper when normal load is close to its optimal value. When x c is
increased, the R x from ELM is smaller than that from the time
integration; i.e., the ELM overestimates the effect of friction
damper when the normal load is much greater than the optimal
value. However, the ELM is shown to be well suited in the study
of normal load optimization because the trend of analytical results
are fairly close to that of simulation results. The error of ELM
near the minimum value of R x for the narrowband excitation is
greater than that for white noise. Furthermore, the maximum er-
rors are approximately 10% and 25% for white noise and narrow-
band excitation, respectively.
4 Optimization of the Normal Load

The optimal value of the normal load, which minimizes the
response of the system, depends on the intensity and the type of
external force. Therefore, the optimum normal load can be deter-
mined easily if the external force is known. However, in the actual
situation, the intensity and frequency band of external force are
not known.
First, the normalized variables, ␴ x / 冑Q 0 and ␮ f F N / 冑Q 0 for
random excitation and A p / f 0 and ␮ f F N / f 0 for sinusoidal excita-
tion, are used to eliminate the effect of intensity of external force.
Note that Q 0 is the intensity of white noise and f 0 is the amplitude
of sinusoidal forcing function. Three different levels of external
forces are selected for each type of excitation. The intensity, Q 0 ,
are selected as 1, 5 and 10 N2 for white noise excitation and 10⫺5 ,
5⫻10⫺5 , and 10⫺4 N2 for narrow band excitation. The ampli-
tudes of sinusoidal excitation are chosen as 1, 5 and 10 N. In Fig.
7, the resulting curves are independent of the level of external
force. They can be referred to as system performance curves to
determine the optimum normal load when the type of external
force is only known. However, the optimum value of normalized
slip load is considerably different with each type of external force.
Next, the nondimensionalized variables, ␴ x / ␴ st and x c / ␴ st for
random excitation and rms(x p )/rms(x st ) and x c /rms(x st ) for
sinusoidal excitation, are used to determine the optimum normal
load. The definition of each new variable is given as follows:
共i兲 random excitation:
␴ st
2
⫽E b x st
2
c (24a)
x st ⫽response when the friction damper is fully stuck
In the case of narrow band excitation, ␻ F is chosen such that
␴ st is maximum.
(24b)
共ii兲 sinusoidal excitation:
rms共 • 兲 ⫽root mean square value (25a)
x p 共 t 兲 ⫽A p cos共 ␻ p t⫹ ␾ p 兲 (25b)
Fig. 7 „a… ␴ x Õ 冑Q 0 as a function of ␮ f F N Õ 冑Q 0 „W.N.…; „b… peak x st 共 t 兲 ⫽A st cos共 ␻ st t⫹ ␾ st 兲 (25c)

value of ␴ x Õ 冑Q 0 as a function of ␮ f F N Õ 冑Q 0 „N.B.…; „c… A p Õ f 0 as
a function of ␮ f F N Õ f 0 „S.… A p ⫽peak amplitude at a given normal load (25d)
A st ⫽resonant amplitude when the friction

damper is fully stuck (25e)
white noise excitation. To study the effect of the normal load only,
the peak values of R x in Fig. 6共a兲 are estimated for many different It should be noted that the standard deviation and the rms value
values of x c , and are plotted in Fig. 6共b兲 as a function of x c . It are mathematically equivalent when the mean value is equal to
shows a trend similar to the results for white noise excitation, Fig. zero. Therefore, the nondimensionalized variables are similar in
5. The optimal value of x c is close to 0.15⫻10⫺3 m, and corre- nature for both the random and sinusoidal excitations.
Fig. 8 ␴ x Õ ␴ st versus x c Õ ␴ st for white noise excitation, peak Fig. 9 ␴ x Õ ␴ st versus x c Õ ␴ st for white noise excitation, peak
value of ␴ x Õ ␴ st versus x c Õ ␴ st for narrow band excitation, value of ␴ x Õ ␴ st versus x c Õ ␴ st for narrow band excitation,
rms„ x p …Õrms„ x st … versus x c Õrms„ x st … for sinusoidal excitation rms„ x p …Õrms„ x st … versus x c Õrms„ x st … for sinusoidal excitation,
„ K G Ä43,000 NÕm… „ K G Ä64,500 NÕm…
Table 4 The performance of friction damper

In Fig. 8, the resulting curves show similar scales. It is also Minimum Value Near-Optimal Value
seen that the system performance with x c / ␴ st or x c /rms(x st )
equal to 0.5 is fairly good for any types of excitation. Hence, the KG Excitation x c / ␴ st ␴ x / ␴ st x c / ␴ st ␴ x / ␴ st
near-optimal normal load can be calculated as white noise 0.5388 0.7256 0.7295
43,000 共N/m兲 narrowband 0.4953 0.5843 0.5 0.5848
K G ␴ st K G rms共 x st 兲 periodic 0.2906 0.4094 0.4749
F N* ⫽ ␣ xc or ␣ xc (26) white noise 0.5509 0.6584 0.6663
␮f ␮f 64,500 共N/m兲 narrowband 0.3362 0.4930 0.42 0.5596
where ␣ xc ⫽0.5. periodic 0.2431 0.3216 0.3910
When the stiffness of friction damper, K G , is increased to
64,500 N/m, the near-optimal value of ␣ xc can be chosen as 0.42
which provides good performance for any types of excitation, as
shown in Fig. 9. The minimum value of ␴ x / ␴ st and those with the mentally, the normal load obtained by Eq. 共26兲 will yield
near-optimal values of x c / ␴ st are represented in Table 4. These near-optimal performance regardless of the external force
results indicate that the near-optimum normal load can be deter- characteristics.
mined by the factor ␣ xc which is independent of the intensity and It is found that the friction damper becomes less effective with
frequency band of excitation. Since ␴ st can be measured experi- increase of frequency band of external force and more effective
with increase of friction damper stiffness, K G , Table 4, Fig. 8 and
9. The effect of frequency band of narrow band excitation is
shown in Fig. 10 when ␰ F ⫽0.01, 0.05, and 0.1. With the increase
of ␰ F , the bandwidth of excitation increases and resulting curves
approach the curve of white noise excitation.
5 Conclusions
A new set of nondimensionalized variables, ␴ x / ␴ st and x c / ␴ st
or rms(x p )/rms(x st ) and x c /rms(x st ), is used to compare the per-
formance of a friction damper under three different types of exci-
tation: white noise excitation, narrowband random excitation, and
deterministic sinusoidal excitation. The resulting curves of re-
sponses with these nondimensionalized variables are independent
of the level of external force and the optimal value of x c / ␴ st or
x c /rms(x st ), ␣ xc , is not sensitive to the bandwidth of excitation.
If ␴ st or rms(x st ) is measured experimentally, the optimal normal
load can be obtained based on the optimal value of ␣ xc , without
any information about the external force. The performance of
blade-to-blade 共B-B兲 and B-G dampers in a multi-degree-of-
freedom model of a rotor stage is currently being examined.
Nomenclature
A ⫽ system matrix in the state-space analysis
Fig. 10 Peak value of ␴ x Õ ␴ st versus x c Õ ␴ st „ ␰ F Ä0.01, 0.05, A p ⫽ peak amplitude of sinusoidal response
and 0.1… A x ⫽ amplitude of sinusoidal response
B ⫽ input matrix in the state-space analysis ␴ 2y G ⫽E b y G2 c (A5b)
BR ⫽ input matrix of Lyapunov equation
c ⫽ damping coefficient E 关 ẋy G 兴
x
cG x
,k G , ␳ G⫽ . (A5c)
␴ ẋ ␴ y
c Gy ,k Gy ⫽ equivalent linear coefficients of damper
dG ⫽ displacement of damper The coefficient c Gy requires a numerical integration. However, it
E关兴 ⫽ expected value can be approximated for small values of ␳ G as
冋冉冊册冉冊
F(t) ⫽ external force vector
FN ⫽ normal load of friction damper 1 xc 1 ␳G ⫺x 2c
c Gy ⫽ 1⫹erf ⫺ exp .
fG ⫽ hysteretic force of damper 2 ␴ y G 冑2 冑␲ 冑1⫺ ␳ G2 2 ␴ 2y G
f0 ⫽ amplitude of sinusoidal excitation (A6)
g(t) ⫽ narrowband excitation
KG ⫽ stiffness of damper Appendix B
k ⫽ modal stiffness of blade
mt ⫽ modal mass of blade Calculation of A p and ␻ p for Sinusoidal Excitation. The
P ⫽ correlation matrix of state variables peak response of the blade, A p , and the corresponding frequency,
Q0 ⫽ intensity of white noise excitation ␻ p , can be obtained by the algorithm available in the literature,
Qf ⫽ quality factor of the bandpass filter 关1,4兴.
R w( ␶ ) ⫽ correlation matrix of white noise excitation 1 Define nondimensional parameters as
R x( ␶ ) ⫽ correlation matrix of blade displacement
U(•) ⫽ unit step function KG
␧ d⫽ , (B1)
w(t) ⫽ white noise excitation k⫹K G
xc ⫽ slip distance of damper
xi ⫽ displacement of ith blade f0
f d⫽ (B2)
xp ⫽ sinusoidal response with peak amplitude at a given K Gx c
normal load
x st ⫽ response of a blade when the friction damper is fully c
stuck, defined in Eq. 共24b兲 ␰ d⫽ . (B3)
2 冑m 共 k⫹K G 兲
yG ⫽ relative displacement of damper
z ⫽ state variables 2 The nondimensional peak response, A m , and the correspond-
␣ xc ⫽ defined in Eq. 共26兲 ing frequency, ␻ m , are obtained by solving the following nonlin-
␴ st ⫽ standard deviation of x st , defined in Eq. 共24a兲 ear algebraic equations:
冑
␴x ⫽ standard deviation of blade displacement
␮f ⫽ friction coefficient 2␧ d
A m ⫽⌫⫹ ⌫⫹ (B4)
␰F ⫽ damping ratio of bandpass filter ␲␰ d ␻ m
␻F ⫽
冑
resonant frequency of bandpass filter
␧d
␻ m⫽ 1⫺ 共 ␲ ⫺ ␪ d ⫹0.5 sin共 2 ␪ d 兲兲 (B5)
␲
Appendix A
x x y ␧ d 共 f d ⫺4/␲ 兲
Expressions for Equivalent Linear Coefficients c G , kG , cG , ⌫⫽ (B6)
y 4 ␰ d␻ m
and k G . Assuming a joint Gaussian probability distribution of
the multidimentional response process, the equivalent linear coef-
ficients are obtained as follows, 关6兴: ␪ d ⫽cos⫺1 1⫺ 冉 2
.冊 (B7)
冉冊
Am
K Gx c xc It should be noted that the allowable ranges for the parameters
x
cG ⫽ erfc (A1)
␴ ẋ 冑2 ␲ ␴ y G 冑2 共 1⫺ ␳ G2 兲 are A m ⭓1 and (1⫺␧ d )⭐ ␻ m ⭐1.
冉冊
3 The peak response of the blade, A p , and the corresponding
xc frequency, ␻ p , are calculated as
x
kG ⫽K G erf
␴ y G 冑2 A p ⫽A m x c , (B8)
冉冊冉冊冑 k⫹K G
K Gx c ⫺x 2c ␳ Gx c ␻ p⫽ ␻ m . (B9)
⫺ exp erfc (A2) m
␴ y G 冑2 ␲ 2 ␴ 2y G ␴ y G 冑2 共 1⫺ ␳ G2 兲
冋冉冊册冕冉冊
References
1 xc 1 ⬁
2 ␳ G␯
c Gy ⫽ 1⫹erf ⫺ e ⫺ ␯ erf d␯ 关1兴 Griffin, J. H., 1980, ‘‘Friction Damping of Resonant Stresses in Gas Turbine
2 ␴ y G 冑2 冑␲ x c / ␴ y 冑2 冑1⫺ ␳ G2 Engine Airfoils,’’ ASME J. Eng. Gas Turbines Power, 102, pp. 329–333.
G
关2兴 Srinivasan, A. V., and Cutts, D. G., 1983, ‘‘Dry Friction Damping Mechanisms
(A3)
冉冊冋冉冊册
in Engine Blades,’’ ASME J. Eng. Gas Turbines Power, 105, pp. 332–341.
关3兴 Menq, C.-H., Griffin, J. H., and Bielak, J., 1986, ‘‘The Influence of a Variable
⫺ ␳ G x c ␴ ẋ ⫺x 2c ␳ Gx c Normal Load on the Forced Vibration of a Frictionally Damped Structure,’’
k Gy ⫽ exp 1⫹erf
冑2 ␲ ␴ y G 冑2 共 1⫺ ␳ G2 兲
ASME J. Eng. Gas Turbines Power, 108, pp. 300–305.
␴ 2y G 2 ␴ 2y G 关4兴 Cameron, T. M., Griffin, J. H., Kielb, R. E., and Hoosac, T. M., 1990, ‘‘An
冉冊
Integrated Approach for Friction Damper Design,’’ ASME J. Vibr. Acoust.,
␴ ẋ 冑1⫺ ␳ G2 ⫺x 2c 112, pp. 175–182.
关5兴 Sanliturk, K. Y., Imregun, M., and Ewins, D. J., 1997, ‘‘Harmonic Balance
⫺ exp (A4)
␴ y G␲ 2 共 1⫺ ␳ G
2
兲 ␴ 2y G Vibration Analysis of Turbine Blades With Friction Dampers,’’ ASME J. Vibr.
Acoust., 119, pp. 96 –103.
关6兴 Asano, K., and Iwan, W. D., 1984, ‘‘An Alternative Approach to the Random
where Response of Bilinear Hysteretic Systems,’’ J. Earthquake Eng. Struct. Dyn.,
12, pp. 229–236.
␴ ẋ2 ⫽E b ẋ 2 c , (A5a) 关7兴 Sinha, A., 1990, ‘‘Friction Damping of Random Vibration in Gas Turbine
Engine Airfoils,’’ International Journal of Turbo and Jet Engines, 7, pp. 95– 关12兴 Chen, S., and Sinha, A., 1990, ‘‘Probabilistic Method to Compute the Optimal
102. Slip Load for a Mistuned Bladed Disk Assembly With Friction Dampers,’’
关8兴 Roberts, J. B., and Spanos, P. D., 1990, Random Vibration and Statistical ASME J. Vibr. Acoust., 112, pp. 214 –221.
Linearization, Chichester: John Wiley and Sons, Chichester. 关13兴 Socha, L., and Soong, T. T., 1991, ‘‘Linearization in Analysis of Nonlinear
关9兴 Whitehead, D. S., 1960, ‘‘The Analysis of Blade Vibration due to Random Stochastic Systems,’’ Appl. Mech. Rev., 44, pp. 399– 422.
Excitation,’’ Reports and Memoranda R & M 3253, Cambridge University,
关14兴 Griffin, J. H., and Sinha, A., 1985, ‘‘The Interaction Between Mistuning and
Cambridge, UK.
Friction in the Forced Response of Bladed Disk Assemblies,’’ ASME J. Eng.
关10兴 Sogliero, G., and Srinivasan, A. V., 1980, ‘‘Fatigue Life Estimates of Mistuned
Blades via a Stochastic Approach,’’ AIAA J., 18共83兲, pp. 318 –323. Gas Turbines Power, 107, pp. 205–211.
关11兴 Minkiewicz, G., and Russler, P., 1997, ‘‘Dynamic Response of Low Aspect 关15兴 MATLAB Manual, 1995, The MathWorks, Inc.
Ratio Blades in a Two Stage Transonic Compressor,’’ AIAA Paper No. 97- 关16兴 Deo, N., 1980, System Simulation With Digital Computer, Prentice-Hall,
3284. Englewood Cliffs, NJ, pp. 153–154.
Engine Torque Nonuniformity
Evaluation Using Instantaneous
Crankshaft Speed Signal
The paper presents the development of a methodology for evaluating the torque nonuni-
formity between the various cylinders of an internal combustion engine (ICE). This non-
uniformity can be due, for example, to pathological operating conditions such as misfires
or misfuels, as well as to other abnormal operating conditions. Between the nominal
torque production and the one corresponding to the absence of combustion there exist, in
N. Cavina fact, a series of possible intermediate conditions. Each of them corresponds to a value of
produced torque that lies between the nominal value and the one corresponding to the
F. Ponti lack of combustion (due for example to statistical dispersion in manufacturing or aging in
the injection system). The diagnosis of this type of nonuniformity is a very important issue
DIEM—University of Bologna, in today’s engine control strategies design. The use of the developed methodology should
Viale Risorgimento 2, in fact allow the control strategy to adopt the appropriate interventions if the diagnosed
40136 Bologna, Italy nonuniformity is related to different behavior of the injectors. In order to evaluate this
torque production variability between the various cylinders, information hidden in the
instantaneous crankshaft speed fluctuations has been processed using a suitable method-
ology. The procedure has been validated running a supercharged 2.0 liters V6 engine, and
a 1.2 liters L4 engine, in a test cell. During the tests, the in-cylinder pressure signal has
been acquired together with the instantaneous engine speed, in order to determine a
correlation between speed fluctuations and the indicated torque produced by each cylin-
der. The actual cylinder-by-cylinder torque nonuniformity can then be evaluated on-board
by processing engine speed. The procedure is able to diagnose the absence of combustion
(due for example to a misfire or a misfuel) as well as abnormal combustions that do not
necessarily involve lack of combustion, with, the accuracy needed for on-board use.
Control interventions to injection and ignition time commands of one or more cylinders
should, in most cases, be able to re-establish torque production uniformity.
关DOI: 10.1115/1.1581892兴
Introduction uniformity that the engine torque production shows when the
number of combustions within an engine cycle increases.
The great interest of industry and research centers on the crank-
On the other hand, the information hidden in the engine speed
shaft instantaneous speed analysis for the misfire detection and
torque estimation in internal combustion engines is due to the fluctuation signal has been used in order to perform engine torque
positive results obtained using different approaches based on this estimation, as demonstrated by previous studies, 关8 –11兴. These
signal, 关1–7兴. algorithms usually require fast fourier transform 共FFT兲 computa-
A simple misfire diagnosis methodology developed by Bosch tion and are time-consuming. Their use on-board a vehicle is still
research, 关2– 4兴, is based on the evaluation of the time periods limited by the capability of the engine control unit 共ECU兲.
spent by the crankshaft to perform the rotation of a pre-defined There exist, therefore, two main approaches for the evaluation
angular sector, related to the combustion phase of each cylinder. of the torque, or the lack of it, using the instantaneous engine
Knowledge of the combustion periods allows evaluating the speed information: the first focusing on diagnostic purposes, the
crankshaft kinetic energy variation caused by each combustion by second one on control. This second approach is usually able to
means of the following expression presented in 关3兴: give also diagnostic information, but is more complex and in-
volves algorithms that require high computing power.
This paper investigates if and how it is possible to obtain infor-
␶ i⫹1 ⫺ ␶ i
LU i ⫽ (1) mation useful for engine management and control even using a
␶ i3 rough and simple approach based on engine speed analysis, a
methodology normally used only for diagnostic purposes.
where ␶ i and ␶ i⫹1 are the time periods relative to two consecutive The algorithm has been tested on an L4 and a V6 engine in a
combustions. LU i can be evaluated for each combustion; it con- test cell. It showed a good capability to diagnose the presence of
tains information on the instantaneous crankshaft kinetic energy accidental misfires even for the 6-cylinder engine. Furthermore, it
variations caused by abnormal combustions, such as misfires and has been demonstrated that it is possible to enlighten the presence
other sources of torque production nonuniformity, or by gear of torque production nonuniformity 共due for example to aging of
shifts, cutoff. the injectors兲 between the different cylinders for the L4 engine.
The extension of these algorithms to engines with a high num- This could allow, for example, the adoption of control interven-
ber of cylinders is a challenging application, due to the higher tions on injection and ignition time commands of one or more
cylinders, in order to reestablish the torque production uniformity
Contributed by the Internal Combustion Engine Division of THE AMERICAN SO- 共due, for example, to aging of the injectors兲.
CIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF
ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received by the ICE
Other control applications could arise from the capability of
Division, Oct. 2001; final revision received by the ASME Headquarters, Dec. 2002. detecting and quantifying the presence of torque nonuniformity.
Associate Editor: D. Assanis. For example, the cycle reference information could be indirectly
Fig. 2 Engine speed waveform for an accidental misfire in cyl-
inder 4 of a V6 engine running in a test cell at 3000 rpm
Fig. 1 Engine-brake dynamic model

