Kuo1996 PDF

Downloaded by RMIT BUNDOORA LIBRARY on August 16, 2015 | http://arc.aiaa.org | DOI: 10.2514/4.
866432
Purchased from American Institute of Aeronautics and Astronautics
Recent Advances in Spray

Combustion: Spray Combustion
Downloaded by RMIT BUNDOORA LIBRARY on August 16, 2015 | http://arc.aiaa.org | DOI: 10.2514/4.866432
Measurements and Model

Simulation
Volume II
This page intentionally left blank

Recent Advances in Spray

Combustion: Spray Combustion
Measurements and Model
Simulation
Volume II
Edited by
Kenneth K. Kuo
Pennsylvania State University
University Park, Pennsylvania
Volume 171
PROGRESS IN
ASTRONAUTICS AND AERONAUTICS
Paul Zarchan, Editor-in-Chief

Charles Stark Draper Laboratory, Inc.
Cambridge, Massachusetts
Published by the
American Institute of Aeronautics and Astronautics, Inc.
1801 Alexander Bell Drive, Reston, Virginia 22091
Copyright © 1996 by the American Institute of Aeronautics and Astronautics, Inc. Printed in the United
States of America. All rights reserved. Reproduction or translation of any part of this work beyond that
permitted by Sections 107 and 108 of the U.S. Copyright Law without the permission of the copyright
owner is unlawful. The code following this statement indicates the copyright owner's consent that
copies of articles in this volume may be made for personal or internal use, on condition that the copier
pay the per-copy fee ($2.00) plus the per-page fee ($0.50) through the Copyright Clearance Center,
Inc., 222 Rosewood Drive, Danvers, Massachusetts 01923. This consent does not extend to other kinds
of copying, for which permission requests should be addressed to the publisher. Users should employ
the following code when reporting copying from this volume to the Copyright Clearance Center:
1-56347-181-7/96 $2.00 + .50
Data and information appearing in this book are for informational purposes only. AIAA is not respon-
sible for any injury or damage resulting from use or reliance, nor does AIAA warrant that use or
reliance will be free from privately owned rights.
ISBN 1-56347-181-7
Progress in Astronautics and Aeronautics
Editor-in-Chief
Paul Zarchan
Charles Stark Draper Laboratory, Inc.
Editorial Board
John J. Berlin Leroy S. Fletcher

U.S. Air Force Academy Texas A&M University
Richard G. Bradley Alien E. Fuhs

Lockheed Martin Fort Worth Company Carmel, California
William Brandon Ira D. Jacobsen

MITRE Corporation Embry-Riddle Aeronautical University
Clarence B. Cohen John L. Junkins

Redondo Beach, California Texas A&M University
Luigi De Luca Pradip M. Sagdeo

Politechnico di Milano, Italy University of Michigan
Martin Summer-field
Lawrenceville, New Jersey

Preface
In view of convenience in transport and flexibility in storage of liquid fuels, spray

combustion processes have been utilized in many engineering applications, in-
cluding energy sources for propulsion and transportation systems, electrical power
generation in power plants, waste disposal and energy recovery in incinerators,
and furnaces for material processing purposes. In the propulsion and transportation
areas, spray combustion has been used in various engines, including liquid rocket
engines (rocket motors), diesel engines (cars and trucks), gas turbine combus-
tors (aircraft), hybrid rocket motors (space launch vehicles), regenerative liquid
propellant guns (advanced gun systems), and ramjets (air-breathing propulsion
systems).
For propulsion studies, this two-volume set of spray combustion books serves
as complementary material to the excellent AIAA Progress Series Volume 147
entitled, Modern Engineering for Design of Liquid-Propellant Rocket Engines,
edited by Dieter K. Huzel and David H. Huang, and further updated and enlarged
by many liquid rocket engine experts of the Rocketdyne Division of Rockwell In-
ternational. While their book stresses the design aspects of liquid propellant rocket
engines, these volumes emphasize the fundamental aspects of spray atomization
and combustion, covering experimental measurements, theoretical modeling, and
numerical solutions.
In general, spray combustion processes have been studied for years to achieve
more economical use of fuels, better control of pollutants in combustion products,
and longer lifetime of engineering devices. However, due to the complex nature
of the spray atomization and combustion processes, many practical devices were
designed based upon the trial-and-error approach, which is very expensive.
Over the last decade, researchers have made significant improvements in non-
intrusive diagnostic methods (laser-based techniques, x-ray radiography, high-
resolution imaging, phase Doppler particle analysis, planar laser-induced fluores-
cence, etc.) that enable more detailed observations and measurements of spray
atomization and combustion processes. The development of supercomputers with
large memories and high-speed processors enables theoreticians to formulate and
numerically solve comprehensive models with more detailed consideration of
physical and chemical processes involved in spray combustion.
Many areas of spray combustion research have attained major advancements
in recent years. Some of these areas include: drop size measurements, liquid
jet breakup mechanisms, characterization of dense spray bahavior, supercritical
evaporation and combustion phenomena, numerical solution of comprehensive
models with complex chemical kinetics, quantitative measurements of sprays using
modern experimental techniques, and externally induced excitation on atomization
processes. The recent advancements in these areas form the major topics addressed
in these volumes, the full scope of which is too broad to be constrained to a single
author or a portion of a combustion textbook. Therefore, many active researchers
with international reputations in these areas were invited to contribute to this
specially planned set.
The objectives of these volumes are: 1) to present various areas of progress

associated with spray atomization and combustion in a systematic manner, 2) to
familiarize readers with the state of the art in this important field, 3) to identify
remaining technological gaps in order to promote further research of unresolved
problems in spray combustion, and 4) to provide a useful document for young
engineers and scientists to enhance their knowledge of up-to-date advances in the
area of spray combustion.
This two-volume set is aimed at individuals in industry, government, or uni-
versity research laboratories who have a technical background in mechanical,
chemical, aerospace, aeronautical, or computer engineering. Engineers and sci-

entists working in chemical processes, thermal energy generation, propulsion,
environmental control, or food processing industries utilizing liquid and solid
sprays should find these volumes useful and informative.
These Progress Series books contain a total of eight subject areas divided into
two volumes. The first volume is dedicated to the study of spray atomization
and drop burning phenomena, covering four specific areas: drop sizing tech-
niques, breakup processes of liquid jets and sheets, dense spray behavior, and
supercritical evaporation and burning of liquid propellants. The second volume
emphasizes spray combustion measurements and model simulation. It includes
four areas: spray combustion measurements, modeling and numerical simulation,
externally induced excitation and wave interaction on atomization processes, in-
stability of liquid fueled combustion systems, and spray combustion in certain
practical systems.
Some articles are intended to be more tutorial in nature and include discussions
of historical developments in spray combustion. These articles are organized in
the front portion of the subject area. Each contributing article was subjected
to a rigorous peer review process, similar to the AIAA Journal or the Journal
of Propulsion and Power. Many authors also served as reviewers in their areas
of expertise, thus reducing duplication of subject material, providing uniform
nomenclature, and ensuring high quality and coherence in this book set.
It is my understanding that there still exist many unresolved problems in spray
combustion, including: combustion instability behavior of propulsion devices,
measurement of supercritical combustion behavior of ligaments and droplets of
liquid fuels and propellants, jet impingement-induced breakup phenomena, and
micro-explosion of fuel droplets with multiple ingredients. Hopefully, some of
these processes will be investigated and become more fully developed in the future.
I should take this opportunity to thank all authors and reviewers for their precious
time and great effort in the preparation of this book set. Also acknowledged is
the support of unnamed agencies who sponsored work by the authors. I would
like to take this opportunity to thank Professor Martin Summer-field, who served
devotedly as the AIAA Progress in Astronautics and Aeronautics Editor-in-Chief
for many years, for both his confidence in my editorial role and recommendation
for this book set. The cooperative assistance of the AIAA staff, in particular of
Kim Walters and Rodger Williams of the editorial department in Washington, DC,
is gratefully acknowledged. I would also like to thank Debra Kimble and Connie
Peters for their technical assistance, and my family for their patience and support.
Finally, I would like to dedicate my editorial work on this book set to my beloved
sister, Katherine K. Lu.
Kenneth K. Kuo
January 2, 1996
Table of Contents
Preface
VOLUME II: Spray Combustion Measurements
and Model Simulation
Subject Area 1. Spray Combustion Measurements
Chapter 1. Continuous- and Dispersed-Phase Structure of Pressure-

Atomized Sprays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
L. K. Tseng, Purdue University, West Lafayette, Indiana, G. A. Ruff, Drexel

University, Philadelphia, Pennsylvania, P.-K. Wu, Taitech, Inc., Dayton, Ohio, and
G. M. Faeth, University of Michigan, Ann Arbor, Michigan
Chapter 2. Behavior of Droplets in Pressure-Atomized Fuel Sprays with

Coflowing Air S w i r l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1
C. Presser, National Institute of Standards and Technology, Gaithersburg, Maryland,
C. T. Avedisian, Cornell University, Ithaca, New York, J. T. Hodges, National
Institute of Standards and Technology, Gaithersburg, Maryland, and A. K. Gupta,
University of Maryland, College Park, Maryland
Chapter 3. Spray Diagnostics with Lasing and Stimulated Raman

Scattering....................................................63
J. C. Swindal and W. P. Acker, Texaco, Inc., Beacon, New York, and G. Chen, A.
Serpengiizel, and R. K. Chang, Yale University, New Haven, Connecticut
Chapter 4. Feasibility of Droblet Size and Concentration Measurements in

a Hypergolic Prepellant Burning Spray . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1
R. Lecourt, R. Foucaud, G. Lavergne, P. Berthoumieu, and P. Millan, Office National
d'Etudes et de Recherches Aerospatiales, Chdtillon, France
Chapter 5. Laser Sheet Visualization of Spray Structure.............113

D. G. Talley and A. T. S. Thamban, Phillips Laboratory, Edwards Air Force Base,
California, and V. G. McDonell and G. S. Samuelsen, University of California,
Irvine, California
Chapter 6. Use of Fluorescence Methods for Measurement of Droplet and

Vapor T e m p e r a t u r e s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 4 3
L. A. Melton, University of Texas at Dallas, Richardson, Texas
Chapter 7. Assessing the Physics of Spray Behavior in Complex

Combustion S y s t e m s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 5 9
V. G. McDonell and G. S. Samuelsen, UCI Combustion Laboratory, University of
California, Irvine, California
Chapter 8. Structure of a Coflow Laminar Spray Diffusion Flame.... .187

G. Chen and A. Gomez, Yale University, New Haven, Connecticut
Chapter 9. Coaxial Airblast Atomizers with Swirling Air Stream......201

Y. Hardalupas and J. H. Whitelaw, Imperial College of Science, Technology and
Medicine, London, United Kingdom
Chapter 10. Pressure Oscillation Effects on Jet B r e a k u p . . . . . . . . . . . . . . 2 3 3

,S. Pal, H. Ryan, D. Hoover, and R. J. Santoro, Pennsylvania State University,
University Park, Pennsylvania
Subject Area 2. Spray Combustion Modeling and

Numerical Simulation
Chapter 11. Simulation of High-Pressure Spray Field Dynamics . . . . . . . 2 6 3

J. C. Oefelein and V. Yang, Pennsylvania State University, University Park,
Pennsylvania
Chapter 12. Numerical Simulation of Deformed Droplet Dynamics and

Evaporation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 0 5
S.-M. Jeng, University of Cincinnati, Cincinnati, Ohio, and Z. T. Deng, Alabama
A&M University, Normal, Alabama
Subject Area 3. Instability of Liquid Fueled Combustion Systems
Chapter 13. Numerical Simulation of High-Frequency Combustion

Instabilities in Liquid Rocket Engines . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3 1
J. M. Grenda and C. L. Merkle, Pennsylvania State University, University Park,
Pennsylvania
Chapter 14. Performance and Stability Characterization of Liquid

Oxygen/Kerosene Injectors at A e r o j e t . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 4 9
J. Pieper, T. Nguyen, and J. Hulka, PRA-SA-APP-OSD 94-S-0469
Chapter 15. Acoustic Sensitivity Measurements of Injectors in a Small

Liquid Propellant Rocket Engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 6 9
R. Lecourt and R. Foucaud, ONERA, Chdtillon, France
Subject Area 4. Spray Combustion in Certain Practical Systems
Chapter 16. Spray Combustion Processes in Internal Combustion

Engines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 9 5
S.-C. Kong and R. D. Reitz, University of Wisconsin—Madison, Madison, Wisconsin
Chapter 17. Spray Combustion Processes in Regenerative Liquid

Propellant G u n s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2 5
G. P. Wren, J. D. Knapton, and T. P. Coffee, Weapons Technology Directorate, U.S.
Army Research Laboratory, Aberdeen Proving Ground, Maryland
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 5 5
List of Series V o l u m e s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 5 7


Chapter 1
Continuous- and Dispersed-Phase Structure

of Pressure- Atomized Sprays

L.-K. Tseng*
Purdue University, West Lafayette, Indiana 47907-1003
G. A. Rufft
Drexel University, Philadelphia, Pennsylvania 19104
P.-K. Wu>
Taitech, Inc., Dayton, Ohio 45433-0830
and
G. M. Faeth§
University of Michigan, Ann Arbor, Michigan 48109-2118
Nomenclature
C, = empirical coefficient
d = injector exit diameter
dp = drop diameter
ep = drop or liquid element effective ellipticity
/ = mixture fraction
L = injector passage length
Lb = length to onset of breakup
LB = liquid core length
t = characteristic eddy size
MMD = mass median drop diameter
Oh — jet Ohnesorge number, ^f/(pfda}^
p = pressure
r = radial distance
Reij — Reynolds number based on phase / and dimension j,
SMD = Sauter mean diameter
Copyright © 1995 by the American Institute of Aeronautics and Astronautics, Inc. All
rights reserved.
*Research Fellow, School of Mechanical Engineering.
t Associate Professor, Mechanical Engineering and Mechanics Department.
^Senior Research Engineer.
§
Arthur B. Modine Professor, Department of Aerospace Engineering.
4 L.-K. TSENG ET AL.
u — streamwise velocity
v = radial velocity
\)t = cross-stream velocity of size i eddy
Wetj = Weber number based on the density of phase / and
dimension y, piU20j/cr
Weg = Favre-averaged spray Weber number
x = streamwise distance
of = volume fraction
A = radial integral scale at jet exit
/z = molecular viscosity
p = density
a = surface tension
rb = drop breakup time
TI = characteristic breakup time, Eq. (4)
TR = characteristic Rayleigh breakup time
Subscripts
c = centerline value
/ = liquid-phase property
g = gas-phase property
/ = onset of breakup condition
p = drop property
0 = injector exit condition
Superscripts
(~) = time-averaged mean property
(~) = mass-averaged mean property
(")' = time-averaged rms fluctuating property
I. Introduction
M OST existing studies of sprays have emphasized the dilute spray region far
from the injector exit, where both experimental observations and modeling
are relatively tractable because liquid volume fractions are small and the flow
is reasonably approximated as a dilute dispersed multiphase flow. As a result,
many features of dilute sprays are understood reasonably well (See the reviews
of Refs. 1-20 and references cited therein). With many features of dilute sprays
understood, greater attention is now being directed to the more problematical
dense-spray region near the injector exit. The main interest in dense sprays is to
determine how the injector design and flow properties, and the spray environment,
combine to influence the properties of the flow entering the dilute-spray region.
Naturally, another important objective is to develop reliable ways to prescribe the
initial conditions needed for rational modeling of dilute spray properties. Thus, the
objective of the present paper is to briefly review recent studies of the properties of
the dense-spray region and to identify areas where additional research is needed.
Two major aspects of dense sprays are reviewed, as follows: 1) the defini-
tion and specification of spray breakup regimes, which is a crucial first problem
of dense sprays similar to the importance of flow regimes for other single and
multiphase flows; and 2) the structure of the near-injector dense-spray region, in
order to help define the environment of various unit processes in dense sprays,
CONTINUOUS- AND DISPERSED-PHASE STRUCTURE 5
e.g., breakup, interphase transport, collisions, etc. Information about other spray
processes and injector configurations can be found in earlier review articles,1"20 and
in references cited therein. Results from a variety of sources will be considered,
however, a series of studies by the present authors will be emphasized,21"36 because
they provide a comprehensive body of information about round pressure-atomized
sprays. Because of space limitations, however, present considerations will be lim-
ited to processes directly relevant to the classical problem of nonevaporating round
pressure-atomized sprays in still gases. Ignoring evaporation (and combustion) is
reasonable because the dense-spray region of practical sprays generally involves
the cool portions of the flow where rates of interphase heat and mass transfer are
small.5'6 In addition, round jetlike flows in still gases are a simple classical flow
configuration that exhibits most of the important features of dense sprays while
only requiring a few defining parameters.
II. Spray Breakup Regimes

A. Introduction
Similar to the importance of identifying whether a single-phase flow is laminar
or turbulent, specification of the topography, or spray breakup regime, is a crucial
issue for dense sprays. Recent work has shown that the definition of the spray
breakup regime is significantly influenced by the degree of vorticity, and by the
presence or absence of turbulence, in the flow leaving the injector passage. In order
to address these points, the following discussion will be divided into three phases:
nonturbulent breakup (or classical breakup regimes), effects of jet exit conditions
(or vortical flow), and turbulent breakup.
B. Nonturbulent Breakup
Several spray breakup regimes have been identified for nonturbulent liquids in
still gases, as follows: nonjetting or drip, stable liquid jet, Rayleigh breakup, first
and second wind-induced breakup, and atomization breakup.37"44 The drip regime
involves low flow rates with We/d < 3-4, which ends when the momentum of
the fluid at the injector exit becomes large enough to overcome surface tension
forces acting over the entire passage exit; this region generally involves low jet
exit velocities where flow properties are significantly influenced by effects of
buoyancy.37 The stable liquid jet regime is present for Oh > 2-4, where liquid
viscous forces become large enough to damp all disturbances that would lead to
breakup.38 Naturally, the degree of atomization, or primary breakup, is negligible
in both these regimes; therefore, these flows do not produce dense sprays and they
will not be considered any further here.
The remaining spray breakup regimes for nonturbulent liquids, along with a
qualitative indication of the length L/, where the liquid jet begins to break up, are
illustrated in Fig. 1. As the flow rate through the injector passage increases, breakup
regimes are observed in the following order: Rayleigh, first wind-induced, second
wind-induced, and atomization breakup. Rayleigh breakup involves interactions
between liquid inertia and surface tension forces, yielding drop diameters that are
somewhat larger than the injector diameter with the distance to the point where
breakup begins progressively increasing with increasing flow rates. The remaining
breakup regimes involve gas-phase aerodynamic effects with the distance to the
L-K. TSENG ET AL
SECOND WIND-INDUCED
ATOMIZATION
Fig. 1 Spray breakup lengths and regimes for nonturbulent round liquid jets injected
into still gases, from Faeth.7
point where breakup begins decreasing with increasing flow rates. First wind-
induced breakup is caused by twisting or helical instability of the liquid column
as a whole, with breakup yielding drops that have diameters comparable to the
injector diameter. Second wind-induced breakup involves both helical (near the end
of the liquid column) and liquid surface instabilities of the liquid column, yielding
a wide-range of drop sizes extending up the injector diameter. With increasing
liquid flow rate, breakup along the sides of the jet occurs progressively nearer to
the injector exit. Finally, the atomization breakup regime is reached when breakup
at the liquid surface begins immediately at the injector exit, yielding drops that
are generally small in comparison to the injector diameter. The dense-spray region
during atomization breakup consists of a liquid core, analogous to the potential
core of a single-phase jet, surrounded by a developing multiphase mixing layer
CONTINUOUS- AND DISPERSED-PHASE STRUCTURE
1o i
STABLE JET
N\Vs> XXXV
10°
103 p f /p f l = 1Q2
Oh 10' 1
DRIP RAYLEIGH WIND- ATOMIZATION

INDUCED
10-2
10-3
10-3 100 10 102 103
W6g
Fig. 2 Spray breakup regime transitions for nonturbulent round liquid jets injected
into still gases, from Faeth.7
(see Fig. 1). Nevertheless, the end of the liquid core is still affected by large-
scale helical instabilities, similar to the behavior of first and second wind-induced
breakup, and still yields relatively large drops downstream of the end of the
liquid core due to this final breakup process. Since practical sprays require small
drops for adequate mixing, the atomization breakup regime is the most important
topography for dense sprays.
Figure 2 represents a convenient illustration of conditions required for the
various breakup regimes in terms of Ohnesorge number and Wegd because the
important transitions to the stable jet, wind-induced, and atomization breakup
regimes involve nearly constant values of these parameters (see Ref. 7 for a
tabulation of properties of breakup regimes). In contrast, the transition between the
drip and the Rayleigh breakup regimes is a function of the liquid/gas density ratio
because Wefd = (pf/pg)Wegd = 4 at this transition. This implies a progressively
narrower Rayleigh breakup regime as p//pg decreases. Ranz38 suggests transition
to wind-induced breakup at Wegd & 0.4 for low Ohnesorge number, whereas
Sterling and Sleicher44 observe transition at Wegd = 1.2 + 3AlOhQ'9 for Oh < 0.4
(Ref. 42). Transition between first and second wind-induced breakup is gradual
and a transition criterion has not yet been found.42 Transition to the atomization
breakup regime also is gradual, with Ranz38 and Miesse41 suggesting Wegd = 13
and 40, respectively.
Transitions to the stable jet regime have received little attention, and are not very
important at low pressures where values of the Oh tend to be small for most liquids
and typical injector diameters.7 Nevertheless, heavy fuels have larger Ohnesorge
number, whereas all liquids have large Ohnesorge number during high/pressure
combustion processes where the liquid surface approaches the thermodynamic
critical point.4'6 In particular, the surface tension of the liquid becomes small,
while the liquid viscosity remains finite, as the thermodynamic critical point
is approached, which implies large Ohnesorge number. Thus, in view of the

importance of high-pressure spray combustion processes to contemporary power
and propulsion systems, additional study of large Ohnesorge number transitions
is needed. Additionally, it is likely that the criteria of Fig. 2 are incomplete, e.g.,
aerodynamic effects will be shown later to affect transitions for turbulent liquid
jets and it is likely that they affect nonturbulent liquid jets as well, especially
transitions involving aerodynamic effects such as second wind-induced breakup.
Finally, the criteria of Fig. 2 were established using small practical injectors that
problems of flow development and disturbances that can affect primary breakup
and thus the criteria for the presence of particular breakup regimes; these issues
will be addressed next.
C. Effects of Jet Exit Conditions