the crank-slider mechanism angle-dependent inertia, 关13,14兴, is
taken into account by introducing the reciprocating torque T r ( ␪ ),
obtained, for a sequential phased injection system, without using a following a well-known approach, 关11,14兴:
冋册
magnetic pickup facing the camshaft. In fact, the instantaneous Z
d f 共␪兲 ˙ 2
crankshaft position is usually measured by placing a magnetic
pickup in front of a gear wheel connected to the crankshaft. One
T r共 ␪ 兲 ⫽ 兺
m⫽1
M eq r 2 f 共 ␪ 兲 f 共 ␪ 兲 ␪¨ ⫹
d␪
␪
共or more兲 teeth of the wheel are usually missing to provide a
where Z is the number of cylinders, and M eq is the equivalent
reference position within the crankshaft revolution. Another
mass of the sliding part of the crank-slider mechanism, 关11,14兴.
pickup is normally placed in front of the camshaft, to differentiate
The approach presented considers the crankshaft as a unique
one revolution to the other 共within the same engine cycle兲, 关12兴.
inertia, thus neglecting its elasticity and damping. A more com-
By using a torque nonuniformity detection algorithm it is possible
plex model, involving two inertias, could be designed in order to
to cause an abnormal injection and determine when the given
take into account also the elasticity and the damping of the crank-
cylinder is affected by the abnormal torque production. The crank-
shaft. The higher complexity of this model, anyway, is not com-
shaft angular distance between the instant in which the abnormal
patible with the development of a simple algorithm for misfire and
injection is caused, and the instant in which the abnormal torque
torque production nonuniformity detection. For this reason the
production is revealed gives the information regarding the actual
simple single-inertia model will be used for the development of
cycle reference of the engine. In this way, the reference signal
this work. The considerations that can be made using this simpler
coming from a magnetic pickup facing the camshaft is no longer
model will require some adjustments in order to take into account
needed, and the cost of the sensor can be saved.
also the effects of the shaft elasticity on the instantaneous engine
speed fluctuations, as it will be mentioned in the following.
Effects of Engine Torque Production on Instantaneous
Engine Speed
To evaluate the effects of the torques acting on the crankshaft Misfire Diagnosis and Torque Production Nonunifor-
over the instantaneous engine speed, a suitable model of the in-
teractions between engine and vehicle dynamics is needed. The
mity Evaluation
development of such a model has been carried out considering the The presence of an accidental misfire in one cylinder will result
engine connected to an eddy current brake, since this was the in a lack of torque acting on the crankshaft, thus influencing the
layout of the experimental tests. The results that have been ob- engine speed waveform. In particular, the instantaneous engine
tained may be extended to on-board applications: In this case the speed will show a characteristic behavior during the engine cycle
vehicle equivalent inertia and the transmission line elasticity have portion in which the misfire occurs: The expansion stroke of the
to be taken into account. misfiring cylinder. A torsional damped vibration will follow due to
Figure 1 depicts the engine-brake model considered in this the torque impulse on the crankshaft caused by the misfire.
study: T i represents the overall indicated torque, T r the overall Figure 2 shows the speed waveform for two engine cycles dur-
reciprocating torque, T f the friction torque, including in this term ing which a misfire occurs in cylinder number 4 of a 2-liter V6
also the torque absorbed by the auxiliaries 共oil pump, cooling engine, where the effects of the misfire can be clearly seen.
system, . . . 兲, and T L the load torque. Newton’s second law re- In order to evaluate the effects of the misfire on the instanta-
quires that neous engine speed, Eq. 共2兲 has been considered. The attention
has been focused on the speed trend during the expansion stroke
J ␪¨ ⫽T i 共 ␪ 兲 ⫺T r 共 ␪ 兲 ⫺T L 共 ␪ 兲 ⫺T f 共 ␪ 兲 (2) of the various cylinders. When a lack of combustion takes place,
where the engine speed will decrease. Integrating Eq. 共2兲 over the expan-
sion stroke of each cylinder shows this effect.
J⫽J crankshaft⫹J flywheel⫹J brake Since the expansion strokes of engines with more than four
and cylinders overlap, for this kind of engines the integration is lim-
ited to a portion of the expansion stroke of each cylinder 共for a
Z
6-cylinder engine this portion will be 120 deg wide兲.
T i 共 ␪ 兲 ⫽A p r 兺 p m共 ␪ 兲 f 共 ␪ 兲 . The integration yields:
冕冕
m⫽1
1 ⌬␪ 1 ⌬␪
In Eq. 共2兲, the term J is assumed constant, since it models only J ␪¨ d ␪ ⫽ 共 T i 共 ␪ 兲 ⫺T f 共 ␪ 兲 ⫺T r 共 ␪ 兲 ⫺T L 共 ␪ 兲兲 d ␪
the effect of the rotating inertias. The torque contribution due to ⌬␪ 0 ⌬␪ 0
Fig. 4 Intervals investigated for optimal MD evaluation
Fig. 3 Indicated torque mean value evaluation over different

integration intervals for a V6 engine with cylinder 4 misfiring at
the beginning of the second engine cycle In order to evaluate M D i over Intervals 1 and 2 in fact the
engine speed instantaneous values at the beginning and at the end
of such intervals should be considered 共see Eq. 共4兲兲. In order to
minimize the effects of noise on the evaluation of the engine
冕 T
speed at the beginning and at the end of each of these intervals,
1 the mean values of the speed over two engine cycle portions cen-
J ␪¨ ␪˙ dt⫽T i 共 ␪ 兲 ⫺T f 共 ␪ 兲 ⫺T r 共 ␪ 兲 ⫺T L 共 ␪ 兲 (3)
⌬␪ 0 tered on these two points and having an amplitude of 120 deg
have been used. Figure 5 shows how the parameter M D i can be
J 关 ␪˙ 2 兴 T0 evaluated, for example, over Interval 2.
⬵T i 共 ␪ 兲 ⫺T f 共 ␪ 兲 ⫺T L 共 ␪ 兲 The 共mean兲 engine speed values before and after each interval
2⌬ ␪
to be used for M D evaluation respectively over Interval 1 and 2
where the mean values of the acting torques over the engine cycle are reported in Fig. 6.
portion taken into account are considered (T̄ r ⬵0 over the portion From Fig. 6 it is possible to observe that M D i evaluated for the
of the expansion stroke considered兲, 关11,13兴. misfiring cylinder 4 over Interval 2 will show a high positive
If the engine is running at steady-state operating conditions for value, while M D i evaluated over Interval 1 will show a small
a nonmisfiring cylinder, the corresponding value of the misfire absolute value and will be negative. This is due to the presence of
detection parameter defined as the speed increase related to the torsional vibration that follows
the misfire event. Therefore, even if the previous considerations
J 关 ␪˙ 2 兴 T0 J ˙2
M D i ⫽⫺ ⫽ 关 ␪ 共 0 兲 ⫺ ␪˙ 2 共 T 兲兴 (4)
2⌬ ␪ 2⌬ ␪
will be approximately equal to zero, while for a misfiring cylinder
it will assume positive values, since T i ( ␪ ) is negligible with re-
spect to ⫺(T f ( ␪ )⫹T L ( ␪ )) under such conditions.
The difference between the M D values for misfiring and non-
misfiring cylinders is therefore related to the lack of indicated
torque in the misfiring cylinder. In order to maximize this differ-
ence 共i.e., to optimize the signal-to-noise ratio for misfire detec-
tion兲 a proper placement of the integration interval for M D evalu-
ation has to be done.
Figure 3 reports the indicated torque mean values over engine
cycle portions of amplitude 4 ␲ /Z for a 6-cylinder engine. Two
different intervals have been chosen: the first one centered on the
expansion stroke of each cylinder 共Interval 1兲, the second one
starting 30 deg before and ending 90 deg after the top dead center Fig. 5 Engine speed evaluation before and after Interval 2
共TDC兲 of each cylinder 共see Fig. 4兲.
The difference between the indicated torque mean values of the
misfiring cylinder and of the other cylinders is maximized choos-
ing Interval 1.
On the other hand, it has to be noticed that in Fig. 3 the effects
of the torsional vibrations following the misfire event are not con-
sidered. These torsional vibrations modify the situation described
in Fig. 3, adding an oscillating torque component acting on the
crankshaft after the misfire occurrence. The mean torque over
each interval will be then different from those reported in Fig. 3.
The effects of the oscillating torque are expected to be less rel-
evant on the mean torque during Interval 2, since in this case the
torque is evaluated at the beginning of the expansion stroke of the
misfiring cylinder, and the torsional vibration is not yet com-
pletely excited.
This consideration is made clearer by looking at the values
of the engine speed to be used to evaluate M D over Intervals 1 Fig. 6 Engine speed values to be used for MD i evaluation over
and 2. Intervals 1 and 2
made on the torque waveform suggest to choose Interval 1 for Table 1 Main engine and sensor characteristics
misfire diagnosis, better results are expected by choosing Interval
2. In fact, as already mentioned, the influence of the torsional
vibration is less relevant over this last interval. For this reason, the
detection algorithm has been designed using M D i evaluated over
Interval 2, with the engine speeds before and after such Interval
obtained as described in Fig. 5. This will be the choice adopted
from here on in this work.
The engine speed values before and after each Interval 2 have
been obtained in this work from the knowledge of the time peri-
ods ␶ i necessary for the crankshaft to cover the same angular
sectors over which these engine speeds have to be evaluated 共for
example, by looking at Fig. 5, ␶ i is the time interval needed to
cover the angular interval ⫺90 deg→⫹30 deg, with respect to
cylinder 4 TDC兲.
Once the time periods ␶ i have been acquired and preprocessed
to filter the effects of sensor-wheel interferences 共such as me-
chanical tolerances of the teeth edges, eccentricity of the wheel
and air gap variation during one revolution兲, the values of the
engine speed at the beginning and at the end of Interval 2 can be
obtained from
⌬␪ ⌬␪
␪˙ 共 0 兲 ⫽ ; ␪˙ 共 T 兲 ⫽ . (5)
␶i ␶ i⫹1
When a misfire occurs, M D i will suddenly increase assuming
values related to the quantity T f ( ␪ )⫹T L ( ␪ ) 共considering approxi-
mately T̄ i ⬵0 in case of misfire—see Eq. 共3兲兲. In order to diagnose introducing an abnormal injection in one cylinder immediately
the presence of a misfire event it is then necessary to define after the engine start, and by determining the engine cycle portion
threshold values to distinguish whether it occurred or not. These at which the torque nonuniformity appears on the engine speed.
threshold values are strictly related for each engine operating con-
dition to the quantity T f ( ␪ )⫹T L ( ␪ ) and can be mapped over the Experimental Tests
engine operating range as a function of manifold pressure and
engine speed. Once this map has been stored in the ECU memory, Some experimental tests have been conducted to identify and
misfire detection is possible comparing the actual value of M D i to validate the misfire detection and torque production nonunifor-
the threshold that can be obtained by entering the map with the mity evaluation methodology previously presented. The tests have
actual values of manifold pressure and engine speed. been conducted running two different SI engines 共a 2.0-liter V6
The same parameter M D used to detect misfiring cylinders and a 1.2-liter L4 engine兲 in a test cell, both in steady-state and in
could be suitable also to evaluate torque production nonuniformi- transient conditions.
ties between the various cylinders. In this case, the interest is Both the engines have been equipped with pressure sensors
focused on the actual value assumed by M D, rather than on its inside the cylinders, in order to evaluate the real torque production
comparison with a threshold value. In fact it can supply informa- of the single cylinders. The main characteristics of the engines
tion on the difference between the indicated torque developed by used and of the sensors employed are reported in Table 1.
the various cylinders. Therefore, this parameter does not allow Two magnetic pickups were used, one facing the flywheel teeth
obtaining information on the absolute value of the torque pro- and another one mounted in front of the camshaft pulley, to gen-
duced by the cylinders but only on the relative difference that erate a cycle reference signal.
exists between them. An important role in realizing these tests was performed by a
The evaluation of these nonuniformities should present more Virtual ECU 共VECU兲: This device, based on a particular configu-
difficulties than the misfire detection, since the signal-to-noise ra- ration of the dSPACE Real Time System, allows controlling the
tio should be in this case lower 共a misfire could be thought, in engine actuators 共ignition coils and/or injectors, for example兲 in
fact, as the torque nonuniformity of the maximum possible ampli- any cylinder, independently of the original ECU. The VECU has
tude兲. On the other hand, it has to be noted that the torsional been used to control ignition and injection of one single cylinder,
vibrations caused by the torque production nonuniformity 共that is inducing misfires 共an example is shown in Fig. 7兲, misfuels or
to be detected兲 are surely less evident than in the case of a misfire, varying the injection time in order to cause also small torque
and therefore the effect of these vibrations over the instantaneous nonuniformities.
engine speed waveform should be less relevant. The Virtual ECU used lookup tables 共containing the original
As already mentioned, the evaluation of the torque nonunifor- spark advance or injection time values兲 to generate the ignition or
mity allows adopting control strategies 共i.e., variations of injection injection logical signals, which were sent to the drivers and then
and ignition commands of one or more cylinders兲, able in most to the actuators.
cases to re-establish the desired torque production uniformity. The
capability of detecting torque production nonuniformities can be Results
used also to obtain a cycle reference information, without using a The misfire detection algorithm previously presented, with
magnetic pickup facing the camshaft. In fact, cycle reference M D i evaluated over Interval 2 as explained in Fig. 5, has been
identification is usually performed using two pickups facing, re- applied to the V6 engine whose characteristics are reported in
spectively, the crankshaft and the camshaft. If the second pickup Table 1. The steady-state operating conditions that have been in-
共the one facing the camshaft兲 is absent, the information obtained vestigated in order to determine the misfire detection threshold
by the first one allows identifying the crankshaft position, but it is map are reported in Fig. 8.
not sufficient to distinguish between the two-cycle positions at For each operating condition of Fig. 8, one misfire has been
which that same crankshaft position is reached. In this case, the induced every 30 cycles in cylinder #4. Using the instantaneous
complete cycle reference information can be obtained by properly engine speed signal, the parameter M D has then been evaluated.
Fig. 7 Example of the Virtual ECU actuation signals to gener- Fig. 10 MD evaluation for the test at 4000 rpm and 160 Nm in
ate a misfire the vicinity of a misfire event
As an example M D evaluated for the test at 4000 rpm engine

speed and 160 Nm load torque has been reported in Fig. 9. Larger misfire event, especially at higher loads, as it can be seen in Fig.
dots represent the combustions where a misfire has been induced. 10. The damped oscillation that follows the misfire causes, in fact,
A threshold value of 25 Nm has been chosen for M D in this the overtaking of the chosen threshold one more time, two com-
operating condition in order to distinguish misfiring from nonm- bustion periods after the one where the misfire has been induced.
isfiring combustions. When M D is higher than the chosen thresh- In order to avoid these false alarms the effects of the torsional
old a misfire is diagnosed in the cylinder whose combustion stroke vibration on the engine speed waveform have to be taken into
is associated to the M D i value that fell beyond the threshold. This account. This can be done, for example, by properly filtering the
threshold seems to enable the algorithm to detect all the misfires, M D waveform after a misfire occurrence, as it will explained in
while it causes a series of false alarms. These false alarms are the following.
related to the presence of the torsional vibration that follows the The M D waveforms after each misfire in Fig. 9 are always very
similar to the one described in Fig. 10; these waveforms are de-
termined by the crankshaft dynamic behavior that follows the mis-
fire torque impulsive excitation. A unique pattern can then be
isolated to represent the effects on M D related to a misfire event,
computing the average of all the waveforms obtained after each
misfire in Fig. 9. From Fig. 10 it is possible to notice that the
misfire effect lasts, in this engine operating condition, for at least
six combustions after the misfire event, and therefore the average
waveform has been computed considering six M D values imme-
diately after each missing combustion; the average waveform has
been reported in Fig. 11.
The waveform identified can be used to avoid the false alarms
previously described; in fact, once a misfire event has been diag-
nosed, the following M D values can be filtered by subtracting the
corresponding values in Fig. 11. The result of such a signal pro-
cessing algorithm is reported in Fig. 12.
In the case of a single misfire, such a procedure allows avoiding
false alarms, while a multiple misfire detection would need the
Fig. 8 Steady-state operating conditions investigated to deter- application of the same filtering technique once for each misfire
mine the misfire detection threshold map
Fig. 9 MD evaluation for the test at 4000 rpm engine speed

and 160 Nm load torque „larger dots represent the induced Fig. 11 MD waveform after a misfire for the test at 4000 rpm
misfires… engine speed and 160 Nm load torque
Fig. 12 MD evaluation for the test at 4000 rpm and 160 Nm in Fig. 14 MD and indicated torque values in cylinder #4 for the
the vicinity of a misfire event, after the torsional vibration ef- test at 1150 rpm and 24 Nm
fects have been filtered away
oped by cylinder #4; the strict correlation between these two sig-
detected. The case of multiple misfires has not been deeply inves-
nals shows that it is possible to detect torque production nonuni-
tigated in this work, and will be the subject for future work on this
formities using M D.
topic.
The same analysis has been conducted also for cylinder #6,
Figure 13 shows the M D pattern for the engine running steady-
where no misfires have been induced by the VECU. Figure 16
state at 1150 rpm engine speed and 24 Nm load torque 共1 misfire
shows that M D overtakes twice the threshold chosen for the spe-
induced every 30 combustions in cylinder #4兲. Also in this case
cific operating condition that is being considered, unveiling the
the choice of a threshold value able to detect all the misfires 共3.5
presence of two misfires due to the cyclic variability of the engine.
Nm, for example兲 causes many false alarms. In this case they are
Figure 17 shows that the same correlation, as for cylinder #4,
not related to the presence of the torsional vibration, and therefore
exists between M D and the indicated torque developed by cylin-
it is not possible to avoid them as previously described.
In order to better understand the nature of these false alarms the
M D values related to the combustions inside cylinder #4 have
been considered separately in Fig. 14, together with the indicated
torque 共obtained using in-cylinder pressure measurement兲 devel-
oped by that cylinder. When a misfire occurs a negative indicated
torque value is measured. Such negative values are not located
only in correspondence of induced misfire events 共see Fig. 14兲.
This means that in the operating condition considered 共low load
and low engine speed兲 the normal cyclic variability of the engine
causes the presence of abnormal combustions that present effects
similar to the induced misfires. It can be stated that under that
operating conditions the engine presents partial burns cycles with
very low torque produced. The false alarms are then the detection
of the torque production nonuniformity related to the natural cy-
clic variability of the engine. The in-cylinder pressure signal ac-
quired for cylinder #4 confirmed this statement.
Figure 15 reports M D versus the indicated torque values devel-
Fig. 15 MD versus indicated torque values in cylinder #4 for
the test at 1150 rpm and 24 Nm
Fig. 13 MD evaluation for the test at 1150 rpm engine speed