It has been recognized for some time that flow properties at the exit of the
injector passage have a significant effect on the atomization properties of pressure-
atomized sprays in still gases (See Refs. 21-25,31-35,45-60, and references cited
therein). For example, even the early studies of De Juhasz et al.45 and Lee and
Spencer46 showed that atomization quality differed for nonturbulent (laminar) and
turbulent jet exit conditions. Subsequently, Phinney48 and Grant and Middleman49
found that jet stability and the onset of breakup were affected by the presence of
turbulence at the jet exit. In fact, more recent work shows that jet exit conditions
dominate atomization properties at the large liquid/gas density ratios typical of
pressure atomization processes in air at atmospheric pressure. For example, Hoyt
and Taylor51~53 observed little effect of relative velocities in primary breakup when
jet exit conditions were turbulent for relative velocities comparable to injection
velocities. Supporting evidence from Arai et al.54'55 and Karasawa et al.57 shows
that breakup can be suppressed entirely for supercavitating flows (where the liquid
jet separates from the injector passage wall and does not reattach), which have a
relatively uniform and nonturbulent flow at the jet exit. Such behavior is really not
surprising because jet exit conditions of this type are widely used for liquid jet
cutting systems, where avoiding breakup is a major design objective. See Yokota
et al.58 and references cited therein.
The recent application of pulsed holography to dense sprays has helped to quan-
tify some effects of jet exit conditions and aerodynamic forces on the properties
of pressure atomization in still gases, see Refs. 21-25 and 32-34. These studies
involved two types of jet exit conditions: 1) so-called nonturbulent flow, where a
large contraction based on Smith and Wang61 passage shapes yielded a nonturbu-
lent flow having uniform velocities at the jet exit (except for laminar boundary
layers along the passage wall); and 2) turbulent flow, where the contraction was
followed by a long length-to-diameter ratio passage to yield fully developed tur-
bulent pipe flow at the jet exit. It was found that mixing rates were faster, that
the length of the potential-core-like liquid column near the jet exit was reduced,
and that drop sizes after primary breakup were increased, for turbulent rather than
nonturbulent jet exit conditions. Furthermore, similar to an earlier suggestion of
Reitz and Bracco,43 the classical aerodynamic theories of primary breakup due to
Taylor62 and Levich2 were not effective. For example, for liquid/gas density ratios
greater than 500, drop sizes after nonturbulent primary breakup were successfully
correlated by an expression that did not involve surface tension, in contrast to the
expectations of the aerodynamic breakup theories of Taylor62 and Levich2; instead,
consideration of vorticity due to the presence of boundary layers along the injector
passage wall yielded a reasonably successful correlation of nonturbulent primary
breakup.35 Similarly, for liquid/gas density ratios greater than 500, drop sizes after
turbulent primary breakup were best explained by detailed consideration of liquid
turbulence properties, as will be discussed later. Aerodynamic effects were ob-
served for turbulent primary breakup at lower liquid/gas density ratios; however,
this behavior was best explained by enhanced effects of secondary breakup.34
Taken together, these observations present a rather strong case that most existing
measurements of primary breakup properties, and definitions of apparent spray
breakup regimes, have been dominated by effects of vorticity and turbulence at

the jet exit. In a sense, such behavior should not be surprising due to the strong
role played by initial disturbance levels on even classical descriptions of primary
breakup processes.2'62 Thus, effects of turbulence, or even the presence of vorticity
in a nonturbulent liquid, caused by fluid motion in the injector passage, tends to
dominate primary breakup processes for liquid injection into air at normal pressure
and temperature. This behavior is enhanced because the competing effects of mo-
mentum exchange between the phases are relatively weak, due to large liquid/gas
density ratios, so that the gas phase is relatively ineffective for developing the vor-
ticity in the liquid that is needed to create drops at the liquid surface in comparison
to the passage walls. Unfortunately, the use of practical injectors for most studies
of spray processes, where small and frequently irregular passages preclude precise
specification or measurement of flow properties at the passage exit, has been a sub-
stantial impediment for quantifying effects of flow properties at the injector exit
on the subsequent properties of the near-injector dense-spray region. Thus, several
recent studies have considered experimental conditions where flow properties at
the injector exit can be defined reasonably well, in order to circumvent these dif-
ficulties. See Refs. 21-25 and 32-35 and references cited therein. Aspects of this
work relevant to the definition of spray breakup regimes will be discussed next.
Consideration of effects of vorticity (nonuniform viscous flow) on the breakup
of liquid jets at large liquid/gas density ratios will begin with a discussion of the
observations of nonturbulent liquid jets in still air at normal temperature and pres-
sure found in Wu et al.35 In this study, the importance of vorticity at the jet exit was
observed by the pulsed shadowgraph photographs of the liquid jet near the injector
exit. Two injector conditions were considered: one involving nonturbulent flow
(with a uniform velocity distribution across the inviscid part of the flow) leaving
the contraction section without removal of the boundary layer near the passage
walls, the other involving the same contraction section and flow rate but with the
boundary layer removed by a sharp-edged cutter followed by a short constant-area
section having a length-to-diameter ratio of 0.15. When the cutter was absent, the
liquid surface became distorted near the injector exit followed by the formation
of ligaments which subsequently broke up at their tips to form drops. In contrast,
when the boundary layer at the passage exit is removed by the cutter, followed by
a constant diameter section that was too short to allow significant development of
subsequent boundary layers along the passage walls, primary breakup along the
liquid surface was entirely suppressed. In this latter case, a solid liquid stream,
similar to those used for liquid cutting jets (see Yokota et al.58) was observed,
which only involved the appearance of relatively large-scale surface irregulari-
ties [varicous (axisymmetric) and sinuous (helical) disturbances] at x/d > 30.
In addition, the absence of significant aerodynamic effects for liquid/gas density
ratios greater than 500 was confirmed by repeating these observations in a helium
environment, with the results indicating negligible effects of varying liquid/gas

density ratios over this range.35 Thus, similar to the earlier observations of Arai
et al.54'55 and Karasawa et al.57 for supercavitating injector flows, the importance
of vorticity at the injector exit for pressure atomization of liquids in gases and
liquid/gas density ratios greater than 500 seems clearly established.
Injector passage design, including the inlet contraction, the presence of trips and
other turbulence-promoting devices, and the roughness and length of the constant
area section after the contraction, all can modify the nature of vorticity at the
injector exit.63-64 Thus, extensive study will be required in order to define spray
breakup regimes in terms of jet exit vorticity properties. Some progress along these
lines, however, has been made for turbulence generated within a developing pipe
flow.35 These experiments involved a Smith and Wang61 contraction followed by
a cutter at the exit of the contraction section to generate a uniform, nonturbulent
slug flow. The resulting slug flow was then passed through constant area sections
of given L/d to generate progressive variations of turbulence levels at the jet exit.
Depending on the L/d of the constant area passage, and the Reynolds number of the
flow, various types of behavior were observed, including: nonturbulent (laminar)
liquid surface with no breakup of the surface (typical of the results illustrated for
L/d — 0.15 in Fig. 3), sinuous (both varicous and sinuous surface disturbances)
leading eventually to breakup of the entire liquid column, and turbulent liquid
\ AA AA
V 4* 4*
DB
»02 — o o oooooo O o Jar® o o -
LAM S U
- pf/,ofl d(mm) TYPE fu R'g \
- WATER
623OC 62, 3.6 FDf A A \
867&.C 9.5,64,6.2, FDF
SINUOUS \ TURBULENT
" 867K 4X> CUTTER
2l3b,c 9.5,6.2,3.6 FDF 2 J JETS \ JETS
IO4&.C 9.5,6.2,3.6 FDF * * I
1
.o _ 867° 0.62
- GLYCEROL42% ^^ i
. 957 6.4? 6.2 FDF ., _ OO^y1! of *t • • • • •
4.0 CUTTER 7
i
t ————————>———————————————————
- GLYCEROL 63% o o o o o o -
7240 62 FDF d *
1007 6.4,62 FDF D •
4.O CUTTER 0 *
247 6.2 FDF g B LAMINAR JETS
120 6.2 FDF
n -HEPTANE
.0° ~~ 594C 6.4 FDF O * ~~
- o DATA FROM GRANT AND MIDDLEMAN
(1966).
- b DATA INCLUDES RESULTS FROM O O O O O O
RUFF et ot.(l99!) AND TSENG «t al.
(1992).
_ c DATA INCLUDES RESULTS FROM
WU.t4UMe.IM3). o 0 Q 0 Q 0 Q
1
id | | 1 1 . 1 1 1 1 1 1 1 1 1 1 I I I
4 5
2
1C) 10* io io io6
R
«fd
Fig. 3 Primary breakup regime map for round liquid jets injected into still gases,
from Wu et al.35
surface with irregular distortion of the liquid surface and breakup of the liquid
surface into drops typical of the results illustrated without the cutter in Fig. 3.
The spray regime map found by Wu et al.35 for constant area passages having
various L/d with a nonturbulent uniform flow entering the passage is illustrated in
Fig. 3. The map indicates the regions of nonturbulent, sinuous, and turbulent spray
behavior plotted as a function of passage L/d and Rcfj. In addition to the results
of Wu et al.,35 earlier observations of Ruff et al.,22 Tseng et al.,25 Wu et al.,33 and Wu
and Faeth34 are shown on the plots; these measurements involved simple rounded
inlets but with relatively large L/d constant area sections (L/d > 40) so that the
effect of the inlet on flow properties was negligible. Finally, the observations of
Grant and Middleman49 are shown on the figure: these measurements involved
sharp-edged inlets (Borda type inlets) that promote separation and reattachment
(cavitation) at the inlet of the constant-area section and were more disturbed than
the other measurements.
The primary breakup regimes for Smith and Wang61 contractions in Fig. 3 yield
only laminar jets for L/d < 4-6, even for Refd, up to 106. For L/d > 6, transition
to turbulent jets is observed for Refd in the range 1-4 x 104, with Refd at transition
decreasing as L/d increases, which is expected from the tendency of large L/d
passages to exhibit turbulent flow at lower Reynolds numbers.64 The relatively
large Re/d required for turbulent pipe flow is also typical of the behavior observed
by others for relatively disturbance-free inlet conditions.64 The measurements of
Grant and Middleman49 highlight behavior at the other extreme where strong inlet
disturbances are present: these conditions yielded turbulent jets at Refd ~ 3000,
which is comparable to the lowest Reynolds numbers where turbulent pipe flow
has been observed.63'64 The results of Arai et al.54'55 also suggest that the minimum
L/d for turbulent jets may be reduced from L/d = 4-6 when inlet conditions
are highly disturbed, e.g., they observed jet exit flows typical of turbulent jets at
L/d = 4 with square or slightly rounded inlet contractions. Thus, the limits of the
turbulent breakup regime in Fig. 3 are undoubtedly conservative and representative
of disturbance-free nonturbulent uniform (slug) flows at the inlet. Clearly, much
remains to be done to define turbulent jet exit conditions for practical injectors,
where for the present only direct experiments can provide reliable guidance.
The final issue with respect to breakup regimes involves the influence of aero-
dynamic effects on transition to the turbulent primary breakup regime. Wu et al.35
studied these effects for p//pg in the range 104-72,400 for smooth inlet condi-
tions. No effect on transition to the turbulent primary breakup regime was found
over this range of conditions, which is reasonable because processes within the
injector passage dominate transition to turbulent jet exit conditions.
D. Turbulent Breakup
Turbulent jet exit conditions are a necessary, but not sufficient, condition for
turbulent primary breakup along the liquid surface. Thus, even for turbulent flow
at the jet exit, behavior analogous to the spray breakup regimes for nonturbulent
jets shown in Fig. 1 can still be found, including: a regime analogous to first
wind-induced breakup where the liquid column breaks up as a whole at some
distance from the jet exit, with no prior breakup at the liquid surface; a regime
analogous to second wind-induced breakup where breakup at the liquid surface
begins at some distance from the injector exit and is followed later by breakup of
the liquid column (after some liquid removal due to surface breakup) as a whole;
12 L-K. TSENG ET AL.
DROP
(TYP.)
STREAMLINES
(TYP.)
LTURBULENT
EDDY (TYP.)
Fig. 4 Sketch of process of turbulent primary breakup at a liquid surface, from Wu

and Faeth.34
and a regime analogous to atomization breakup where breakup at the liquid surface
begins right at the jet exit followed by later breakup of the liquid column as a whole
similar to second wind-induced breakup. Photographs of the appearance of the
sprays for these turbulent breakup regimes can be found in Ruff et al.21 The key
to specifying transitions between these regimes is to determine the location where
turbulent primary breakup begins along the liquid surface; work completed thus
far to define this condition will be discussed next.
The onset of turbulent primary breakup along liquid surfaces has been addressed
by Wu et al.33 for p//pg > 500 where aerodynamic effects are small and by
Wu and Faeth34 for p//pg < 500 where aerodynamic effects are significant. The
phenomenological analysis used to find drop properties at the onset of turbulent
primary breakup was based on the configuration illustrated in Fig. 4. Thus, the
onset mechanism is assumed to involve the formation of a drop from a turbulent
eddy having a characteristic size I and a characteristic cross-stream velocity
relative to the surrounding liquid of v^. The eddy is shown to be elongated in
shape because length scales in the streamwise direction are larger than in the
cross-stream direction for turbulent pipe flow.63'64 The eddy is assumed to be
convected in the streamwise direction at the local mean velocity that is taken to
be based on measurements.34 The drop formed by the eddy is assumed to have a
diameter comparable to €.
Drops formed at the onset of turbulent primary breakup are the smallest drops
that can be formed by this mechanism, since drop sizes produced by turbulent
primary breakup progressively increase with increasing distance from the injector
exit.33'34 The smallest drops that can be formed are either comparable to the
smallest (Kolmogorov) scales of turbulence, or the smallest turbulent eddy that
has sufficient mechanical energy to form a drop—whichever is larger. For test

conditions considered thus far, however, Kolmogorov scales have been much
smaller than the smallest observed drop sizes; therefore, only the second criterion
has been considered. Similarly, effects of liquid viscosity were neglected because
conditions of interest were not near Kolmogorov scales where dissipation of
turbulence by viscous effects is important.
The second criterion for the smallest drop that can be formed can be found from
energy considerations. The mechanical energy available to form a drop includes the
kinetic energy of the eddy relative to its surroundings plus the added mechanical
energy due to the pressure drop caused by acceleration of the surrounding gas
over the tip of the protuberance as illustrated in Fig. 4. Equating these sources of
mechanical energy to surface energy required to form a drop then yields34:
(p/vl + CsapguM^Cstalt (1)
where various empirical factors have been absorbed by the coefficients Csa and
CSi, e.g., effects of ellipticity, nonuniform velocities within the eddy, nonuniform
pressure distributions over the protuberance, efficiency of conversion of mechani-
cal energy to surface energy, etc. Liquid-phase turbulence properties decay slowly
in the liquid core and can be approximated by jet exit conditions, whereas for
drops to be formed, €/ must be less than the largest scales present, which are
comparable to the radial integral scale A. Thus, it is reasonable to assume that ij
is in the inertial range of the turbulence spectrum, where
^•~ir o (4/A)i (2)

Finally, it is assumed that SAfD/ ^ •£/ and that the proportionality constants were
selected to best fit measurements of drop sizes at the onset of turbulent primary
breakup to yield
(SMDt/A)[l + 0.04(p,/p / )(«oK) 2 (A/WA) i ] 3/5 = 76 Werf'69 (3)
where VQ/UQ = 0.058 for fully developed turbulent pipe flow.47 This result provides
a reasonable confirmation of the approximate theory; for example, the predicted
power of We/h is —3/5, which is close to the measurements, whereas the large
coefficient on the right-hand side of the equation is expected because it is propor-
tional to (z7o/t>o)6/5 which is large for turbulent pipe flow, i.e., Csi — 2.5 based on
pipe flow turbulence properties, which is of order unity as expected for a parameter
of this type.
Given SAfD/, the location of the onset of turbulent primary breakup, jc/, is found
by assuming that the drop-forming eddy convects along the liquid surface with
a stream wise velocity UQ, which was reasonable based on measurements of the
velocities of protuberances. Then the time of growth of the protuberance until it
breaks up into a drop was taken to be proportional to the Rayleigh breakup time
of a liquid jet having a diameter €/ and jetting velocity i>#. Then, neglecting liquid
viscous effects as before, the characteristic breakup time becomes66:
<rf (4)
L.-K. TSENG ET AL
itf •
LIQUID/GAS d(mm) SYM. "
Pl'Pi
WATER /HELIUM 623O 6.2 <5
™
WATER/AIR 0 667 9.5 0
WATER /AIR 0 667 6.4 A
WATER /AIR 867 6.2 o
"
WATER/ AIR0 867 3.6 *
GLYCEROL 42% /AIR0 957 6.4 0
n- HEPTANE /AIR 0
3 io2 -
594 6.4 Q
WATER/ FREON 12 ( 1 ATM) 213 6.2 «
WATER/FREON 12 (2 ATM) 104 6.2 •

0
FROM WU «t o/. (1992).
A
V^
^ X * .-CORRELATION
•J - w—
-
-
- Q <£x
-
0
5D
* A
- O
\
X
i- I , . .i , , I i i i i
? •o4 icP IO6
Fig. 5 Streamwise location of the onset of turbulent primary breakup for round
liquid jets injected into still gases, from Wu and Faeth.34
Setting SMDi « it in the expression for breakup time, as before, yields the cor-
relation illustrated in Fig. 5. Conditions where aerodynamic effects are important
are denoted by filled and half-filled symbols. The measurements provide a fair
correlation when plotted in the manner of Fig. 5, with a best fit of the results, as
follows34:
= 2570 (5)
Comments about the comparison between the terms on the right-hand side
(RHS) of Eq. (5) and the simplified theory are the same as before.34 The results
of Fig. 5 indicate that the onset of breakup moves progressively toward the jet
exit with increasing liquid flow rates, or W?/A, similar to past observations of
behavior in the second wind-induced breakup regime, e.g., Ref. 21. Nevertheless,
a limitation of the results illustrated in Fig. 5 is that measurements do not extend
very far from the jet exit. This limitation is a concern because the small-scale
portion of the turbulence spectrum, which tends to involve drops at the onset of
breakup, decays with increasing distance from the jet exit: this can affect onset
of turbulent primary breakup properties. Clearly, additional study of the onset of
turbulent primary breakup is needed, emphasizing behavior at greater distances
from the jet exit than past work, as well as conditions where aerodynamic effects
are important.
With decreasing flow rates through the injector passage, but still in the turbulent
jet regime of Fig. 3, the onset of turbulent primary breakup occurs at progressively
larger distances from the injector exit. Thus, a condition is eventually reached
where the liquid column breaks up as a whole before the onset of breakup at the
liquid surface, which is analogous to the onset of the first wind-induced breakup
regime illustrated in Fig. 1 . The length of the liquid core (see the bottom illustration
of Fig. 1) has been studied by Grant and Middleman,49 Hiroyasu et al.,67 Arai
et al.,68 Chehroudi et al.,69 and Andrews.70 The earliest measurements of liquid

core length, carried out by Grant and Middleman,49 were for conditions where
effects of aerodynamic breakup processes were negligible while the flow leaving
the injector exit was turbulent. These results gave the following expression for the
liquid core length49:
8.51Hfe°f (6)
Not surprisingly, this correlation only depends on liquid properties, with
playing a strong role similar to the correlation for the onset of turbulent pri-
mary breakup along the liquid surface of Eq. (5). In contrast, the results of Refs.
67-70 involved conditions where aerodynamic effects were important, yielding
expressions for the length of the liquid core, as follows:
LB/d = CB(pf/p8r> (7)
Hiroyasu et al.67 found CB = 15.8 using an electrical conductivity probe whereas

Chehroudi et al.69 found CB = 7.0 by the same technique and attribute the differ-
ence to the fact that Hiroyasu et al.67 did not account for effects of electric current
carried by drops in the spray. Andrews70 reviews all of the data and suggests
CB = 5 based on the measurements of Lee and Spencer,46 Hoyt and Taylor,51 and
Arai et al.55 Note, however, that values of LB for conditions where aerodynamic
effects are important are strongly dependent on the degree of flow development (or
L/d) at the exit of the injector passage. Thus, additional measurements of LB/d
for conditions where aerodynamic effects are both absent and present would be
desirable to further assess the correlations of Eqs. (6) and (7). Until such results
are available, however, Eqs. (6) and (7) at least provide a rough estimate of LB
and the potential for turbulent sprays to be in the first or second wind-induced
turbulent breakup regimes, based on the comparison between LB and the value of
xi found from Eq. (5).
The preceding discussion of turbulent primary breakup regimes has involved
various approaches, depending on the relative importance of aerodynamic effects.
There are three issues of importance concerning aerodynamic effects for turbulent
primary breakup, as follows: whether aerodynamic effects influence either the
onset or the subsequent properties of turbulent primary breakup, whether aero-
dynamic effects enhance the onset of breakup, and whether aerodynamic effects
enhance turbulent primary breakup once it has begun. Enhancement of the onset
of breakup has already been discussed in connection with Fig. 4. The aerodynamic
enhancement of turbulent primary breakup once it has begun will be discussed
in more detail later. This process basically involves merging of turbulent primary
breakup and secondary breakup, i.e., the aerodynamic breakup of the tip of a
ligament before it can separate into a drop by the Rayleigh breakup mechanism.
Definition of the conditions where these turbulent primary breakup regimes are
observed has been considered by Wu and Faeth,34 as is discussed next.
The measurements of Wu and Faeth34 suggested that liquid/gas density ratio, and
the relative magnitudes of characteristic Rayleigh and secondary breakup times
of ligaments, fix the boundaries of the three turbulent primary breakup regimes.
The resulting breakup regime map, based on available data from Refs. 22, 25,
33, and 34, is illustrated in Fig. 6. Expressing the ratio between the Rayleigh and
secondary breakup times as follows34:
(8)
the measurements illustrated in Fig. 6 yield the following criteria for transitions
between regimes for aerodynamic/nonaerodynamic transition
= 500 (9)
Id1 i « i i i 1 • I I I
NON-AERODYNAMIC
PRIMARY BREAKUP
-
I03 —
%g&&®
\ o o
AERODYNAMIC PRIMARY | AERODYNAMIC
BREAKUP |RJUfi_fl „ PRIMARY
(ENHANCED) oaoadBaflPCg<* a BREAKUP -
(MERGED)
ioz "T
- Pl/Pg d (mm) Up(m/t) pt/pq d (mm) u^m/t) -
C
623O 6.2 <>30-O47<>67 213 9.5 *4O
36° O38 6.2 D 30 a 47 a 67 -
957 6/4? * 32 A 4 3 A 57 4 7 9 499 3.6 022038
867 9.5^,0021 O40O47 i04 6.2 B22$47»67
- 6.4° 0 16 0-19 9 2 2 028 040 36 *22
Q 58 0 8 1 0 109 a
6.2,, 0 30 S 47 a 67 FROM WO f t of. (1992)
b
3.6 « <*35*38 FROM RUFF tt of. (1991)
594 6,4°
c
V20V45 c
FROM TSENG »t at. (1992) -
10' 426 9.5 O40
' . . 1
10' 10'
|/2
Fig. 6 TUrbulent primary breakup regime map, from Wu and Faeth.34

and for enhanced aerodynamic/merged transition
TR/Tb = 4 (10)
These breakup regimes refer to the turbulent primary breakup along a liquid surface
and they should not be confused with the definitions of overall breakup regimes
for pressure-atomized sprays as a whole, such as those of Miesse41 and Ranz39
discussed in connection with Figs. 1 and 2 for nonturbulent jet exit conditions. In
particular, the transition between the enhanced and merged aerodynamic regimes
is crossed as the distance from the jet exit increases for a particular spray and is
not a transition relevant to the spray as a whole.
E. Conclusions
The major conclusions about spray breakup regimes for round pressure-atom-
ized sprays in still gases can be summarized as follows:
1) Spray breakup regimes are strongly influenced by the presence of vorticity
due to turbulence or to variations of mean velocities caused by viscous effects
within injector passages. Thus, the classical breakup regime transitions discussed
in connection with Figs. 1 and 2 should be reconsidered in order to provide better
definition of flow properties at the injector exit for the measurements.
2) The turbulent jet regime discussed in connection with Fig. 3 should be
reconsidered to quantify effects of the disturbed inlet conditions that are typical
of practical injectors.
3) The available database for the onset of turbulent primary breakup along
liquid surfaces, and for the presence of significant aerodynamic effects, discussed
in connection with Figs. 4-6, is rather limited and more study of these effects is
clearly warranted.
4) Available information concerning the length of the liquid core, discussed in
connection with Eqs. (6) and (7), is very limited; additional study of the effects
of flow development at the injector exit and aerodynamic phenomena are needed
in order to better quantify this important property of the dense spray region of
pressure atomization processes.
III. Spray Structure

A. Introduction
The discussion up to this point has demonstrated the importance of flow prop-
erties at the jet exit on the subsequent structure of sprays. As a result, present
considerations of spray structure will focus on fully developed turbulent pipe flow
at the injector exit, so that experimental conditions are well defined and repeatable.
Finally, as noted earlier, present considerations of spray structure will be limited
to the atomization breakup regime, which is of the greatest practical interest.
The streamwise extent of the dense spray region of round pressure-atomized
sprays is generally taken to be comparable to the length of the liquid core.5'6
This region is rather extended, even for turbulent jet exit conditions, based on the
estimates of liquid core lengths discussed in connection with Eqs. (6) and (7). For
example, adopting the estimate of Eq. (7), which is appropriate for the elevated
pressures relevant to practical combustion processes used in power and propulsion
systems, yields Ls/d in the range 150-400 at atmospheric pressure for typical
gases and liquids (allowing for variations of CB in the range 5-16, based on the
18 L-K. TSENG ET AL
estimates given in Refs. 67-70). At elevated pressures, LB/d would subsequently

decrease inversely proportional to the square root of pressure, which still implies
a long liquid core at typical combustion chamber pressures. Thus, liquid cores,
and their associated dense spray regions, are very prominent features of round
pressure-atomized sprays.
A sketch of the near-injector dense region of round pressure-atomized sprays
in the atomization breakup regime appears at the bottom of Fig. 1. The dispersed
flow region adjacent to the liquid core involves a developing multiphase mixing
layer. Downstream of the end of the liquid core, the flow is a multiphase jet that
evolves into a dilute round spray. It will be seen subsequently that the presence
of irregular drops and ligaments in the dispersed flow, rather than a dense array
of drops having a large liquid fraction, is a general characteristic of the dispersed
flow region within dense sprays. Thus, the term "dense spray" itself is something
of a misnomer when applied to the dispersed flow region near the injector exit.
B. Mixing Properties
Flow structure will be illustrated using the measurements of Ruff et al.21~23
and Tseng et al.24'25 for the injection of water into still air at various pressures.
Nonintrusive measurements were used in both studies: deconvoluted gamma ray
absorption measurements for cordlike paths through the flow to find distributions
of liquid volume fractions, single- and double-pulsed holography to find liquid
element and drop size and velocity distributions, and phase-discriminating laser
velocimetry to measure velocities within the continuous (air) phase. Baseline
predictions also were prepared for these measurements, using a mass weighted
(Favre) averaged turbulence model under the locally homogeneous flow (LHF)
approximation, where relative velocities between the phases are assumed to be
small in comparison to mean flow velocities. Ruff et al.21 provide a complete
description of the LHF model for present experimental conditions, whereas Faeth5'6
provides a more general description and evaluation of the LHF methods for sprays
and other dispersed flows.
Measured and predicted time-averaged liquid volume fractions along the axis
of the test sprays at various ambient pressures are illustrated in Fig. 7. The region
near the jet exit (x/d < 3—8) exhibits mean liquid volume fractions near unity. Just
beyond this region, however, mean liquid volume fractions decrease rapidly. The
initial reduction of liquid volume fractions occurs at progressively smaller values
of x/d as the pressure increases, indicating faster mixing rates at higher ambient
gas densities; this behavior is analogous to effects of flow density ratio for single-
phase turbulent jets. See Ricou and Spalding.71 There is good agreement between
measurements and predictions, suggesting that the LHF approach correctly treats
effects of the density ratio of the flow on mixing properties. It should be noted,
however, that conditions illustrated in Fig. 7 represent low levels of mixing; in
particular, all of these results involve Favre-averaged mixture fractions greater
than 0.85 (Ref. 24). For such low levels of mixing, predictions based on the LHF
approximation have been reasonably good in the past,21 because separated flow
effects due to relative velocity differences between the gas and liquid are not
very significant when the mass of the flow is mainly liquid. Finally, although the
variation of liquid volume fraction suggests a relatively short liquid core, this is
not the case when viewed in terms of a mixture fraction that shows that even
1.2
FULLY-DEVELOPED
FLOW
IOO
x/d
Fig. 7 Time-averaged liquid volume fractions along the axis of round pressure-
atomized water sprays injected into still air at various pressures for atomization
breakup with fully developed turbulent pipe flow at the jet exit, from Tseng et al.24
low levels of flapping of the liquid core can explain the observed liquid volume
fraction reductions.24
Predicted and measured radial profiles of mean liquid volume fractions at at-
mospheric pressure are illustrated in Fig. 8 for the same conditions as Fig. 7.
The independent measurements of Ruff et al.21 and Tseng et al.24 agree within
experimental uncertainties, except at x/d = 100 where the greater confinement
of the flow studied by Tseng et al.24 might be a factor. The measurements show
a progressive increase of flow width as the distance from the jet exit increases.
The apparent flow radii based on off/off c, however, are much smaller than for the
single-phase jets as a result of the sensitivity of a to the extent of mixing due
to the large density ratio of the flow. Notably, corresponding flow widths based
on mean void fraction distributions for gas jets in liquids are unusually large for
similar reasons.72 In contrast, for both turbulent^liquid jets in gases and gas jets in
liquids it is found that flow widths in terms of / are relatively normal far from the
jet exit. The comparison between LHF predictions and the measurements is rea-
sonably good, except at large values of x/d. Ruff et al.21 show that the difficulties
20 L-K. TSENG ET AL.

10
FULLY-DEVELOPED
FLOW
DATA,PRESENT
• OATA.RUFFet.al
— THEORY
2r/d
Fig. 8 Radial profiles of time-averaged liquid volume fractions for round pressure-
atomized water sprays in still air at atmospheric pressure for atomization breakup
with fully developed turbulent pipe flow at the jet exit, from Tseng et al.24
with the LHF predictions at large x/d are due to effects of separated flow as the
flow becomes more dilute. Direct experimental evidence of significant effects of
separated flow will be discussed subsequently.
Results considered thus far suggest some qualitative similarities between dense
sprays for pressure-atomization processes and single-phase turbulent jets: the pres-
ence of a potential corelike region, the sensitivity of the flow to jet exit conditions,
the progressively increasing flow widths with increasing streamwise distance, and
the increasing mixing rates as the ratio of the injected to the ambient fluid densities
decreases. These features have prompted the advocacy of LHF methods to esti-
mate spray properties in the past.4"6 Nevertheless, results considered here, and in
more detail in Refs. 21-25, provide clear evidence of the deficiencies of the LHF
flow approximation for quantitative estimates of dense spray properties; therefore,
the separated flow effects responsible for these problems will be considered next.
C. Separated Flow Properties

The first issue concerning separated flow properties involves the methodol-
ogy to be used to report results for various drop sizes. Fortunately, drop size
distributions in sprays have been found to be relatively universal, satisfying the
universal root normal size distribution function with MMD/SMD =1.2, proposed
by Simmons73 based on extensive observations of industrial sprays (see Belz74 for
a detailed discussion of the properties of the root normal distribution function).
Results illustrating this behavior for various positions in the dispersed-flow region
of round pressure-atomized sprays, for both nonturbulent and turbulent jet exit
conditions, are reported by Ruff et al.23 This finding was confirmed by later mea-
surements of drop sizes after primary breakup by Wu and coworkers32"36 and after
secondary breakup due to shock wave disturbances by Hsiang and Faeth.29~31 Then,
since the universal root-normal distribution function only has two parameters, and
one parameter is fixed because MMD/SMD = 1.2, a single parameter serves to
define the entire drop size distribution for present purposes, this parameter will be
taken to be the SMD.
Measurements of liquid element shapes, sizes, and velocities for turbulent jet
exit conditions and atomization breakup at atmospheric pressure are illustrated in
Fig. 9. These results were obtained from Ruff et al.22 at x/d = 12.5. The range
of positions where the edge of the liquid core was observed is marked on the plot
for reference purposes; the measurements extend from the liquid core to the outer
edge of the multiphase flow region where r/x is in the range 0.15-0.20. Although
these results are for a particular spray, they are typical of available measurements
for the dense spray region of pressure-atomized sprays.21'25
The results illustrated in Fig. 9 show that the region near the liquid surface
consists of large, irregular, ligamentlike elements (large SMD and ep), even though
this spray had good atomization properties, whereas the dilute spray region near
the edge of the flow involves smaller round drops. This provides some evidence
of secondary breakup in the dense spray region near the liquid surface. Additional
evidence will be considered later, however, because effects of turbulent dispersion
of the liquid phase also would be expected to preferentially transport the small
drops to the outer edge of the multiphase mixing layer.
A very important finding of the measurements of Refs. 21-25 was that the
dispersed flow region, exterior to the liquid core, was surprisingly dilute with
mean liquid volume fractions of less than 0.1% (see Ruff et al.23 and Tseng
et al.25); therefore, the large liquid volume fractions observed in Fig. 7 near the
injector exit are mainly due to the presence of the liquid core. The low liquid
volume fractions within the dispersed flow region imply that collisions between
liquid elements are improbable; see Refs. 4-6. Thus, these findings support the
conventional picture of atomization within dense sprays as discussed by Giffen
and Muraszew,1 where sprays form by primary breakup into ligaments and large
drops at the liquid surface, followed by secondary breakup into smaller round
drops, whereas effects of collisions are negligible.
The distributions of drop velocities in Fig. 9 show that they vary considerably
with drop diameter at each point in the flow, providing direct evidence of sig-
nificant separated flow effects in dense sprays. The largest drops have velocities
comparable to mean liquid injection velocities near the liquid core, however, ve-
locities decrease with both decreasing drop size and increasing radial distance.
FULLY-DEVELOPED
FLOW_____
x/d »I2.5
1500 LIQUID SURFACE

(TYP)
-g IOOO
â,
O
2E
10
500
60
20
0.20
Fig. 9 Radial profiles of dispersed-phase properties for a round pressure-atomized

water spray in still air at atmospheric-pressure for atomization breakup with fully
developed turbulent pipe flow at the jet exit xld = 12.5, from Ruff et al.22
This implies relatively ineffective momentum exchange between the phases be-
cause the large drops contain most of the momentum, and they respond slowly
to drag forces due to their relatively large inertia. Finally, the LHF predictions
illustrated in Fig. 9 are poor because separated flow effects are important within
most of the dense spray region.
The final issue considered here involves further examination of the importance
of secondary breakup, as well as the related process of drop deformation, within the
dense-spray region. Criteria for drop deformation and secondary breakup generally
are stated in terms of the drop Weber and Ohnesorge numbers.28"31 Drops formed
by primary breakup in dense sprays are abruptly placed in a low-velocity gaseous
environment; therefore, it is reasonable to evaluate their propensity for secondary
breakup based on the criteria for secondary breakup for shock wave disturbances.
These criteria have recently been discussed by Hsiang and Faeth,29"31 finding
Wepg & 1 for the onset of significant deformation and Wepg ~ 13 for the onset
of secondary breakup, for the low Ohnesorge number conditions of the spray
measurements of Refs. 21-25. Given these results, the potential for secondary
drop breakup in the dispersed flow region of the pressure-atomized sprays was
evaluated by computing mass-weighted spray Weber numbers Weg at various
points in the flow.23 The resulting distributions of mass-weighted spray Weber
numbers demonstrated significant potential for drop deformation and breakup
near the liquid surface for pressure-atomized sprays. The region of significant
deformation and secondary breakup extended across the entire dispersed flow
region at small x/d, but became progressively more constricted (in terms of r/jc)
as xfd increased.23 In summary, the potential for deformation and secondary

breakup agrees with the observations discussed in connection with Fig. 9, with
the progressive contraction of the region where these phenomena are important
(in terms of r/x) illustrating the natural evolution from the dense-spray region
(where irregular liquid elements and drop deformation and secondary breakup are
dominant features) to the dilute spray region (where smaller round drops stable to
deformation and secondary breakup are dominant features).
D. Breakup
The results illustrated in Fig. 9 clearly show the importance of primary and
secondary breakup processes. In particular, primary breakup at the surface of the
liquid core is the source of liquid for the dispersed phase, so that this process
controls the rate of removal of liquid from the liquid core and thus its extent. In
addition, secondary breakup is clearly a natural consequence of primary breakup
in dense sprays and also influences interphase transport because these rates are
strongly coupled to liquid element (or drop) sizes. Processes of secondary breakup
have recently been reviewed in Lee and Faeth,28 Hsiang and Faeth,29~31 and Wu
et al.32 Thus, the following discussion will be limited to the main features of
primary breakup, emphasizing turbulent primary breakup.
Aspects of turbulent primary breakup discussed in the following will emphasize
drop sizes and velocities resulting from turbulent primary breakup. As discussed
earlier, the universal root normal drop size distribution function of Simmons73 was
satisfactory for this process so that drop size properties are completely prescribed
by the SMD alone. In addition, Wu et al.33 showed that drop velocities after tur-
bulent primary breakup approximated the velocities in the liquid phase before
breakup. Thus, mean streamwise drop velocities are nearly equal to mean stream-
wise liquid velocities, or UP/UO & 0.9, whereas mean cross-stream drop velocities
approximate the rms radial velocity fluctuations in the liquid, or VP/VQ ~ 1. The
only exception to this behavior was a small region near the injector exit (x/d < 4)
where UP/UQ progressively increased from 0.4 to 0.9: this behavior was attributed
to the passage wall locally retarding streamwise velocities near the liquid surface
and the jet exit.
The variation of SMD along the liquid surface due to turbulent primary breakup
was initially studied for conditions where aerodynamic effects were small, e.g.,
Pf/Pg > 500; see Wu et al.33 The approach used to interpret and correlate the
measurements was an extension of the method used to find *,-, described earlier.
It was assumed that the SMD was proportional to the largest drop that could be
formed at a particular position, x, that drops formed at the tip of growing ligaments
due to Rayleigh breakup, and that ligament sizes and radial velocities could be
related in the same manner as the length and velocity scales within the inertial
24 L.-K. TSENG ETAL
range of turbulence. This yielded the following expression for the variation of
SMD with distance from the jet exit:
SMD IA = Csx[x/(AWelA)~\* (11)
Available measurements of the variation of SMD with distance from the jet exit
are plotted in Fig. 10 according to the variables in Eq. (11). The experimental
results shown were obtained from Ruff et al.,23 Tseng et al.,25 and Wu et al.33 In
general, these results are limited to p//pg > 500 where aerodynamic effects on
turbulent primary breakup have been shown to be small34; exceptions have been
corrected for aerodynamic effects as discussed by Wu et al.33 The correlation of the
measurements in terms of Eq. (11) is seen to be quite good and can be represented
by the following empirical fit:
SMD/A = 0.65[*/(AHfe}A)]? (12)

Notably, the good agreement between the measurements and the rather complex
power relationships of Eq. (11), coupled with a coefficient of order of magni-
tude unity as anticipated for a parameter of this type, suggests that the physical
principles used to derive Eq. (11) are reasonable.
The results illustrated in Fig. 10 show that drop sizes after turbulent primary
breakup progressively increase with increasing distance from the jet exit. An-
other interesting feature of these results is that the SMD approaches the order of
magnitude of the diameter of the liquid core itself as the end of the liquid core
10' 1
' ' 1 ' ' ' ' 1 ' .... . . .., ^ , .,
x/d
Pl'Pq <*l"»"> ini. 1 4 6 10 25 5O IOO
426 9.5° - - - c - <> - -
2 13 9.5° - - - » - + - -
6.2
3.6
10° 104 62 • - • - • • - rs ^ £v^^ /INTACT
COR
^ «fr-4i i * ^ E
3.6 , ® e W +$ LENGTH
°FROM TSENG •fol. (1992) ^ J
b
FROM WU if al. (1992) C&x, F <D
o C
FROM RUFF »t 01(1991)
*
fsty x/d
A^/A'O ovmmi jnj , 4 6 IO 25 5O IOO
I01 62 3O 6.2 B - - O - - O - 6 ^> -

0 3.6b - - O - 0 $ * -
" ..e&P' ' 957 6.4b B A A - A A fy. -
867 9.5°'C H - - < > 6 - 0 9 0-
b
6.4 o o o o e o o
B^ L THEORETICAL
6.2 « - o - a > e e -
Id2
*r PREDICTION
3.6b n - v - T ^ 7 - _
594 6.4b o - _ - - - - -
i i t i • , i i » t , . ! , , , ,
3
id 10* I0 1
Fig. 10 SMD after turbulent primary breakup as a function of distance from the jet
exit for round liquid jets injected into still gases, after inverting merged secondary
breakup effects, from Wu and Faeth.34
is reached, which is clearly compatible with the liquid column breaking up as a

whole in this region. This behavior is illustrated by comparing the estimates of
liquid core length for turbulent liquids when aerodynamic effects are negligible,
given in Eq. (6) from Grant and Middleman,49 with the measured drop size cor-
relation. In particular, it is interesting to note that specifying the end of the liquid
core when SMD/ A is on the order of unity, and recalling that A/d is a constant
for fully developed turbulent pipe flow, implies from Eq. (1 1)
LB/d = CBWe}d (13)

Clearly, the form of Eq. (13) is identical to the empirical correlation of Grant
and Middleman49 given by Eq. (6), with the powers of We^(0.32 and 0.50) being
reasonably close to each other in view of the difficulties of measuring a property
like the unbroken length of the liquid core.
Another important issue concerning turbulent primary breakup involves the con-
ditions where aerodynamic effects are important for this process. The influence of
aerodynamic effects on breakup regimes has already been discussed in connection
with Fig. 6, whereas a relatively complete discussion of this phenomenon is pro-
vided by Wu and Faeth.34 For conditions where aerodynamic effects influence tur-
bulent primary breakup (pf/pg < 500), the aerodynamic secondary breakup time
of a ligament of characteristic size €,- scale according to £,- ( p f / p g ) i /UQ, whereas
the Rayleigh breakup time of the same ligament is proportional to (p/^/a)2. As
a result, Rayleigh breakup times increase more rapidly than secondary breakup
times as i-L increases. This means that there is a tendency for primary and sec-
ondary breakup to merge as distance from the jet exit increases. These conditions
were analyzed by using Eq. (12) to define initial drop sizes, and then applying
the secondary breakup results of Hsiang and Faeth29'30 to find the final SMD after
merged primary and secondary breakup. The resulting best fit correlation to find
the SMD after merged primary and secondary breakup is as follows:
PgSMDul/a = \2.9(x/A^(Pg/pf^We\JRe}^ (14)
Available measurements of the SMD after merged primary and secondary