and 24 Nm load torque „larger dots represent the induced Fig. 16 MD and indicated torque values in cylinder #6 for the
misfires… test at 1150 rpm and 24 Nm
der #6. As already mentioned this could allow detecting the entity
of the torque production nonuniformities by monitoring the values
of the parameter M D.
Once a threshold value has been chosen for each steady-state
operating condition investigated, it is possible to obtain a map of
the threshold values as a function of the engine operating condi-
tions 共in terms of engine speed and manifold pressure兲, to be used
for misfire detection.
Figures 18 and 19 show the misfire detection capability during
two speed and load transients where one misfire every 30 com-
bustions has been induced in cylinder #4. Engine speed and load
torque waveforms during these transient tests are reported, respec-
tively, in Figs. 20 and 21. The methodology is able to detect all
the misfires that have been induced, since M D falls beyond the
mapped threshold value for every misfire. It has to be noted also
Fig. 17 MD versus indicated torque values in cylinder #6 for the presence of false alarms, due to the presence of the torsional
the test at 1150 rpm and 24 Nm vibration. They can be avoided, as previously discussed, by filter-
ing the effects of this torsional vibration on the engine speed
waveform.
In order to verify the existence of a correlation between the
torque production nonuniformities and the parameter M D also for
another engine architecture, some steady-state tests have been per-
formed on the L4 engine whose characteristics are reported in
Table 1 共see Fig. 22兲.
For each steady-state operating condition the injection has been
controlled using the Virtual ECU. The injectors have been con-
trolled maintaining the same injection time as that imposed by the
original ECU for all the cylinders except for cylinder #1, in order
to simulate an injector malfunctioning condition in that cylinder.
Smaller injection time adopted for cylinder #1 causes a different
torque production between cylinder #1 and the other cylinders. In
this case M D can be used to diagnose not only an accidental
partial torque production 共or a misfire in the worst case兲 in one
Fig. 18 MD evaluation for transient test #1 compared to the
cylinder, but also a systematic torque production difference be-
mapped threshold value „larger dots are the induced misfires… tween two cylinders. Figure 23 shows, as an example, the differ-
ence between the M D evaluated for cylinder #1 and #4, versus the
difference between the indicated torque developed in those
cylinders.
As it can be observed, the parameter M D gives in this case
information related to the different torque production associated to
each cylinder. This different behavior can be for example the re-
sult of a different air filling into the cylinders taken into account
during the intake stroke, or can be due for example to a malfunc-
tion in one of the two injectors. The use of M D in such situations
could help in developing a cylinder-by-cylinder torque production
management and control strategy.
It has to be noted that the results presented here have been
obtained with the engine mounted in a test cell. The approach that
has been developed is still valid also on-board a vehicle. As pre-
viously mentioned, anyway, the different elasticity and inertia of
the transmission line has to be considered for on-board applica-
tions, thus modifying the entity of the torsional vibration that
Fig. 19 MD evaluation for transient test #2 compared to the
mapped threshold value „larger dots are the induced misfires… follows a torque production nonuniformity.
Fig. 20 Engine speed and load torque trends during transient test #1
Fig. 21 Engine speed and load torque trends during transient test #2
cylinder overlap is higher than the case of an engine with the same
number of cylinders and combustions evenly spaced.
Finally, the paper introduces the problem of the crankshaft tor-
sional vibration effects on the engine speed fluctuations, that make
the misfire detection critical in particular when multiple misfires
are present in the same cycle. This topic will be further analyzed
in future works.
The engine speed signal is already present in modern engine
control systems, and the computing time required for the evalua-
tion of the parameter M D seems to be compatible with the actual
ECU computing power. This should allow an easy application on
board a vehicle of the methodology presented, thus giving a ben-
efit to actual engine control strategies. Rather than giving only a
diagnostic information the procedure is, in fact, able to give in-
formation useful to design control strategies able to improve on-
Fig. 22 Steady-state tests performed on the 1.2 liters L4
engine
board engine torque management.
Acknowledgments
The authors would like to thank Dr. Gabriele Serra 共Magneti
Marelli兲, Prof. Piero Azzoni and Borghi & Saveri Engine Brakes
for the support and the contribution in the development of the
present work.
Nomenclature
␶ i ⫽ combustion period relative to the ith
combustion 关s兴
LU i ⫽ misfire detection parameter 关l/s2兴
M D i ⫽ misfire detection parameter 关Nm兴
␪ ⫽ crank angle 关rad兴
Fig. 23 Difference between the MD evaluated for cylinder #1 ␪˙ ⫽ crankshaft speed 关rad/s兴
and #4, versus difference between the indicated torque devel-
oped in the same cylinders
T i ( ␪ ) ⫽ indicated torque of the engine 关Nm兴
T L ( ␪ ) ⫽ load torque 关Nm兴
T f ( ␪ ) ⫽ friction and auxiliaries torque 关Nm兴
T r ( ␪ ) ⫽ reciprocating torque 关Nm兴
Conclusions J ⫽ crankshaft, flywheel, and brake equiva-
The results obtained demonstrate the feasibility of detecting lent inertia 关kg m2兴
torque production nonuniformity and misfires from engine speed M eq ⫽ equivalent mass of the sliding part of
fluctuations. The M D parameter can in fact be used both as a the crank-slider mechanism 关kg兴
misfire index 共diagnosis is performed by comparing M D values, r ⫽ crank-slider mechanism radius 关m兴
calculated for each combustion, with threshold values that depend l ⫽ connecting rod length 关m兴
on the engine operating condition兲, and to evaluate torque produc- A p ⫽ piston surface 关m2兴
tion differences between the cylinders 共by comparing M D values p m ( ␪ ) ⫽ pressure inside mth cylinder 关Pa兴
related to the combustion of each cylinder, at the same engine ␭ ⫽ r/l
operating condition兲. The developed methodology has been ap- Z ⫽ number of cylinders
plied both to a V6 and an L4 engine, yielding good results in both f ( ␪ )⫽sin(␪)
cases. Therefore the extension of this approach to other engine ⫹␭ sin(2␪)/
architectures seems to be feasible. Anyway, the applicability of the 2冑1⫺␭ 2 sin2(␪) ⫽ crank slider mechanism function
methodology could be particularly challenging for engines that T im ( ␪ ) ⫽ indicated torque for mth cylinder 关Nm兴
present combustions not evenly spaced. In this case, in fact, the T ⫽ expansion stroke portion duration 关s兴
angular portions over which the expansion strokes of different ⌬␪ ⫽ expansion stroke angular portion 关rad兴
References in High Performance SI 12 Cylinder Engine,’’ SAE Technical Paper 980521.
关8兴 Cavina, N., Ponti, F., and Rizzoni, G., 1999, ‘‘Fast Algorithm for On-Board
关1兴 California Air Resources Board, 1991, ‘‘Technical Status Update and Proposed Torque Estimation,’’ SAE Technical Paper 1999-01-0541.
Revisions to Malfunction and Diagnostic System Requirements Applicable to 关9兴 Azzoni, P., Moro, D., Ponti, F., and Rizzoni, G., 1998, ‘‘Engine and Load
1994 and Subsequent California Passenger Cars, Light-Duty Trucks, and Torque Estimation With Application to Electronic Throttle Control,’’ SAE
Medium-Duty Vehicles—共OBD II兲,’’ CARB Staff Report.
Technical Paper 980795.
关2兴 Forster, J., Lohmann, A., Mezger, M., and Ries-Muller, K., 1997, ‘‘Advanced
关10兴 Citron, S. J., O’Higgins, J. E., and Chen, L. Y., 1989, ‘‘Cylinder by Cylinder
Engine Misfire Detection for SI-Engines,’’ SAE Technical Paper 970855.
Engine Pressure and Pressure Torque Waveform Determination Utilizing
关3兴 Klenk, M., Moser, W., Muller, W., and Wimmer, W., 1993, ‘‘Misfire Detection
by Evaluating Crankshaft Speed—A Means to Comply With OBD II,’’ SAE Speed Fluctuations,’’ SAE Technical Paper 890486.
Paper 930399. 关11兴 Moro, D., Ponti, F., and Cavina, N., 1999, ‘‘In-Cylinder Pressure Reconstruc-
关4兴 Plapp, G., Klenk, M., and Moser, W., 1990, ‘‘Methods of On-Board Misfire tion Based on Instantaneous Engine Speed Signal,’’ Proceedings of ICE Divi-
Detection,’’ SAE Technical Paper 900232. sion, Oct. 16 –20, 1999, Ann Arbor, MI, ASME, New York.
关5兴 Azzoni, P. M., Moro, D., Porceddu-Cilione, C. M., and Rizzoni, G., 1996, 关12兴 Letenturier, P., and Benning, J., 1999, ‘‘Enhanced Engine Position Acquisition
‘‘Misfire Detection in a High Performance Engine by the Principal Compo- and Treatment,’’ SAE Technical Paper 1999-01-0203.
nents Analysis Approach,’’ SAE Technical Paper 960622. 关13兴 Heywood, J. B., 1988, Internal Combustion Engine Fundamentals, McGraw-
关6兴 Azzoni, P. M., Cantoni, G., Ceccarani, M., Mazzetti, S., Minelli, G., Moro, D., Hill, New York.
and Rizzoni, G., 1995, ‘‘Measurement of Engine Misfire in a Lamborghini 关14兴 Moskwa, J., Wang, W., and Bucheger, D. J., 1998, ‘‘A New Methodology for
V-12 Engine Using Crankshaft Speed Fluctuations,’’ SAE Technical Paper Engine Diagnostics and Control Utilizing ‘‘Synthetic’’ Engine Variables: The-
950837. oretical and Experimental Results,’’ Proceedings of the ASME Dynamic
关7兴 Azzoni, P. M., Minelli, G., and Moro, D., 1998, ‘‘Misfire Pattern Recognition Systems and Control Division, DSC-Vol. 64, ASME, New York.
S. L. Yang
Associate Professor
Y. K. Siow
Numerical Study of Lean-Direct
Graduate Research Assistant
Injection Combustor With
C. Y. Teo
Graduate Student Discrete-Jet Swirlers Using
Department of Mechanical Engineering—
Engineering Mechanics, Reynolds Stress Model
Michigan Technological University,
Houghton, MI 49931-1295 The flowfield in a lean-direct injection (LDI) combustor with discrete-jet swirlers is de-
scribed and analyzed using a computational fluid dynamics (CFD) code with a Reynolds
stress turbulence model (RSTM). The results from the RSTM are compared to time-
R. R. Tacina averaged laser-Doppler velocimetry (LDV) data, as well as results from the National
Combustion Code (NCC) that has a cubic nonlinear ␬-␧ turbulence model, and from the
A. C. Iannetti KIVA code using the standard ␬-␧ model. The comparisons of results indicate that the
RSTM accurately describes the flow details and resolves recirculation zones and high
P. F. Penko velocity gradients while the ␬-␧ models are unable to capture most flow structures. This
confirms that, within the Reynolds averaging approach, the higher-order RSTM is pre-
Combustion Technology Branch ferred for simulating complex flowfields where separations, strong anisotropy, and high
NASA Glenn Research Center, swirl are present. 关DOI: 10.1115/1.1610012兴
Cleveland, OH 44135
Introduction more accurate calculations of the complex flowfield typical of an

engine, or a combustor with high degree of swirl, use of a higher-
Flame stability over a wide range of operating conditions is a
order turbulence model is more appropriate.
basic requirement for through-flow combustion devices, such as
Unlike the ␬-␧ models, the Reynolds stress turbulence model
gas turbine combustors. One common practice used to stabilize
共RSTM兲 naturally resolves flow anisotropy, streamline curvature,
combustion is generation of swirling flow that entrains and recir-
sudden changes in strain rate, secondary motions, flow in rotation
culates a portion of the hot combustion products. For direct injec-
共swirl flow兲, etc. This is mainly due to the fact that each Reynolds
tion engine, recirculation zones also enhance the mixing of the
stress component has its own transport equation. In addition, the
incoming air with fuel. This effect is readily accomplished by the
production source terms in the RSTM are exact and need not be
use of swirlers, such as discrete-jet swirlers, surrounding the fuel
modeled. This greatly improves the accuracy and robustness of
injector, 关1兴.
the solver. All in all, the RSTM represents the most comprehen-
Air swirlers are often used in many types of aircraft engines. In
sive description of turbulent flows within the framework of the
the lean direct injection 共LDI兲 concept, a key feature is the injec-
Reynolds averaging approach, 关8兴. Despite the added complexity
tion of finely atomized fuel into the high-swirling airflow at the
due to the number of equations in the modeling, the applicability
combustor dome that provides a homogeneous, lean fuel-air mix-
of the RSTM has been made feasible by present-day computer
ture. This allows for a better combustion efficiency and the reduc-
capacity.
tion of NOx and other pollutants. Another advantage of the LDI
In addition to obtaining a better understanding of the complex
concept is that, with proper design of swirler configuration, a
flow physics inside the combustor, the other purpose of this paper
shorter combustor length is possible. Accordingly, to fully realize
is to present cold-flow calculations of the flowfield in an LDI
the potential of the LDI concept, a good understanding of the
combustor with a discrete-jet swirler array using the KIVA-3V
complex, turbulent swirling flow inside the combustor becomes
code, 关7兴 with an RSTM 关9兴. It also aims to compare the RSTM
essential.
results with the same calculations by KIVA’s built-in standard ␬-␧
To effectively calculate flowfield where swirl, wall effect, flow
model, and the NCC’s cubic nonlinear ␬-␧ model, which has been
separation, and recirculation are present, a high-level turbulence
tuned for swirl flow calculations, 关10,11兴. All calculated results
model should be used. The most widely used turbulence model in
are compared to laser-Doppler velocimetry 共LDV兲 data, 关12兴, for
current advanced CFD codes is the ␬-␧ model, such as the cubic
the same configuration and operating conditions.
nonlinear model, 关2,3兴, used in the National Combustion Code
共NCC兲, 关4,5兴, and the standard ␬-␧ model, 关6兴, in the KIVA code,
关7兴. Its popularity is due to the relative ease of implementation and The National Combustion Code
its low CPU and memory overhead. It does, however, have con- The NCC is a research code under development at the NASA
siderable drawbacks for use in resolving flow with high gradients Glenn Research Center. It is a four-stage Runge-Kutta, precondi-
of velocity and swirl. tioned, massively parallel CFD code with super linear scalability,
Regardless of the advantages of the ␬-␧ model, it has a ten- 关4,5兴. The NCC includes various advanced chemistry models,
dency to yield inconsistent and diffusive results for complex flows such as the intrinsically low-dimensional manifold and a general-
because of its isotropic nature in modeling eddy viscosity. Without ized finite rate kinetic model. In addition, turbulence-chemistry
ad hoc modifications, the ␬-␧ model often cannot completely de- interaction is modeled via the Magnussen model and a joint-scalar
scribe the large gradients in swirling and recirculating flows. For PDF model for turbulent reacting flow. Furthermore, the NCC
uses a dynamic wall function for the effect of adverse pressure
Contributed by the Internal Combustion Engine Division of THE AMERICAN SO-
gradient, a low Reynolds number wall treatment, and unstructured
ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received by the ICE grids for computations. The turbulence models in the code include
Division, Feb. 2002; final revision received Mar. 2003. Associate Editor: D. Assanis. the standard, quadratic, and cubic ␬-␧ models.
Fig. 2 Orientation of the discrete-jet swirler array
Fig. 1 Computational mesh cedure not only allows for the construction of any complex geom-
etry, but is also efficient in the use of time and resources.
To insure that the calculated results represent a converged so-
The cubic nonlinear ␬-␧ turbulence model is a second-order lution, a grid dependency study was performed. Three grid densi-
closure-scheme turbulence model, which uses both a nonlinear ties were chosen for this purpose, namely coarse, medium, and
Reynolds stress-strain model, 关13兴, and a cubic Reynolds stress- dense grids. The grid densities are given in Table 1, along with the
strain model, 关14兴, for mean flow calculations. The model has combustor length used in each computational model. Notice that,
been tuned and has fewer deficiencies than the standard ␬-␧ for the dense-grid case, the rectangular burner duct has a length of
model, particularly in simulating the effects of flow rotation and 7.62 cm. This length was chosen for CPU time consideration and
curvature. was based on the LDV measurements. From the LDV data, there
is no reversed flow at this location and the measured recirculation
bubbles are no more than 5.3 cm in length. Comparisons of the
The KIVA-3V Code With Reynolds Stress Turbulence LDV data with the numerical results among the three computa-
Model „RSTM… tional meshes show that the solution converges with grid density.
The KIVA-3V computer code is the latest version of the KIVA- Therefore, the dense-grid result is chosen for presentation.
family CFD codes developed by the Los Alamos National Labo-
ratory and is in use worldwide among the engine research com-
munities, 关7兴. It solves transient, two and three-dimensional, Discrete-Jet Swirler Array Flow Physics
chemically reactive fluid flow with fuel sprays. The turbulence Swirl flow is created when air is forced through the small flow
models in the code include both the standard and RNG-variant ␬-␧ passages, inclined at an angle with respect to the cross-flow plane.
models. To better simulate complex, turbulent engine flow, the As air enters and contracts through the small passages, air velocity
Speziale-Sarkar-Gatski 共SSG兲 RSTM, 关15兴, has been incorporated increases, producing the so-called discrete jets. As each jet enters
into the KIVA code, 关9兴. The SSG RSTM is considered a quasi- the combustor, it spreads, expands, and interacts with the neigh-
linear model, since it contains one quadratic term in the pressure- boring jets and the bounding walls. Due to the orientation of the
strain model. One advantage of using the SSG model is that, un- flow passages, azimuthal and radial velocity components, which
like most other RSTMs available in the literature, no wall-distance are responsible for the creation of swirl, are developed. When the
parameter is needed to account for wall reflection. This is a desir- swirl reaches certain strength, reversed or recirculating flow re-
able feature for engine flow simulations, since the computational gions will form near the top of the combustor.
domain is often of a complex shape and may also be time- Based on the arrangement of the swirler array shown in Fig. 3,
dependent. Details of the RSTM implementation, numerical is- the nine groups of swirlers can be classified, according to their
sues, and code validation can be found in Yang et al. 关9兴 and will geometric locations, into three types of swirler, namely ‘‘center,’’
not be repeated here. ‘‘corner,’’ and ‘‘side,’’ 关12兴. Since the ‘‘center’’ swirler is not
bounded by any solid walls but is instead surrounded by other
Problem Description and Computational Mesh swirlers, it is expected that it will receive the highest level of
interactions with other swirlers. Two solid walls and two neigh-
The multipoint LDI combustor has a rectangular inlet section boring swirlers, on the other hand, bound the four ‘‘corner’’ swirl-
1.27 cm long and is upstream of nine groups of swirlers arranged ers. The remaining four ‘‘side’’ swirlers have three sides bounded
in a 3⫻3 pattern, as shown in Fig. 1. Each group of swirlers by other swirlers and one side by a solid wall. On the basis of flow
comprises eight flow passages, all of which orientated at 35 deg in physics, the same swirler type is expected to behave similarly.
the tangential direction, Fig. 2. Figure 3 illustrates the 3⫻3 swirler Further discussion of the complex flow pattern inside the combus-
array arrangement and the relative location of the flow passages. A tor will be given in the Results and Discussion section through the
rectangular burner duct, either 7.62 cm or 17.78 cm long— aid of both measured and CFD results.
depending on the grid density, is downstream of the swirler sec-
tion. The duct has a cross section of 7.62 cm⫻7.62 cm. The inlet
surface of the combustor lies in the XY -plane and the inlet veloc- Flow Conditions and Laser-Doppler Velocimetry „LDV…
ity is in the positive Z 共axial兲 direction.
The computational grids are constructed by using Pro- Measurements
Engineer® to create a three-dimensional model, which is then For comparison, the operating conditions in this study are iden-
exported in IGES format to the Gridgen®, 关16兴, database. A Car- tical to the LDV measurements by Cai et al. 关12兴 and an NCC
tesian multiblocked structured mesh is then generated using Grid- study reported by Iannetti et al. 关10,11兴. The inlet axial velocity is
gen®. Next, the grids are exported from Gridgen® into the KIVA 6.6 m/s. As in the KIVA code, wall functions are used for all solid
code where all blocks are appropriately interconnected. This pro- walls. At the inlet, the turbulence intensity and length scale are
Fig. 3 Swirler array configuration and classification
assumed 10% of the inlet mean flow kinetic energy and 3% of the ⫽0.3 cm, 1.3 cm, 2.3 cm, and 5.3 cm. These cutting-plane loca-
inlet opening, respectively. The outlet pressure is at the ambient tions are shown in Fig. 1. In addition to contour plots, line plots
condition. along two axial lines downstream of the swirlers are also pre-
The experimental method and assumptions reported in Ref. 关12兴 sented for quantitative comparison. The first line is located at X
are now briefly described. The measurements were made with a ⫽2.54 cm, Y ⫽⫺2.54 cm 共at the center of the bottom, right
two-component LDV system, at 12 axial stations downstream of corner-swirler兲, and the second at X⫽1.27 cm, Y ⫽⫺1.27 cm 共be-
the discrete-jet swirlers. Due to geometric symmetry of the swirler tween four swirlers at the bottom, right corner兲.
array, only 4/9 of each cross section 共Fig. 3兲 was covered during The results in the XY -planes show that the RSTM yields a
the experiment. In other words, the LDV system measured the better and more consistent agreement with the experiment than do
axial 共W兲 and a cross-flow velocity component 共U兲 within the both the ␬-␧ models. At Z⫽0.3 cm, Fig. 4, although the RSTM
volume downstream of four neighboring swirlers. The third veloc- 共Fig. 4共c兲兲 calculated a slightly stronger forward velocity 共red
ity component, V, was derived by assuming that the flowfield was spots兲 at the jets, the size, shape, and strength of the reversed-
antisymmetrical about the diagonal. velocity contours compare considerably well with the LDV data.
The ␬-␧ models, Figs. 4共a–b兲, despite showing acceptable con-
Results and Discussion tours of forward velocity, had difficulty computing the reversed
flow. In particular, the NCC ␬-␧ result displays a strong presence
The primary objective of this study is to investigate the flow- of recirculation zone 共the blue regions兲 surrounding the center
field of an LDI combustor with multipoint swirler array in a swirler. This suggests an overcompensation for the diffusiveness
square duct, and to compare results among the SSG RSTM, the inherited in the standard ␬-␧ formulation.
standard ␬-␧ model in the KIVA code, the cubic nonlinear ␬-␧ Further downstream at Z⫽1.3 cm, Fig. 5 contains interesting
model in the NCC, and the LDV measurements. As the measure- yet serious differences among the turbulence models. The experi-
ments were made in a nonreacting flow, only cold-flow calcula- mental measurements, Fig. 5共d兲, show kidney-shaped reversed-
tions are presented. In the NCC study, 关10兴, tetrahedral grids were flow regions around the bounding walls, and a circular region at
used with a total of 2.26 million cells and 2.28 million vertices. the center. This is caused by the high shear induced by the inter-
Comparisons of results are presented in XY -planes, i.e., at Z actions among the co-rotating swirlers. The RSTM result, Fig.
5共c兲, agrees very well with the LDV data. The locations and mag-
nitude of the recirculation zones are essentially resolved by the
Table 1 Grid density
RSTM. The ␬-␧ models, on the other hand, each computed an
Coarse Medium Dense inaccurate yet distinct profile. While the standard ␬-␧ model re-
sult, Fig. 5共b兲, may seem reasonable, the locations of the reversed-
Number of cells 357,257 527,687 965,905 flow regions have shifted. More importantly, the size of these
Number of vertices 360,450 528,618 967,312
Burner length modeled 17.78 cm 17.78 cm 7.62 cm regions is much smaller compared to the LDV plot. This is an
Fig. 4 Axial velocity „ W … contours at Z Ä0.3 cm
Fig. 5 Axial velocity „ W … contours at Z Ä1.3 cm
Fig. 6 XY „ UV … velocity vectors at different Z -planes, RSTM results
indication of the well-known diffusive nature of the standard ␬-␧ zones appear larger than the measured ones. This again suggests
model. In contrast, the NCC’s ␬-␧ model, Fig. 5共a兲, failed to re- that the cubic nonlinear ␬-␧ model may have been overadjusted to
solve the recirculation structures. It instead calculated rather well- counteract the diffusiveness found in the standard ␬-␧ model. To-
behaved and circular reversed flow regions, without much swirler- gether, the results from both ␬-␧ models strongly indicate the
swirler interactions. In the NCC studies, 关10,11兴, it was assumed downside of ad hoc modifications in turbulence modeling. Modi-
that the combustor geometry was rotationally periodic every 90 fying the two-equation formulation on a case-by-case basis is
deg. While this is true in geometric terms, the LDV data in both problematic and therefore should be used with caution, particu-
Figs. 4 and 5 show that the flowfield is not symmetrical about larly when large gradients, strong shear, and high swirl are ex-
both the X and Y-axes. Therefore, a periodic boundary condition pected.
may not be appropriate for this particular geometry and swirler While Fig. 5 characterizes the complex flow physics associated
array configuration. Nevertheless, since the corner kidney-shaped with the discrete-jet swirlers, Fig. 6 shows with velocity vectors
recirculation is located away from the symmetry boundaries, the how the swirling jets interact with each other and with the solid
flow structure should be resolved regardless of the symmetry as- walls along the axial direction (Z⫽1.3 cm, 2.3 cm, and 5.3 cm兲.
sumption. In addition, some of the NCC-computed recirculation In Fig. 6共b兲, all the ‘‘corner’’ and ‘‘side’’ swirler vortex centers
Fig. 7 U velocity
have shifted relative to each swirler group’s center due to wall sents the U-velocity at different axial locations downstream of the
effect and swirling jet interactions. Smaller, secondary induced bottom, right swirler. Immediately below the swirler, the LDV
vortices due to these interactions are also visible. The center data shows small, negative U-velocity, which quickly becomes
swirler, on the other hand, is able to preserve its characteristics positive further downstream, reaches a peak, and gradually tapers
while gaining strength from its neighbors, hence maintaining a off. This implies an axial shift of momentum from the presence of
nearly circular reversed-flow structure. Further downstream at Z neighboring swirlers. Comparing the LDV measurements with the
calculated results provides an obvious distinction among the dif-
⫽2.3 cm, Fig. 6共b兲, due to jet spreading, expanding, and interact- ferent turbulence models. The RSTM agrees exceptionally well
ing processes and wall effect, those swirlers neighboring the wall with the LDV data, while the two ␬-␧ models give contrasting
gradually merge and eventually form a single, albeit weaker, profiles, particularly before Z⫽3.0 cm where recirculation is
swirl, Fig. 6共c兲. strong. At Z⫽1.4 cm, the KIVA’s standard ␬-␧ model calculated a
The line plots in Figs. 7–10 quantitatively show, in further de- peak value approximately 175% higher than the measurements,
tail, how the RSTM computations give better agreement with the while the NCC, unable to resolve the peak, computed much lower
measurements than do the ␬-␧ calculations. Since in the LDV values.
experiment two velocity components (U,W) were measured, it is Figure 7共b兲 contains a line-plot comparison along axial loca-
appropriate to present comparisons in terms of these components. tions between the four bottom-right swirlers. The RSTM shows a
Figure 7 contains two line plots of U-velocity comparison among close agreement with the LDV data, with a maximum offset near
the measured and computed results. The first plot, Fig. 7共a兲, pre- Z⫽1.0 cm. The ␬-␧ results exhibit similar behavior as in Fig.
Fig. 8 Axial „ W … velocity
Fig. 9 Swirl velocity
Fig. 10 Turbulence kinetic energy „␬…
7共a兲, except the standard ␬-␧ model calculated much lower values results and experimental data. The SSG RSTM was used to de-
and the cubic nonlinear model, much higher. Both plots in Fig. 7 scribe the turbulence inside the combustor. The RSTM-computed
illustrate the inconsistency of ␬-␧ models when the flow is results were compared to LDV measurements, as well as results
complex. from the cubic nonlinear ␬-␧ turbulence model in the NCC, and
The line plots in Fig. 8 quantitatively present the axial velocity the standard ␬-␧ model in the KIVA code: The result comparisons
component 共W兲. In Fig. 8共a兲, from the LDV data zero axial- show that the RSTM accurately captured most of the flow char-
velocity occurs at approximately Z⫽2.2 cm downstream of the acteristics. Both ␬-␧ models exhibited contrasting behaviors, ei-
swirler, which may be assumed to be the size of the recirculation ther over or undercalculating eddy viscosity in regions of high
bubble. The RSTM profile closely follows the LDV data, although swirl and strong recirculation. Similar conclusion was also
a maximum offset of 25% occurs below the recirculation region. achieved in a previous study of an LDI combustor with an axial
The NCC’s ␬-␧ model computed a bubble size nearly twice the swirler array, 关17兴.
measured one, while the KIVA’s model calculated only half, Accurate calculation of swirl and recirculation zones is impor-
clearly displaying its diffusive nature. Figure 8共b兲 on the other tant in deciding fuel spray locations as well as in the design of
hand, does not show as distinct a comparison as does Fig. 8共a兲. combustors. Within the Reynolds averaging approach, the RSTM
The RSTM result exhibits the general trend of the LDV profile for has proven to be a proper choice in accurately calculating com-
Z⬍1.0 cm as well as Z⬎3.0 cm. The NCC calculates a substan- plex flow characteristics, given the general capacity and availabil-
tially stronger reversed flow in the Z⬍1.0 cm region. ity of current-generation computers. The RSTM computations re-
Although the V-velocity component was not directly measured quired 178.4 ␮sec per cycle per grid point on an SGI Origin 3400
in the LDV experiment, it would still be useful to show the line platform, while the computation with KIVA’s ␬-␧ model took 87.6
plots of swirl velocity and turbulence kinetic energy, ␬. The plots ␮sec. The extra computational overhead from using the RSTM is
that follow, therefore, serve as comparisons of trend rather than justified by its accuracy and robustness. No ad hoc modifications
accuracy. The swirl velocity in Fig. 9共a兲 carries a characteristic are necessary for complex geometry and flow conditions. Future
similar to the U-velocity plot in Fig. 7共a兲. The swirl quickly peaks work may include study of similar combustor geometry with dif-
at about Z⫽1.4 cm, then gradually settles. The RSTM basically ferent swirl angles as well as with combustion and fuel spray.
follows the trend and stays within 20% of the measured values.
The highest swirl computed by the standard ␬-␧ model occurs at a
Z value close to the measured one, and the cubic nonlinear ␬-␧ Acknowledgment
model was unable to resolve any swirl structure within the recir- The authors would like to thank NASA for their financial assis-
culation zone. tance under NASA Grant Contract #: NAG3-2088.
Figure 9共b兲 shows a more complex profile, particularly for Z
⬍3.0 cm. The measured swirl values are generally lower than
those in Fig. 9共a兲, and fluctuate considerably inside the bubble. References
Both ␬-␧ models generally yield excessive swirl up to Z 关1兴 Lefebvre, Arthur H., 1983, Gas Turbine Combustion, Hemisphere, Washing-
⫽2.5 cm. Halfway downstream and onward, however, all com- ton, DC.
puted results agree quite well with the LDV measurements. 关2兴 Shih, T.-H., Chen, K.-H., Liu, N.-S., and Lumley, J. L., 1998, ‘‘Modeling of
The last set of figures contains the ␬-levels at the same loca- Turbulent Swirling Flows,’’ NASA-TM—113112.
关3兴 Shih, T.-H., Chen, K.-H., and Liu, N.-S., 1998, ‘‘A Non-Linear k-epsilon
tions as in the preceding line plots. Figure 10共a兲 shows that the Model for Turbulent Shear Flows,’’ AIAA Paper 98-3983.
NCC’s ␬-␧ model fails to capture the steep profile of the measure- 关4兴 Stubbs, R. M., and Liu, N.-S., 1997, ‘‘Preview of the National Combustion
ments, near Z⫽1.5 cm. Instead, the model gives a generally flat Code,’’ AIAA Paper 97-3314.
line that falls below the LDV data. Since eddy viscosity, ␮ T , is 关5兴 Quealy, A., Ryder, R., Norris, A., and Liu, N.-S., 2000, ‘‘National Combustion
Code: Parallel Implementation and Performance,’’ NASA TM-2000-209801.
directly proportional to the square of ␬, a small ␬ value results in 关6兴 Launder, B. E., and Spalding, D. B., 1974, ‘‘The Numerical Computation of
an even smaller ␮ T . By the same token, the standard ␬-␧ model, Turbulent Flows,’’ Comput. Methods Appl. Mech. Eng., 3, pp. 269–289.
having ␬ values much higher than the LDV data, has excessively 关7兴 Amsden, A. A., 1999, ‘‘KIVA-3V, Release 2, Improvements to KIVA-3V,’’ Los
large ␮ T throughout the entire combustor length. In Fig. 10共b兲, all Alamos National Laboratory, LA-13608-MS, Los Alamos, NM.
关8兴 Hanjalic, K., 1999, ‘‘Second-Moment Turbulence Closures for CFD: Needs
calculated results fall below the LDV data, although exhibiting and Prospects,’’ Int. J. Comput. Fluid Dyn., 12, pp. 67–97.
similar trend. 关9兴 Yang, S. L., Peschke, B. D., and Hanjalic, K., 2000, ‘‘Second-Moment Closure
Model for IC Engine Flow Simulation Using KIVA Code,’’ ASME J. Eng. Gas
Turbines Power, 122, pp. 357–365.
Conclusion 关10兴 Iannetti, A., Tacina, R., Jeng, S.-M., and Cai, J., 2001, ‘‘Toward Accurate
Prediction of Turbulence, Three-Dimensional, Recirculation Flows With the
An LDI combustor with a discrete-jet swirler array was mod- NCC,’’ Paper No. AIAA-2001-0809.
eled for cold-flow investigation. The complex flow physics asso- 关11兴 Iannetti, A., Tacina, R., Cai, J., and Jeng, S.-M., 2001, ‘‘Multi-Swirler Aero-
ciated with the swirler array was described with the aid of CFD dynamics: CFD Predictions,’’ AIAA Paper No. 2001-3575.
关12兴 Jeng, S.-M, Cai, J., and Tacina, R., 2001, ‘‘Multi-Swirler Aerodynamics: Ex- 关15兴 Speziale, C. G., Sarkar, S., and Gatski, T. B., 1991, ‘‘Modeling the Pressure-
perimental Measurements,’’ AIAA Paper No. 2001-3574. Strain Correlation of Turbulence: An Invariant Dynamical Systems Approach,’’
关13兴 Craft, T. J., Launder, B. E., and Suga, K., 1993, ‘‘Extending the Applicability J. Fluid Mech., 227, pp. 245–272.
of Eddy Viscosity Models Through the Use of Deformation Invariations and 关16兴 Gridgen CFD and FEA Pre-Processing, V.13.3, Pointwise, Inc.
Non-Linear Elements,’’ 5th IAHR Conference on Refined-Flow Modeling and
关17兴 Yang, S. L., Siow, Y. K., Peschke, B. D., and Tacina, R., 2003, ‘‘Numerical
Turbulence Measurement, Paris, Sept. 7–10.
关14兴 Shih, T.-H., and Lumley, J. L., 1993, ‘‘Remarks on Turbulent Constitutive Study of Nonreacting Gas Turbine Combustor Swirl Flow Using Reynolds
Relations,’’ Math. Comput. Modell., 18共9-6兲. Stress Model,’’ ASME J. Eng. Gas Turnbines Power, 125, pp. 804 – 811.
Development of an Exhaust Gas
Recirculation Distribution
Prediction Method Using
K. Yoshizawa
K. Mori
Three-Dimensional Flow Analysis
Y. Matayoshi and Its Application
S. Kimura A multidimensional computational fluid dynamics (CFD) method has been used to im-
prove the exhaust gas recirculation (EGR) distribution in the intake manifold. Since gas
Powertrain and Environment Research flow in the intake system is affected by intake pulsation caused by the gas exchange
Laboratory, process, a pulsation flow simulation is used. A one-dimensional gas exchange calculation
Nissan Research Center, is combined with three-dimensional intake gas flow calculation to simulate pulsation flow.
Nissan Motor Company, Ltd., This pulsation flow simulation makes it possible to predict the EGR distribution. The gas
1, Natsushima-cho, flow in the intake system was analyzed in detail. It was found that a reverse flow region
Yokosuka 237-8523, Japan formed downstream of the throttle valve. The size and shape of the reverse flow region
greatly depend on the engine operating conditions. With a conventional EGR system, it is
difficult to distribute EGR uniformly under various engine operating conditions. A new
EGR system that uses a spiral flow to mix the fresh air and EGR gas has been developed
to obtain a uniform EGR distribution. As a result of adopting this system, a uniform EGR
distribution is obtained regardless of the engine operating conditions. This spiral flow
EGR system was applied to a low-emission vehicle (LEV) put on the Japanese
market. 关DOI: 10.1115/1.1581896兴
Introduction change process. This pulsation flow simulation makes it possible