breakup are plotted in Fig. 11, according to the variables of Eq. (14). Measure-
ments shown on the plot were obtained from Tseng et al.25 and Wu and Faeth.34
Equation (14) is plotted on the figure as well and is seen to provide an excellent
correlation of the measurements, tending to support the physical ideas used in its
derivation. The slightly improved correlation also shown on the plot can be found
in Wu and Faeth.34
Although some progress has been made concerning the properties of turbulent
primary breakup, several aspects of primary breakup merit additional attention.
First of all, the influence of vorticity at the injector exit for nonturbulent flows
merits study: Wu et al.35 present some limited findings about the influence of
wall boundary layers on nonturbulent primary breakup but much more work is
needed in this area. Finally, liquid removal rates by primary breakup for both
nonturbulent and turbulent jet exit conditions are not known beyond some simple
ad hoc geometrical rules, based on estimates of L#; see Ref. 29 and references
cited therein. More information about primary breakup rates clearly is needed for
a full description of dense-spray properties.
10*
d(mm) D0(m/s)
426 9.5° €4O
2I3 9.5° «40
6.2 <?30 D47
3.6 022 A 38
1 04 6.2 *30 "47
3.6 022 A 38
°FROM TSENG ft oL (I992)
10°
\6* io IOU io1
3/4 .*, 5/6 __ -1/2
) WefA RefA
Fig. 11 SMD after aerodynamically enhanced turbulent primary breakup for round
liquid jets injected into still gases, from Wu and Faeth.34
E. Conclusions
The major conclusions about the structure and primary breakup properties of
the dense-spray region of round pressure-atomized sprays in still gases can be
summarized as follows.
1) The large liquid volume fractions observed in dense sprays are generally
due to the presence of the liquid core; in contrast, liquid volume fractions in the
dispersed flow region beyond the liquid surface are low (less than 0.1 %) so that the
flow in this region corresponds to a dilute spray but with the added complications
of irregular liquid elements and secondary breakup.
2) Measurements generally support the traditional view of pressure atomization
expressed by Giffen and Muraszew,1 among others; namely, primary breakup at
the liquid surface is followed by secondary breakup in a dilute spray environment
where separated flow effects are important, while effects of drop collisions are
negligible.
3) Drop size distributions at various points within dense sprays, as well as
after both primary and secondary breakup, satisfy the universal root normal distri-
bution function of Simmons73 with MMD/SMD =1.2; therefore, the entire drop
size distribution can be conveniently characterized by a single moment, like the
SMD.
4) Extension of the principles used to find the onset of turbulent primary breakup
yielded effective methods for estimating the effect of distance from the jet exit
and the liquid/gas density ratio on drop sizes after turbulent primary breakup.
Both measurements and predictions exhibit increased drop sizes with increasing
distance from the jet exit while the end of the liquid core (for conditions where
aerodynamic effects are small), is associated with conditions where drop sizes
after turbulent primary breakup are comparable to the diameter of the liquid core
itself. Aerodynamic effects on turbulent primary breakup become important for
< 500 and involve the merging of primary and secondary breakup to yield
finer atomization after primary breakup.
Several major issues about dense spray structure and breakup remain to be
investigated, as follows: the rate of liquid removal from the liquid core by either
nonturbulent or turbulent primary breakup must still be resolved, ways to relate
the vorticity at the injector exit and aerodynamic effects to the properties of non-
turbulent primary breakup must still be resolved, and effects at elevated pressures
where liquid viscosity (or large Ohnesorge numbers) modifies drop deformation
and breakup processes must be determined.
Acknowledgments
The authors' research described here was sponsored by the Air Force Office of
Scientific Research, Grants 89-0516 and F49620-92-J-0399, under the technical
management of J. M. Tishkoff; and by the Office of Naval Research, Grants
N00014-89-J-1199 and N00014-95-1-0001, under the technical management of
G. D. Roy and E. P. Rood, respectively. The U.S. Government is authorized to
reproduce and distribute copies for governmental purposes notwithstanding any
copyright notation thereon.
References
îffen, E., and Muraszew, A., Atomisation of Liquid Fuels, Wiley, New York, 1953.
2
Levich, V. G., Physicochemical Hydrodynamics, Prentice-Hall, Englewood Cliffs, NJ,
1962, pp. 639-646.
3
Harrje, D. T., and Reardon, F. H., Liquid Rocket Combustion Instability, NASA SP-194,
1972, pp. 49-55.
4
Faeth, G. M., "Current Status of Droplet and Liquid Combustion," Progress in Energy
and Combustion Science, Vol. 3, No. 2, 1977, pp. 191-224.
5
Faeth, G. M., "Evaporation and Combustion in Sprays," Progress in Energy and Com-
bustion Science, Vol. 9, No. 1, 1983, pp. 1-76.
6
Faeth, G. M., "Mixing, Transport and Combustion in Sprays," Progress in Energy and
Combustion Science, Vol. 13, No. 4, 1987, pp. 293-345.
7
Faeth, G. M., "Structure and Atomization Properties of Dense Turbulent Sprays"
Twenty-Third Symposium (International) on Combustion, Combustion Institute, Pittsburgh,
PA, 1990, pp. 1345-1352.
8
Clift, R., Grace, J. R., and Weber, M. E., Bubbles, Drops and Particles, Academic, New
York, 1978, pp. 26 and 339-347.
9
Lefebvre, A. H., "Airblast Atomization," Progress in Energy and Combustion Science,
Vol. 6, No. 3, 1980, pp. 223-246.
10
Lefebvre, A. H., Gas Turbine Combustion, Hemisphere, New York, 1983.
H
Lefebvre, A. H., Atomization and Sprays, Hemisphere, New York, 1989, pp. 27-78
and 201-272.
12
Elkotb, M. M., "Fuel Atomization for Spray Modelling," Progress in Energy and
13
Drew, D. A., "Mathematical Modeling of Two-Phase Flow," Annual Review of Fluid
Mechanics, Vol. 15, 1983, pp. 261-291.
14
Law, C. K., "Recent Advances in Droplet Vaporization and Combustion," Progress in
Energy and Combustion Science, Vol. 8, No. 3, 1982, pp. 169-199.
l5
Sirignano, W. A., "Fuel Droplet Vaporization and Spray Combustion Theory," Progress
in Energy and Combustion Science, Vol. 9, No. 4, 1983, pp. 291-322.
16
Bracco, F. V., "Structure of High-Speed Full-Cone Sprays," Recent Advances in Gas
Dynamics, edited by C. Casci, Plenum, New York, 1983.
17
Bracco, F. V., "Modeling of Engine Sprays," Society of Automotive Engineers, SAE
Paper 850394, 1985.
18
Wierzba, A., and Takayama, K., "Experimental Investigation of the Aerodynamic
Breakup of Liquid Drops "A/AA Journal, Vol. 26, No. 11, 1988, pp. 1329-1335.
19
Chigier, N. A., "The Physics of Atomization," Proceedings of the Fifth International
Conference on Liquid Atomization and Spray Systems, National Inst. of Standards and
Technology, Washington, DC, 1991, pp. 1-15, NIST SP-813.
20
Annamalai, A., and Ryan, W., "Interactive Processes in Gasification and Combustion.
Part 1: Liquid Drop Arrays and Clouds," Progress in Energy and Combustion Science,
Vol. 18, No. 3, 1992, pp. 221-295.
21
Ruff, G. A., Sagar, A. D., and Faeth, G. M., "Structure of the Near-Injector Region of
Pressure-Atomized Sprays " A/AA Journal, Vol. 27, No. 7, 1989, pp. 901-908.
22
Ruff, G. A., Bernal, L. P., and Faeth, G. M., "Structure of the Near-Injector Region
of Non-Evaporating Pressure-Atomized Sprays," Journal of Propulsion and Power, Vol. 7,
No. 2, 1991, pp. 221-230.
23
Ruff, G. A., Wu, P.-K., Bernal, L. P., and Faeth, G. M., "Continuous- and Dispersed-
Phase Structure of Dense Non-evaporating Pressure-Atomized Sprays," Journal of Propul-
sion and Power, Vol. 8, No. 2, 1992, pp. 280-289.
24
Tseng, L.-K., Ruff, G. A., and Faeth, G. M., "Effects of Gas Density on the Structure
of Liquid Jets in Still Gases," A/AA Journal, Vol. 30, No. 6, 1992, pp. 1537-1544.
25
Tseng, L.-K., Wu, P.-K., and Faeth, G. M., "Dispersed-Phase Structure of Pressure-
Atomized Sprays at Various Gas Densities," Journal of Propulsion and Power, Vol. 8, No.
6, 1992, pp. 1157-1166.
26
Lee, T.-W., Gore, J. P., Faeth, G. M., and Birk, A., "Analysis of Combusting High-
Pressure Monopropellant Sprays," Combustion Science and Technology, Vol. 57, Nos. 4-6,
1988, pp. 1281-1290.
27
Lee, T.-W., Tseng, L.-K., and Faeth, G. M., "Separated-Flow Considerations for
Pressure-Atomized Combusting Monopropellant Sprays," Journal of Propulsion and Power,
Vol. 6, No. 4, 1990, pp. 382-391.
28
Lee, T.-W., and Faeth, G. M., "Structure and Mixing Properties of Combusting Mono-
propellant Sprays," Journal of Propulsion and Power, Vol. 8, No. 2, 1992, pp. 271-279.
29
Hsiang, L.-R, and Faeth, G. M., "Near-Limit Drop Deformation and Secondary
Breakup," International Journal of Multiphase Flow, Vol. 18, No. 5, 1992, pp. 635-652.
30
Hsiang, L.-P, and Faeth, G. M., "Drop Properties After Secondary Breakup," Interna-
tional Journal of Multiphase Flow, Vol. 19, No. 5, 1993, pp. 721-735.
31
Hsiang, L.-P., and Faeth, G. M., "Drop Deformation and Breakup due to Shock Wave
and Steady Disturbances," International Journal of Multiphase Flow, Vol. 21, No. 4, 1995,
pp. 545-560.
32
Wu, P.-K., Ruff, G. A., and Faeth, G. M., "Primary Breakup in Liquid/Gas Mixing
Layers," Atomization and Sprays, Vol. 1, No. 4, 1991, pp. 421-440.
33
Wu, P.-K., Tseng, L.-K., and Faeth, G. M., "Primary Breakup in Gas/Liquid Mixing
Layers for Turbulent Liquids," Atomization and Sprays, Vol. 2, No. 3, 1992, pp. 295-317.
34
Wu, P.-K., and Faeth, G. M., "Aerodynamic Effects on Primary Breakup of Turbulent
Liquids," Atomization and Sprays, Vol. 3, No. 3, 1993, pp. 265-289.
35
Wu, P.-K., Miranda, R. F., and Faeth, G. M., "Effects of Initial Flow Conditions on
Primary Breakup of Nonturbulent and Turbulent Round Liquid Jets," Atomization and
Sprays, Vol. 5, No. 2, 1995, pp. 175-196.
36
Wu, P.-K., Hsiang, L.-R, and Faeth, G. M., "Aerodynamic Effects on Primary and
Secondary Breakup," Liquid Rocket Engine Combustion Instability, edited by V. Yang and
W. Anderson, Progress in Astronautics and Aeronautics, Vol. 169, AIAA, Washington, DC,
1995, pp. 247-279.
37
Sheele, G. F., and Meister, B. J., "Drop Formation at Low Velocities in Liquid-Liquid
Systems," American Institute of Chemical Engineers Journal, Vol. 14, No. 1, 1968, pp.
9-19.
38
Ranz, W. E., "On Sprays and Spraying," Pennsylvania State University, Engineering
Research Bulletin No. 65, University Park, PA, 1956.
39
Ranz, W. E., "Some Experiments on Orifice Sprays," Canadian Journal of Chemical
Engineering, Vol. 36, No. 8, 1958, pp. 175-181.
4()
Ohnesorge, W., "Formation of Drops by Nozzles and the Breakup of Liquid Jets,"
Zeitschriftfur Angewandte Mathematik und Mechanik, Vol. 16, 1936, pp. 355-358.
41
Miesse, C. C., "Correlation of Experimental Data on the Disintegration of Liquid Jets,"
Industrial and Engineering Chemistry, Vol. 47, No. 9, 1955, pp. 1690-1697.
42
Reitz, R. D., "Atomization and Other Breakup Regimes of a Liquid Jet," Ph.D. Disser-
tation No. 1375-T, Princeton Univ., Princeton, NJ, 1978.
43
Reitz, R. D., and Bracco, F. V, "Mechanism of Atomization of a Liquid Jet," Physics
of Fluids, Vol. 25, No. 10, 1982, pp. 1730-1742.
44
Sterling, A. M., and Sleicher, C. A., "The Instability of Capillary Jets," Journal of
Fluid Mechanics, Vol. 68, Pt. 3, 1975, pp. 477^95.
45
De Juhasz, K. J., Zahm, O. F, Jr., and Schweitzer, P. H., "On the Formation and
Dispersion of Oil Sprays," Engineering Experimental Station, Pennsylvania State Univ.,
Bulletin No. 40, University Park, PA, 1932, pp. 63-68.
46
Lee, D. W., and Spencer, R. C., "Preliminary Photomicrographic Studies of Fuel
Sprays," NACA TN 424, 1932; Lee, D. W., and Spencer, R. C., Photomicrographic Studies
of Fuel Sprays, NACA Rept. No. 454, 1933.
47
Hinze, J. O., "Fundamentals of the Hydrodynamic Mechanism of Splitting in Disper-
sion Processes," American Institute of Chemical Engineers Journal, Vol. 1, No. 3, 1955,
pp. 289-295.
48
Phinney, R. E., "The Breakup of a Turbulent Jet in a Gaseous Atmosphere," Journal
of Fluid Mechanics, Vol. 60, Pt. 4, 1973, pp. 689-701.
49
Grant, R. P., and Middleman, S., "Newtonian Jet Stability," AIChE Journal, Vol. 12,
No. 4, 1966, pp. 669-678.
5()
McCarthy, M. J., and Malloy, N. A., "Review of Stability of Liquid Jets and the
Influence of Nozzle Design," Chemical Engineering Journal, Vol. 7, No. 1, 1974, pp. 1-20.
51
Hoyt, J. W., and Taylor, J. J., "Waves on Water Jets," Journal of Fluid Mechanics,
Vol. 88, Pt. 1, 1977, pp. 119-123.
52
Hoyt, J. W., and Taylor, J. J., "Turbulence Structure in a Water Jet Discharging in Air,"
Physics of Fluids, Vol. 20, Pt. II, No. 10, 1977, pp. S253-S257.
53
Hoyt, J. W., and Taylor, J. J., "Effect of Nozzle Boundary Layer on Water Jets Dis-
charging in Air," Jets and Cavities, edited by J. H. Kim, O. Furuya and B. R. Parkin,
ASME-FED, Vol. 31, American Society of Mechanical Engineering, New York, 1985, pp.
93-100.
54
Arai, M., Shimizu, M., and Hiroyasu, H., "Break-Up Length and Spray Angle of High
Speed Jet," Proceedings of the 3rd International Conference on Liquid Atomization and
Spray Systems, 1985, pp. IB/4/1-IB 74/10.
55
Arai, M., Shimizu, M., and Hiroyasu, H., "Break-Up Length and Spray Formation
Mechanisms of a High Speed Liquid Jet," Proceedings of the 4th International Conference
on Liquid Atomization and Spray Systems, 1988, pp. 177-184.
30 L.-K. TSENG ET AL
56
Hiroyasu, H., Aral, M., and Shimizu, M., "Break-Up Length of a Liquid Jet and Internal
Flow in a Nozzle," Proceedings of the 5th International Conference on Liquid Atomization
and Spray Systems, 1991, pp. 275-282.
57
Karasawa, T., Tanaka, M., Abe, K., Shiga, S., and Kurabayashi, T., "Effects of Nozzle
Configuration on the Atomization of a Steady Spray," Atomization and Sprays, Vol. 2, No.
4, 1992, pp. 411-426.
58
Yokota, M., Ito, Y., and Shinoke, T., "High Speed Photographic Observations of Cav-
itation Arising in the High-Speed Oil-Flow Through a Very Small Long Orifice," 9th
International Symposium on Jet Cutting Technology (Sendai, Japan), 1988, pp. 13-21.
59
Ramamurthi, K., and Nandakumar, K., "Disintegration of Liquid Jets from Sharp-
Edged Nozzles," Atomization and Sprays, Vol. 4, No. 5, 1994, pp. 551-564.
60
Mansour, A., and Chigier, N., "Effect of Turbulence on the Stability of Liquid Jets and
Resulting Droplet Size Distributions," Atomization and Sprays, Vol. 4, No. 5, 1994, pp.
583-604.
61
Smith, R. M., and Wang, C.-T. (1944), "Contracting Cones Giving Uniform Throat
Speeds," Journal of Aeronautical Sciences, Vol. 11, No. 4, 1944, pp. 356-360.
62
Taylor, G. I., "Generation of Ripples by Wind Blowing Over a Viscous Liquid,"
The Scientific Papers of Sir Geoffrey Ingram Taylor, edited by G. K. Batchelor, Vol. Ill,
Cambridge Univ. Press, Cambridge, England, 1963, pp. 244-254.
63
Schlichting, H., Boundary Layer Theory, 7th ed., McGraw-Hill, New York, 1979,
p. 599.
64
Hinze, J. O., Turbulence, 2nd ed., McGraw-Hill, New York, 1975, pp. 427 and 724-
742.
65
Tennekes, H., and Lumley, J. L., A First Course in Turbulence, MIT Press, Cambridge,
MA, 1972, pp. 113-124.
66
Weber, C, "Zum Zerfall eines Flussigkeitsstrahles," Zeitschrift fur Angewandte Math-
ematik undMechanik, Vol. 2, 1931, pp. 136-141.
67
Hiroyasu, H., Shimizu, M., and Arai, M., "The Breakup of a High Speed Jet in a
High Pressure Gaseous Atmosphere," Proceedings of the 2nd International Conference on
Liquid Atomization and Spray Systems, Univ. of Wisconsin, Madison, WI, 1982, p. 69.
68
Arai, M., Tabata, M., Hiroyasu, H., and Shimizu, M., "Disintegration Process and
Spray Characterization of Fuel Jet Injected by a Diesel Nozzle," Society for Automotive
Engineers, SAE Paper 840275, 1984.
69
Chehroudi, B., Onuma, Y, Chen, S.-H., and Bracco, F. V, "On the Intact Core of Full
Cone Sprays," Society of Automotive Engineers, SAE Paper 850126, 1985.
70
Andrews, M. J., "The Large-Scale Fragmentation of the Intact Liquid Core of a Spray
Jet," Atomization and Sprays, Vol. 3, No. 1, 1993, pp. 29-54.
71
Ricou, F. P., and Spalding, D. B., "Measurements of Entrainment by Axisymmetrical
Turbulent Jets," Journal of Fluid Mechanics, Vol. 11, Pt. 1, 1961, pp. 21-32.
72
Loth, E., and Faeth, G. M., "Structure of Underexpanded Round Air Jets Submerged
in Water," International Journal of Multiphase Flow, Vol. 15, No. 4, 1989, pp. 589-603.
73
Simmons, H. C., "The Correlation of Drop-Size Distributions in Fuel Nozzle Sprays,"
Journal of Engineering for Power, Vol. 99, No. 3, 1977, pp. 309-319.
74
Belz, M. H., Statistical Methods in the Process Industries, Wiley, New York, 1973,
pp. 103-104.
Chapter 2
Behavior of Droplets in Pressure-Atomized Fuel

Sprays with Coflowing Air Swirl
C. Presser*
National Institute of Standards and Technology, Gaithersburg, Maryland 20899
C. T. Avedisian*
Cornell University, Ithaca, New York 14853
J. T. Hodges*
National Institute of Standards and Technology, Gaithersburg, Maryland 20899
and
A. K. Gupta§
University of Maryland, College Park, Maryland 20742
I. Introduction
S PRAY flames are commonly stabilized by using swirl burners or a variety of

other mechanisms to establish a stable flame by recirculation of combustion
gases and fuel vapors. Swirl burners have attracted considerable interest in recent
years because of their ability to improve combustion efficiency and reduce the
emission of pollutants in the waste disposal and power generation industries. The
aerothermochemistry of swirl-stabilized flames is partly dependent on the degree
of mixing between the fuel droplets and coflowing air stream that, in turn, is
determined by the droplet transport and gas stream turbulence. The importance
of matching the fuel spray pattern with the associated aerodynamic flowfield
has been recognized through the direct application of laser sheet photography.
Studies have shown that droplets pass through the flame sheet relatively unburnt
when commercially available atomizers are used.1 The effect of liquid penetration
through the recirculation zone of swirl-stabilized spray flames leads to changes
in flame structure and stability.2 Information dealing with the effect of the gas
This paper is declared a work of the U.S Government and is not subject to copyright
protection in the United States.
*Group Leader, Chemical Science and Technology Laboratory. Associate Fellow AIAA.
^Professor, Sibley School of Mechanical and Aerospace Engineering. Associate Fellow
AIAA.
^Research Engineer, Chemical Science and Technology Laboratory.
^Professor, Department of Mechanical Engineering. Fellow AIAA.
31
32 C. PRESSER ET AL
flowfield (both mean and fluctuating quantities) on the drag and trajectory of
individual droplets of various sizes is critical for determining the dispersion of
droplets within the surrounding airstream.
Several studies have focused on the effect of fuel spray pattern3 and combustion
air swirl4"9 on the structure of spray flames. Owen10 has carried out spray flame
measurements using laser velocimetry (LV). Larger fuel droplets were found to be
transported along the spray cone with velocities determined initially by the fuel
nozzle characteristics. For droplets with large inertia relative to the turbulence,
it was found that the initial conditions and early droplet displacement history
dictate the dispersion of droplets farther downstream.11 Unlike larger droplets, the
smaller droplets lose their inertia and respond relatively quickly to the ambient
gas velocity. The inertia of the larger droplets enables them to penetrate through
the airstream with little interaction. For the smaller droplets, significant droplet
velocity fluctuations about the mean occur as a result of their interaction with the
airflow.
Large-scale motion of the flowfield plays a key role in the turbulent mixing pro-
cesses, energy release, and pollutants emission.12 The size, strength, and turbulence
levels of the recirculation zone are important for affecting droplet transport.13'15 It
is apparent that increased levels of turbulence enhance droplet/air mixing, and thus
promote near-stoichiometric combustion and flame stability, and reduce pollutant
formation processes.
The effect of turbulence on droplet dispersion can be investigated using phase
Doppler interferometry (PDI)16 in different droplet-laden environments. The effect
of operating parameters of the phase Doppler system on the mean and fluctuating
velocities has been investigated.17'18 In a previous paper by Presser and Gupta,19
droplet size and velocity distributions were obtained using PDI in a nonswirling
nonburning spray to examine shear layer effects near the spray boundary on droplet
transport.
This paper examines droplet transport in a swirling kerosene spray using PDI.
Droplet mean properties (namely, diameter, number density, and velocity) were
measured to characterize the global features of the spray, whereas size and ve-
locity distributions provided a more detailed picture of the behavior of individual
droplets. Also, measured droplet velocity components were used to determine
the direction of droplet motion throughout the spray from which conclusions are
drawn concerning droplet transport. The results are compared with data for a
nonswirling kerosene spray19 for the same conditions of fuel and airflow rates to
assess the influence of swirl on droplet transport.
Data are presented at several locations in the spray. Locations are chosen where
gas recirculation near the nozzle results in significant droplet dispersion (i.e., near
the central region of the spray) and where the droplet concentration is relatively
high (i.e., near the spray boundary). In the latter case, larger size droplets are
transported in a ballistic fashion, as their motion is not influenced strongly by the
surrounding airflow pattern.
II. Experimental Apparatus

Experiments were carried out in a spray combustion facility that has been de-
scribed elsewhere.4'20 The facility includes a swirl burner with a movable 12-vane
swirl cascade (see Fig. 1). The vanes rotate simultaneously to impart the desired
BEHAVIOR OF DROPLETS 33
Fuel Nozzle
Primary Air Passage
Combustion Air Passage

Fig. 1 Schematic of the movable-vane swirl burner.
degree of swirl intensity to the airstream that surrounds the fuel nozzle. The swirl
number S refers to the ratio of the axial fluxes of angular momentum to linear
momentum; the nominal value of S at the burner exit is calculated from theory
developed for guide-vane cascades.21 Experiments were carried out for swirling
(S == 0.53) and nonswirling (S = 0) airstreams under nonburning conditions.19
The value of S = 0.53 was chosen because of the good flame stability that is pro-
vided under burning conditions. A propane-fueled ring, located at approximately
300 mm downstream of the nozzle exit, was used as an afterburner to burn the
droplets before they were exhausted to the atmosphere.
A simplex pressure-jet fuel nozzle was located along the centerline of the burner
(at the burner exit). The fuel nozzle was operated at total air and fuel flow rates of
64.3 kg/h and 3.2 kg/h, respectively; these flow rates provided an input equivalence
ratio of approximately 0.7. The nozzle generates a nominal 60-deg (full-angle)
hollow-cone kerosene spray under unconfmed conditions that is injected vertically
upwards from the nozzle. A schematic of the hollow-cone spray, which includes
the measurement grid, is illustrated in Fig. 2. In this study, measurements were
obtained at axial positions of 10, 25.4, 50.8, and 76.2 mm. For each axial station,
data were recorded in increments of 2.54 mm across the profile (at z = 10 mm the
increment was 1.27 mm). Also shown are coordinates for several specific locations
(designated L1-L5) where results are discussed in more detail. Locations LI, L2,
and L4 are representative of the central region of the spray where recirculation
34 C. PRESSER ET AL.
spray
nozzle
air passage
units: mm
nominal spray
boundary
62 10.2 20.3
Fig. 2 Schematic of the spray boundary position; the numbers identify particular
locations where the flow parameters are discussed in detail.
of the surrounding gases and droplet entrainment is expected to be significant.

Locations L3 and L5 correspond to the spray boundary.
A stepper-motor-driven, three-dimensional traversing arrangement translates
the burner assembly in the vertical and horizontal directions (see Fig. 3). All
optical diagnostics are fixed in position about the burner assembly, and the burner
translates independently of the optical equipment. Enough symmetry is assumed
to exist about the spray axis so that measurements of spray properties in any
particular plane containing that axis are considered representative of the entire
spray. Measurements are reported in the plane in which the burner is traversed (X
in Fig. 3). Additional details can be found elsewhere.4-20
The air flowfield was determined in a previous study,4 using a single-component
laser velocimetry system. Under swirling conditions, the mean axial air velocity
is of the Rankine vortex type across the coflowing air passage at the burner exit
BS - Beam stop
Traverse _
Assembly PR - PDI receiving optics
PT - PDI transmitting optics
M -Mirror
Micro- PDI
Computer Processor
Fig. 3 Schematic of the experimental droplet size/velocity facility.
(see Fig. 4 of Ref. 4), and has a maximum value of approximately 2.3 m/s. The
air velocities decay with increasing axial position. The tangential component of
air velocity was also of the Rankine vortex type for these experimental conditions
(the radial velocity component was negligible).
A two-channel phase Doppler interferometer16 was used to obtain droplet size
and velocity information on individual droplets passing through the measurement
volume (see Fig. 3). The measurement technique employed is similar to a con-
ventional dual-beam laser velocimeter, except that four detectors are located in
the receiver assembly.22 Measurement of the temporal frequency of scattered light
from the probe volume interference fringe pattern is used to determine droplet
velocity. Measurement of the spatial frequency of scattered light, obtained from
two separated detectors, is used to infer the phase difference and to provide in-
formation on particle size. Mean properties were based on the statistical analysis
of the ensemble sizes and velocities. The influence of droplet trajectory through
the probe volume on inferred droplet size can lead to the measured droplet size
distribution being broader than the actual size distribution within the spray.23 The
sensitivity of measured diameter to droplet trajectory is minimized by ensuring
that the characteristic width of the probe volume is much greater than the parti-
cle diameter, and by measuring the scattered light intensity in the near-forward
direction.24 In the present setup, the scattering measurements were made in the
near-forward direction (30-deg scattering angle) and the minimum probe volume
width was on the order of 100 /zm.
The phase Doppler system provides information on particle size between 1 /zm
and 300 /xm with the optical arrangement employed in this investigation. The
measurement volume is defined by the 119- and 113-^m laser beam waists at
514.5 and 488.0 nm, respectively (with a fringe spacing of approximately 6.6 /zm
for both laser beams). The focal lengths of the transmitting and receiving optics
were 495 and 500 mm, respectively. The focal length of the collimating lens was
300 mm. The receiving optics has a 100-/xm slit. For the experiments carried
out in this investigation, the photomultiplier detector voltages were optimized to
provide the greatest sensitivity to the wide range of droplet sizes typically found
in these sprays.18 Near the spray boundary, which is defined as the location where
the volume flux is largest in the swirling case, data acquisition rates were high
(10,000 data points were collected within 120 s). In other regions of the spray,
however, considerably more time was required to collect the same number of data
points.
Near the spray boundary, the required system gain was significantly lower than
near the spray centerline for the swirling case. The percent validation (i.e., percent
difference between detected and validated signals) for the swirling case were at best
as high as 75% and as low as 50%. For the nonswirling case, the percent validation
reached a maximum of 80% and a minimum of 50%. The percent validation was
not correlated to the spatial position. The effect of percent validation on number
density and mass flux is discussed in more detail elsewhere.4'5 Measurements were
repeated at several selected positions to ensure measurement repeatability that was
generally better than 5% near the spray boundary.
III. Results and Discussion

A. Droplet Mean Properties
Droplet mean properties (i.e., size, number density, and velocity) are presented
to describe the global features of the kerosene spray. The effects of swirl on
droplet transport are emphasized by comparing the spray under swirling conditions
(S = 0.53) to the nonswirling case (S = 0). The effect of combustion on droplet
transport is presented elsewhere.25 The results for the nonswirling case were taken
from an earlier investigation.19 For a hollow-cone spray, it is reasonable to define
the location of the spray boundary at the radial position where most of the fuel
mass resides. This is accomplished by presenting the droplet volume flux, which
is defined as the volume of the droplets passing a unit cross-sectional area per
unit of time.26 Thus, radial profiles of the volume flux are presented in Fig. 4 at
axial locations z of 10, 25.4, 50.8, and 76.2 mm downstream of the nozzle exit.
The solid boxes along the abscissa indicate the position of the burner passage
walls, with the fuel nozzle located at the axis of the burner (r = 0). Coflowing
air is introduced through the outer coaxial passage. The profiles in Fig. 4 (for
both the nonswirling and swirling cases) indicate that radial positions exist at each
axial position where the volume flux exhibits a maximum or peak and thus are
used to define the spray boundary. The results show that the presence of swirl
redistributes the droplets to much larger radial positions (by a factor of 2-3 at
z — 10 mm) and the peaks are farther apart than in the nonswirling spray. The
nonswirling case is indicative of the fuel mass distribution that results from the
nozzle design. At z = 10 mm, the peak values for the nonswirling case are an order
of magnitude higher than the swirling case (about 2 x 10"1 cm3 s"1 cm"2 for S = 0
and 2 x 10~2 cm3s~1cm"2 for S = 0.53). At z = 76.2 mm, the magnitude of both
the nonswirling and swirling cases are of similar order (5 x 10~~3 cm3s~1cm~2 for
S = 0 and 8 x 10~3 cm3s~!cm~2 for S = 0.53). This result indicates that swirl
promotes spray uniformity closer to the nozzle than in the nonswirling case where
shear-induced drag disperses the droplets farther downstream.
-60 -40 -20 0 20 40 60

I I i I i i i I i i i I i i i I i i i I i i t a jnQ
rUT1?
^10'2>
eêeeeo€ -io-'f
4i
L
-io-6 i
10°- Z = 50.8 mm io- J
los
= 2 5 . 4 mm
lO'-i Z = 1 0 mm
-10
io-'4
-r ios
10 n"i i i i i i i ~T i i i i i T i i i i i i i
-60 -40 -20 0 20 40 60
(mm)
Fig. 4 Variation of volume flux with radial position r measured at different axial
positions z for the nonswirling and swirling sprays.
-60 -40 -20 0 20 40

i i i I i i i I i i i I i i t I i i i I
i I i | I l l |T I I | I I r | I I I I I
-60 -40 -20 0 20 40

r (mm)
Fig. 5 Variation of Sauter mean diameter D32 with radial position r measured at
different axial positions z for the nonswirling and swirling sprays.
Radial profiles of the droplet Sauter mean diameter D^ at the same four axial
locations are presented in Fig. 5. The vertical bars identify the nominal spray
boundary, which is defined as the radial location (at a given axial position) where
the volume flux exhibits a peak (see Fig. 4). At each axial position, the spray
boundary for the swirling case is located at a larger radial position than for the
nonswirling case. The spray boundary also becomes ambiguous with increasing
axial distance because of droplet dispersion. These peaks in Sauter mean diameter
appear to match the peaks in volume flux for the swirling case at each axial
position. For the nonswirling case, the peaks for Dî and volume flux are not
coincident at z — 50.8 mm and 76.2 mm because the droplet mean size increases
monotonically with increasing radial position. Note again that the radial position
of the peaks in volume flux are different for both cases, with the peaks for S = 0
being closer to the spray centerline than for S = 0.53. This difference is attributed
to the centrifugal forces (present in the swirling case) displacing the bulk of the
droplets radially outward from the spray centerline.
The profiles of Sauter mean diameter and volume flux for both cases indicate
the expected broadening of the spray with increasing axial distance. The presence
of relatively small droplets near the centerline and large ones near the spray
boundary is attributed to the characteristic design of hollow-cone spray nozzles.