to predict the EGR distribution under the influence of intake pul-
Continuing concerns about air quality have necessitated further
sation. With this analysis method, the EGR distribution was accu-
reductions of automotive emissions such as nitrogen oxides (NOx)
rately predicted to within 3–5%.
and carbon dioxide (CO2 ). Recently, a heavy Exhaust Gas Recir- After validating the EGR distribution predicted with the pulsa-
culation 共EGR兲 system has attracted much interest as a key tech- tion simulation method, gas flow in the intake system was ana-
nology for reducing exhaust emissions of internal combustion en- lyzed in detail. It was found that a reverse flow region formed
gines, because it has the potential to lower both NOx and CO2 downstream of the throttle valve. The size and shape of the re-
levels simultaneously, 关1兴. Heavy exhaust gas recirculation 共EGR兲 verse flow region greatly depend on the engine operating condi-
refers to an EGR rate of more than 30% in the cylinder. This kind tions. With a conventional EGR system, it is difficult to distribute
of large EGR rate has become possible with development of EGR uniformly under various engine operating conditions, be-
direct-injection gasoline 共DIG兲 engines, 关2兴, since the fuel distri- cause the distribution is greatly influenced by the reverse flow
bution in the cylinder is controlled more accurately than in port- region, 关6兴.
injection gasoline engines. Under the application of heavy EGR, A new EGR system has been developed to obtain a uniform
the EGR distribution in the intake system is an important factor. If EGR distribution. The EGR passage is connected to the intake
the EGR distribution is not uniform, a cylinder with a large EGR manifold tangentially and recirculated exhaust gas flows into the
gas concentration would cause unstable combustion. That would manifold in a spiral manner. This spiral flow enhances the mixing
generate large emissions and also deteriorate drivability. There- of fresh air and recirculated exhaust gas and is less influenced by
fore, achieving a uniform EGR distribution is essential. However, changes in the reverse flow region. The EGR distribution is im-
time-consuming experimental research has so far been the only proved by this newly developed mixing system. As a result of
way to evaluate the EGR distribution in the intake manifold. De- adopting this system, a uniform EGR distribution is obtained re-
sign and development lead time must be shortened in order to gardless of the engine operating conditions.
meet consumers’ rapidly changing needs, 关3兴. Hence, there have This spiral flow EGR system was then modified for production
been strong expectations for the development of a simulation vehicle application, taking into account layout limitations and pro-
method for predicting the EGR distribution. duction cost. It has been applied to a low-emission vehicle 共LEV兲
In this connection, computational fluid dynamics 共CFD兲 has put on the Japanese market.
been used. Since gas flow in the intake system is affected by
intake pulsation caused by the gas exchange process, a pulsation Exhaust Gas Recirculation „EGR… Distribution in
flow simulation method is used, 关4,5兴. A one-dimensional gas ex-
change calculation and a three-dimensional intake gas flow calcu- Conventional EGR Systems and Engineering Tasks for
lation are used to simulate pulsation flow caused by the gas ex- Heavy EGR System
Figure 1 shows a schematic diagram of a conventional EGR
Contributed by the Internal Combustion Engine Division of THE AMERICAN SO- system. The engine system comprises a collector, an intake mani-
ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received by the ICE
fold and an exhaust manifold. The throttle body, incorporating the
Division, Feb. 7, 2002; final revision received Aug. 2002. Associate Editor: D. As- throttle valve, is connected upstream of the collector. An EGR
sanis. passage connects the exhaust manifold and the throttle body. In
Fig. 3 Influence of average EGR rate on EGR variance
Fig. 1 Schematic diagram of conventional EGR system

velop a heavy EGR system. Therefore, as a first step, CFD simu-
lation was applied to predict the EGR distribution.
this conventional system, EGR gas is introduced downstream of Exhaust Gas Recirculation „EGR… Distribution Predic-
the throttle valve, i.e., far upstream of the collector, to allow
enough time for mixing with the fresh air so as to attain a uniform
tion With Pulsation Flow Simulation
ERG distribution. Pulsation Flow Simulation Method. When an engine is run-
Figure 2 shows the EGR distribution for different engine oper- ning, the gas exchange in the cylinders takes place intermittently.
ating conditions. These results were obtained experimentally us- The gas flow in the collector and intake manifold is greatly af-
ing the conventional EGR system shown in Fig. 1. The EGR rate fected by intake pulsations. Therefore, it is necessary to simulate
of each cylinder is calculated by the EGR gas flow and its diffusion under pulsation flow in order
EGRR共cyl.#)⫽M EGR共cyl.#)/ 关 M EGR共cyl# 兲 ⫹M air共cyl#)] (1) to predict the EGR distribution accurately, 关7兴. That requires a
model of the entire engine system, including the intake and ex-
where EGRR is the EGR rate, M EGR is the EGR gas mass flow, haust systems. However, it would be impractical to simulate the
and M air is the air mass flow. The CO2 concentration in the air gas flow in the entire engine system with a three-dimensional
inducted into each cylinder was measured and the EGR gas mass model. Therefore, a pulsation flow simulation method was used,
flow was calculated. The EGR rate in each cylinder and its ten- in which a one-dimensional gas exchange calculation and a three-
dency vary depending on the engine operating conditions. These dimensional exhaust gas flow calculation were combined.
results suggest that it is difficult to achieve a uniform EGR distri- Figure 4 shows an outline of the pulsation flow simulation
bution under different engine operating conditions. method. In the pulsation simulation, the engine system model for
Figure 3 shows the EGR rate in each cylinder for different calculating pulsation flow caused by the gas exchange process is
average EGR rates. The EGR rate is increased assuming that one-dimensional. In the one-dimensional calculation, the entire
heavy EGR is applied. The variance in the EGR rate becomes engine system model is described as a combination of one-
larger as the average EGR rate increases. These results suggest dimensional pipes and dimensionless containers. In the three-
that if heavy EGR is applied with a conventional EGR system, the dimensional calculation, an intake system model that includes the
EGR rate variance would become unacceptably large. If the EGR throttle body, collector and intake manifold is used. First, a one-
gas concentration difference in each cylinder reached this level, dimensional gas exchange simulation is run to obtain velocity,
the cylinder with a large EGR gas concentration would cause pressure, temperature, and density data for one cycle 共720 crank
unstable combustion. That would generate large emissions and angle degrees兲 that are necessary for a three-dimensional calcula-
also deteriorate drivability. Accordingly, it would be difficult to tion. Next, the three-dimensional calculation is executed using the
apply heavy EGR with a conventional EGR system. Therefore, it results of the one-dimensional calculation as the boundary condi-
is necessary to develop a robust EGR system that makes it pos- tions. It is possible to run both the one-dimensional gas exchange
sible to achieve a uniform EGR distribution under different engine calculation and the three-dimensional flow calculation together at
operating conditions even when heavy EGR is applied. one time as shown in Fig. 4. The difference between the separate
However, time-consuming experimental research has so far calculation method and the combined calculation method is neg-
been the only way to evaluate the EGR distribution in the intake
system. Considerable time and cost are needed to study and de-
Fig. 2 Influence of engine operating conditions on EGR

distribution Fig. 4 Schematic diagram of pulsation flow simulation method
Fig. 6 Computational grid and boundary conditions
Fig. 5 One-dimensional gas exchange calculation results
culated by multiplying the EGR gas concentration, the gas mass

ligible. Therefore, in this study, the separate calculation method flow rate and a calculation time step at each crank angle, and
was used, since it is easier to analyze both the one-dimensional totaling the results for one cycle. Specifically, the following ex-
and three-dimensional results in detail. pressions were used to calculate the EGR gas mass and the air
mass.
One-Dimensional Gas Exchange Calculation. An in-house
i⫽1,720
gas exchange calculation code has been adopted for performing
one-dimensional calculations, 关8,9兴. The data needed for the cal-
culations are the engine specifications, intake and exhaust system
M EGR共cyl.#)⫽ 兺关 C EGR共cyl.# 兲 •M gas共cyl.#)•⌬t] (2)
dimensions and engine operating conditions. Figure 5 shows an i⫽1,720

example of the one-dimensional gas exchange calculation results.
The velocities calculated at the EGR gas inlet and outlet of the
M air共cyl.#)⫽ 兺关关 1⫺C EGR共cyl.# 兲兴 •M gas共cyl.#)•⌬t] (3)
intake manifold for one cycle are presented. Air was introduced where C EGR is the EGR gas concentration, M gas is the total gas
into each cylinder in turn according to the ignition timing. Yet the mass flow rate and ⌬t is the calculation time step 共equivalent to
EGR gas velocities are almost stable for the entire cycle, though one crank angle degree兲.
some effect of the pulsation flow can be observed. These velocity At the start of the calculation, namely, in the initial condition,
data were used in the three-dimensional calculation as the bound- there is no EGR gas in the intake manifold. Therefore, a certain
ary conditions. amount of time is required for the EGR gas to diffuse through the
Three-Dimensional Gas Flow Calculation. The general intake manifold. In this case, eight to ten calculation cycles were
STAR-CD code was used for the three-dimensional calculation, needed before the results converged to show cycle-to-cycle stabil-
关10兴. Table 1 shows the flow analysis conditions. The flow was ity, though it depended on the number of grid points and engine
assumed to be compressible. The conservation equations of mass, operating conditions. One of the factors influencing the calcula-
momentum and energy, and the turbulence energy and its dissipation accuracy is the boundary conditions obtained by the one-
tion 共␬-␧ turbulence model兲 were used. The law of the wall was dimensional gas exchange calculation. It is also important to
used at the wall boundary. The differencing scheme was upwind. model the throttle valve with a fine grid to prevent calculation
Figure 6 shows the computational grids and the boundary con- divergence.
ditions. The computational grids consist of the collector and in- Figure 7 shows examples of the calculated results for the EGR
take manifold, including the intake plenum, throttle body, throttle gas concentration distribution. In this calculation, the EGR gas
valve and the EGR gas introduction port. The number of grid was introduced upstream of the collector. It is seen that the EGR
points depends on the intake system configuration and it is gener-
ally between 150,000–200,000. The intake plenum is used for
making the calculations stable. Boundary conditions are set at the
inlet of the intake plenum, the inlet of the EGR gas introduction
port and at the four outlets of the intake manifold runners.
Unsteady flow that varied with the crank angle degree was
simulated using the one-dimensional gas exchange calculation re-
sults as the boundary conditions 共Fig. 5兲. In this calculation, the
EGR gas was marked to distinguish it from the fresh air by using
a scalar definition as its physical properties. The EGR rate of each
cylinder was calculated at the four outlets of the intake manifold
runners by Eq. 共1兲. The EGR gas mass of each cylinder was cal-
Table 1 Flow analysis conditions
Fig. 7 Calculated results for EGR gas concentration

distribution
where ERROREGR is the calculation error of the EGR rate,
EGRRpre is the predicted EGR rate, and EGRRmea is the measured
EGR rate. The calculation error was within 3–5% in both cases,
though the predicted EGR rate was a little smaller than the mea-
sured one. These results confirmed the validity of the pulsation
flow simulation. It is possible to predict the EGR rate quantita-
tively with this method.
Physical Phenomena in the Intake System and Their

Effect on Exhaust Gas Recirculation „EGR… Distribu-
tion Characteristics
Gas Flow in the Intake System. After validating the EGR
rate predicted with the pulsation flow simulation method, physical
phenomena in the intake system that influence the EGR distribu-
tion were examined. As noted earlier, the EGR gas is introduced
Fig. 8 Calculation error of EGR rate compared with experi- downstream of the throttle valve in conventional EGR systems
mental data 共Fig. 1兲. In that case, the EGR rate variance in each cylinder and
its tendency vary depending on the engine operating conditions
共Fig. 2兲. To analyze this factor, physical phenomena in the intake
system were investigated in detail.
gas mixes with the fresh intake air coming from upstream of the Figure 9 shows gas flow trace lines in the throttle body obtained
throttle valve and diffuses in the collector. The EGR gas concen- by the steady flow simulation. There is a reverse flow region
tration decreases as the gas flows downstream. downstream of the throttle valve. In this region, the fresh intake
An experiment was then conducted to validate the EGR rate air is recirculating. In conventional EGR systems, the EGR gas
predicted by the pulsation flow simulation method. To investigate would be induced into the reverse flow region and would show
the effect of the engine specifications, two types of four-cylinder complex flow characteristics in this region under the strong influ-
engine were used. Figure 8 shows the calculation error of the EGR ence of the recirculating flow. It is assumed that the intake flow
rate. The calculation error was calculated by in the throttle body has a large effect on the EGR distribution
downstream.
ERROREGR⫽ 共 EGRRpre⫺EGRRmea兲 /EGRmea•100 (4) A detailed analysis was then made of the reverse flow region.
First, a steady-flow simulation using a simple throttle body model
with a straight pipe was run to study the geometric features of the
reverse flow region. A pulsation flow simulation was then con-
ducted to analyze the effect of the engine operation conditions.
Reverse Flow Region Downstream of Throttle Valve. Fig-
ure 10 shows the steady-flow simulation results obtained with the
simplified throttle body model. The gas velocities are shown in a
side view section. The reverse flow region on the downstream side
of the throttle valve edge is larger than that on the upstream side.
The vortex center is located on the upstream side of the throttle
valve edge 共lower side in Fig. 10兲. Figure 11 shows the reverse
flow region in a cubic form. The region of the computational grid
having an upstream flow velocity was selected to show the reverse
flow region.
The shape of the reverse flow region was analyzed for different
throttle body sizes and engine operating conditions. In all cases,
Fig. 9 Gas flow trace lines
Fig. 10 Section view of reverse flow region
Fig. 13 Trace lines of EGR gas flow „position of EGR introduc-
tion port: 4.0 D…
Fig. 11 Plan view and side view of reverse flow region
the reverse flow region was always just downstream of the throttle
valve. Schematically, the location of the reverse flow region was
similar for different throttle body sizes. The shape and size of the
reverse flow region changed depending on the opening angle of
the throttle valve and the gas mass flow rate. Figure 12 compares
reverse flow regions for different engine loads. The reverse flow
region became larger under a low load condition with a small
Fig. 14 Schematic gas flow pattern „EGR gas introduced out-
throttle valve opening angle.
side reverse flow region…
The size of the reverse flow region was then investigated quan-
titatively. It was found that the length between the throttle valve
and the downstream end of the reverse flow region was similar for and the schematic flow pattern, respectively. The EGR gas was
different throttle body sizes. The length was between 1.3–1.8 introduced horizontally to the throttle valve axis at the center. The
times the throttle body diameter 共D兲, though the shape of the location of the EGR introduction pipe was 4.0 D downstream.
reverse flow region varied depending on the throttle valve opening In this case, the EGR gas is caught between the upper and
angle. lower main streams of fresh air, and there is not enough turbu-
In a conventional EGR system, the EGR gas is introduced lence for the EGR gas and fresh air to mix. As a result, the EGR
downstream of the throttle valve. As explained above, the size of gas is quickly carried away toward the collector before it diffuses
the reverse flow region varies depending on the engine operating sufficiently. This results in poor mixing and the EGR gas is con-
conditions. In some cases, the EGR gas would be introduced into centrated in some parts of the collector. Therefore, it is assumed
the reverse flow region and, in other cases, it would be introduced that the EGR rate to each cylinder would differ substantially.
outside the region. It is assumed that the EGR gas flow in the The EGR gas flow was then analyzed when the EGR gas was
intake system and the EGR distribution characteristics are signifi- introduced into the reverse flow region. Figures 15 and 16 show
cantly affected by changes in the reverse flow region. the EGR gas trace lines and the schematic flow pattern, respec-
Effect of EGR Gas Introduction Location on EGR tively. Again, the EGR gas was introduced horizontally to the
Gas Flow. The effect of the EGR gas introduction location on throttle valve axis at the center. The location of the EGR introduc-
the EGR gas flow was examined in detail. First, the EGR gas flow tion pipe was 1.0 D downstream of the throttle valve. In this case,
was analyzed when the EGR gas was introduced outside the re- the EGR gas circulates in the reverse flow region, allowing more
verse flow region. Figures 13 and 14 show the EGR gas trace lines opportunity for it to mix with the fresh air. However, as explained
earlier, the size and shape of the reverse flow region vary depend-
ing on the engine operating conditions. That means the mixing
rate of the EGR gas and fresh air also varies accordingly. These
Fig. 15 Trace lines of EGR gas flow „position of EGR introduc-

Fig. 12 Effect of throttle opening angle on reverse flow region tion port: 1.0 D…
Fig. 16 Schematic gas flow pattern „EGR introduced in re-
verse flow region…
results explain why the EGR rate variance in each cylinder and its
tendency vary depending on the engine operating conditions as
shown in Fig. 2. Conventional EGR systems are thus not robust Fig. 18 EGR rates with lower side blow-type EGR system
enough. It is difficult to achieve a uniform EGR distribution under
a wide range of engine operating conditions where heavy EGR is
applied, because the degree of mixing is greatly dependent on the
reverse flow region. with a small EGR rate. However, the EGR rate variance in each
cylinder became as large as 5% when a large EGR rate such as
50% was applied. It is assumed that the direction of the upward
New Exhaust Gas Recirculation „EGR… System for Uni-
flow due to the main streams of the fresh air was still slightly
form EGR Distribution influenced by the throttle valve opening angle.
Lower Side Blow-Type EGR System. New EGR systems There is still room to improve the lower side blow-type EGR
were investigated based on the foregoing analytical results. The system by optimizing the location and size of the EGR introduc-
key factor was to develop an EGR system that would not be tion ports. However, one disadvantage of this EGR system is that
susceptible to the reverse flow region even though the EGR gas the momentum of the main gas flow coming from the downstream
was still introduced downstream of the throttle body. A lower side side free edge of the throttle valve is not used for mixing energy.
blow-type EGR system has been developed by applying the It is concluded that it is possible to achieve a uniform EGR dis-
knowledge obtained about the gas flow in the throttle body. Figure tribution with a small EGR rate but difficult to do so under heavy
17 shows the location of the EGR introduction ports and sche- EGR conditions with this lower side blow-type EGR system.
matic gas flow pattern. The EGR gas is introduced on the lower Spiral Blow-Type EGR System. As explained above, EGR
side of the throttle body 共upstream edge side of the throttle valve兲, distribution characteristics are substantially influenced by the re-
not in the center. As indicated in the figure, two EGR introduction verse flow region generated downstream of the throttle valve.
ports are facing each other. As shown in Fig. 14, the main stream Based on the analytical data presented here, it is concluded that
of the fresh air coming from the upstream edge of the throttle the requirements for achieving a uniform EGR distribution under
valve progresses toward the upper side. The lower side blow-type a wide range of engine operating condition are as follows:
EGR system makes effective use of this upward stream. The EGR • to avoid interference with the reverse flow region 共enhance
gas coming from both sides of an EGR introduction port uses the robustness兲,
momentum of the upward main stream to create a twin-vortex • to increase the travel time of the EGR gas from its introduc-
flow. This twin-vortex flow enhances the mixing of the EGR gas tion port to the intake manifold 共enhance the mixing effect兲,
and fresh air to achieve a uniform EGR distribution even when and
heavy EGR is applied. • to mix the EGR gas with both main streams of fresh air 共make
Figure 18 shows the EGR rate in each cylinder with the lower effective use of the momentum and turbulence of the main
side blow-type EGR system. Both the calculated and experimental streams兲.
results are shown. The EGR rate variance in each cylinder was One practical way to meet these requirements is to adopt a
much smaller than with the conventional EGR system 共Fig. 3兲, spiral blow-type EGR system like that outlined in Fig. 19. One
especially when the EGR rate was increased to about 30%. With EGR gas introduction port each is located just at the rear of the
the lower side blow-type EGR system, a uniform EGR distribu- downstream and upstream side free ends of the throttle valve. The
tion was obtained under most of the engine operating conditions two introduction ports are opened in a cross-flow 共or counter-
flow兲 manner in opposite directions. Each port is connected to the
throttle body tangentially and EGR gas flows into the passage in a
spiral manner. The cross-sectional shape of each port is elongated
along the direction of the main streams of fresh air to avoid inter-
ference with the reverse flow region. The mixing of the EGR gas
and fresh air is thus enhanced by creating a spiral flow.
The EGR gas flows into both main streams of fresh air in a
tangential manner, creating a strong spiral flow in the downstream
direction by using the momentum of the main streams. Each in-
troduction port is inclined downstream so as to form a predeter-
mined angle ␪ 共lead angle with respect to the fresh air flow direc-
tion兲. The optimum lead angle ␪ depends on the intake system
configuration and it is generally between 5–15°. The port is ex-
tended along a line tangent to the circular cross section of the
throttle body. This extended pipe serves as a guide for the main
stream of the gas, causing it to flow in a spiral manner. Both the
Fig. 17 EGR gas introduction ports and gas flow pattern of lead angle and extended port design strengthen the spiral flow.
lower side blow-type EGR system The EGR gas mixes with the fresh air using this spiral flow gen-
Fig. 22 Shapes of EGR gas introduction ports
that seen for the conventional EGR system 共Fig. 3兲, especially
when the EGR rate was increased. With the spiral blow-type EGR
Fig. 19 Schematic diagram of spiral blow-type EGR system system, a uniform EGR distribution is obtained under a wide
range of engine operating conditions even when a large EGR rate
is applied.
erated intentionally. In addition to that, the EGR gas travels a
greater distance along the cylindrical wall surface toward the col- Application of Spiral Blow-Type Exhaust Gas Recircu-
lector than in a conventional EGR system. This longer distance
lation „EGR… System to Nissan QR Engine
allows more time for the EGR gas to mix with the fresh air. The
EGR gas diffuses from the wall surface toward the center of the As seen in the preceding section, the spiral blow-type EGR
collector, achieving a uniform EGR distribution by the time the system makes it possible to achieve a uniform EGR distribution
gas goes into the collector. As a result, the EGR rate in each for a wide range of engine operating conditions, even when heavy
cylinder becomes equal. Figure 20 shows the trace lines of the EGR is applied. As the next step, this EGR system was modified
EGR gas flow. It is clearly seen that the EGR gas flows into the for application to the Nissan QR engine. There were two engineer-
collector in a spiral manner. ing tasks involved in that application. One was to improve the
Figure 21 shows the EGR rate in each cylinder with this spiral discharge coefficient of the intake system and the other was to
blow-type EGR system. Both the calculated and experimental re- reduce the production cost.
sults are shown. The EGR rate difference was much smaller than As shown in Fig. 19, in the spiral blow-type EGR system, the
ports for introducing EGR gas are extended into the throttle body
to intensify the spiral flow. These extended ports generate pressure
loss in the throttle body and reduce the discharge coefficient. That
would cause volumetric efficiency to decline, thereby reducing
engine power. Therefore, the discharge coefficient must be im-
proved. One simple way would be to remove the guide of the
EGR gas introduction ports. However, without a port guide, it is
assumed that the spiral flow would become weak and the mixing
effect would be smaller. It is necessary to enhance the spiral flow
without using a port guide. Therefore, the cross-sectional area of
the EGR introduction port was reduced to increase the gas flow
velocity.
Figure 22 shows the shapes of the EGR gas introduction ports
that were examined. Two modified types without a port guide
Fig. 20 Trace lines of EGR gas flow with spiral blow-type EGR
system
were compared with the original type with the port guide. Case 1
had the same sectional area and the guide was removed. In case 2,
the sectional area was reduced. Figure 23 shows the calculated
EGR rates. As expected, with case 1, the EGR rate variance was
increased compared with the base type, since the mixing effect
was smaller. However, with case 2, the level of EGR rate variance
was similar to that of the original type. These results confirmed
that it is possible to achieve a uniform EGR distribution with the
spiral flow-type EGR system without port guides, if the size and
shape of the port cross section is optimized.
The possibility of reducing the production cost was then inves-
tigated. As shown in Fig. 1, conventional EGR systems have only
one EGR introduction port. By contrast, the spiral blow-type EGR
system requires two ports, one each on the upper and lower sides,
as shown in Fig. 19. The use of two ports would increase the
production cost. In addition to that, there are layout limitations
around the throttle body and it would be difficult to install two
ports in some engine types. To reduce the production cost, the
Fig. 21 EGR rates with spiral blow-type EGR system number of ports should be reduced to one.
Fig. 23 Effect of EGR gas introduction port shape on EGR rate
„calculated…
Therefore, the effect of the location of the EGR gas introduc-

tion port was examined. Figure 24 shows the port locations that
were investigated. Both upper and lower side locations were ana-
lyzed. In the case of one port, the port location and the lead angle Fig. 26 Spiral blow-type EGR system applied to Nissan QR
␪ 共Fig. 19兲 were optimized. Figure 25 shows the calculated EGR engine
rates. With an upper side port location, the EGR rate variance
increased compared with the base type. As shown in Fig. 10, the
cross-sectional area where the upper main flow of air passes is though the EGR rate in the #4 cylinder was a little smaller than
smaller than that of the lower side. The velocity of the main air that of the other cylinders. As explained above, the velocity of the
flow on the upper side is larger than that on the lower side. There- main air flow on the lower side is smaller than that on the upper
fore, the spiral flow generated by the EGR gas flow weakens, side. It is possible to generate sufficient spiral flow with just one
reducing the mixing effect. For that reason, the EGR rate variance EGR gas introduction port. These results confirmed a uniform
increased. On the other hand, with the lower side port location, the EGR distribution can be achieved with the spiral flow-type EGR
EGR rate variance was at the same level as that of the base type, system having only one EGR gas introduction port, if the port
location and lead angle ␪ 共Fig. 19兲 are optimized.
Figure 26 shows the spiral blow-type EGR system applied to
the Nissan QR Engine. The EGR gas introduction passage is at-
tached to the throttle body. The single EGR introduction port is
connected to the lower side of the throttle body tangentially.
Therefore, the EGR gas flows into the throttle body in a spiral
manner.
Figure 27 shows the EGR rate variance obtained experimentally
with this system under different engine operating conditions. The
EGR rate variance indicates the difference between the maximum
and minimum EGR rates among the four cylinders. The variance
was less than three percentage points under all engine operating
conditions with the application of heavy EGR. This result demon-
strates that the spiral blow-type EGR system makes it possible to
Fig. 24 Location of EGR gas introduction port
Fig. 25 Effect of EGR gas introduction port location on EGR

rate „calculated… Fig. 27 EGR rate variance „experimental…
achieve a uniform EGR distribution regardless of the engine op- Acknowledgments
erating conditions even when heavy EGR is applied.
The author would like to thank all of the people who contrib-
This spiral blow-type EGR system is robust against the size and
uted to the development of the new EGR system and to the writ-
shape of the intake system. Therefore, it is possible to apply this
ing of this paper.
technology to many types of engine with little optimization work
required. As one example, this spiral blow-type EGR system has
been applied to small diesel engines. Variations of this spiral Nomenclature
blow-type EGR system for application to different engines are
C EGR ⫽ EGR gas concentration
now under development.
D ⫽ throttle body diameter
Moreover, it is now possible to predict the EGR distribution
EGRR ⫽ EGR rate
with the pulsation flow simulation method that combines a one-
EGRRmea ⫽ measured EGR rate
dimensional gas exchange calculation and a three-dimensional gas
EGRRpre ⫽ predicted EGR rate
flow calculation. Much of the development work for EGR systems
ERROREGR ⫽ calculation error of EGR rate
can now be done by simulation, thereby greatly reducing the de-
M air ⫽ air mass flow rate
velopment lead time and cost.
M EGR ⫽ EGR gas mass flow rate
In this study, pulsation flow simulation was used to analyze the
M gas ⫽ total gas mass flow rate (M air⫹M EGR)
EGR gas flow. It is also possible to apply this pulsation flow
␪ ⫽ lead angle of EGR introduction port
simulation method to an analysis of other gas flows, such as
blow-by gas from the cylinder and purge gas from the canister.
Either gas must be distributed uniformly in the intake system. This References
method is an effective tool for analyzing this type of intake gas 关1兴 Kimura, S., Muranaka, S., and Aoki, O., 1999, ‘‘Possibility of Attaining High-
flow. In future work, it will be applied for analyzing these other Efficiency, Ultra-Clean Diesel Engines With Low-Temperature Premixed
intake gas flows. Combustion,’’ Journal of JSAE, 53共4兲, pp. 17–23 共in Japanese with English
abstract兲.
关2兴 Takagi, Y., Itoh, T., Muranaka, S., Iiyama, A., Iwakiri, Y., Urushihara, T., and
Conclusion Naitoh, K., 1998, ‘‘Simultaneous Attainment of Low Fuel Consumption, High
A pulsation flow simulation method that combines a one- Output Power and Low Exhaust Emissions in Direct Injection SI Engines,’’
SAE Paper 980149.
dimensional gas exchange calculation and a three-dimensional gas 关3兴 Mamiya, N., and Chuubachi, M., 1997, ‘‘Application of Three-Dimensional
flow simulation was used to investigate the EGR gas flow in the CAD System to Engine Development,’’ Journal of JSAE, 51共12兲, pp. 50–56
intake system. This analysis was done with the aim of improving 共in Japanese with English abstract兲.
the EGR distribution and the following results were obtained in 关4兴 Yoshizawa, K., Takeyama, S., Sakai, E., and Tanzawa, K., 1997, ‘‘Numerical
Analysis of Exhaust Gas Flow during the Gas Exchange Process and Design
this study. Optimization,’’ Proceedings of the JSAE Annual Autumn Meeting, 975, pp.
129–132 共in Japanese with English abstract兲.
1. With this analysis method, the EGR rate in each cylinder 关5兴 Yoshizawa, K., Mori, K., Arai, K., and Iiyama, A., 1999, ‘‘Numerical Analysis
was accurately predicated to within 3–5%. of Exhaust Gas Flow and Its Application for Lambda Control Improvement,’’
2. When the EGR gas was introduced downstream of the Proceedings of the ASME Technical Conference, ICE-Vol. 33-3, pp. 151–159.
throttle valve, the distribution was significantly influenced 关6兴 Ahn, H., Cho, K., and Choi, J., 1999, ‘‘Analysis of EGR Distribution Charac-
teristics in Intake Manifolds and Plenum Chamber With 1-D and 3-D Hybrid
by the engine operating conditions. This was due to the in- Calculations,’’ JSAE and KSAE ICE Symposium.
fluence of the reverse flow region that formed downstream 关7兴 Bolton, B., Takenaka, Y., Aoyagi, Y., and Joko, I., 1993, ‘‘Analysis of Intake
of the throttle valve. Manifold Flow in an Inline 6-Cylinder DI Diesel Engine,’’ Proceedings of the
3. Simulation and experimental results confirmed that the EGR JSAE Annual Spring Meeting, 933, pp. 159–162 共in Japanese with English
abstract兲.
distribution was improved by a newly developed mixing 关8兴 Takeyama, S., Ishizawa, S., Yoshikawa, Y., and Takagi, Y., 1987, ‘‘Gas Ex-
system that uses a spiral flow to mix the fresh air and EGR change Simulation Model for Improving Charging Efficiency of 4-Valve Inter-
gas. As a result of adopting this system, a uniform EGR nal Combustion Engine,’’ I.M.E. The First Conference of Computers in Engine
distribution is obtained, regardless of the engine operating Technology, pp. 123–129.
关9兴 Kimura, S., and Yokoyama, T., 1994, ‘‘Reduction of Intake Pulsation Noise by
conditions. Reducing Noise Sources,’’ Proceedings of the JSAE Annual Autumn Meeting
4. This spiral blow-type EGR system has been applied to a 共in Japanese with English abstract兲.
low-emission vehicle 共LEV兲 put on the Japanese market. 关10兴 STAR-CD Version 3.05 Manual, 1998.
Energy and Exergy Analyses of
the Pulse Detonation Engine
T. E. Hutchins Energy and exergy analyses have been performed on a pulse detonation engine. A pulse
detonation engine is a promising new engine, which uses a detonation wave instead of a
M. Metghalchi deflagration wave for the combustion process. The high-speed supersonic combustion
wave reduces overall combustion duration resulting in an nearly constant volume energy
release process compared to the constant pressure process of gas turbine engines. Gas
Mechanical, Industrial and Manufacturing
mixture in a pulse detonation engine has been modeled to execute the Humphrey cycle.
Engineering Department,
The cycle includes four processes: isentropic compression, constant volume combustion,
Northeastern University,
isentropic expansion, and isobaric compression. Working fluid is a fuel-air mixture for
Boston, MA 02115
unburned gases and products of combustion for burned gases. Different fuels such as
methane and JP10 have been used. It is assumed that burned gases are in chemical
equilibrium states. Both thermal efficiency and effectiveness (exergetic efficiency) have
been calculated for the pulse detonation engine and simple gas turbine engine. Compari-
son shows that for the same pressure ratio pulse detonation engine has better efficiency
and effectiveness than the gas turbine system. 关DOI: 10.1115/1.1610015兴
Introduction eral introduction to PDEs can be found in Bussing and Pappas’s