Swirl, resulting in the formation of a toroidal recirculation region, increases the
droplet radial dispersion (i.e., droplets were detected at larger radial positions)
well beyond the radial position of the spray boundary. Without swirl, the values of
D32 tend to increase monotonically with increasing radial position toward the edge
of the spray at downstream axial positions. This result is attributed to the nozzle
design in which the droplets are preferentially dispersed with the smaller droplets
generally detected toward the spray centerline and larger droplets found at the
largest radial positions. For the nonswirling case, there is also a gradual increase
in the value of D32 along the spray boundary (from about 45 /xm at z = 10 mm
to 71 /im at z = 76.2 mm); see Fig. 5. This variation is attibuted to the shear
layer mixing that occurs between the spray and the surrounding airstream, which
results in radial dispersion of the smaller droplets. Another possible parameter to
consider is some shear-enhanced vaporization of the smaller droplets.
Note also that the profile for the nonswirling spray at z = 10 mm exhibits a
minimum in the value of Z)32 within the spray boundary. This is attributed to
the fact that droplets at the periphery of the spray boundary (i.e., inner and outer
surface of the hollow-cone spray structure) are in more direct contact with the
surrounding airstream than those droplets within the spray boundary. The shear
experienced at the spray periphery (both inner and outer surfaces) has a larger
influence on the smaller droplets and disperses these droplets away from the
boundary. Consequently, smaller droplets are less abundant at the spray periphery,
which results in the formation of slight peaks for the droplet mean size.
The results presented in Fig. 5 indicate that the swirling spray has greater spatial
uniformity and enhanced droplet mixing than for S = 0. As mentioned earlier, the
nozzle is designed to disperse the larger droplets along the spray boundary and the
smaller droplets near the spray centerline, which is apparent from the nonswirling
spray results. Thus, in the swirling case, enhanced mixing would result in larger
mean sizes near the center of the spray and smaller sizes toward the periphery.
The weak radial variation of D32 for S = 0.53 (see Fig. 5) occurred at all of the
measured axial positions, emphasizing the penetrating influence of swirl that is
experienced by the surrounding airstream.
The variation of droplet mean size with axial position along the centerline
and spray boundary is shown in Figs. 6a and 6b, respectively. Compared to the
swirling case, the droplet mean diameter is smaller near the spray centerline (see
Fig. 6a) and larger near the spray boundary (see Fig. 6b) in the nonswirling
case. This feature results from the large variation in the droplet mean diameter
for the nonswirling spray and the relatively small variation in mean diameter
(greater spatial uniformity) for the swirling spray. The gradual increase in the
values of D32 along the spray boundary (from approximately 31 /xm at z = 10
mm to 48 /im at z = 76.2 mm for S = 0.53, see Fig. 6b) is again attributed to the
40 C. PRESSER ET AL
TO 20 30 40 50 60 70 80
50- -50
Centerline
40—
CL> -••5 = 0
•O- S = 0.63
"E 3 ...---o —30
°- .. o- .
o 20— —20
to- h-10
CO
10 20 30 40 50 60 70 80
z (mm)
10 20 30 40 50 60 70 80
100-
i i I i i -100
Boundary
90— —90
80— —80
70— —70
60— —60
50— -O —50
o 40— 10
CD
30— —30
20— —20
13
O 10— —10
CO
0 0
10 20 30 40 50 60 70 80
z (mm)
Fig. 6 Variation of Sauter mean diameter Z>32 with axial position along a) the
centerline and b) the nominal spray boundary for the nonswirling and swirling
sprays.
entrainment and transport of smaller droplets by the swirling flowfield; and the
associated shear layer mixing that occurs between the spray and the surrounding
airstream. This mixing promotes the depletion of smaller droplets in the spray
boundary by means of such possible mechanisms as coalescence between droplets,
droplet vaporization (both of which are expected to be insignificant due to the low
probability of collisions between small droplets and the absence of combustion in
this study), and radial dispersion away from the spray boundary. It is recognized,
however, that the data obtained in this study cannot identify with certainty the
precise mechanisms. The variation of droplet size with increasing axial and radial
position is even more pronounced for S = 0 in which the values of D^ increase
from about 45 /xm at z = 10 mm to 71 /zm at z = 76.2 mm (see Fig. 6b).
Near the spray centerline, there is little change in the value of D32 with increasing
axial position for both cases (see Fig. 6a). For example, for S = 0 the change in the
value of D32 is from 12 /xm to 17 /zm, and for 5 = 0.53 the change is from 24 /zm
to 29 /jim. An increase in droplet mean size may be expected for the same reasons
that there is an increase near the spray boundary. Elucidation of this transport
phenomenon is made by examining the arithmetic mean diameter (/)io), which
has a greater sensitivity to the smaller size droplets. The value of D\Q increases
slightly near the centerline for S = 0.53 from approximately 8 /xm at z = 10 mm
to 12 /xm at z = 76.2 mm. This small variation in the values of D\Q with axial
position along the centerline is also small when S = 0. This small change may be
explained by the lack of radial droplet dispersion in the nonswirling case, but would
not describe the centrifugal effects on droplet transport along the spray centerline in
the swirling spray. The small change in D\Q for the swirling case may be attributed
to the inward transport of smaller droplets by the toroidal recirculation zone, which
replenishes the central region of the spray with smaller droplets from the spray
periphery.
Results for droplet number density N for both the nonswirling and swirling
sprays are presented in Fig. 7 for z = 10, 25.4, 50.8, and 76.2 mm downstream
of the nozzle exit. The data indicate an increase in radial spread of the spray with
axial position for both the swirling and nonswirling cases, with the swirling spray
extending to much larger radial positions as compared to the nonswirling case. In
the swirling spray, the maximum value of N occurred near the spray boundary
where the larger droplets exist. In contrast, the maximum value of N for the
nonswirling case is in the center of the spray despite the hollow-cone nature of the
fuel nozzle. These profiles are indicative of the larger concentrations of relatively
small droplets that remain confined to the center of the spray. The results for
volume flux (see Fig. 4), however, indicate that the bulk of the mass does not
reside near the spray centerline for S = 0. Based on the number density at z = 10
mm, there is no indication of the hollow-cone nature of the spray for which one
would expect to find peaks in the values of N toward the spray boundary (see
z = 10 mm in Fig. 8 of Ref. 4). This result may be attributed to the difficulty
in detecting relatively small droplets (in the presence of larger ones) with the
phase Doppler system because of the limited instrument dynamic range of 35:1;
the spatial resolution was estimated to be sufficient because of the small probe
volume size employed in this study.
The variation in number density with axial position is presented in Fig. 8 at both
the centerline and spray boundary. The number density decreases as axial position
increases for both cases. The values of N are larger for the nonswirling spray near
the centerline but are significantly smaller toward the spray boundary. Near the
centerline, the number density decreased slightly for S = 0 from N &9 x 103
particles/cm3 at z — 10 mm to N & 2 x 103 particles/cm3 at z = 76.2 mm. For
S = 0.53, number density changes from N ^ 4 x 103 particles/cm3 atz = 10 mm
to/V ^ 1 x 103 particles/cm3 atz = 76.2 mm. Near the spray boundary (as defined
by the peaks in volume flux from Fig. 4, note vertical bars), the number density
is similar to that near the centerline for S = 0.53 (N « 5 x 103 particles/cm3 at
z = 10 mm to N ^ 1 x 103 particles/cm3 atz == 76.2mm), but significantly lower
for S = 0 at downstream positions (N « 5 x 103 particles/cm3 at i — 10 mm
to N ^ 6 x 101 particles/cm3 at z = 76.2 mm). These differences between the
nonswirling and swirling cases are attributed to the presence of significant air
-60 -20 0 20
i i I i I i i i
0 20 40 60
r (mm)
Fig. 7 Variation of number density N with radial position r measured at different
axial positions z for the nonswirling and swirling sprays.
swirl in the latter case that promotes the radial dispersion and spatial uniformity
of the fuel droplets.
The maximum values of N are significantly smaller than those measured with
the polarization ratio technique4 that is more sensitive to smaller size droplets. For
these measurements, number densities were as high as N & 1 x 106 particles/cm3
in the upstream portion of the spray. This difference may be attributed to the lack
of sensitivity of the phase Doppler system to detect submicron droplets, as well
as other factors that lead to rejection of data (e.g., presence of multiple droplets
10 20 30 40 50 60 70 80
io e -d i i I
Centerline
i
e
^
o 1
' Un -*- S = 0 -10s
-0- S = 0.53
104- -10"
•f 103- -10J
102 -10 2
<o 20 30 40 50 60 70 80
2 (mm)
10 20 30 40 50 60 70 80
i
104-d
i \ i I I
Ho 4
103- -105
2
tosi ^ r io
O>
f -10'
10 C •10 C
10 20 30 40 50 60 70 80
z (mm)
Fig. 8 Variation of number density W with axial position along a) the centerline and
b) the nominal spray boundary for the nonswirling and swirling sprays.
in the measurement volume). Also note that for 5 = 0, the bulk of the mass does
not reside in the center of the spray where transport of the smaller droplets occurs
(see the volume flux in Fig. 4). The volume flux is larger near the spray boundary
since it is proportional to (D^ despite the smaller values of N. For S = 0.53
the mass resides near the spray boundary where both the size and number density
are higher.
The spatial profiles of droplet mean axial u and radial v velocity components at
z — 10, 25.4, 50.8, and 76.2 mm are shown in Figs. 9 and 10, respectively. Results
for the swirling spray are again compared with those for the nonswirling spray.19
Positive values of u correspond to droplets moving in the downstream direction
-60 -40 -20 0 20 40 60

-=-20
-60 -40 -20 0 20 40

r (mm)
Fig. 9 Variation of axial velocity u with radial position r measured at different axial
positions z for the nonswirling and swirling sprays.
(away from the atomizer) and positive values of v signify radially outward motion
(that is, droplet motion away from the spray centerline). The results in Fig. 9 show
that the magnitude of the droplet mean axial velocity decays with increasing axial
position for both the nonswirling and swirling sprays. In contrast to the nonswirling
spray, droplets in the swirling spray have negative axial velocities (and negligible
radial velocities) at positions near the axis of the burner (see Fig. 9). These data
indicate that droplet transport is upstream toward the fuel nozzle for the swirling
spray.
60 -40 -20 0 20 40 60
.1 i i I i i i I i i i I i i » I i i i I » i i -20
2 = 76.2 mm =
HO"
20- ^5
15- Z = 50.8 mm
r-20
2 = 25.4 mm
H0~
20-
S = 0
S = 0,53
Z= 1 0 mm
10-E
-fr i i i i i i i i ~T i i i i i i i I I I I I i i
-60 -40 -20 0 20 40 60

r (mm)
Fig. 10 Variation of radial velocity v with radial position r measured at different
axial positions z for the nonswirling and swirling sprays.
The axial and radial velocity components of the droplets for the swirling and
nonswirling sprays exhibit maxima near the spray boundary (see Figs. 9 and
10), with the mean flow corresponding to a direction that is consistent with
the nominal spray cone half-angle (a in Fig. 11). When compared to the non-
swirling case, the values of u and v are smaller for the swirling case since the
centrifugal effects and droplet recirculation tend to diminish mean axial momen-
tum and enhance spatial uniformity. For example, in the nonswirling case, given
the peak values of u « 13 m/s and v « 6 m/s at z = 10 mm (as shown in Figs.
46 C. PRESSER ET AL
spray centerline
nominal spray
boundary
Fig. 11 Coordinate system used in the definition of droplet trajectory angle 9:

coordinate a is replaced by b at the radial location where the magnitude of radial
velocity component is a minimum (A = 0 for a perfectly symmetric spray).
9 and 10), the mean flow direction 0 of the droplets (see Fig. 11) is approxi-
mately 33 deg whereas the nominal spray angle a, (as specified by the nozzle
manufacturer) is about 30 deg. Near the spray boundary, the values of u for the
nonswirling spray are at least a factor of two higher than those of the swirling
spray (see Figs. 9 and 10); the values of v are similar near the spray center-
line. Assuming that the droplet velocities at the nozzle are the same for both
cases, this result suggests that the presence of swirl increases significantly the
mean rate of droplet deceleration in the axial direction and increases the radial
spread.
At z = 10 mm the respective rms values of the velocity components for both
the nonswirling and swirling sprays indicate that the velocity distributions are
narrower toward the spray centerline (i.e., smaller values of rms velocity) and
wider toward the spray boundary (i.e., larger rms velocities) where the mean
values reach a maximum. The spatial profiles of mean and rms velocity are similar
at downstream positions of the spray, i.e., for z > 50.8 mm. To summarize, near
the center of the spray, the droplet sizes are smaller and have a narrower velocity
distribution (for both the axial and radial components). Toward the spray boundary,
both the size and velocity distributions are wider.
B. Direction of Droplet Motion

The entrainment of droplets by the surrounding gaseous medium and secondary
breakup of droplets in a spray are complicated phenomena involving aerodynamic
drag forces and the creation of waves at the surface of droplets. Somewhat simplis-
tic measures of these phenomena are provided by the Weber number (=V^pD/cr)
for droplet breakup, the Stokes number (=piD2VTQ\/18/x5) for dispersion and en-
trainment, and the Reynolds number (=VTe\Dp/^) for drag and internal circulation
within droplets. In the definition of these parameters, VK\ is the relative velocity
between the droplet and surrounding gas, p the gas density, p\ the liquid density,
D the droplet diameter, a the droplet surface tension, /z the gas phase viscosity,
and 8 a length scale characteristic of a vortex or eddy in the coflowing swirling
gas stream. Two practical difficulties arise when evaluating these parameters for
sprays. First, the droplet velocity relative to the surrounding gas phase is not easily
obtained, and, second, the characteristic length scale for the recirculating vortex in
the definition of the Stokes number is essentially unknown. Therefore, it would be
inappropriate in this study to evaluate the Weber, Reynolds, and Stokes numbers.
Information related to droplet transport, however, can still be obtained indirectly
from the droplet velocity components.
The droplet velocity components were used to obtain a droplet mean trajectory
angle 0, which is defined in Fig. 11. The symbol A is a measure of the asymmetry
of the spray. The spray centerline is therefore defined to correspond to the radial
position (at any axial location) where the radial velocity component is a minimum
(see convention, defined earlier, for the velocity components). When A = 0, the
spray is symmetric about the axis of the atomizer. The value of A is determined
at any axial location by first identifying the radial location at which the radial ve-
locity component reaches a minimum. The symbol £ is referenced to that location
and f = 0 is the centerline. For £ > 0, coordinate system b is used, and for £ < 0
coordinate system a is used. With this choice of an axisymmetric coordinate sys-
tem and sign convention for the velocity components mentioned earlier, negative
values of 6 will correspond to velocity vectors that are directed inward toward
the centerline. For the conditions of the spray with swirl, the droplet tangential
velocity component decays rapidly and becomes negligible within a few millime-
ters downstream of the nozzle.4 Therefore, the u and v velocity components alone
determine the direction of droplet motion in the two-dimensional (z-r) plane,
where the droplet mean trajectory angle is 9 = tan"1 ( v / u ) .
Variation of the droplet mean angle is presented in Fig. 12 at the various
aforementioned locations within both the nonswirling and swirling sprays. In
the outer radial regions of the spray, there is little difference in the direction of
droplet motion between the two sprays at downstream positions (i.e., z > 25.4
mm), with the mean direction of droplet motion being in the nominal direction of
the spray cone half-angle, which is about 30 deg. At i = 10 mm, however, there
is a departure between the two cases. For 5 = 0, the spray cone angle (i.e., the
trajectory angle near the edge of the spray) is preserved, namely, the mean droplet
angle remains at approximately 30 deg. For S = 0.53, recirculated droplets are
transported radially outward into the surrounding airstream and thus results in an
increase in the mean angle (namely, the mean direction of motion is altered when
considering the presence of the recirculated droplets in addition to the droplets
that emanate from the nozzle). Near the spray centerline, droplets in the swirling
-60 -40 -20 0 20 40 60

I I I I I I I I I
-90 ^-
— 180
50.8 mm
9(H
(H
H80 _
_ OT»
= 25.4 mm
-90
—
3.
o>
-o f
— cr>
o>
-90 I
_ 180-: Z = 10 mm
h-180 ^
CT> _
s o>
90- —
f o-
cr» _
o ~
1 -90-
1
-180
-60 -40 -20 0 20 40 60
r (mm)
Fig. 12 Variation of droplet mean trajectory angle 6 (see Fig. 10) with radial position
r at different axial positions z for the nonswirling and swirling sprays.
spray move upstream in a direction opposite to those in the nonswirling spray (as
a result of significant gas phase recirculation). For example, in the nonswirling
spray 0 & 0 at the axis of the burner, whereas for the swirling spray 9 ~ 180 deg.
Reasons for these differences in mean direction near the center of the spray are
explored further by examining the size-classified distributions of droplet angle.
Distributions of droplet angle are presented in Figs. 13-15 at positions LI
(r = 0 mm), L2 (r = 6.4 mm), and L3 (r = 10.2 mm) from the centerline of the
spray (A = 0) to the spray boundary at z = 10 mm for both the nonswirling and
swirling sprays. Since there is a slight asymmetry to the spray (e.g., see z = 10 mm
-180 -go 90 180
U.UJ - 0.03
Da = 18.9 jim
L'r =o r U = 6.385 m/s
i = 10,0 mm V = 0,281 m/s
.^ 0.02 - 4 = -2.5 deg -0,02
15 _
o
f~i
S o.oi - -0.01
J
Q_
n nn - I
0,00
-180 -90 0 90 180
Ang I e (deg)
-180 -90 90 180

U.U* -
OB = 45.4 pm
^ r = 6.4 mm U = 13.098 m/s
0.03 - 2 = 10,0 mm V = 4,972 m/s -0.03
* = 20 r 7 deg
o 0.02 - : 0,02
_0
o
v_
__ i__
Q_ O.Of - -0,01
0.00:
1_ L
0,00
80 -90 0 90 U JO
A n g le ( d e g )
-1 80 -90 0 90 U JO
0.10 - -0,10
Da = 41.8 urn
0.08 -j
^ r = I0r2 mm U = 7,606 m/s -0,08
z = 10,0 mm n V = 5,812 m/s
0.06 -j
* = 37.3 deg ;0,06
O
jQ 0.04 -m J-0.04
O
Q.
0.02 -m -0.02
n nn - _T T
1 n nn
-180 -90 0 90 180
Ang le (deg)
S = O
Fig. 13 Probability distributions (L1-L3) for droplet trajectory angles at three dif-
ferent locations in a nonswirling and swirling spray for all droplet sizes. (Continued)
in Figs. 9 and 10), location LI is positioned at r = 0 mm, A = 2.54 mm, and £ < 0
(i.e., £ = —2.54 mm). The results presented in Fig. 13 show the distributions for
all size classes, whereas the data in Figs. 14 and 15 are for specified size ranges:
1 fjim < D < 5 /xm (hereafter referred to as 3-//,m-size droplets) for Fig. 14, and
48 /xm < D < 52 /xm (referred to as 50-/xm-size droplets) for Fig. 15.
Differences between nonswirling and swirling conditions are evident from
Fig. 13. In the swirling case, there is a wide distribution of angles (i.e., larger
variance) at locations LI and L2. This distribution becomes narrower (i.e., smaller
variance) toward the spray boundary at L3. The wide distributions at locations LI
50 C. PRESSERETAL
-180 -90 0 90 180

0.03 0.03
Dj, = 24.0 /im
U = -1.88 m/s
V = -0.142 m/s
-^ 0.02 - = 175.6 deg -0.02
O.Ot - n -0.01
o.oo 0.00
-180 -90 0 90
A n g l e (deg)
-180 -90 90 180

u.uz u.uz
DM = 23 f 8 /im
r = 6.4 mm U = 1.851 m/s
2 = 10.0 mm n V = 0.93 m/s
i) = 26,6 deg
a O.Of - -0.01
^0
o
al 1 J--
[Th^ r», „ , nlffl,

-180 -90 0
IJlrnT^^
90
1 n nn
180
A n g l e (deg)
-180 -90 90 180

u. i u - - u. i u
Oj, = 31.0 jim
0.08 -
^ r = 10.2 mm U = 7.359 m/s -0.08
2 = 10.0 mm V = 5.228 m/s
0.06 -
1) = 35.3 deg -0.06
o
o 0.04 - 70. 04
j
0_
0.02 - 70.02
n nn - NhT-K,_ 1 n nn
-180 -90 0 90 180
A n g l e (d eg)
S = 0.53
Fig. 13 (Continued).
and L2 for the swirling spray indicate the presence of two counter flowing streams,
namely, droplets emanating directly from the nozzle (angle = ±90 deg) and re-
circulated droplets (±90 deg < angle < ±180 deg). In the nonswirling spray, all
droplets are moving downstream away from the nozzle at each location, whereas
in the swirling spray at LI, and to a lesser extent at L2, most droplets are moving
upstream toward the nozzle. The fact that droplets tend to move in a ballistic fash-
ion for the nonswirling spray has also been confirmed by laser sheet visualization
of droplet trajectories in similar sprays.4-27
The size-classified angle distributions shown in Figs. 14 and 15 provide a
more detailed depiction of individual droplet transport. Figure 14 shows that
whereas the 3-^m-class droplets are generally moving in the downstream direction
-180 -90 90 180
U.UJ - - u.u j
•j n 1 * 0 * 5 /im
U = 6.1 m/s
z = 10,0 mm V = 0,5 m/s
0,02 - tJ = -5. deg -0,02
-g
o
0.01 -
1 -0.01
n nn -
r I - n nn
-180 -90 0 90 180

An g le (deg )
-180 -90 90 180

u,u^ - - u,u^
^ r = 6.4 mm U = 13.3 m/s

>-N
0.03 - z = 10,0 mm V = 6 m/s -0,03
* = 24 r 2 deg
lo 0.02 - r -0,02
0
-a
0
ol 0.01 - -0,01
n nn - - n nn
90
An g le ( d e g)
-180 -90 90 180

u. i u - - u, i u
o.oa - ^ r = 10,2 mm U = 6,1 m/s -0,08

z = 10,0 mm V = 4,4 m/s
0.06 -
1) = 35.5 deg ;0,06
o
L
-JO
o 0.04 - 0,04
0.02 - 70.02
n nn - - n nn
-180 -90 x 90 180
An g I e (deg)
S = O
ferent locations in a nonswirling and swirling spray for droplets in the size range of
1-5 p,m. (Continued)
along the centerline (see frame LI) for the nonswirling spray (9 = —5.0 deg), the
range of directions is comparatively narrow. By contrast, in the swirling spray
a significant number of 3-/zm-size droplets are moving upstream at locations
LI and L2. This trend reflects the effect of recirculating gas phase motion on
droplet relative velocity and drag. At location L3 (close to the spray boundary)
for the swirling case, the few 3-/xm-size droplets that were detected moved in
the downstream direction. For the nonswirling case, 3-/>tm-size droplets were not
detected at location L3 since the smaller droplets tend to be concentrated at the
spray centerline for hollow-cone spray nozzles.
-180 -90 0 90 180
u.uj - -
1Ll1 1 * D $ 5 pm
r =0 U = -1.7 m/s
z = 10,0 mm V = 0 m/s
0,02 - * = -178.9 deg -0.02
0
0.01 - -0.01
CL. I
n nn _ . Hm rfHTr^nrflnrJTInrTHr^rrUlUI - n nn
-180 -90 0 90 180

Angl e (deg)
-180 -90 90 180

u,u/ -
^ r = 6.4 mm U = 0.5 m/s

z = 10.0 mm V = 0.7 m/s
iJ = 54,6 deg
0.01 - -0.01
" -
n nn - 1 HI h D-i film m n r] HI
-180 -90 0
Ilif^ yif
90 180
I n nn
Angl e (deg)
-180 -90 90 180

U. ! U - - u. i u
0.08 -|
" r = 10.2 mm U = 3.6 m/s ^0.08
=>^ z = 10.0 mm V = 4.1 m/s
0.06 -^
iJ = 48,4 deg
0
_o 0.04 -| -0.04
0
„ Ih rnfl
Q_
0.02 -m -0.02
n nn - 1 n nn
-180 -90 0 90 180
Angl e (deg)
S = 0.53
It is expected that smaller droplets will have a higher propensity to be car-

ried along with the surrounding flowfield than the larger droplets that are rela-
tively unaffected by the gas phase flow pattern. As shown in Fig. 15, 50-ju,m-size
droplets in the nonswirling spray have a relatively narrow distribution of angles
with only a few droplets deviating from the mean direction of droplet motion.
Closer to the spray boundary (at locations L2 and L3) the droplet mean trajec-
tory angle increases, from 20.5 deg at r = 6.4 mm to 36.8 deg at r = 10.2 mm
at z = 10 mm. At location LI (spray centerline), 50-/xm-size droplets were not
detected. The fact that larger droplets are not present at the center of the spray
-180 -90 0 90 180

U.UJ - U.U J
Ij 48 i 0 $ 52 urn
Ll
r =0 11 = 11.1 m/s
z = 10,0 mm V = -0,2 m/s
0,02 - * = 1.4 deg -0,02
-O
o 0.01 - -0,01
n nn - - n nn
-180 -90 0 90 180
An g le (deg)
-180 -90 90 180

U,IM - - U,IM
^ r = 6.4 mm U = 17,9 m/s

0.03 \ 2 = 10,0 mm V = 6,7 m/s -0,03
* = 20,5 deg
0.02 - :0,02
O
i—
Q_ 0.01 -j r -0,01
n nn -I Hit] 1 1 n nn
-180 -90 0 90 180
An g l e (deg )
-180 -90 90 180

u. i u - U. 1 U
0.08 -j
^ r = 10,2 mm U = 11,2 m/s -0,08
2 = 10,0 mm V = 8,4 m/s
0.06 -_
tJ = 36,8 deg -0.06
O
_o 0.04 -j J-0.04
o
0.02 -j
n nn -
-180 -90
Angle
0
j (deg)
1
90
70.02
- n nn
180
S = O
ferent locations in a nonswirling and swirling spray for droplets in the size range of
48-52 p,m. (Continued)
and smaller droplets are not found near the spray boundary is consistent with the
hollow-cone nozzle design.
For the swirling spray, several 50-/zm-size droplets are detected moving up-
stream as illustrated in Fig. 15 at locations LI and L2. Such upstream motion
of large droplets is surprising as it shows the comparatively strong entrainment
of droplets by the surrounding air. Close to the spray boundary (at location L3),
virtually all of the 50-/zm-size droplets are moving in the direction of the nominal
spray cone angle with no upstream droplet motion.
-180 -90 0 90 180

u.uj - - -- u.uj
nil 48 i 0 I 52 urn
Ll
r=0 U = -1.5 m/s
z = 10,0 mm V = 0 m/s
£? 0.02 - * = 177.7 deg -0.02
o -, |-| r.
8 0.01 - -0.01
0.00 -• ——- _ _ n nn
-180 -90 0 90 180
Angle (deg)
-180 -90
1
0 90
1
180 f\ f\ A
^ r = 6.4 mm U = 8.2 m/s

0.03 -j z = 10,0 mm V = 0.6 m/s J-0.03
i
4 = 4,2 deg
•g 0.02- : 0.02
o _
Q- 0.01 - : 0.01
n nn --
-90 90
1 .1 n nn
Angle (deg)
-180 -90 0 90 180
0.08 -j " r = 10,2 mm U = 16.1m/s -0.08

z = 10.0 mm V = 7.6 nn/s
0.06 -j * = 25.3deg -0.06
•
0.04 -j -0.04
0.02 -j ^-0.02
n nn ~
-180
, . , , rfl
-90 0 90
- n nn
180
Angl e (deg)
S = 0.53
C. Velocity and Droplet Size Correlation

The correlation among the droplet diameter and velocity components provides
additional information relevant to understanding mechanisms of droplet transport
in sprays. Figures 16 and 17 present such correlations at locations Ll and L3 under
nonswirling and swirling conditions, respectively. The symbols in Figs. 16c, 16f,
17c, and 17f are coded to the droplet size as shown in the inset of Figs. 16f
and 17f. At location Ll (burner centerline), the axial droplet velocity component
is weakly correlated with diameter in both cases (i.e., weak linear dependence
between diameter and axial velocity with a correlation coefficient r2 of —0.2 for
*o - — xo
20 - — 2O
la - .. -16
jfr£ :"
10 - -10
a -
o -
9f*& * • • — a
-0
-a -
A
-10 -
0 20 40 QO BO 1C 0
Ola m e t e r (M™ )
D 20 4O OO BO 1C 0
- 25
20 - - 20
1a - - 15
to - - 10
a - - 5
JM&j£i •
o -
H?9r-" "* • - 0
- -5
B
2O 4O 60 BO 1C 0
Dfameter (/2,m)
10 -5 O B 1O 15 2O 2 5
20 - - 20
15 - - 15
10 - - 10
B - - 5
•
0 - - O
** *^^^v^ *o * ^
-B -
c - -5
0 3 10 16
Axfal Velocity (m/s)
Non b u rn f n g Dm — 18.9 fj.n

S - O U — 6 . A m/s
r — O mm (LI) V — O . 3 m/s
2 •=« 1O mm
Fig. 16 Axial velocity/diameter, radial velocity/diameter, and axial/radial velocity

correlations, respectively, at the centerline (location LI, frames a-c) and spray bound-
ary (location L3 frames d-f) for the nonswirling spray; axial/radial velocity correla-
tions are coded to represent the change in droplet diameter. (Continued)
5 = 0, and 0.5 for S = 0.53) as shown in Figs. 16a and 17a, though there is a
slight increase in the values of u as diameter increases. Furthermore, for any given
diameter, the axial velocity varies over a relatively large range. For example, the
axial velocity of 10-/xm-diameter droplets varies between about 3 m/s and 11 m/s
for S = 0, and -5 m/s and 7 m/s for 5 = 0.53, as shown in Figs. 16a and 17a,
respectively. By contrast, the radial velocity component is much smaller and shows
much less scatter for a given droplet diameter than the axial component, as shown
56 C. PRESSER ET AL
2O -^ 720
1 o -£ j-18
/ * » • * * • •
• **\* a^L T** • ^ •
10
10 -=
**i^3flK53?» y** * • ** r
a -^ r 8
• -^R^"- "
1
o -^ r °
-a -I r ~s
D
<3 20 4O 6O SO 1C 0
Diameter (/am)
(3 20 40 60 BO 1C 0
20 -^ -20
ia -j - 1a
10 -j • «'*:•"• V • " • * • . ' - 10
a -jj •flm^^K-^ - 0
o-i 1 • - 0
-a -_ - -5
E
40 60
Dfometer
o a 10
26 - r 26
O — d = 1 OO /Arm
20 ~ - 20
O — d = 6O /u.m
13 -; • — d = 2O /^m - 15
10 -_ - 10
ii (tf&^f^
- ^^SSjiKfv
a -^ jtAJUf^v** - fi
0 -j «* o r°
-a -^ r -s
F
-3 0 6 10 16 20
Axfol Velocity (m/s)
Nonburnlng DM — -4-1.8 /U.IT

S — O U — 7.6 m/s
r — 1 O.2 mrh (L3) V —. 5.8 m/s
2 —» 1O m m
in Figs. 16b and lib (with the scatter decreasing as diameter increases). Figures
16c and lie show the correlation between the two velocity components at location
LI. The axial component is weakly dependent on the radial component, which
is close to zero as shown in both Figs. 16c and lie. The size-coded symbols
in Figs. 16c and lie show the tendency of larger droplets to have larger axial
velocity components than smaller droplets. It is interesting to note that the effect
of swirl near the centerline is to affect the velocity range (i.e., generate recirculated
droplets), and that somewhat larger size droplets are present at the spray centerline
than detected in the nonswirling case.
20 - - 20
V)
"E* 10 - - 10
^ 10 - - 10
o a - - a
*• "*•
Jo
0 - .-^^-V i^ \*,\ - o
-*. -5 - T^&f"-- - ' A - -B

-10 - - - 10
43 20 40 00 BO 1C0
Dfameter (/zm)
(D 20 40 00 BO 1C)0
26 -
20 - - 20
e/»
J^ 15 - - 15
^ 10 - - 10
"o
o
6 - - fi
> V* •
"5 O - ÎJft^^^.Vv-'-jiu « ... . — 0
-o 1*
o — -6
ct:
B
-1O- - - 10
<D 2O 40 60 OO 1C O
Diameter (/zm)
10 -5 0 B 1O 15 2O 2 5
26 - - 26
2O — — 2O
"«/>"
jt 15 - - 15
.>- 10 - - 10
o
o
a - - a
•v • '
Jo o - < - 0
TO
^4f$^^,*.a o
c
O
ct: -6 - - -B
20 26
A x f a l Velocity (m/s)
Nonburnfng DM — 2.4-, O >um

S - O.53 U - — 1 . 9 m/s
r — O mm ) V — — O . I m/s
2 •— 1O mm
Fig. 17 Axial velocity/diameter, radial velocity/diameter, and axial/radial velocity

correlations, respectively, at the centerline (location LI, frames a-c) and spray bound-
ary (location L3 frames d-f) for the swirling spray; axial/radial velocity correlations
are coded to represent the change in droplet diameter. (Continued)
At location L3 (near the spray boundary, see Figs. 16d and 16e for S = 0 and
Figs. 17d and 17e for S = 0.53) the results show a different trend than along the
spray centerline. Whereas the axial velocity component does not vary strongly
with droplet diameter at location LI (see Figs. 16a and 17a), the correlation be-
tween axial velocity and diameter is more pronounced at location L3 (r2 & 0.7
for S = 0 and r2 « 0.75 for S = 0.53). For the nonswirling case, the value of u
is well correlated with diameter over the range 5 /u,m < D < 60 /zm, whereas for
20 -^ - 20
-,. ^ ^ -• ;r :
' .* "•*
CO
^ 10 -^ - 1 O
_>^ 10 -^
• " • •&**£!!**•*' " "" - 10
*o I
* **%tf4E'*^* * ••
jO a -7 ^^JTRV* ^ - a
OJ
r*- - **l^tP5*B • "
15
o -f - v**-^ - o
-«c -a -: - -a
D
-10 -
<D 20 40 60 80 1C10
Diameter (>um)
<) 2O 4O QO BO 1C30
20 -^ - 20
e/i •
1? 1 5 -j - 15
>-»
1O -^ •%• * * * .* — • * " . " . • • • * . - 10
"tj . • ' * ^ < . *j **• *« • %/ •* *
o
o - > - fi
>
OJ
* «(wr£kj'?* " "^ •" *
~ •••^••^rT" • • • %
Ja Q -_ . •• "- - 0
-a
o -fi -^ - -5
Qe:
E
C) 20 4O 60 BO 1C30
Diameter (/xm)
— O -5 O 5 1O 15 2O 2 5
26 - r 26
O — d = 1 OO /^m
2O -^ - 20
o — d = 6O /zm
15 -_ , — d = 2.O >um - 15
10 -^
• ;." • *a * "o.^ "cp^l^0®^ - 10
Bo< O
6 -•
O -^
•*^!$fii&iffi*t \ ffl
" •% • • • *
°° - a
- o
- -5
"1 F
fi 10 15 26
A x t a l V e l o c i t y (m/s)
onburnlng a — 3 1 .O ^rr
— O,53 — 7.4. m/s
— 10. 2. mm (L3) — 5.2 m/s
«* 1O mm
Z) > 60 /zm, the dependence of diameter on axial velocity appears to weaken

sharply. For D > 60 /zm, the radial velocity component also increases at a slower
rate (with increased variance) and appears to approach an asymptotic value. Near
the spray boundary, where the droplets are moving at comparatively higher veloc-
ity, a simple force balance on droplets shows that droplet deceleration is inversely
proportional to diameter squared.28 Since larger droplets will decelerate at a rate
lower than that of the smaller droplets, the velocity of larger droplets is expected
to be influenced less by droplet diameter than is the velocity of smaller droplets.
For larger droplets, which would experience minimal deceleration, the droplet
velocity should be independent of diameter in the spray (assuming that their initial
velocities at the nozzle exit are independent of diameter). This is consistent with
the trends shown in Figs. 16d, 16e, 17d, and 17e.
The correlation between velocity components near the spray boundary shown
in Figs. 16f and 17f is consistent with the mean trajectory angle shown in Fig. 12.
Although these correlations are similar for both the swirling and nonswirling
cases, there are some notable differences. The size-coded symbols in Figs. 16f
and 17f again show the tendency of larger droplets to have larger axial velocity
components than smaller droplets. The effect of swirl near the spray boundary
is to affect the velocity range by widening the range of both the axial and radial
velocity components as compared to the nonswirling case. This is again evident in

Fig. 13 from the wider range of angles that is associated with the droplet motion
in the swirling case.
IV. Summary
Droplet transport was examined in a kerosene spray using a phase Doppler
interferometry system. The effect of swirl was studied by comparing a swirling
spray to a nonswirling spray for the same operating conditions. For both the
nonswirling and swirling cases, the larger droplets were found near the spray
boundary and the smaller ones near the spray centerline, a characteristic of hollow-
cone spray nozzles. Near the center of the spray, the size and velocity distributions
are narrow, and there is little correlation between droplet size and velocity. Near
the spray boundary, the width of the distributions are wider, and droplet size and
velocity are better correlated near the nozzle exit. Near the nozzle exit, droplet
momentum is far too large near the spray boundary for the surrounding turbulent
airstream to have an influence on the droplets (including the smaller size droplets).
The presence of air swirl has several effects on droplet transport. Near the
centerline, droplet transport is primarily upstream in the axial direction (i.e.,
reverse flow) as a result of the swirling air pattern. Under swirling conditions,
the distribution of droplet trajectory angles is wider at each radial position across
the spray than in the nonswirling case. This feature is most prominent near the
spray centerline where a significant number of droplets (representing a wide range
of droplet sizes) counterflow in the direction opposite to the droplets emanating
directly from the nozzle. These results indicate that a moderate degree of swirl
imparted to the airstream has a significant influence on individual droplet transport
and on the overall spray structure.
Acknowledgments
The contributions of CTA were supported by the New York State Center for Haz-
ardous Waste Management (A. Scott Weber, Director). The technical assistance
of Michael J. Carrier and James D. Alien is greatly appreciated.
References
'Presser, C, Gupta, A. K., Avedisian, C. T, and Semerjian, H. G., "Fuel Property
Effects on the Structure of Spray Flames," Twenty-Third Symposium (International) on
Combustion, Combustion Inst., Pittsburgh, PA, 1990, pp. 1361-1367.
2
Plee, S. L., and Mellor, A. M, "Flame Stabilization in Simplified Prevaporizing, Par-
tially Vaporizing, and Conventional Gas Turbine Combustors," Journal of Energy, Vol. 2,
No. 6, 1978, pp. 346-353.
3
Rosfjord, T. J., and Russell, S., "Influences on Fuel Spray Circumferential Uniformity/'
AIAA Paper 87-2135, June 1987.
4
Presser, C., Gupta, A. K., and Semerjian, H. G., "Aerodynamic Characteristics of
Swirling Spray Flames: Pressure-Jet Atomizer," Combustion and Flame, Vol. 92, No. 1/2,
1993, pp. 25-44.
5
McDonell, V. G., Cameron, C. D., and Samuelsen, G. S., "Symmetry Assessment of a
Gas Turbine Air-Blast Atomizer," AIAA Paper 87-2136, June 1987.
6
Mao, C. P., Wang, G., and Chigier, N., "An Experimental Study of Air-Blast Atomizer
Spray Flames," Twenty-First Symposium (International) on Combustion, Combustion Inst.,
Pittsburgh, PA, 1986, pp. 665-673.

7
Ghaffarpour, M, and Chehroudi, B., "Experiments on Spray Combustion in a Gas
Turbine Model Combustor," Combustion Science and Technology, Vol. 92, Nos. 1-3, 1991,
pp. 173-200.
8
Bulzan, D. L., "Velocity and Drop Size Measurements in a Swirl-Stabilized, Combusting
Spray," SPIE—The International Society for Optical Engineering, edited by L.C. Liou,
Vol. 1862, 1993, pp. 113-122.
9
Edwards, C. K, and Rudoff, R. C., "Structure of a Swirl-Stabilized Spray Flame by
Imaging, Laser Doppler Velocimetry, and Phase Doppler Anemometry," Twenty-Third
Symposium (International) on Combustion, Combustion Inst., Pittsburgh, 1990, pp. 1353-
1359.
10
Owen, F. K., "Measurements in Combustion System," Laser Velocimetry and Particle
Sizing, edited by H. D. Thompson and W. H. Stevenson, Hemisphere, New York, 1978,
pp. 123-135.
H
Call, J. C., and Kennedy, I. M., "Particle Dispersion and Velocity Statistics in a Tur-
bulent Shear Flow: Measurements and Simulations," Proceeding of the Western States
Section Meeting of the Combustion Institute (Boulder, CO), Combustion Inst., Pittsburgh,
PA, March 1991, Paper WWS/CI 91-38.
12
Moreau, P., and Borghi, R., "Experimental and Theoretical Studies of Nitrogen Oxide
Production in a Turbulent Premixed Flame," Journal of Energy, Vol. 5, No. 3, 1981,
pp. 152-157.
13
Crowe, C. T, Chung, J. N., and Troutt, T. R., "Particle Mixing in Free Shear Flows,"
Progress in Energy and Combustion Science, Vol. 14, No. 3, 1989, pp. 171-194.
14
Uthuppan, J., Aggarwal, S. K., Grinstein, F. F., and Kailasanath, K., "Particle Dispersion
in a Transitional Axisymmetric Jet: A Numerical Simulation," AIAA Paper 93-0105, 1993.
15
Hardalupas, Y, Taylor, A. M. K. P., and Whitelaw, J. H., "Particle Dispersion in a
Vertical Round Sudden-Expansion Flow," Philosophical Transactions of the Royal Society
of London A, Vol. 341, No. 1662, 1992, pp. 411-442.
16
Bachalo, W. D., and Rudoff, R. C., "Time-Resolved Measurements of Spray Drop Size
and Velocity," Liquid Particle Size Measurement Techniques: 2nd Volume, ASTM STP
1083, edited by E. D. Hirleman, W. D. Bachalo, and P. G. Felton, American Society for
Testing and Materials, Philadelphia, PA, 1990, pp. 209-224.
17
McDonell, V. G., and Samuelsen, S., "Sensitivity Assessment of a Phase-Doppler
Interferometer to User-Controlled Settings," Liquid Particle Size Measurement Techniques:
2nd Volume, ASTM STP 1083, edited by E. D. Hirleman, W. D. Bachalo, and P. G. Felton,
American Society for Testing and Materials, Philadelphia, PA, 1990, pp. 170-189.
18
Presser, C., Gupta, A. K., Avedisian, C. T, and Semerjian, H. G., "Effect of Dodecanol
Content on the Combustion of Methanol Spray Flames," Atomization and Sprays, Vol. 4,
No. 2, 1994, pp. 207-222.
19
Presser, C., and Gupta, A. K., "Behavior of Droplets in Pressure-Atomized Fuel
Sprays," AIAA Paper 93-0132, January 1993.
2()
Pr'esser, C, Gupta, A. K., Semerjian, H. G., and Santoro, R. J., "Application of Laser
Diagnostic Techniques for the Examination of Liquid Fuel Spray Structure," Chemical
Engineering Communications, Vol. 90, April 1990, pp. 75-102.
21
Gupta, A. K., Lilley, D. G., and Syred, N., Swirl Flows, Abacus, Kent, England, U.K.,
1984.
22
Bachalo, W. D., and Houser, M. J., "Phase/Doppler Spray Analyzer for Simultaneous
Measurements of Drop Size and Velocity Distributions," Optical Engineering, Vol. 23,
No. 5, 1984, pp. 583-590.
23
Sankar, S. V, Buermann, D. H., Bachalo, W. D., and Robart, D. M., "Nonintrusive
Characterization of Liquid-Liquid Mixing in Sprays," AIAA Paper 95-0138, 1995.

24
Schaub, S. A., Alexander, D. R., and Barton, J. P., "Theoretical Analysis of the Effects
of Particle Trajectory and Structural Resonances on the Performance of a Phase-Doppler
Particle Analyzer," Applied Optics, Vol. 33, No. 3, 1994, pp. 473-483.
25
Gupta, A. K., Presser, C., Hodges, J. T., and Avedisian, C. T, "The Role of Combustion
on Droplet Transport in Pressure-Atomized Spray Flames," Journal of Propulsion and
Power, Vol. 13, No. 3, May-June 1996.
26
Zhu, J. Y., Rudoff, R. C., Bachalo, E. J., and Bachalo, W. D., "Number Density and
Mass Flux Measurements Using the Phase Doppler Particle Analyzer in Reacting and
Nonreacting Swirling Flows," AIAA Paper 93-0361, January 1993.
27
Goix, P. J., Edwards, C. F., Cessou, A., Dunsky, C. M., and Stepowski, D., "Structure
of a Methanol/Air Coaxial Reacting Spray Near the Stabilization Region," Combustion and
Flame, Vol. 98, No. 3, 1994, pp. 205-219.
28
Presser, C., Gupta, A. K., Hodges, J. T, and Avedisian, C. T., "Interpretation of Size-
Classified Droplet Velocity Data in Swirling Spray Flames," AIAA Paper 95-0283, January
1995.