关6兴 article in the AIAA journal. They give a description of the
Pulse Detonation Engines 共PDEs兲 have been predicted to be
more efficient than current aeronautical propulsion systems. They basic ideas behind both PDEs and detonation waves. A constant
potentially represent a revolutionary propulsion technology that specific heat efficiency model was used to show that the Hum-
offers advantages in hardware simplicity, operations, and reliabil- phrey cycle has an advantage over the Brayton cycle. The Hum-
ity, 关1兴. PDE’s show performance advantages in both the subsonic phrey cycle is used to model the constant volume combustion
and supersonic flight regimes, 关2兴. PDE’s are unsteady propulsion process present in PDE’s. Finally, a computational fluid dynamics
devices. They use detonation waves, which allow the chemical 共CFD兲 model was constructed to study the unsteady operation
energy of the fuel used to be released in a far more rapid manner, PDEs.
关3兴. The combustion chamber is filled with a fuel/air mixture and An initial experiment, involving prototype PDEs, was con-
detonated. A detonation wave propagates through the chamber ducted by Bussing et al. 关7兴 Results including wall pressure his-
creating high pressures that produce thrust. Products of combus- tory, thrust force histories, detonation wave velocities, and defla-
tion are exhausted and the cycle starts again. Either running this gration to detonation transition 共DDT兲 distances. DDT is the
cycle at high frequencies or coordinating multiple combustion distance necessary for transition from deflagration to detonation
chambers can produce quasi-steady thrust, 关4兴. combustion wave. This parameter is dependant on many different
PDE’s have been sought out by many different organizations. characteristics of the shock tube including tube diameter and wall
The Navy has shown great interest in the engine. The Multi- texture.
disciplinary, Multi-University Research Initiation 共MURI兲, which Multiple combustion tubes may be a viable way to balance out
is sponsored by the Office of Naval Research, has completed re- the unsteady thrust produced by PDE combustion chambers. The
search on many of the issues facing the PDE, 关4兴. The Air Force basic idea behind using multiple tubes is to isolate the steady-state
has also shown interest in the engine. A recent paper summarized operation of the air intake and mixing chamber from the unsteady
their progress towards developing a kerosene-based PDE, 关5兴. operation of the combustion chamber. Tubes are bundled together
Many design challenges lie ahead for researchers in the PDE and fired in an alternating manner. Bussing 关8兴 has addressed one
area. Four major areas need design considerations for the PDE to possible method of coupling multiple combustion tubes. Some of
be a success, 关1兴. The first design challenge is an air intake that the benefits of his design are high-cycle frequency and lower inlet
will perform well over the engine’s full range of Mach numbers. A loss.
mixing chamber must be designed to provide the fuel/air mixture Bratkovich and Bussing 关9兴 developed a PDE performance
to the combustion chamber under conditions favorable to detona-
model. They used detail calculations to analyze all the compo-
tion. A third consideration is an inlet and injection system that
nents necessary for a PDE. A comparison was made to CFD cal-
provides reproducible detonations in multiple chambers. Finally, a
culations as well as experiment results. They were able to predict
nozzle must be designed to optimize the thrust provided from the
unsteady engine. the wall pressure history within a reasonable amount in compari-
Another major hurdle, which needs to be crossed by PDE re- son to the experimental results.
searchers, is to show that 75–100 Hz required to perform well at For detonation to occur at all, the detonation properties of hy-
Mach 1.2 to 3 can be achieved. Experimental data showing 100 drocarbon fuels must be understood. Ting et al. 关10兴 conducted
Hz has been recorded. However, the experiment did not always experiments on the detonation abilities of propane. Parameters
produce a detonation wave causing the shock tube to become hot such as wave velocity, pressure wall history, and deflagration to
and suspend the test, 关3兴. detonation transition 共DDT兲 distances were measured. Shchelkin
spirals were used in the tubes to help induce detonation. This
Previous Pulse Detonation Engine „PDE… Research. A gen- metal coil helps create a turbulent flow through the tube, which
helps to create a detonation wave. They found that using obstacles
Contributed by the Fuels and Combustion Division of THE AMERICAN SOCIETY such as Shchelkin spirals helped to create shorter DDT. In addi-
OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF ENGI-
NEERING FOR GAS TURBINES AND POWER. Manuscript received by the F&C Divi- tion, the article shows that small tube diameter leads to shorter
sion May 2002; final revision received Aug. 2002. Associate Editor: S. Gollahalli. DDT. Another important result was that a mixture of propane and
oxygen is capable of detonating a mixture of propane and air. This
means that using a driver of fuel/oxygen would work to accom-
plish shorter DDT.
Williams et al. 关11兴 studied the initiation of detonation waves.
In particular, JP10 (C10H16) fuel was analyzed in detail with re-
spect to PDEs. JP10 is widely used in volume-limited combustion
chambers. The authors concluded that JP10 breaks down to 176
reactions of 36 different species. A chemical-kinetic model was
developed in order to predict ignition times for PDEs. JP10 igni-
tion is a process of decomposition to smaller hydrocarbons fol-
lowing by ignition of the resulting products.
One of the major design issues is the frequency capability of
the engine. Bussing et al. 关12兴 determined the basic parameters
that govern the cycle time. The fill time was shown to be the
longest portion of the cycle. This highlighted the need for an
efficient valve system to keep the fill time at a minimum. They
also determined the theoretical frequency scaling laws that go
along with the PDE.
Rockets have also been considered as an application for PDEs. Fig. 1 Brayton cycle
The simple mechanics of the PDE offer many advantages over
propulsion systems currently used. Bratkovich et al. 关13兴 have
written an introductory article regarding the pulse detonation open end. These are the basics involved in a pulse detonation
rocket engines 共PDRE兲. Many of the basic components are similar engine. Other components, including fuel and air mixing cham-
to that of standard PDEs. The major difference is requirement of bers, nozzles and diffusers, need to be considered for a working
on board oxidizer. The cycle involved in PDREs is identical to model, 关15兴.
that of a standard PDE. The full PDE process is illustrated in Fig. 2. The fuel/air mix-
A discussion of detonation waves and their applications in pro- ture enters the shock tube. The detonation is initiated either di-
pulsion have been provided by Kailasanath 关3兴. A general review rectly or through deflagration to detonation transition. Direct ini-
of detonation waves was conducted including standing normal
detonations, pulse detonations, rotating detonations, and oblique
shock-induced detonations. This paper includes an extremely in-
depth and complete history of research on PDEs and their com-
ponents. The author also gave two important points regarding
PDEs. First, he suggested that PDE’s operating on multiphase
mixtures should be explored. In addition, the exhausting of the
products of combustion needs to be better understood so that the
products do not interfere with the next detonation cycle.
One component missing from the overall understanding of
PDEs is a full understanding of how much is gained by using
PDEs. The purpose of this paper is to complete both an energy
and exergy analyses of PDEs. Previous work has developed the
constant specific heat model for the Humphrey cycle, 关6兴. A com-
plete thermodynamic analysis based on combustion has not been
completed. The purpose of this paper is to develop this analysis.
The Humphrey cycle is used to model the PDE. A major charac-
teristic of the Humphrey cycle is the constant volume combustion
process. For comparison, the Brayton cycle is also analyzed. The
Brayton cycle represents a standard gas turbine engine, which is
widely used in aeronautics. A pressure versus volume diagram of
the Brayton Cycle is shown in Fig. 1.
A detailed combustion analysis has been used to determine the
efficiency and effectiveness of the Humphrey and Brayton cycles.
Using the Stanjan™ 关14兴 Chemical Equilibrium software, the ef-
ficiency and effectiveness of both cycles is calculated using both
methane and JP10 as the fuel.
Pulse Detonation Engines „PDEs…

Pulse detonation engines are characterized by a supersonic
combustion wave. Supersonic flame propagation is known as a
detonation wave. In contrast, gas engines use a deflagration wave
共subsonic flame propagation兲. A detonation wave will increase the
pressure and density of the unburned fuel/air mixture ahead of the
wave increasing its temperature, 关6兴.
A basic PDE consists of a shock tube, a system designed to
deliver the fuel air mixture and a device to initiate the detonation
waves. The shock tube is closed at one end and open to atmo-
spheric air at the other end. The location of the detonation device
has produced considerable attention. Two possibilities exist. The
detonation wave can be initiated from either the closed end, or the Fig. 2 PDE combustion cycle
Fig. 3 Humphrey cycle Fig. 4 T-s diagram
tiation requires a considerable amount of energy to start

The Brayton cycle, which includes a compressor, a combustor
detonation. Using deflagration and transitioning to detonation
and a turbine, uses a constant pressure combustion process. From
takes less energy to initiate. However, there is a distance required
the p-v diagram shown in Fig. 1, the net work produced by the
to transition to a detonation wave. This distance is known as de-
system is determined by
flagration to detonation transition 共DDT兲 distance, 关10兴.
Rarefaction waves follow the detonation wave. As the wave w net⫽ 共 h 3 ⫺h 4 兲 ⫺ 共 h 2 ⫺h 1 兲 . (1)
propagates through the chamber, the fuel/air mixture ahead is
compressed. High pressure and temperature is reached creating For the Humphrey cycle, the combustion process occurs with con-
engine thrust. As the detonation exits the shock tube, the rarefac- stant volume. From the p-v diagram shown in Fig. 2, the net work
tion waves start back into the tube, exhausting the products of the produced for the closed system is
combustion. This brings the chamber back to the initial pressure w net⫽ 共 u 1 ⫺u 2 兲 ⫹ 共 u 3 ⫺u 4 兲 ⫹R 共 T 1 ⫺T 4 兲 . (2)
where the process can begin again. The engine is mechanically
simple requiring few moving parts. This offers many advantages To determine the thermal efficiency, the net work is divided by the
in reliability, 关6兴. high heating value 共HHV兲 of the fuel.
Processes in pulse detonation engines are very complicated em- w net
ploying subsonic and supersonic combustion waves propagation. ␩⫽ (3)
To analyze the processes involved in this device requires complete HHV
computational fluid dynamic analysis with detail chemical kinetic For the effectiveness or exergetic efficiency, the net work is di-
modeling, which is beyond the scope of this investigation. Jacob vided by the chemical availability of the fuel.
cycle, 关16兴, has been used to model the detonation process while,
historically Humphrey cycle has been used to represent the pro- w net
cess in the pulse detonation engine. While the model is a simpli- ␧⫽ (4)
achfuel
fication of the processes that gas mixtures experience, it provides
information that can be used to compare this engine to the gas Determining Thermodynamic Properties. The Chemkin™
turbine system. 关17兴 system was used to determine the thermodynamic properties
required for this analysis. Chemkin™ provides coefficients for a
The Humphrey Cycle. In order to evaluate thermodynamic polynomial fit for specific heat, enthalpy and entropy of different
characteristics of PDEs, a cycle must be chosen to model the species. The three equations necessary are
engine. The detonation wave allows the combustion to be mod-
eled as constant volume. The Humphrey cycle is chosen as the c op ⫽R 共 a 1 ⫹a 2 T⫹a 3 T 2 ⫹a 4 T 3 ⫹a 5 T 4 兲 (5)
冉冊
model.
The Humphrey cycle consists of four processes. The first pro- a2 a3 a4 a5 a6
h o ⫽RT a 1 ⫹ T⫹ T 2 ⫹ T 3 ⫹ T 4 ⫹ (6)
cess is an isentropic compression. This compression occurs ahead 2 3 4 5 T
冉冊
of the detonation wave in PDEs. Compression is followed by
constant volume combustion. Another isentropic process expands a3 2 a4 3 a5 4
s o ⫽R a 1 ln T⫹a 2 T⫹ T ⫹ T ⫹ T ⫹a 7 (7)
the combustion products back to atmospheric pressure. In the 2 3 4
PDE, the rarefaction waves cause this expansion process. Finally,
The values for a 1 through a 7 were found in a data file used for
an isobaric process brings the cycle back the start. Figure 3 shows
the Chemkin code, which will be discussed later.
this process on a p-v diagram.
A T-s diagram is shown in Fig. 4 comparing the Humphrey and Compression Process. The compression process was as-
Brayton cycles. sumed isentropic. Enthalpy is determined under the initial condi-
tions based on a temperature of 300 K. In the case of the Hum-
phrey cycle, the initial internal energy is also determined from the
Energy and Exergy Analyses following expression:
In order to quantify the expected gains, which a pulse detona-
u o ⫽h o ⫺RT. (8)
tion engine will offer in comparison to current methods of com-
bustion, an energy and exergy model must be established for both The pressure ratio, r, is used to find the pressure just prior to
characteristic cycles. combustion assuming the initial pressure is 1 atm. Initial entropy
is found based on a stoichiometric fuel and air mixture. 共Air is h of 兩 C10H16⫽h o 共 300 K兲 .
assumed to be 79% N2 and 21% O2 .) The temperature prior to
combustion is determined by solving Eq. 共7兲 by the Newton- The high heating value was found to be
Raphson method. Enthalpy is then evaluated from this tempera-
kJ
ture. Again, for the case of the Humphrey cycle, internal energy is HHVC10H16⫽94,224 .
also found. kmolC10H16
Combustion Process. Stanjan™ chemical equilibrium soft- For the chemical availability, the following equation was used:
ware is used to find properties after combustion. In order to find
achC10H16⫽g C10H16⫹14g O2 ⫺10g CO2 ⫺8g H2 O共 l 兲 ⫹10achCO2
the major contributing species in the combustion products, a list
was generated of possible compounds that can be created from ⫹8achH2 O⫺14achO2 . (12)
carbon, hydrogen, oxygen, and nitrogen atoms. The species con-
sidered were The Gibbs function was found by the following:
g o ⫽h o ⫺Ts o . (13)
N2 O2 N O CH O⫹ O⫺ H⫹ H⫺
The chemical availability was found to be
H2 O CO2 CO H HO2 HO H2 O2 H2 NO
kJ
NO2 CH3 CH2 O C2 H6 CH4 CH2 NH NH2 NH3 achC10H16⫽93,700 .
kmolC10H16
NNH
Brayton Cycle. Stanjan™ was used to determine chemical
equilibrium of the products of combustion. Stanjan™ requires two
The chemical equilibrium was found using Stanjan™ based on constant properties to perform the equilibrium calculations. For
these species being the products of combustion. For the purpose the case of the Brayton cycle, constant pressure and constant en-
of a combustion analysis, only the species that have a mole frac- thalpy was chosen. The code calculates all properties after com-
tion greater than 10⫺3 have been considered in the combustion bustion.
calculations. The eight most dominant species are
Humphrey Cycle. The properties of the combustion products
are found in the same manner as with the Brayton cycle. The
N2 H2 O CO CO2 O2 OH H2 NO Humphrey cycle is characterized by a constant volume combus-
tion process. For the Stanjan™ program, constant volume and
This result was found to be the same for both fuels used in this constant internal energy was chosen to conduct the chemical equi-
analysis. librium calculations. Temperature and pressure of the burned
In order to validate the assumption that eight species character- gases were determined by Stanjan™.
izes the chemical reaction sufficiently, the Humphrey cycle was
analyzed based on 20 most dominant species. The following were Expansion Process. The expansion process was assumed isen-
used: tropic. Using the entropy found for the burned gases Eq. 共7兲 was
solved for the temperature after the expansion.
The temperature found after expansion is only valid for frozen
N2 H2 O CO2 CO O2 OH H2 NO H chemistry. However, for this cycle, the chemistry of the products
O NO2 H2 O2 N NH3 NH NNH NH2 CH3 cannot be assumed frozen. Stanjan™ was again employed to find
the chemical equilibrium mixture. The inputs include the chemical
CH4 CH2 molar concentrations of the products, the frozen chemistry tem-
Methane Combustion. Methane was used as a simple fuel to perature at state four and atmospheric pressure. The two property
determine the thermodynamic performance of PDEs. The bal- constants required to run the program were pressure and entropy.
anced stoichiometric chemical reaction for methane combustion Temperature and chemical concentrations at chemical equilib-
with air is rium were determined by Stanjan™. Enthalpy was then deter-
mined for the expanded products. In the case of the Humphrey
cycle, the internal energy was found from the enthalpy.
CH4 ⫹2O2 ⫹7.52N2 →CO2 ⫹2H2 O⫹7.52N2 (9)
The high heating value and the chemical availability for methane Results and Discussion
were found to be, 关18兴,
Methane was the first fuel analyzed for use in pulsed detonation
kJ kJ engines. A range of compression ratios was used to compare the
HHVCH4 ⫽93,522 achCH4 ⫽86,642 . thermal efficiency and effectiveness of the Humphrey and Brayton
kmolCH4 kmolCH4 cycles. Figure 5 shows the Humphrey cycle to have an advantage
over the Brayton cycle with respect to thermal efficiency.
JP10 Combustion. A jet fuel, JP10, was used as the second Similar results are seen in the effectiveness. Figure 6 shows
fuel since the Navy has expressed interest in using JP10. The the advantage held by the Humphrey cycle with respect to
average chemical formula for JP10 is C10H16 , 关19兴. The balanced effectiveness.
chemical reaction for C10H16 combustion with air is A more probable fuel to be used for pulsed detonation engines
would be JP10 jet fuel. For this reason, Figs. 7 and 8 confirm the
C10H16⫹14O2 ⫹52.64N2 →10CO2 ⫹8H2 O⫹52.64N2 (10) same advantage will be present when using JP10 fuel as when
methane is used.
The high heating values and chemical availability for C10H16 These results show a significant advantage in using the Hum-
were calculated. The heating value was calculated by the follow- phrey cycle over the Brayton cycle. An increase of 4 to 12% in
ing: thermal efficiency and effectiveness can be expected over the
Brayton cycle based on methane. Similar results are shown when
HHVC10H16⫽10h of 兩 CO2 ⫹8h of 兩 H2 O共 l 兲 ⫺h of 兩 C10H16 (11) using JP10 fuel.
Figure 9 compares the thermal efficiency between methane and
where JP10 based on the Humphrey cycle. There is an insignificant dif-
Fig. 8 Effectiveness of the Humphrey and Brayton cycles
Fig. 5 Thermal efficiency of the Humphrey and Brayton cycles based on a combustion analysis using JP10
using methane as the fuel
equivalent power that a comparable gas turbine engine produces.

ference between the two fuels. There is, however, a slight differ- Each cycle of combustion produces a set amount of energy based
ence between methane and JP10 based on effectiveness. Figure 10 on fuel used per cycle. Frequency is directly proportional to the
shows methane to have slightly higher effectiveness than JP10. power produced. Further experimentation and research is neces-
Figure 11 compares efficiency of the Humphrey cycle based on sary to determine a frequency that will produce the equivalent
8 species and 20 species. The analysis is conducted using methane power produced by gas turbine engines.
as the fuel. In order to compare these two models, the assumption
was made that chemistry is frozen during the expansion portion of Conclusions
the cycle. There is not a significant difference between the results The pulse detonation engine has been shown to have significant
of the two models. Similarly, Fig. 12 shows the same comparison thermodynamic performance advantages over current air-
based on effectiveness. Again, there is not a significant difference. breathing jet engines. These performance advantages are seen to
Future Considerations. The pulse detonation engine works be similar for both fuels studied. The greatest advantage was seen
on an unsteady basis. In order to make the PDE a viable jet pro- at lower compression ratios.
pulsion engine, a frequency must be achieved to produce the PDEs have many advantages that may make them the choice
for propulsion in the future. Along with performance advantages,
Fig. 6 Effectiveness of the Humphrey and Brayton cycles us-

Fig. 9 Efficiency of methane and JP10 based on the Hum-
ing methane as the fuel
phrey cycle
Fig. 7 Thermal efficiency of the Humphrey and Brayton cycles Fig. 10 Effectiveness of methane and JP10 based on the Hum-
using JP10 as the fuel phrey cycle
T ⫽ temperature
u ⫽ internal Energy
w ⫽ work
Superscripts
o ⫽ standard state
Greek
␧ ⫽ effectiveness
␩ ⫽ thermal efficiency
References
关1兴 Santoro, R. J., Yang, V., Law, C. K., and Shepard, J. E., 1999, ‘‘Multidisci-
plinary study of Pulsed Detonation Engine Propulsion,’’ ONR Contract
Fig. 11 Thermal efficiency of the Humphrey cycle using 8 and N99914-1-0744.
关2兴 Pegg, R. J., Couch, B. D., and Hunter, L. G., 1998, ‘‘Pulsed Detonation Engine
20 species based on methane
Air Induction System Analysis,’’ NASA Langley Research Center and Lock-
heed Martin Tactical Aircraft Systems.
关3兴 Kailasanath, K., 1999, ‘‘Review of Propulsion Applications of Detonation
Waves,’’ AIAA Paper No. 99-1067.
关4兴 ‘‘Pulsed Detonation Engines,’’ Caltech University, 2000, http://
web.galcit.caltech.edu/EDL/projects/pde/pde.html 共Apr. 18, 2001兲.
关5兴 Anthenian, R. A., Schauer, F. R., Stutrid, J. S., Frayne, C. W., Bradley, R. P.,
Chen, L. P., and Katta, V. R., 1999, ‘‘Development of a Pulse Detonation
Research Engine,’’ Air Force Research Laboratory and Innovative Scientific
Solutions, Inc.
关6兴 Bussing, T., and Pappas, G., 1994, ‘‘An Introduction to Pulsed Detonation
Engines,’’ AIAA Paper No. 94-0263.
关7兴 Hinkey, J. B., Bussing, T. R. A., and Kaye, L., 1995, ‘‘Shock Tube Experi-
ments for the Development of a Hydrogen-Fueled Pulse Detonation Engine,’’
AIAA Paper No. 95-2578.
关8兴 Bussing, T. R. A., 1995, ‘‘Rotary Valve Multiple Pulse Detonation Engine
共RVMPDE兲,’’ AIAA Paper No. 95-2577.
关9兴 Bratkovich, T. E., and Bussing, T. R. A., 1995, ‘‘A Pulsed Detonation Engine
Performance Model,’’ AIAA Paper No. 95-3155.
关10兴 Ting, J. M., Bussing, T. R. A., and Hinkey, J. B., 1995, ‘‘Experimental Char-
acterization of the Detonation Properties of Hydrocarbon Fuels for the Devel-
opment of Pulse Detonation Engine,’’ AIAA Paper No. 95-3154.
Fig. 12 Effectiveness of the Humphrey cycle using 8 and 20 关11兴 Williams, F. A., and Varatharajan, B., 2000, ‘‘Theory of Initiation of Detona-
species based on methane tion in Pulse Detonation Engines: Chemical Kinetics and Mechanisms,’’ ONR
Contract N00014-99-1-0745.
关12兴 Bussing, T. R. A., Bratkovich, T. E., and Hinkey, Jr., J. B., 1997, ‘‘Practical
Implementation of Pulse Detonation Engines,’’ AIAA Paper No. 97-2748.
PDEs have a potential mechanically simple design with far less 关13兴 Bratkovich, T. E., Aarnio, M. J., Williams, J. T., and Bussing, T. R. A., 1997,
moving parts than other air-breathing engines. ‘‘An Introduction to Pulse Detonation Rocket Engines 共PDREs兲,’’ AIAA Paper
No. 97-2742.
Nomenclature 关14兴 Reynolds, W. C., 1986, Stanford University Report, ME 270, No 7.
关15兴 Bussing, T. R. A., Hinkey, J. B., and Kaye, L., 1994, ‘‘Pulse Detonation Engine
a ⫽ Chemkin Preliminary Design Considerations,’’ AIAA Paper No. 94-3220.
ach ⫽ Chemical availability 关16兴 Fickett, W., and Davis, W. C., 2000, Detonation Theory and Experiment, Do-
ver, New York.
cp ⫽ specific heat 关17兴 Kee, R. J., Rupley, F. M., Meeks, E., and Miller, J. A., 1996, ‘‘Chemkin III: A
g ⫽ Gibbs free energy Fortran Chemical Kinetics Package for the Analysis of Gas Phase Chemical
h ⫽ enthalpy and Plasma Kinetics,’’ Sandia National Laboratories, http://
hf ⫽ enthalpy of formation nathan.ca.sandia.gov/chemkin/docs/CHEMKIN III.pdf 共Feb. 20, 2001兲.
关18兴 Moran, M. J., and Shapiro, H. N., 1996, Fundamentals of Engineering Ther-
HHV ⫽ high heating modynamics, John Wiley and Sons, New York.
R ⫽ universal gas constant 关19兴 Li, S. C., Varatharajan, B., and Williams, F. A., 2001, ‘‘The Chemistry of JP-10
s ⫽ entropy Ignition,’’ AIAA J., 39共12兲.
Numerical Investigation of the
Effects of Axial Cylinder Bore
Profiles on Piston Ring Radial
Dynamics
Y. Piao The role of cylinder bore shapes in engine performance has been the subject of several
studies in recent years. In particular, the influence of bore distortion on oil consumption
S. D. Gulwadi under high speed conditions has generated significant interest. In this paper, the effect of
an axial bore profile on radial dynamics of a ring is investigated. Radial ring motions
Ricardo, Inc.,
within grooves due to the axial bore profile can generate significant inertial effects and
7850 Grant Street,
also have an impact on ring end-gap sizes and lubrication conditions at the ring-liner
Burr Ridge, IL 60527
interfaces. The magnitude of such effects is dependent on the ring-pack configuration,
engine operating conditions (speed and load) and axial bore profile details. These issues
are investigated in this study due to their implication on engine oil consumption, friction
and blow-by. The authors have developed an analytical expression to account for the
effects of radial ring inertia due to an axial bore profile for implementation in a piston
ring-pack simulation tool RINGPAK. Simulation results from a gasoline engine study are
presented to illustrate the effects of engine speeds, ring tensions, and characteristics of
axial bore profiles on ring radial dynamics and ring-liner lubrication. Relevant qualita-
tive comparisons are made to experimental measurements available in the literature.
关DOI: 10.1115/1.1610016兴
Introduction bore diameter values are different at various axial locations along
the cylinder兲. The mathematical model used to simulate the radial
A desirable piston ring-pack set is one that provides good seal-
dynamics of the ring within the groove, 关7兴, has been extended to
ing performance 共i.e., minimal gas blow-by兲 with low lubricating include the inertial effects arising from the piston ring sliding over
oil consumption and friction loss. This is a challenging task due to an axial profiled bore. Besides modeling details, simulation results
the nature of the phenomena and interactions associated with pis- are also presented for response of piston ring motions to an axially
ton rings. For example, increasing the installed ring tension which profiled bore in a gasoline engine at various speeds. The simula-
is a method to control oil consumption also tends to increase ring tion studies have been carried out using a piston ring-pack simu-
friction, 关1兴. Hence, any attempt to optimize ring-pack perfor- lation code RINGPAK, 关8兴. Qualitative comparisons are made
mance output parameters requires a good understanding of the with experimental results from the literature, 关6兴, that are shown in
dynamics of cylinder-kit components 共e.g., piston, rings, liner兲 the Appendix.
and lubrication at the interfaces 共e.g., ring-liner, ring-groove,
skirt-liner兲 for various operating conditions. To this effect, cylin-
der bores have been the subjects of several investigations. There Mathematical Model
have been extensive studies, 关2–5兴, on the effects of bore surface In an operating engine, piston rings are subjected to forces and
finish, bore out-of-roundness and bore distortion on lubricating oil moments arising from several sources such as ring-liner hydrody-
consumption. The topic addressed in this paper is the effect of namic lubrication, ring-liner boundary lubrication, ring tension,
axial bore profiles on radial ring dynamics. Axial bore profiles inter-ring gas pressures, etc. Figure 1 depicts the various force
occur due to various reasons 共e.g., thermal gradients, liner sup- components acting on a ring in the radial direction under any
ports兲. A particular phenomenon associated with axially profiled operating condition.
bores is that piston rings tend to move away from the bore in the
radial direction as they slide over the heavily distorted portions of Radial Ring Dynamics. The radial force balance on a piston
the cylinder bore, especially at high engine speeds. This has been ring, as depicted in Fig. 1, may be expressed as
observed experimentally by M. Basaki et al. 关6兴, who also re- F oil⫹F asp⫹F gas⫺F gro⫺F ten⫺F ine⫺F frc⫽0, (1)
ported that this ring separation from the bore contributes to in-
creased oil consumption at high engine speeds. The ability of where F oil is force due to oil film pressures at ring-liner interface,
piston rings to slide along an axially profiled cylinder wall and to F asp is force due to contact pressures at ring-liner interface, F gas is
force from land gas pressures at ring-liner interface, F gro is force
conform to a circumferentially distorted cylinder bore is a critical
due to groove gas pressure at back of the ring, F ten is force due to
factor in determining the oil consumption associated with that
installed ring tension, F ine is inertia force on the ring due to its
ring-pack set. Hence, these effects are important and need to be
radial motion, and F frc is friction force at the ring-groove
considered when simulating and/or analyzing ring-pack behavior. interface.
This study places the focus on radial ring motion in an axially At the ring-liner conjunction, based on operating conditions,
profiled bore only with no circumferential distortions 共i.e., only several types of lubrication modes are possible. In general, piston
rings tend to operate in the mixed lubrication regime but there are
Contributed by the Internal Combustion Engine Division of THE AMERICAN SO-
some instances such as ‘‘ring lift’’ where the ring is gas lubricated
ENGINEERING FOR GAS TURBINES AND POWER. Manuscript received by the ICE or ‘‘starved conditions’’ when the ring operates in the boundary
Division, June 2002; final revision received Oct. 2002. Associate Editor: D. Assanis. lubrication regime. In order to capture these phenomena, a mass-
dm i p iV i
dt
⫽ 兺 ṁ and m i⫽
RT i
, (6)
where m i is mass of gas in the subvolume i, ṁ is the gas mass

flow rate across the boundaries of the subvolume, R is the gas
constant, and p i and T i are the gas pressure and temperature in the
subvolume V i .
Gas mass flow-rates between adjacent lands through the ring
end-gap and ring-liner nonconforming areas are calculated using
the isentropic orifice flow equation
Fig. 1 Forces acting on a ring in the radial direction ṁ⫽

C d Ap u
冑RT u 冉冊冉冊冉冉冊冊
2␥
␥ ⫺1
1/2
pd
pu
1/␥
1⫺
pd
pu
␥ ⫺1/␥ 1/2
or
conserving mixed lubrication model is employed. Details of this

model are provided in Ref. 关7兴 and only a brief overview is given
ṁ⫽
C d Ap u ␥ 1/2
冑RT u 冉冊2
␥ ⫹1
␥ ⫹1/2共 ␥ ⫺1 兲
if
pd
pu
⭐
2
冉冊
␥ ⫹1
␥ / ␥ ⫺1
,
in this paper. (7)
The oil film pressure distribution p(y) at the ring-liner interface
where C d and A are the discharge flow coefficient and flow area
is obtained on solving the Reynolds equation
associated with the orifice, T u is the upstream gas temperature, p u
冉
⳵ h共 y 兲3 ⳵ p
⳵ y 12␮ ⳵ y
⫽ 冊
Vy ⳵h共 y 兲 ⳵h共 y 兲
2 ⳵y
⫹
⳵t
, (2)
and p d are the upstream and downstream pressures, and ␥ is the
polytropic exponent.
On the other hand, flow between lands and grooves through the
where h(y) is the clearance profile at the ring face-liner conjunc- ring-groove side clearances is treated as an isothermal compress-
tion defined along the axial coordinate y, V y is the piston ring ible flow through a narrow channel, such that
冉冕冊
axial velocity, and ␮ is the lubricant viscosity. The oil film lubri- ⫺1
cation extents Y 1 and Y 2 are determined by the coupled mass- 共 p 2u ⫺p 2d 兲 dx
ṁ⫽ ␲ 共 D⫺L ring兲 , (8)
conserving lubrication model and the oil transport model de- 24␮ mixRT h共 x 兲3
scribed in Ref. 关7兴.
From the oil film pressures, the force due to oil pressures F oil is where h(x) is the ring-groove side clearance profile along the
computed using radial coordinate x, T is the average temperature of this volume,
␮ mix is the oil-gas mixture viscosity, and L ring is radial thickness
F oil⫽ 冉冕冊
p 共 y 兲 dy ␲ D, (3)
of the ring.
Thus, on calculating the groove and land pressures, the radial
force components F gro and F gas are expressed as
where D is the nominal bore diameter. During boundary lubrica-
tion at the ring-liner interface, local contact pressures p asp are F gro⫽p groB ␲ 共 D⫺2L ring兲 , (9)
calculated using the Greenwood-Tripp model, 关9兴,
F gas⫽ 关 p ul Y 1 ⫹p ll 共 B⫺Y 2 兲兴 ␲ D, (10)
p asp⫽
16冑2
15
␲共 ␴ ␤ ␩ 兲2E
冑␴ h 共 y 兲
冑␤
G
␴
, 冉冊 (4a)
where p gro, p ul , and p ll are gas pressures in the groove, upper
land and lower land surrounding each ring, respectively, B is the
ring axial width, Y 1 and Y 2 are locations on the y-axis along the
where ring face which define the boundaries of the oil lubricated zones at
冕冉冊
the leading and trailing edges, respectively.
1 ⬁ s2 The force component F ten is based on design information avail-
G共 x 兲⫽ 共 s⫺x 兲 2.5 exp ⫺ ds. (4b)
冑2 ␲ 0 2 able from the drawings while the force component F frc is based on
calculation of oil film pressures and contact pressures at the ring-
␴, ␤, and ␩ are surface roughness parameters which represent the groove interfaces, similar to those shown in Eqs. 共2兲, 共4a兲, and
mean r.m.s. height, radius of curvature, and density of the surface 共4b兲. Since the radial inertial force component F ine is an important
asperities, respectively, and E is the composite elastic modulus of aspect of this study, it is treated in detail in the next section.
the contacting materials. Consequently, the contact force F asp is Using the approach described above, the instantaneous radial
calculated using location of a piston ring within its groove is computed over the
冉冕冊
engine cycle. Although a brief summary of the interaction be-
tween the various phenomena is given here, more information on
F asp⫽ p asp共 y 兲 dy ␲ D. (5)
coupling between the various phenomenological models is given
in Ref. 关8兴.
Gas pressures in the groove and land volumes surrounding the The study undertaken in this paper focuses on the effect of an
ring are needed to calculate F gro and F gas . Additionally, land pres- axial bore profile on the radial motion of a piston ring. An axial
sures are also used as boundary conditions in solving the Rey- bore shape alters the instantaneous ring-liner clearance profiles as
nolds Equation given by Eq. 共2兲. As a result, a gas dynamics the rings slide within the cylinder bore. In addition to its impact
model is employed to compute inter-ring pressures 共i.e., pressures on the radial inertia term F ine , the axial bore profile has a signifi-
in a labyrinth of land and groove subvolumes which make up the cant influence in determining the lubricating conditions at the
ring-pack region兲. Flow paths exist between these subvolumes, for ring-liner interfaces 共i.e., full lubrication, partial lubrication or
example, gas flows between adjacent land subvolumes through the starved兲. As a result, calculations of the force components F oil ,
ring end-gap and ring-liner nonconforming areas whereas gas F asp , and F gas are affected along with the radial ring motion
flows between the land and groove subvolumes through the ring- within the groove.
groove side clearances. Hence, to compute these instantaneous
inter-ring pressures ideal gas law in conjunction with mass con- Radial Inertia. Figure 2 shows the schematic of an axially
servation is applied to each subvolume, such that profiled bore and the coordinate system used in the analysis. The
Fig. 3 Effect of axial bore profile on ring-liner clearance
within the piston groove. The terms dz/dy and d 2 z/dy 2 are the
first and second derivatives of the bore profile. In order to obtain
Fig. 2 Schematic layout of a ring in an axially profiled bore
continuous dz/dy and d 2 z/dy 2 , cubic splines 共cubic spline curve-
fitting兲 are used to represent the bore profile based on the input
‘‘points’’ of the bore profile. Hence an analytical expression is
axial bore profile is based on bore diameter values calculated at developed to completely represent the term d 2 z/dt 2 as described
different axial locations along the cylinder wall. In this study, the in Eq. 共15兲.
nominal bore diameter is used as the reference diameter. The Ring-Liner Clearance Profile. The users input the ring face
variation in value of the bore radius at each axial location along profiles along with the axial bore profile and the simulation pro-
the cylinder wall is then represented by z(y) about the nominal gram computes the clearance shape for ring-liner lubrication at
bore. The dots in Fig. 2 represent the resulting axial bore profile. each time-step. The instantaneous clearance profile is dependent
At any given axial location of the ring y, h 0 is the minimum on ring radial motion and toroidal twist within the groove, piston
ring-liner oil film thickness, s cg and s are the radial distances of secondary motion, and bore shape. The general procedure for in-
the ring center of gravity and ring face 共minimum clearance loca- cluding effects of an axial bore shape on the ring-liner clearance
tion兲 from the nominal bore, respectively. Using the nominal bore profile is given here. Axial length of the bore is roughly two
diameter as a reference, the inertial term in Eq. 共1兲 can be written orders of magnitude more than the axial width of a ring. Addition-
as ally, the axial bore profile is generated on a coarser mesh com-
F ine⫽⫺m ringa r , (11a) pared to the ring-face. Therefore, calculation of the instantaneous
ring-liner clearance profile uses a representation of the liner as a
where the acceleration term is given by line that’s tangential to the local bore profile 共i.e., at a location
along the cylinder bore corresponding to the axial location of the
d 2 s cg d 2s d 2 共 s cg ⫺s 兲 ring兲. Figure 3 illustrates the effect of an axial bore profile on
a r⫽ 2
⫽ 2
⫹ . (11b)
dt dt dt 2 ring-liner clearance.
The angle ␣ bore in Fig. 3 is calculated from ␣ bore
m ring is mass of the ring, and (s cg ⫺s) is the radial distance be- ⫽arctan(dz/dy). This angle is used in the calculation of the ring-
tween center of gravity of the ring and minima of the ring face liner clearance profile in a manner similar to the inclusion of ring
profile, which is dependent on instantaneous values of ring toroi- toroidal twist or piston tilt as described in Ref. 关10兴.
dal twist. However, due to the relatively small magnitude of the
twist angle and its rate of change, this term may be neglected in
Eq. 共11b兲. Thus, the acceleration term reduces to
Simulation Results and Discussions
d 2s
a r⫽ . (12) Engine and Ring-Pack Configuration. Simulations are con-
dt 2 ducted on an automotive gasoline engine with an 84.8-mm bore
Since s⫽h 0 ⫺z, Eq. 共12兲 can be rewritten as diameter and 88.0-mm stroke length. The piston ring-pack con-
sists of a standard set of three rings: a barrel faced top ring with a
d 2h 0 d 2z rectangular cross section, a taper faced second ring with a rectan-
a r⫽ ⫺ . (13) gular cross section, and a twin-rail oil ring.
dt 2 dt 2
Results are presented in this investigation from parametric stud-
Further, the term d 2 z/dt 2 can be expanded as ies involving 共a兲 engine speeds, 共b兲 ring tensions, and 共c兲 axial
冉冊冉冊冉冊冉冊
bore profiles. For simplicity, the same cylinder pressure and tem-
d 2z d 2z dy 2
dz d2y perature values 共typical for a medium load and medium speed
⫽ ⫹ , (14)
dt 2
dy 2 dt dy dt 2 condition兲 versus crank angle are used for various configurations
and engine speeds. In addition, focus is placed on radial ring
where dy/dt⫽V y and d 2 y/dt 2 ⫽a y are the velocity and accelera- dynamics during the intake and exhaust strokes. During these por-
tion of a ring in the axial direction. Therefore, Eq. 共14兲 reduces to tions of the engine cycle which have relatively low gas pressures,
冉冊冉冊
ring motions within the grooves are more dependent on engine
d 2z d 2z dz speed and bore profile.
⫽ V 2y ⫹ a . (15)
dt 2
dy 2 dy y
Baseline Case. A baseline model with an axial bore profile is
In this formulation, a y and V y can be expressed analytically in set up as shown in Fig. 4. The discrete user-input points and the
terms of engine speed and configuration 共based on ideal crank- calculated spline-fitted curve used to represent the axial bore pro-
slider motion兲 by neglecting the relative axial motion of the ring file are shown in Fig. 4. For reference purposes, ring locations at
Fig. 4 Baseline axial bore profile Fig. 6 Top ring MOFT at different engine speeds
TDC and BDC reversal positions are included in Fig. 4. It should sensitivity to the bore profile due to the large tension of the ring.
be noted that the two rails of the oil ring are shown along with the During the power stroke, top ring radial motion is also insensitive
upper compression rings. to the bore shape due to high gas pressures.
In a relevant experimental study, 关6兴, a similar axial bore profile In particular, the top ring behavior is of significant interest since
共including a deformed segment or ‘‘depression’’ of 32-mm axial oil flowing past this ring towards the combustion chamber con-
extent and 0.076-mm radial depth兲 is used in a gasoline engine of tributes to oil consumption. Thus, the analysis presented here will
95-mm bore diameter and 86-mm stroke length. The axial bore focus on the top ring for parametric studies involving engine
profile used in that study is shown in Fig. 21 in the Appendix. This speeds, ring tensions, and axial bore shapes.
can be compared to the current bore profile with the deformed
portion or ‘‘depression’’ of 16-mm axial extent and 0.042-mm Effect of Engine Speed. Keeping other input of the ring pack
radial depth, as shown in Fig. 4. The engine and ring-pack model model unchanged, simulations are carried out for engine speeds of
in the present study is ‘‘similar’’ to the setup of the aforemen- 1000, 2000, 3000, 4000, 5000, and 6000 rpm. Figure 6 shows
tioned experimental investigation. This enables some qualitative minimum oil film thickness 共MOFT兲 values of the top ring at
comparisons to be made between the current study and reported different engine speeds.
experimental work, 关6兴. In Fig. 6, the sharp gradients in MOFT values occur when the
An important parameter in ring-liner lubrication is the mini- top ring passes over the ‘‘depression.’’ It is seen that as the speeds
mum oil film thickness 共MOFT兲, which is also a measure of ring increase, these gradients in MOFT values also increase. The
radial motion relative to the cylinder bore. The MOFT calculation ‘‘jump’’ in MOFT value is defined as the ‘‘ring separation dis-
for a ring is based on the radial force balance given by Eq. 共1兲. tance’’ from the bore, as shown in Fig. 6. The variable ‘‘ring
Figure 5 shows the values of MOFT from simulations of all three separation distance,’’ during the intake stroke, is plotted as a func-
rings at an engine speed of 4000 rpm. tion of engine speed in Fig. 7.
The differences in MOFT values between the three rings can be The ‘‘ring separation’’ from the bore is based on radial inertial
attributed to many factors such as groove pressures, ring tensions, force due to the combined effects of ring radial dynamics, piston
ring face profiles, oil availability, etc. Relevant to this study, the axial motion and an axial bore profile as described in Eqs. 共1兲,
axial bore profile and axial spacing between rings also play an 共11a兲, 共13兲, and 共15兲. The contribution to ring radial inertia due to
important role in the calculation of MOFT. For example, consider piston/ring axial motion in an axially profiled bore is represented
the ‘‘depression’’ on the bore shown in Fig. 4; each ring passes by Eq. 共15兲. Here, the terms for axial piston ring velocity and
over it at a different instant of the engine cycle when values of acceleration are proportional to engine speed N and square of the
piston velocity and acceleration are different. As seen in Fig. 5, engine speed N 2 , respectively. As a result, this radial inertia con-
during up-strokes, the top ring slides over the ‘‘depression’’ first tribution due to an axial bore profile is a function of N 2 and the
followed by the second ring evidenced by the sequence of sudden variable ‘‘ring separation distance’’ versus engine speed in Fig. 7
rise and fall in their MOFT values. This is the reverse of down- shows this trend. A similar behavior is also observed in the mea-
strokes, when the second ring first passes over the ‘‘depression’’ surements, 关6兴, of ‘‘ring separation distance’’ versus engine speeds
followed by the top ring. The oil ring motion does not show much and these are shown in Fig. 23 in the Appendix.
Fig. 5 MOFT values for the ring-pack at 4000 rpm Fig. 7 Ring separation distance versus engine speed
Fig. 10 Oil flow rate below the ring face and gas flow rate
Fig. 8 Peak values of piston acceleration versus engine speed through end gap of the top ring in the intake stroke at 6000 rpm
Figure 8 shows a plot of peak piston acceleration as a function increase in land volume between the rings as this region passes
of engine speed and the similar trends observed between Figs. 7 over the ‘‘depression’’ and 共2兲 increase in the ring end gap area for
and 8 are consistent with the above discussions. gas flow. As shown in Fig. 10, gas flow rate changes in a manner
Although Fig. 7 provides a trend of peak ring radial inertia similar to the change in axial bore profile since the end gap area is
effects during intake over engine speeds, more information can be directly related to the shape of bore 共effect of the oil film thick-
obtained by resolving the ‘‘ring separation’’ event at a crank angle ness on the end gap area is generally relatively small兲. Due to the
level for a fixed engine speed. In other words, ring radial motion influence on oil consumption and gas dynamics, analysis of this
as it slides over the ‘‘depression’’ on the liner would provide more type of ring motion merits a detailed investigation.
insight into the phenomenon. Figure 9 shows the top ring radial Figures 6 and 9 show plots of top ring minimum film thickness
motion in the crank angle interval spanning from 400 through 440 共MOFT兲, which are measured from the liner. An alternative
deg 共i.e., early part of the intake stroke兲. For comparative pur- method of analyzing ring radial motion is by tracking the radial
poses, Fig. 9 shows the MOFT of the top ring in this crank angle location of the top ring measured from the cylinder axis. To this
interval for an axially profiled bore as well as a straight bore at a effect, Fig. 11 shows radial location of the top ring versus axial
speed of 6000 rpm. The MOFT values for a top ring moving in an location along the liner when the ring slides over the ‘‘depression’’
axially profiled bore behave in a manner consistent with instanta- during the intake stroke for increasing engine speeds. At higher
neous ring-liner clearance profiles. To demonstrate this aspect, speeds, increasing inertial effects lead to ‘‘ring separation’’ which
Fig. 9 also shows the instantaneous axial location of the top ring is clearly visible and highlighted as the top ring emerges from the
inside the profiled bore in this crank angle interval. In particular, ‘‘depression’’ 共located at a distance of 0.0425 m from top of the
around 425 crank angle deg, as the piston/ring assembly acceler- liner兲.
ates towards the bottom of the liner, the top ring moves out of the The ‘‘ring separation’’ seen in Fig. 11 at high engine speeds
‘‘depression’’ and radially accelerates away from the bore leading 共e.g., 6000 rpm兲 occurs during the downward motion of the
to ‘‘ring separation.’’ As seen from Fig. 9, the top ring eventually piston/ring assembly in the early part of the intake stroke. During
returns to the bore around 430 crank deg. up-strokes 共e.g., exhaust兲, the signs of velocity and acceleration of
In the duration of ‘‘ring separation’’ 共from 425 to 430 crank the piston/ring assembly in Eq. 共15兲 changes compared to the
deg兲, oil flow rate past the ring face and gas flow rate through the intake stroke for the same axial location. Thus, Fig. 12 shows the
end gap also show a corresponding response to the radial ring differences in values of radial location of the top ring between the
motion. Figure 10 shows the oil flow rate below the top ring face intake and exhaust strokes when the top ring slides over the ‘‘de-
and gas flow rate through the end gap for the same period as in pression’’ at an engine speed of 6000 rpm.
Fig. 9. An initial inspection of Fig. 12 seems to suggest that the top
As shown in Fig. 10, there is increased oil flow under the top ring separates from the bore only during the intake stroke and not
ring face towards the combustion chamber creating the potential during the exhaust stroke. However, this is not the case as noted
for higher oil consumption. The implications of the ‘‘depression’’ from a closer look at the ring radial motion provided in Fig. 13.
in the bore profile relevant to the gas dynamics model are 共1兲 Figure 13 shows that during the exhaust stroke ‘‘ring separation’’
Fig. 11 Radial location of top ring face measured from the cyl-
Fig. 9 MOFT of top ring in the intake stroke at 6000 rpm inder axis at different engine speeds during intake stroke
Simulations have been carried out for the baseline axial bore pro-
file depicted in Fig. 4 at a fixed engine speed of 6000 rpm.
Figure 14 displays the top ring radial motion during the intake
stroke when it slides over the ‘‘depression’’ for different ring ten-
sion values. As expected, with increasing tension, the ring con-
forms more to the profiled bore and shows lower tendency to
‘‘separate’’ from the bore.
In the experimental work conducted by M. Basaki et al. 关6兴,
instantaneous end gap sizes were measured for various ring ten-
sion values. The simulated results carried out in this study are
Fig. 12 Radial location of top ring face measured from the