Chapter 3
Spray Diagnostics with Lasing and Stimulated Raman

Scattering
J. Christian Swindal* and W. P. Ackerf

Texaco Inc., Beacon, New York 12508
and
G. Chen,1 A. Serpenguzel,§ and R. K. Chang1
Yale University, New Haven, Connecticut 06520-8284
Introduction
F UEL is introduced in the form of a spray in almost all devices which com-
bust liquid fuel, ranging from internal combustion and jet engines to home
furnaces. The properties of these fuel sprays directly affect the combustion phe-
nomenon, thereby affecting the performance and emissions from the combustor.
An example of the criticality of spray properties may be found in diesel combus-
tion, where the degree of fuel vaporization at the time of combustion determines
the properties of premixed and diffusion combustion. High-premixed portion gen-
erally leads to higher NOX emissions, whereas more diffusion-type combustion
generally increases particulate emission. In-cylinder evaporation is further compli-
cated by the multicomponent nature of the fuel which contains species with a wide
range of boiling points. Droplet size, spatial distribution, and temporal evolution
of the fuel spray are all important parameters in minimizing emissions. Recent ad-
vances in diesel emission reduction have focused on increasing injection pressure,
which affects droplet size, spatial distribution, and the temporal evolution of the
spray. Spray diagnostics are important in understanding the effects of these design
changes to improve combustion efficiency and to minimize undesirable emission
formation from practical combustors. In this chapter, light scattering properties are
discussed along with the use of lasing emission and stimulated Raman scattering
rights reserved.
*Project Physicist, Fuels & Lubricants Research Department, Fuels Research Section.
tGroup Leader, Fuels & Lubricants Research Department, Fuels Research Section.
^Postdoctoral Associate, Department of Applied Physics and Center for Laser
Diagnostics.
^Department of Physics, Bilkent University, Ankara, Turkey,
^Henry Ford II Professor of Applied Physics, Department of Applied Physics and Center
for Laser Diagnostics.
63
64 J. C. SWINDAL ET AL
(SRS) to study the chemical and physical properties of individual droplets: droplet
evaporation including the effects of neighboring droplets and streams and the
shape distortions of flowing droplets caused by inertial forces. Two-dimensional
imaging techniques are also discussed and used to visualize the vapor distribu-
tion around a droplet stream and to discriminate among the constituents of a
spray.
Resonance Mode Properties of Droplets

Light which is emitted and/or scattered from a droplet is dependent on the

droplet size, shape, and composition. Spectral measurements of elastically scat-
tered light, fluorescence, lasing, and stimulated Raman scattering from droplets
have all been used to deduce droplet size, shape deformation, size change (due to
evaporation), chemical species of a multicomponent droplet, and chemical con-
centration changes (due to differential evaporation). Different colors from a laser
illuminated spray have provided discrimination among vapor, liquid ligaments,
large droplets, and small droplets in sprays. The spherical or nearly spherical
morphology of droplets provides several unique features. The input-laser intensity
thresholds for both lasing and SRS are substantially lower for droplets than for bulk
solutions. Additionally, the emitted spectra from droplets differ from that of a bulk
solution because of the optical cavity resonances1"3 associated with the droplets.
The combination of signal enhancement and the discrete resonance peaks in the
lasing and SRS spectra are exploited for diagnostics of single droplets, interacting
droplets, and sprays. The signal enhancement and resonance peaks are caused by
three features that stem from the electromagnetic and quantum mechanical nature
of lasing and SRS emission processes.
First, when the input wavelength of a plane wave does not correspond to a
resonant mode of the droplet (commonly referred to as off input resonance), the
curved interface between the liquid and the surrounding air acts like a lens to
intercept incident radiation from the illuminated side and concentrate it to a small
region just within the droplet shadow side (see Fig. 1). This focusing depends
on the relative index of refraction m(a>) between the liquid and the surrounding
gas. For an ethanol droplet (with 35-/zm radius) in an atmospheric surrounding,
i.e., the real part of the index of refraction Re[m(o>)] = 1.36, the high-intensity
region within the droplet shadow face is increased by ^lOOX relative to the
incident intensity. The intensity enhancement increases with increasing droplet
size as well as m(&>), thereby lowering the input intensity thresholds required
to achieve nonlinear optical processes such as lasing and SRS. When the input
wavelength corresponds to a resonant mode of the droplet (commonly referred to
as on input resonance), the input radiation can be stored inside the droplet and take
on the intensity distribution of the morphology dependent resonances (MDRs) of
the droplet. Each MDR has a quality factor Q associated with it. The Q can be
defined as the amount of energy stored in a particular MDR divided by the amount
of energy dissipated (by cavity leakage and optical absorption) from the droplet
during one optical cycle. Theoretically, MDRs with Q values as high as 1027
have been calculated, but when the unavoidable linear absorption and scattering
losses are taken into account, Q is limited to approximately 108. The enhancement
of the internal intensity can be very large when the input-laser pulse duration is
longer than the radiation storage time, defined as T — Q/co, where a> is the optical
frequency in radians per second.
SPRAY DIAGNOSTICS WITH LASING 65
High Internal Intensity

(a) ________^———^^ Region « 100 I 0
Laser Input
Intensity I o
% ~~— ^^~ i
Droplet Shadow
Side
(b) ________
.^^^^^^^k,
Internally Scattered
Laser Input —— ™ ^, ^ Photons
Intensity Io
Fig. 1 a) Curved surface of the droplet intercepts incident radiation of intensity /<>
and concentrates it near the shadowside of the droplet with an increase in intensity of
approximately 100/o for the off input resonance case and b) fluorescent and Raman
radiation may be generated from the high-intensity region within the droplet. Some
of the light waves may strike the droplet interface at near glancing incidence and
travel around the droplet circumference with many near total internal reflections. If
the light wave returns to its starting point with its initial phase, the light wave is said
to be on resonance with a cavity mode.
Second, the droplet provides optical feedback for internally generated radiation
at specific wavelengths which satisfy the MDRs. A fraction of rays that are emitted
within the droplet may strike the droplet interface at an angle larger than the
critical angle for total internal reflection. Many reflections occur as the rays travel
around the droplet circumference. If the light wave returns to its initial starting
point with its initial phase, the droplet acts as an optical cavity for that particular
wavelength. The higher Q-value MDRs provide more feedback for stimulated
emission processes than MDRs with lower Q values. However, the spatial overlap
of the internal intensity distributions of the MDRs with input-laser radiation is an
additional consideration for stimulated process thresholds. The higher Q MDRs
are located radially closer to the droplet interface and the lower Q MDRs are more
extended toward the droplet center. For off input-resonance radiation, the spatial
overlap is largest for lower Q MDRs, which extend more toward the droplet center.
For on input resonance, the spatial overlap is largest for MDRs with nearly the
same radial distributions.
Third, the presence of the spherical cavity can lead to cavity quantum elec-
trodynamic (QED) effects.4'8 In contrast to bulk liquid in a cell, where the final
radiation states are a continuum of modes, the final radiation states of a droplet
are the discrete cavity modes (MDRs). For the liquid in a large cell, a molecule
undergoing a radiative transition may radiate into the continuum of modes. For
a liquid droplet, however, a molecule undergoing a radiative transition can have
altered spontaneous and stimulated transition rates, depending on the location of
the molecule inside the droplet. If the molecule is located at the droplet center,
the transition rate is the same as that for a large cell, since none of the radiation
is trapped by the droplet morphology. If the molecule is located within the region
of high-field distribution of the MDR, an enhanced probability of emission occurs
at discrete wavelengths corresponding to MDRs, and a decreased probability of
emission occurs at other wavelengths not corresponding to MDRs. The enhance-
ment of emission at wavelengths corresponding to MDRs increases the efficiency
of stimulated processes. Each of these three effects (high-internal intensity, optical
feedback, and QED enhancement) contributes to reducing the thresholds neces-
sary for achieving stimulated emission in droplets. In fact, the SRS threshold for
a water droplet with radius a « 45 //,m is lower than that of water in an 11-cm
optical cell.9
MDRs are described mathematically by a set of vector-spherical harmonics.10-H
The resonances occur at discrete size parameters xn^ = 2na/^n^ where a is the
droplet radius and \n^ are the discrete wavelengths of the MDRs. Each MDR
can be labeled by three indices: a radial mode index i, which corresponds to the
number of intensity peaks in the radial direction as r varies from 0 to a; an angular
momentum index n, which corresponds to 2n number of intensity peaks in the
equatorial distribution as </> varies from 0 to 360 deg; and an azimuthal mode
number m = ±n, ±(n — 1 ) , . . . , 0. MDRs with low I values have high Q values
and are spatially confined near the droplet interface.11'12 MDRs with high t values
have lower Q values and are broadly extended11'12 from r = a to r « a/m(co).
The spatial distribution of an w-mode MDR is mostly confined to a great circle
whose normal is inclined at 0 = cos"1 (m/ri) with respect to the z axis (see Fig. 2).
For a perfect sphere, each m-mode MDR has the same perimeter length and, thus,
is spectrally (2n + l)-fold degenerate. For a spheroid (with rotational symmetry
about the z axis), each m-mode MDR has a slightly different perimeter length and,
therefore, possesses slightly different resonance wavelengths. For a spheroid with
a small deformation amplitude [with e = (rp — re)/a, where \e\ <3C 1, rp is the
polar radius, re is the equatorial radius, and e is positive for a prolate spheroid and
Fig. 2 Spherical coordinate system used in description of droplet morphology

dependent resonances.
negative for an oblate spheroid], the resonance frequency of each m-mode MDR
(for a fixed ri) is given by13
(1)
where COQ is the frequency of the degenerate MDR (at A n> £) in the absence of
any shape deformation. The degeneracy-split m modes (for a MDR of fixed ri)
assume values ranging from m — 0, ±1,..., ±n. Each of the doubly degenerate m
modes of the slightly distorted sphere has nearly the same Q value as the (2n + 1)
degenerate n-mode MDR of a perfect sphere.11 By analogy with the angular
momentum of an electron in a spherical potential, MDRs can be thought as having
a total angular momentum L, with the following properties: L2 = n(n + 1), and
Lz = \L\ cos 9 = m. For a spheroid, Lz is conserved whereas Lx and Ly are not
conserved.
Fluorescence and Lasing in Droplets

The typical fluorescence and absorption spectral profiles of a fluorescent dye
are given in Fig. 3. The lasing threshold is proportional to the difference between
the fluorescence efficiency and the absorption loss, which reaches a maximum on
the red side of the fluorescence maximum. Below the lasing threshold, fluores-
cence emission occurs and appears yellow in color. Above the lasing threshold,
the laser emission appears red shifted relative to the fluorescence emission. Con-
sequently, the well-known red shift of the lasing spectrum of dye containing liquid
is associated with the increasing absorption loss as the dye concentration is in-
creased. For dyes with small Frank-Condon shifts, the fluorescence maximum and
absorption maximum are closely separated in wavelength. Thus for these dyes,
the lasing wavelength region is significantly red shifted relative to the fluores-
cence maximum in order to minimize the absorption tail extending to the longer
wavelength.14
Dye-doped droplets provide significant feedback only at wavelengths which
correspond to MDRs. For lasing to occur in dye-doped droplets, it is necessary,
green
yellow
Fig. 3 Typical absorption and fluorescence profiles of a fluorescent dye.
68 J. C. SWINDAL ET AL.
but not sufficient, that some high- Q MDRs must be within the fluorescence profile.
The number density of MDRs per unit frequency interval11'12 is sufficiently large
(typically, 2 MDRs per inverse centimeter for droplets where a & 40 /xm and
A « 0.5 /zm) that numerous MDRs are spanned by the fluorescence linewidth
(typically, 103 cm"1). Lasing is first supported by the MDRs which possess the
highest Q values and have the greatest spatial overlap with the internal input-
laser intensity distribution. From Lorenz-Mie calculations it is known that the
Q values of MDRs decrease as the droplet radius a decreases, and the internal
intensity of the input radiation also decreases as a decreases.15 Consequently, large
droplets more readily achieve the lasing threshold and will appear redder than
the smaller droplets which only fluoresce. We have used the wavelength-shifted
lasing emission to discriminate the larger droplets from the smaller droplets, liquid
ligaments, and fuel vapor. Droplet lasing radiation will be spectrally restricted to
those particular wavelengths which correspond to MDRs of the droplet. We have
used the discrete lasing wavelengths to measure the droplet size, evaporation rate,
and shape distortions associated with inertial forces. It should also be noted that
shifts of MDR-related elastic scattering peaks have been measured to determine16'17
the evaporation rates of levitated droplets.
Stimulated Raman Scattering in Droplets

The SRS process can also cause a red wavelength shift, even for droplets or
sprays of pure liquids that are not doped with fluorescent dyes. The SRS frequency
cos is red shifted from the incident laser frequency a>input by the frequency of a
vibrational mode a)y^ of the molecules in the liquid (ct}S = <înput — &Mb). For
example, when the input laser frequency is in the green (A.input = 0.532 /xm),
the SRS wavelength of water is in the red (Xs & 0.65 /zm) because the O-H
stretching mode of water has a vibrational frequency of o>Vib ^ 3500 cm"1 and
has the largest Raman scattering cross section. For aliphatic hydrocarbons,15'18"20
such as dodecane and pentane, the Raman cross section of the C-H stretching
mode is large and o>vib ^ 3000 cm"1, which corresponds to %.$ ^ 0.63 /xm. For
aromatic hydrocarbons,15 such as benzene and toluene, the Raman cross section of
the C-C ring breathing mode is large and a>vjb & 1000 cm"1, which corresponds
to Xs & 0.56/xm. Figure 4 shows low-resolution multishot-averaged SRS spectra
from toluene and pentane droplets. The ordinate is the SRS intensity in arbitrary
units, and the abscissa is the Raman shift a)v^ from the input-laser frequency
at &>input- Pentane has a characteristic <wvib ^ 2950 cm"1 whereas toluene has a
characteristic o>Vib ^ 1000 cm"1. Toluene also shows the second-order cascade
SRS at a)2s ^ 2000 cm"1 which is being pumped by the first-order SRS. For
species identification, the unique <wvib associated with different types of molecules
is used.15'18-20
SRS starts from spontaneous Raman scattering intensity ISQ (commonly referred
to as Raman noise at OL>S) resulting from the thermally excited molecular vibrations.
The molecular vibrations can also be excited by the "beating" between the electric
fields at (WinpUt and cos- The larger the molecular vibrational amplitude, the more
Raman scattering occurs, which give rise to larger electric fields at cos- At low-
input intensities, the Raman scattered intensity is in the spontaneous region and
is linear with the input-laser intensity. Although spontaneous Raman signals are
very weak, spectral measurements of the Raman line shapes and intensities can
yield information regarding chemical composition21'22 and temperature.23
800
3
*8
400
2v
toluene
>^
•| 800
•S 400
pentane
1000 2000 3000

1
Raman shift (cm" )
Fig. 4 Low-resolution multishot-averaged SRS spectra from toluene and pentane
droplets with a « 45 p,m.
At high-input intensities, the Raman scattered intensity is in the stimulated

region and is exponentially dependent on the input-laser intensity. At even higher
input intensities, the exponential growth saturates when the input intensity is
partially depleted or when the SRS intensity is partially depleted because it is
serving as the pump radiation for a cascading SRS process. Figure 4 shows the
first-order toluene SRS at o>vib ^ 1000 cm"1 serves as a pump for another first-
order SRS with the second-order cascade SRS at 0*25 = <înput — 2o>vib. At still
higher input intensities, laser-induced breakdown occurs within the droplet, and
the resultant plasma (which is further heated by the subsequent portion of the laser
pulse) causes the droplet to shatter explosively.
For SRS to occur in droplets, it is necessary for MDRs to overlap spectrally with
the spontaneous Raman linewidth (typically, 10 cm"1). These spectrally overlap-
ping MDRs must have sufficiently high Q to provide feedback for the internally
generated spontaneous Raman photons and have significant spatial overlap with
the internal input-laser intensity. Raman gain only occurs at this overlap region,
whereas leakage of the Raman radiation occurs throughout the droplet rim as
the Raman radiation circulates within the circumference. At high-input intensities
/input, the intensity of the SRS /SRS is governed by15
/SRS = /So exP(£/input ~ /-total}(2na)N Ti (2)

where the Raman gain g (in cm/W) is linearly proportional to the species concen-
tration as well as to the spatial overlap between the /input and the MDRs within
the Raman gain profile. The Ltotai is the total round-trip losses in the droplet. The
SRS threshold is reached when the round-trip gain (g/input) exceeds the round-trip
loss (Ltotai). The number of round-trips (around the droplet circumference) during
the input-laser pulse is Nn . The exponential gain factor can be as large as e30, as
in the case of SRS from a 10-cm-long optical cell.
For a multicomponent fuel droplet, the SRS threshold will be reached for the
majority species, and its subsequent 7SRs will deplete 7input to such an extent
that the remaining /input is insufficient for the minority species to reach its SRS
threshold. Typically, SRS is barely detectable for minority components in the 10%
volume fraction range. The gain g or the Raman scattering cross section of the
minority species must be selectively increased in order to increase the detectability

of minority species by SRS. Should the absorption band of the minority species
be near A.input, then the Raman cross section is pre-resonance Raman enhanced.24
Selectively "seeding" ISQ with external radiation at the cos of the minority species is
another way to increase the detectability for minority species by SRS [see Eq. (2)].
Selective seeding of ISo with internal fluorescence radiation that spectrally overlaps
with the minority species cos has been demonstrated.25-26 With both external and
internal seeding methods, it is still important that g/input > ^totai- The extrapolation
of relative species concentration from measured molecule-specific SRS intensities
is inaccurate, because the exponential factor in Eq. (2) is highly sensitive to the
<2 of the MDR and its spatial overlap with the internal input-laser intensity and
MDR distributions. However, the SRS is very intense and can provide a qualitative
indicator of the presence of the majority and minority species.
Flowing Droplets
Droplets which are flowing through another medium, such as an ambient at-
mosphere, are deformed27 to some extent as the result of hydrodynamic inertial
effects. Liquid droplets are deformed into oblate spheroids, with the magnitude
of deformation related to the Weber number We and Reynolds number Re. The
droplets which flow through a gaseous environment with velocity v experience
hydrodynamic inertial effects which deform the droplets as shown in Fig. 5. The
magnitude of deformation is related to the droplet Weber and Reynolds numbers;
typical values for ethanol droplets used in the experiment are shown. Closely
spaced droplets experience different forces than those of an isolated droplet be-
cause of the shielding effect produced by the moving wake of the previous droplet.
A spheroidal droplet still acts as an optical cavity, however, the resonance wave-
lengths are slightly different from those of a perfectly spherical droplet. The
(2n -f l)-degenerate m modes of a spherical droplet are split13 into n 4- 1 nonde-
generate modes of an oblate or prolate droplet. For each degeneracy-split MDR,
The predicted11 frequency shifts Ao> from the single resonant frequency of a
sphere with /i » 1, is Aa> = (e/6){[3m2/n(n + 1)] - 1}, which is independent
Ethanol droplets falling in air
v = 10 m/sec
a = 40.7 |im
Terminal velocity = 0.16 m/sec
Weber number = 0.22
Reynolds number = 27
v
Fig. 5 Deformed ethanol droplets flowing through a gaseous environment.
of polarization (Transverse Electric (TE) or Transverse Magnetic (TM)), radial

mode order •£, and droplet radius a [see Eq. (1)].
Figure 6a depicts the spectral positions of four consecutive n-mode MDRs of
a spherical droplet. Should the droplet uniformly decrease from a to (a — A0),
then each of the n-mode MDRs will shift to shorter wavelength by an amount
AA = (A,/0)Aa. A relative size change of Aa = 1.4 nm can be deduced from
the small spectral shifts of lasing emission from droplets with a « 40 /xm. Exact
droplet sizes can be determined from the computer-intensive exact Lorenz-Mie
theory.11 Accurate absolute droplet size can be determined28'31 from measurements
of the spectral spacings between two consecutive n-mode MDRs. A convenient

asymptotic formula that provides an approximate droplet size (accurate to 1% for
droplets with large size parameters x and large ri) is32
-!]["! _ J_~
(3)
A,2.
where m(a)) is the relative index of refraction, Xi is the wavelength of the n-mode
MDR, and A,2 is the wavelength of the (n + l)-mode MDR.
Figure 6b depicts a slightly spheroidal droplet along with the spectral positions
of four nearly equally spaced MDRs with consecutive ns. Because each n mode of
the sphere is split into n + 1 nondegenerate m modes, the n-mode MDRs exhibit
an apparent broadening when there is insufficient spectral resolution. However,
should each of the nondegenerate m modes be spectrally resolved, then the m-mode
(a) n+1 n+2
(b)
m = ±(n -1)
m =0 m = ±n
m = ±(n - 2) m=0
(c) n-1 n+1 n+2
Fig. 6 MDRs of spherical droplets a) (2/z+l)-fold degenerate in the azimuthal mode

order m, b) slightly distorted droplet causes the m modes to split, with an apparent
broadening the MDRs if the modes are not spectrally resolved, and c) for severely
deformed droplets the splitting can become so large that the spectra appears as a
continuum.
splittings can be used to deduce the amount of deformation of an oblate droplet.