cylinder axis during intake and exhaust strokes at 6000 rpm
occurs just after the top ring enters the ‘‘depression’’ 共at a location
approximately 0.04 m measured from top of the liner兲. On the
other hand, Fig. 13 also shows ‘‘ring separation’’ during early part
of the intake stroke when the ring emerges from the ‘‘depression’’
in the bore profile.
Effect of Ring Tension. Ring tension plays a key role in the
conformance of a ring to a distorted bore. Hence, it has been
selected as one of the variables in the parametric study undertaken
here. In particular, the installed tension of the top ring and its
effect on radial ring motion in an axially profiled bore are of
interest. For purposes of this study, three values of the top ring
tension 共tangential兲 are used; a baseline tension of 9.39 N, a high
tension value of 12.21 N and a low tension value of 6.58 N.
Fig. 13 Enlarged view of highlighted region in Fig. 12
Fig. 15 Top ring end gap sizes during intake and exhaust
Fig. 14 Effect of ring tension on radial location of top ring face strokes for different tensions; „a… 9.39 N, „b… 12.21 N, and „c…
during the intake stroke at 6000 rpm 6.58 N at 6000 rpm
Fig. 16 ‘‘Ring separation distance’’ versus engine speeds for
Fig. 18 Various axial bore profiles used in this study
different values of top ring tension
presented for comparison purpose. Figure 15 shows the simulated section is to illustrate how the shape of the ‘‘depression’’ and its
top ring end-gap behavior during the intake and exhaust strokes, location on the liner affects the radial dynamics of the top ring.
for different tension values, when the top ring is moving over the Figure 18 shows the three axial bore profiles used 共a兲 baseline
‘‘depression’’ in the bore profile. profile with the ‘‘depression’’ located in the upper part of the
The changes in end-gap size as the ring slides over the ‘‘depres- cylinder 共also shown in Fig. 4兲, 共b兲 bore profile designated ‘‘A’’
sion’’ was also measured in 关6兴 and the results from that study are with the identical ‘‘depression’’ but located in the lower portion of
shown in Fig. 22 in the Appendix. From Fig. 15 it is noted that, the cylinder, and 共c兲 a bore profile designated ‘‘B’’ with a ‘‘bump’’
during ‘‘ring separation,’’ there are smaller changes in the end-gap instead of a ‘‘depression’’ with identical geometry and at the same
size between the intake and exhaust strokes with increasing ring axial location of the baseline case in the cylinder.
tension values. The instantaneous response of ring end-gaps to a Figure 19 shows the top ring radial motion when it slides over
bore profile is critical to gas blow-by, blow-back, and oil con- the ‘‘depression’’ associated with bore profile A during the exhaust
sumption in engines. The computed results show good qualitative and intake strokes at 6000 rpm. Comparisons in top ring radial
comparison with the measurements and underscore the need for motions between the baseline bore profile and bore profile A dur-
such effects to be incorporated in ring-pack models. ing the intake stroke can be made from Figs. 12 and 19. In par-
Figures 16 and 17 show the combined effects of engine speed ticular, the ‘‘ring separation’’ event during the intake stroke, ob-
and ring tension on the variable ‘‘ring separation distance.’’ As served with the baseline bore profile, is not seen with profile A,
expected, for a fixed engine speed and decreasing ring tension, the from Fig. 19. Also, ‘‘ring separation’’ during the exhaust stroke in
top ring has a higher tendency to move away or ‘‘separate’’ from Fig. 13 is not seen in Fig. 19. The differences in ring radial mo-
the liner as it slides across the ‘‘depression’’ in the bore profile. tions at locations L1 and L2 shown in Fig. 18, arise due to differ-
For a given ring tension and with increasing engine speeds the top ent values of piston/ring velocity and acceleration at those loca-
ring shows the appropriate influence of higher inertial effects as tions on the cylinder bore. Such analysis is useful in determining/
discussed in the prior section. The predicted radial motion of the evaluating the limits of bore distortion so that they do not
top ring in response to the axial bore profile and varying values of adversely affect the ring-pack performance.
engine speed and ring tension compares favorably with the mea- Figure 20 shows the top ring radial motion when sliding in a
sured trends reported in 关6兴 and shown in Figs. 23 and 24 in the bore with profile B that has a ‘‘bump’’ instead of a ‘‘depression.’’
Appendix. Comparing Figs. 12 and 20 shows the main difference; the radial
location of the top ring measured from the cylinder axis decreases
Effect of Axial Bore Profiles. In order to study the influence when the ring moves over the ‘‘bump’’ compared to the increase
of axial bore profiles on top ring radial dynamics; three different in radial location when moving over the ‘‘depression’’ 共i.e., mo-
axial bore shapes are used in the simulation. The intent of this tion at locations L1 and L3兲.
Fig. 19 Radial location of top ring face during the intake and
Fig. 17 ‘‘Ring separation distance’’ versus top ring tension exhaust strokes while sliding over the ‘‘depression’’ in bore
values for different engine speeds profile A at 6000 rpm
Fig. 20 Radial location of top ring face during the intake and
exhaust strokes while sliding over the ‘‘bump’’ in bore profile B
at 6000 rpm
Fig. 23 Ring separation from bore versus engine speed
Fig. 21 Measured cylinder bore deformation Fig. 24 Ring separation from bore versus ring tension
Conclusions 2. The ‘‘ring separation distance’’ can be substantial so as to

The effect of an axial bore profile on ring radial dynamics has allow oil to flow past the ring towards the chamber, potentially
been accounted in the ring-pack phenomenological models and leading to higher oil consumption. Large predicted values of ring
numerically investigated. The predictions from simulations show minimum oil film thickness 共MOFT兲 during ‘‘ring separation’’ are
good qualitative comparisons with experimental results on a simi- accompanied by substantial oil flow rates below the ring-face dur-
lar engine. Some conclusions can be drawn from this study which ing this event.
are listed below: 3. The ring end-gap sizes change continuously as the rings
slide in an axially profiled bore. This has an impact on inter-ring
1. Radial ring motion is significantly influenced by axial bore gas dynamics, blow-by, and blow-back. Additionally, since oil is
profiles. An analytical expression has been developed to represent entrained in the blow-back gases, this oil consumption contribu-
the radial inertia of a ring due to an axial bore profile, which tion mode may also be affected.
provides an insight to the phenomena. There can be situations of 4. The influence of axial bore profiles on ring radial dynamics
‘‘ring separation’’ from the bore dependent on 共a兲 bore profile has been demonstrated in this investigation. These effects should
geometry, 共b兲 engine speed, 共c兲 ring tension, and 共d兲 gas loads. be taken into consideration in ring-pack analysis due to their im-
Fig. 22 Measured ring end gap at 6000 rpm
plication on the performance parameters such as oil consumption, Vi ⫽ volume of subvolume i (m3 )
friction and blow-by. Such analysis helps to 共a兲 evaluate/ x ⫽ radial coordinate 共m兲
determine limits for acceptable bore distortion, 共b兲 optimize y ⫽ axial coordinate 共m兲
location of liner supports without adverse effects on ring-pack z ⫽ radial deformation of cylinder bore 共m兲
behavior. Y1 ⫽ axial location of leading edge of ring oil lubricated
zone 共m兲
Acknowledgments Y2 ⫽ axial location of trailing edge of ring oil lubricated
zone 共m兲
The authors acknowledge the support and permission of ␣ bore ⫽ slope of axial bore profile
Ricardo, Inc. for publishing this paper. ␤ ⫽ radius of curvature at the peak of surface asperities
共m兲
Nomenclature ␥ ⫽ polytropic exponent
A ⫽ flow area of orifice 共m2兲 ␩ ⫽ density of the surface asperities 共l/m2兲
ar ⫽ ring radial acceleration 共m/s2兲 ␮ ⫽ lubricant dynamic viscosity 共Pa•s兲
ay ⫽ ring axial acceleration 共m/s2兲 ␮ mix ⫽ dynamic viscosity of oil-gas mixture 共Pa•s兲
B ⫽ axial ring width 共m兲 ␴ ⫽ mean r.m.s. height of surface asperities 共m兲
Cd ⫽ discharge flow coefficient of an orifice
D ⫽ nominal bore diameter 共m兲 Appendix
E ⫽ composite elastic modulus of contacting materials
共N/m2兲 Figures 21 to 24 in this Appendix are selectively extracted from
F asp ⫽ force due to asperity contact pressures at ring-liner an experimental study on a similar engine by M. Basaki, K. Saito,
interface 共N兲 T. Nakashima, and T. Suzuki 关6兴 in order to make relevant quali-
F frc ⫽ friction force at the ring-groove interface 共N兲 tative comparisons with the current numerical investigation. This
F gas ⫽ force from land gas pressures at ring-liner interface experimental work was conducted on a gasoline engine of 95-mm
共N兲 bore diameter and 86-mm stroke length with a three-ring ring
F gro ⫽ force due to groove gas pressure at back of the ring pack set. In their study, instantaneous end gap size was measured
共N兲 by a visualization method. Using the relation between radial
F ine ⫽ force on the ring due to its radial inertia 共N兲 movement and change in end gap size of a ring, ‘‘ring separation’’
F oil ⫽ force due to oil pressures at ring-liner interface 共N兲 from the bore was calculated as shown in Figs. 23 and 24.
F ten ⫽ force due to installed ring tension 共N兲
h(x) ⫽ ring-groove side clearance 共m兲 References
h(y) ⫽ clearance at the ring face-liner conjunction 共m兲
关1兴 Takiguchi, M., Watanabe K., Kato, T., Sato, M., and Ueno, H., 1999, ‘‘Effects
h0 ⫽ minimum radial distance from ring face to liner 共m兲 of Piston Ring Tension on Oil Consumption and Piston Friction in Diesel
L ring ⫽ ring radial thickness 共m兲 Engines,’’ ASME ICE Spring Technical Conference, Vol. 32-3, Paper No. 99-
ṁ ⫽ gas mass flow rate across boundaries of a subvolume ICE-199.
关2兴 Hill, S. H., Kantola, T. C., Brown, J. R., and Hamelink, J. C., 1995, ‘‘An
共kg/s兲 Experimental Study of the Effect of Cylinder Bore Finish on Engine Oil Con-
mi ⫽ mass of gas in the subvolume i (kg) sumption,’’ SAE Paper No. 950938.
m ring ⫽ ring mass 共kg兲关3兴 Schneider, E. W., Blossfeld, D. H., Lechman, D. C., Hill, R. F., Reising, R. F.,
N ⫽ engine speed 共rpm兲 and Brevick, J. E., 1993, ‘‘Effect of Cylinder Bore Out-of-Roundness on Pis-
ton Ring Rotation and Engine Oil Consumption,’’ SAE Paper No. 93796.
p ⫽ oil film pressure at ring-liner interface 共Pa兲关4兴 Hitosugi, H. et al., 1996, ‘‘Study on Mechanism of Lubricating Oil Consump-
p asp ⫽ asperity contact pressure at ring-liner interface 共Pa兲 tion Caused by Cylinder Bore Deformation,’’ SAE Paper No. 960305.
pd ⫽ downstream pressure of an orifice 共Pa兲关5兴 Goto, T., Arai, T., Goto, T., and Hamai, K., 1989, ‘‘Influence of Cylinder Bore
p gro ⫽ gas pressure in a ring groove 共Pa兲 Distortion on Engine Performances 共Part 1: Static Mode Analysis of Bore
Distortion for Amount of Oil Consumption,’’ JSAE Paper No. 891010.
pi ⫽ gas pressure in subvolume i (Pa) 关6兴 Basaki, M., Saito, K., Nakashima, T., and Suzuki, T., 2000, ‘‘Analysis of Oil
p ll ⫽ gas pressure in the lower land of a ring 共Pa兲 Consumption at High Engine Speed by Visualization of the Piston Ring Be-
p ul ⫽ gas pressure in the upper land of a ring 共Pa兲 haviors,’’ SAE Paper No. 2000-01-2877.
pu ⫽ upstream pressure of an orifice 共Pa兲关7兴 Gulwadi, S. D., 1998, ‘‘A Mixed Lubrication and Oil Transport Model for
Piston Rings Using a Mass-Conserving Algorithm,’’ ASME J. Eng. Gas Tur-
R ⫽ gas constant 共J/kg•K兲 bines Power, 120, pp. 199–208.
s ⫽ radial distance the minima of ring face profile 共m兲关8兴 Gulwadi, S. D., 2000, ‘‘Analysis of Tribological Performance of a Piston Ring
s cg ⫽ radial distance of center of gravity of a ring 共m兲 Pack,’’ Tribol. Trans., 43-2, pp. 151–162.
t ⫽ time 共s兲关9兴 Greenwood, J. A., and Tripp, J. H., 1971, ‘‘The Contact of Two Nominally Flat
Rough Surfaces,’’ Proc. Inst. Mech. Eng., 185, pp. 625– 633.
Ti ⫽ gas temperature in subvolume i (K) 关10兴 Piao, Y., and Gulwadi, S. D., 2000, ‘‘Effect of Piston Secondary Motion on the
Tu ⫽ upstream gas temperature of an orifice 共K兲 Numerical Modeling of Ring Pack Performance,’’ ASME-ICE Fall Technical
Vy ⫽ piston ring axial velocity 共m/s兲 Conference, ICE-Vol. 35-3, Paper No. 2000-342.
Journal of
Engineering
for Gas Turbines
and Power
Erratum
Erratum: ‘‘Effect of Biodiesel Utilization of Wear of Vital Parts in Compression

Ignition Engine’’ „ASME J. Appl. Mech., 2003, 70, pp. 604–611…
A. K. Agarwal, J. Bijwe, and L. M. Das
The title of this paper should read ‘‘Effect of Biodiesel Utilization on Wear of Vital Parts in Compression Ignition Engine’’ 共‘‘Effect of
Biodiesel Utilization on Wear . . . ’’ instead of ‘‘Effect of Biodiesel Utilization of Wear . . . ’’兲.

JEGTP 2003 Vol 125 N4 PDF

Uploaded by

Copyright:

Available Formats

JEGTP 2003 Vol 125 N4 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

JEGTP 2003 Vol 125 N4 PDF

Uploaded by

Copyright:

Available Formats

Transactions Journal of

of the ASME Engineering for Gas ®

Turbines and Power

Journal of Engineering for Gas Turbines and Power OCTOBER 2003

1091 Author Index

„Contents continued on facing page…

Journal of Engineering for Gas Turbines and Power OCTOBER 2003

ANNOUNCEMENTS AND SPECIAL NOTES

1100 Information for Authors

The ASME Journal of Engineering for Gas Turbines and Power is

1104 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Introduction stages of air temperature, including a relatively low temperature

862 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Chemical FBS FBN

864 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

866 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

868 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Fuel Modeled Measured Conclusions and Recommendations

870 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

position within the combustor. The diffuser formed as quarter-

PMT signal shows significant contributions only at the fundamen-

Fig. 7 OH* chemiluminescence of a forced combustion oscil-

874 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Test objects MTU–Burner 1

over the oscillation period are shown. To separate the periodic

Experimental Results with

876 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

878 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

now-lean reactive mixture are then achieved in the downstream

880 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

fuel-rich mixture in contact with the catalyst is oxygen-

Single-Injector Catalytic Pilot

Baseload inlet temperature 705 K 共432°C/810°F兲

Table 2 Air splits in the combustor

Catalytic Pilot Configuration Standard Pilot Configuration

882 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

replacement of the standard diffusion pilot with the catalytic pilot

884 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Introduction Theory of Operation: The Ramgen Engine

Experimental Results: The Pre-Prototype Engine

Fig. 3 Ramgen engine flow details

886 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Fig. 4 Vortex generator flameholder

888 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Fig. 9 Chemical reactor modeling network

Fig. 10 High-speed data trace for point 1

890 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Fig. 12 High-speed data trace for point 2

affords the combustor with the largest possible combustion vol-

Fig. 15 High-speed trace for the 1 cm tall VGs

892 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Fig. 18 VFD comparisons between all VG and pylon runs

894 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Y 共 x,t 兲 ⫽ 具 Y 共 x,t 兲 典 ⫹y 共 x,t 兲 (2) h⬅h s ⫹ 兺h i

896 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

898 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

900 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

902 Õ Vol. 125, OCTOBER 2003 Transactions of the ASME

Mean Residence Times

Y 共 x,t 兲 ⫽ 具 Y 共 x,t 兲典 ⫹y 共 x,t 兲 (2) h⬅h s ⫹ 兺h i

Composition 关%兴共Mass/Volume兲