Note that the radiation confined to the droplet equatorial plane is red shifted most,
whereas and the radiation confined to the droplet poles is blue shifted most.
Figure 6c depicts a more deformed droplet (e.g., with e « 0.1). The deformation
of the droplet is now so large that the splittings of the w-mode MDRs are larger
than the spacing of the consecutive n modes and, thereby, the m modes of one
n-mode MDR overlaps with the split m modes of the adjacent n-mode MDRs. In
this case, it may be difficult to deduce the droplet deformation because the spectral
positions of the m modes would be too convoluted.
Evaporation Rate Measurements of a Segmented Droplet Stream

Figure 7 shows a typical experimental arrangement for droplet size, size change,
and shape distortion measurements. The equipment consist primarily of a high-
intensity illumination source (Q-switched Nd:YAG laser), an image preserving
spectrograph to collect and image the light scattered from the droplets, a two-
dimensional CCD detector to record the spectrally and spatially resolved in-
formation, and a source of highly monodispersed droplets. A linear stream of
monodispersed ethanol droplets is produced with a Berglund-Liu vibrating-orifice
generator. Because of vibrations induced by a piezoelectric transducer (PZT) at
50 kHz, the liquid column emerging from the orifice is periodically perturbed.
By approximately 5 mm downstream of the orifice, a continuous linear stream of
monodispersed droplets with a & 40.7 /zm, droplet-droplet spacing of &4a and
a velocity of approximately 10 m/s is produced. A segment of droplets can be
isolated from the continuous stream of droplets by electrically charging and de-
flecting all but the desirable segment of droplets.33 Figure 8 shows that a charging
collar (^3 mm in length) is placed beneath the electrically grounded orifice of
the droplet generator. By applying a positive voltage (e.g., 130 V) to the charging
collar, the electric field between the charging collar and the orifice induces a neg-
ative charge onto the tip of the liquid column that extends into the collar. As the
liquid column breaks into droplets, the induced negative charge is retained on the
droplet surface and, therefore, the droplets are negatively charged. After the charg-
ing voltage is switched to 0 V, eventually the liquid column becomes uncharged.
Switching from 130 V to 0 V is controlled by a charging-voltage circuit that is
synchronized with the laser Q-switch pulse (at 10 Hz). Two deflecting plates are
SHG
. 532 nm
Q-switched NdrYAG Laser
Cylindrical Lens
1
x imaging Lens
Stream
3
X5nt= -—-~—. Spectrograph
Color
Filter 7 CCD
Fig. 7 Schematic view of the experimental arrangement used for droplet size, evap-
oration rate, and shape distortion measurements.
nLiquid feed
PZi
+ 700 v
For lead droplet

Droplet velocity = 10 m/s
Terminal velocity = 16 cm/s
Fig. 8 Experimental arrangement used to selectively charge droplets for the produc-
tion of an isolated segment of uncharged droplets.
placed 5 mm below the charging collar and are permanently connected to +700 V
and —700 V. The electric field between these plates deflects the negatively charged
droplets and allows the uncharged droplets to pass undeflected. The length of the
isolated uncharged droplet segment is controlled by the duration of the 0-V pulse
at the charging collar.
The ethanol droplets used in the experiment contain a laser dye (5 x 10~4
M Rhodamine-6G) that is selected for its low-lasing threshold and for lasing
emission in the orange-red wavelength region. The droplets are illuminated by
the second-harmonic output of a Q-switched Nd:YAG laser (înput = 0.532 /xm)
which has a pulse duration of 10 ns and a pulse repetition rate of 10 Hz. The
laser beam is focused into a vertical sheet by a cylindrical lens and illuminates
several droplets in the linear stream. The illuminated droplets in the stream are
imaged through a red color filter to the entrance slit of a 0.5-m spectrograph
which contains a 2400 groove/mm dispersion grating. Collection of the droplet
lasing emission at 90 deg with respect to the input-laser propagation direction
minimizes the elastically scattered green light, whereas the red filter serves to block
the residual elastically scattered light and pass the wavelength-shifted orange-
red lasing emission from the droplets. A two-dimensional charge-coupled device
(CCD) is placed at the spectrograph output plane to record the spectral information
from individual droplets in the stream which are flowing parallel to the input slit.
A single 10-ns green-laser pulse pumps the dye molecules in the droplet beyond
the lasing threshold. Because the Q of MDRs are finite, the lasing light which is
trapped in the droplet will eventually leak out of the droplet for some 10-20 ns
Lasing spectrum from segmented droplet stream
(a) (b)
20 mm below the orifice 25 mm below the orifice
X(nm) X (nm)
588 590 592
Fig. 9 CCD of the lasing emission spectra from an isolated segment of uncharged
droplets at a position a) 20 mm and b) 25 mm below the droplet generator orifice.
after the green pump radiation has ended. During this short time period, the droplet
position can be considered to be static. Figure 8 also shows a gray-scale photograph
of dye-lasing droplets from the leading portion of an isolated uncharged droplet
segment. A CCD image shows the droplet lasing emission leaking from the edges
of the first five droplets of the isolated droplet segment. The photograph shows that
the orange-lasing emission is confined mainly around two arcs centered about the
droplet equatorial plane, where the green pump radiation is concentrated toward
the droplet shadow face (see the bright dot on the right side of each droplet). The
observation that the dye-lasing emission is confined to two arcs centered about
the droplet equatorial plane is consistent with the properties of MDRs. In our
experiment, the lead droplet is always two or three times larger than the trailing
droplets. We believe the larger lead droplet is a result of coalescence of two or
three droplets, occurring after the trailing droplets catch up with the lead droplet
which experiences a larger drag coefficient.
Figure 9 shows CCD images of the lasing emission spectra from an isolated
segment of uncharged droplets which are spatially resolved. Figure 9a records
the MDR-related lasing spectrum of some 18 uncharged droplets at a position
20 mm below the droplet generator orifice. By further delaying the green pump
laser pulse for 0.5 ms, Fig. 8b records the dye-lasing spectra of the droplets at a
position 25 mm below the orifice. For each droplet along the vertical z axis, the
three bright lasing peaks correspond to three MDRs having the same mode order
I but with three consecutive mode numbers, n,n + \, and n -h 2. For droplets that
are progressively farther from the orifice, the MDRs of a specific n are slightly
shifted to shorter wavelengths. The same decreasing wavelength shift is evident
with the other MDRs with two different ns. For a MDR of a fixed n and i and,
hence, a fixed size parameter xn^ a change in wavelength A A corresponds to a
change in droplet radius, i.e., AX = (X/a)Aa. This progressive shift of MDRs to
shorter wavelengths indicates that the droplets that have been out of the orifice for
a longer time, or farther from the orifice, are smaller because those droplets have
had a longer time to evaporate.
In Figs. 9a and 9b, the progressive wavelength shift for those droplets farther
upstream (e.g., droplet 7-18) is the same as that for a continuous stream of droplets
where no charging voltage is applied and all droplets pass undeflected through the
deflection plates. Consequently, we conclude that the evaporation rate for these
upstream droplets in an isolated segment (shown in Figs. 8a and 8b) is equal
to the evaporation rate of a continuous stream of droplets.34 From the spectral
shift of the MDRs, the droplet radius change A# ^ 10 A between successive
droplets. The droplet evaporation rate for these upstream droplets is dA/dt =
—7.2 x 10~4cm2/s. Because of the vaporizing wake effect of the downstream
droplets, this evaporation rate is lower than that of a single ethanol droplet falling
in a quiescent environment.35-36
For the droplets farther downstream toward the lead droplet (e.g., droplet 6-
2), Figs. 9a and 9b show that the MDRs are progressively shifted more toward
shorter wavelengths, implying that the droplet radii for these five droplets are
even smaller than the upstream droplets. The implication is that the evaporation
rates for these five droplets are larger than the evaporation rates of the droplets
farther upstream.34 The deduced34 evaporation rates of a segmented stream of
ethanol droplets is summarized in Fig. 10. The evaporation rate of the lead droplet
could not be determined due to its large size and its large shape deformation. The
evaporation rates of the successive droplets behind the lead droplet decrease until
droplet 7. For droplets farther upstream the evaporation rate approaches that of
droplets in a continuous stream.
The lasing spectrum of the lead droplet appears as a continuum (see Fig. 9). A
continuum spectrum is expected for very large droplets, where the mode density
Evaporation rate from experiment
#7 Q
#6 Q
#5 Q Aa
#4 O Aa
#3 Q Aa
#2 O Aaz=(^f-) 2 At2 OA =-11.2x10-" cm %ec
?
« O
Fig. 10 Evaporation rates of ethanol droplets in a segmented stream deduced from
the lasing spectral shifts of MDRs.
76 J. C. SWINDALETAL
is higher, and for more deformed droplets, where the degeneracy-split m modes of
adjacent n-mode MDRs merge together (see Fig. 6c). The gray-scale photograph
of the lasing droplets (Fig. 7) confirms that the lead droplet is larger and more de-
formed than the trailing droplets. It is well known in the ink-jet field,37 that during
the transition from a positive voltage on the droplet charging collar, charge com-
pensation is needed to prevent the unintentional induction of an opposite charge on
several of the trailing droplets. These negatively charged droplets will induce pos-
itively charged droplets. Two oppositely charged droplets will be electrostatically
attracted, e.g., the positively charged droplet will be attracted to the last negatively
charged droplet that was produced before the charging voltage was switched from
130 to 0 V. The electrostatic attraction also results in the coalescence of two or three
droplets and produces a large droplet with two or three times the volume of other
droplets in the segmented stream. The net charge of two or three coalesced droplets
is nearly zero and, hence, become the lead droplet since it is undeflected by the two
deflection plates. The large lead droplet is also considerably more distorted than
the trailing droplets because the lead droplet is flowing in quiescent air whereas the
trailing droplets are flowing in a moving wake of the vaporizing lead droplet. For
the same reason, the drag coefficient of the lead droplet is larger than the trailing
droplets and, therefore, the trailing droplets can catch up and coalesce with the lead
droplet. We were unable to deduce from the continuum lasing spectrum (see Fig. 9)
any information about the lead droplet size, evaporation rate, or shape distortion
amplitude.
Evaporation Rate Measurements of Multiple Droplet Streams

Since the evaporation rates of droplets in the segmented stream are influ-
enced by the vapor generated by neighboring droplets, a similar experiment was
performed to examine the influence of other closely spaced droplet streams;
a first approximation to a spray. A droplet generator containing nine 34-/zm-
diam orifices arranged in a 3 x 3 square array was used to produce droplets
with radius a ^ 3 0 / x m and a velocity of approximately 10 m/s. The spac-
ing between nearest neighbor orifices, and thus droplet streams, is nominally
500 /xm measured center to center. The experimental setup is the same as that
shown in Fig. 7, but no charging collar is used, so that the droplet streams are
continuous.
The lasing spectra obtained indicates that the evaporation rates of droplets
within a particular stream is constant. However, the evaporation rate depends on
the stream location within the 3 x 3 square array of streams. Figure 11 summarizes
the lasing-spectra deduced evaporation rates of droplets in the different streams
for both ethanol droplets and for a 2:9 volumetric ratio acetone:ethanol mixture
where the streams have a nominal nearest neighbor distance of 500 /xm. The re-
sults for ethanol indicate that the droplets in the corner streams possess the largest
evaporation rates, dA/dt « —6.4 x 10~4 cm2/s, which is lower than the evapo-
ration rate of ethanol droplets in a single continuous stream. This is reasonable
considering that the droplets are affected by the ethanol vapor trails produced by
the other eight streams. The droplet stream in the center of the 3 x 3 square array
exhibits no measurable evaporation within the sensitivity of the spectral shift de-
tection technique. The center droplet stream exhibits the lowest evaporation rate,
the side streams exhibit a higher evaporation rate, and the corner streams exhibit
the highest evaporation rate.

(a) Ethanol droplet streams
(b) 2:9 acetone:ethanol droplet streams
O O O— £=-«-
oo
Fig. 11 Droplet deduced evaporation rates for a) ethanol and b) 2:9 ace tone rethanol.
The addition of acetone to ethanol (acetone has boiling point of 56°C and
ethanol has a boiling point of 79°C), was intended to increase the evaporation rate
of the center stream above the measurable limit for ethanol droplet streams. The
evaporation rates of the acetone: ethanol droplets are also found to depend on the
stream position within the 3 x 3 square array. The center droplet stream exhibits
the lowest evaporation rate. The edge droplet streams have a higher evaporation
rate, and the corner droplet streams have the highest evaporation rate. This is
consistent with the notion that the center droplet stream experiences the least
amount of ambient air flow and the maximum of fuel vapor from the eight other
streams. The corner droplet streams experience the greatest exchange of ambient
air, whereas the exchange of ambient air for the edge droplet stream is less than
the corner stream but more than the center stream.
Vapor Visualization
Direct visualization of the fluorescence from the vapor emanating from the
evaporating droplet can be performed. Acetone molecules absorb ultraviolet light
and, subsequently, fluoresce in the blue region of the spectrum, between approxi-
mately 420 and 480 nm. We attempted to visualize the acetone vapor distribution
around a single continuous stream of evaporating acetone droplets. The acetone
droplets were spontaneously generated in a Berglund-Liu droplet generator, with
the orifice voltage turned off, causing the droplets to break up in a random manner
with random sizes and droplet-droplet spacings. The droplet illumination source
is an excimer laser operating in the ultraviolet at 308 nm. The blue fluorescence
from the acetone droplets and surrounding acetone vapor is then directly imaged
with a CCD camera. The imaging glass lenses are opaque to the 308-nm laser light
and, subsequently, act as a blocking filter for the elastic scattering and function
as lenses for the acetone fluorescence image. The major problem we encountered
is the large fluorescence signal from the liquid droplets causing saturation of the
Droplet Generator
O
Fig. 12 Gray-scale CCD image of acetone fluorescence from an acetone droplet

stream with random sizes and droplet-droplet spacings.
CCD and obscuring the weaker fluorescence image from the surrounding acetone
vapor.
Figure 12 is a gray-scale CCD image of the randomly spaced acetone droplets
and acetone vapor surrounding the droplet stream. The fluorescence is strongest
from the droplets, but fluorescence from the vapor can be seen for many droplet
diameters on both sides of the droplet stream as it diffuses into the surrounding
air. A magnified portion of the droplets shows a greater amount of acetone vapor
trailing behind the flowing droplet than in front of the droplet. This is the case for
any of the droplets which are significantly separated from the preceding droplet. If
the droplet-droplet spacing is too small, the bright acetone droplet image blurs the
surroundings and, hence, the variations in vapor concentration are not resolvable.
The observed spatial extent of the vapor produced by the evaporation from a
continuous stream of acetone droplets supports the observed dependence of droplet
evaporation rates on their local environment; i.e., the location of the droplet within
the segmented stream as well as the position of the stream among the 3 x 3 array
of streams.
Spectroscopic Measurements of Shape Deformation

The shape (oblate or prolate) and magnitude of droplet deformation due to
inertial forces can also be obtained from the lasing spectra. Note that in Fig. 9
each of the n-mode lasing peaks exhibits a "comet tail" extending toward shorter
wavelengths. Because each of the degeneracy-split m-mode MDRs is mainly
confined to different regions around the droplet rim (see Fig. 6), the radiation
leakage from the m modes and, hence, from various portions of the droplet rim
has different wavelengths. However, it is not necessary to spectrally resolve each
of the different m modes, e.g., with an interferometer, if we are able to determine
with a spectrograph the lasing emission wavelength from different portions of the
entire droplet rim.
Figure 13a depicts the magnified imaging of half of an oblate (re > rp) droplet
rim onto the spectrograph entrance slit to spectrally resolve the light which leaks
from the various m-mode MDRs of deformed droplets. The m — ±n mode, which
is mainly confined around the droplet equator possesses the longest wavelength.
The m < ±n mode, which is confined within a great ellipse with its normal
inclined at 6 = cos~1(m/n) possesses a shorter wavelength. The m = 0 mode,
which is mainly confined within a great ellipse that encompasses the polar regions,
possesses the shortest wavelength. The wavelength from various points on the
(a) Oblate r e >r p
(b) Prolate r e <r p
Spectrograph Wavelength
entrance slit
(c)
591.5 592 592.5

Wavelength (nm)
Fig. 13 Imaging of a droplet on the entrance slit of a spectrograph: a) oblate and b)
prolate.
droplet rim can be related to the vertical displacement Az from the droplet center
as follows38:
(4)
where e is negative for an oblate spheroid and is positive for a prolate spheroid.
Therefore, when the different wavelengths from the rim of an oblate spheroid are
dispersed, resulting in a D-shaped image, i.e., the larger equatorial circle is shifted
to a longer wavelength, the smaller polar ellipse is shifted to a shorter wavelength,
and all the wavelengths for various great ellipses inclined at different angles
between the equator and poles have a quadratic dependence on Az. Figure 13b
shows that for a prolate spheroid, the wavelength shifts are of the opposite sense,
and the resulting dispersed emission from the droplet rim would result in an
oppositely directed C-shaped figure.
These D-shaped figures can be seen from the comet-tail lasing emission from
each of the n-mode MDRs shown in the CCD images of Fig. 9. However, the
droplet images on the spectrograph entrance slit are too small, and the CCD
can barely resolve the dependence on Az. We magnified the droplet image in
Fig. 13c and now that D-shaped figure (belonging to one n mode) can readily be
seen in the lasing emission from various portions of the droplet rim. From the
D-shaped figure and by fitting the (Az) 2 dependence in Eq. (4), we know the
droplet is an oblate spheroid. The orientation of the dispersed image is indicative
of whether the droplet is oblate or prolate. The deformation amplitude for a linear
stream of ethanol droplets with a « 35.6/xm, at 10 mm downstream from the
orifice, a droplet-droplet separation of &4a, and a flow velocity of ^10 m/s has
been determined38 to be e = — 4 x 10~3.
Had we used a one-dimensional detector (e.g., a linear array of photodiodes),
we would not be able to preserve any of the (Az) information and, hence, A,(Az)
would appear as a broadened lasing emission spectra, as shown in Fig. 6b. An
apparent broadened spectral peak (because of poor spectral resolution) may be
misconstrued as being on a low Q MDR as a result of shape distortion. Small
shape distortions actually do not lower the Q values of MDRs but only cause
(2n + 1) degeneracy lifting of the m modes into their various spectral shifts [see
Eqs. (1) and (4)].39
There is a second contribution to the D-shaped spectra when a spectrograph
is used. Since the droplet image is itself semicircular, the spectral output should
also exhibit some degree of this geometric semicircular shape. Depending on
whether the left side or right side of the droplet is imaged at the spectrograph,
the D-shaped geometric aberration will subtract or add to the D-shaped spectral
emission emanating from the spheroidal droplet rim. The geometric semicircular
shape can be taken into account by turning the spectrograph grating to the zeroth
order and then digitally removing the geometric semicircular image from the CCD
spectral image. Measurements of the D-shaped or c-shaped spectral image alone
can permit the determination of the true droplet shape deformation as well as its
distortion amplitude.
Temporal Measurements of Shape Deformation

A time-resolved measurement technique40 has been developed which can also
be used to determine the deformation amplitude of single droplets. By focusing

Fig. 14 a) Spherical droplet with internally generated SRS which circulates around
the droplet rim in a great ellipse that has a normal incline at 9 and b) oblate droplet
with the normal of the great ellipse precessing around the axisymmetric z axis at
frequency 17.
the incident laser beam propagating along the —x direction to a small spot on
the droplet surface, a selected group of m-mode MDRs, that overlap with the
small externally focused high-intensity region, can provide the needed feedback
for the SRS to reach threshold. For a perfect sphere, Fig. 14a shows that the
two counter-propagating SRS waves will be confined to the inclined great circle
which has the best spatial overlap with the focused input-laser beam. This inclined
great circle can be considered as the equator, since the assignment of the z axis
for a sphere is arbitrary. The two counterpropagating propagation vectors of the
internally circulating SRS must be tangent to the droplet surface. The radiation
reaching the detector must be from two wave vectors pointing toward the detector
and, hence, from two spots that are diametrically opposite on the inclined great
ellipse. The subsequent leakage of the SRS from the droplet would then appear
as a two short arcs, rather than emission from the entire rim. There will be one
bright arc on one side of the droplet, and this bright arc will be stationary in time
and space.
For a spheroid, the z component of the MDR angular momentum, L z , would be
conserved but Lx and Ly would not be. Unlike the sphere, the z axis of a spheroid
must be defined along the axisymmetric direction. In analogy to a spinning top,
when \L\ ^ L z , L will have a precession frequency £2 around the z axis. Hence, the
great ellipse formed by a selected group of m-mode MDRs will, therefore, precess
(see Fig. 14). The normal to the great ellipse will be inclined (with respect to the
z axis) with 0 = cos"1 (m/ri). After half of the precession period ^p = n/ £2, the
normal of the MDR plane, inclined at polar and azimuthal angles (9, 0), will be
inclined at (0, 0 -f n). Consequently, if the MDR leakage is first observed from
Laser Input Pulse

1.0
0.5
§
CO .0.0 - .
•200 200 400 600 600 1000 1200
Time (ps)
Ethanol Droplets a « 50 pm
600 - (b) SRS

400
I 200
0
1000 (c) SRS

8 = 30°
< 500
<o o
c
Si 1.0 '(d) SRS
6 = 0°
0.5
0.0
200 400 600 800 1000 1200 1400
Time (ps)
Fig. 15 Time-resolved profiles of three different input-laser inclinations: a) the
incident-laser beam, b) 40 deg below, c) 30 deg below, and d) on the equatorial plane.
the lower part of the droplet P/ower of Fig. 14b, then after a time TT/^ the light
leakage from a precessed MDR will be observed from the upper part of the droplet
ûpper- As the MDR plane continues to precess, the two bright arcs on one side
of the droplet appear to be modulated at high frequency, commensurate with the
MDR precession frequency.
Figure 15 shows the temporal profile of a single mode-locked 100-ps laser pulse
which is focused near the surface of ethanol droplets with a & 50 /xm. Figure 15
also shows the oscillatory behavior of the SRS emission from three different input-
laser inclinations relative to the equator and, hence, different inclination angles
6 of the MDRs supporting the SRS. The dashed lines correspond to the SRS
emission from /Yower whereas the solid lines correspond to the SRS emission from
PUP^T . Note that the temporal behavior of the SRS signal from Pf and P^S are
180-deg out of phase. The reason for observing no SRS oscillation for the 0 = 0
equatorial modes is that the two SRS arcs near the equator spatially overlap, and
the two rapidly oscillating out-of-phase SRS signals are effectively summed at
the detector. The resultant exponential decay is due to the SRS cavity lifetime r
associated with leakage and other losses. From the measured r, an effective Q
value for the MDR (Q = t u>$) can be determined. The MDR precession frequency
£2 is related to the droplet deformation amplitude40:
dco coQ\e\
as ^3 — ^ ———— cos 9 (5)
dm m((o)x
where m((o) is the relative index of refraction, x is droplet size parameter, and 0
is the inclination of the normal to the MDR plane, relative to the droplet z axis. In
Eq. (5), the precession frequency £2 = 8co/8m is analogous to the group velocity
for linear propagation vg = 8a)/8k, where k is the linear propagation wave vector.
From the measured £2 at several 0s, the deduced40 droplet deformation amplitude
is e = — 1 x 10~3, a value that is larger than the deformation determined by the
spectrally resolved lasing MDRs. This is consistent with the larger droplets used
in the precession measurements which, therefore, experienced larger deformation
because of the larger hydrodynamic inertial forces.
Spray Measurements
The quantitative measurement techniques discussed to characterize droplets,
droplet streams, and multiple droplet streams can conceivably be extended and
applied to sprays, resulting from pressurized liquid that emerges from a noz-
zle. Near the nozzle, the spray diverges, forms liquid sheets, and then evolves
into liquid ligaments. These ligaments in the diverging spray subsequently break
up into droplets that evaporate as the air entrainment increases. Important char-
acteristics of the spray include breakup length, penetration depth, spray angle,
and the size and spatial distribution of the droplets. The constituents of a spray
emerging from a nozzle are either in the vapor or the liquid phase. The liquid
can assume several shapes such as sheets, ligaments, small droplets, and large
droplets.
Elastic scattering, fluorescence, lasing, and SRS have been used to selectively
image various liquid phase constituents and to discriminate among the various
constituents.41 Both the liquid and the vapor give rise to elastic scattering of the
input-laser radiation. By introducing an ionic dye that has a low concentration and
poor quantum efficiency in the vapor phase,42 such as Rhodamine-6G, the liquid
can be made to fluoresce whereas fluorescence from the vapor is greatly reduced.
Consequently, when Rhodamine-6G is added to the liquid emerging from a nozzle,
the yellow fluorescence (excited by a green input-laser beam) is mainly from the
liquid sheets, ligaments, and droplets in the spray.
Lasing requires the gain to be larger than the total losses of the fluorescence
radiation within the droplet. The larger droplets have higher gain because of greater
pumping by the input-laser intensity and more effective optical feedback for the
fluorescence radiation within the droplet. Hence, only the large droplets achieve
the lasing threshold. The intense lasing emission from large droplets is red shifted
relative to the yellow fluorescence emission, because the dye absorption loss is
decreasing toward the red side of the yellow fluorescence maximum (see Fig. 3).
Similarly, for a liquid spray that does not contain fluorescent dye, SRS occurs only
from large droplets that can provide greater optical pumping and more effective
feedback for the Raman radiation within the droplet.
The intense SRS from large droplets is also red shifted from the green input-
laser wavelength. It is also possible to increase (or decrease) the intensity of the
input-laser to decrease (or increase) the critical droplet size that will achieve the
SRS threshold. At very high input-laser intensities (above 1 GW/cm2), however,
the laser-induced breakdown threshold of the droplet will be reached. By suit-
ably illuminating and imaging a spray, the different wavelengths of the scattered
light can be used to selectively image various liquid phase components of the
spray.
INPUT LASER
X = 0.532 nm
/?
AEROSOL ^fV
GENERATOR Jfa)
Image field-of-view
Fig. 16 Schematic of experimental set-up used to discriminate among the various
constituents of a hollow-cone spray.
Figure 16 shows the experimental arrangement for selective imaging of a

Rhodamine-6G dye-doped water spray from a hollow-cone nozzle. The illumi-
nation source is the green second-harmonic output of a multimode <2-switched
Nd: YAG laser with A.input = 0.532 /xm, a pulse repetition rate of 10 Hz, and a pulse
duration of ^15 ns. The direction of the incident laser beam is along the horizontal
x axis, and the laser polarization direction is along the vertical z axis. The green
input beam is formed into a sheet (^10 x 0.1 mm) with the combination of a
diverging spherical lens and a converging cylindrical lens. The green illumination
sheet with /input ^ 0.1 GW/cm2 intercepts a select portion of the nozzle spray. The
hollow-cone nozzle has a flow rate of 13.8 m/s at a back pressure of 412 kPa, an
orifice diameter of 0.4 mm, and a cone angle of 30 deg. A water spray containing
10~4 M Rhodamine-6G dye emerges from the vertically directed nozzle (along
the — z axis) and is photographed with a 35-mm camera located along the ;y axis.
Figure 16 schematically indicates the field of view in the spray.
Figure 17a shows a gray-scale image of the green elastic scattering photograph
of the spray imaged through a neutral density filter (ND 2) and recorded on
a 200 ASA color film. The photograph extends from the tip of the nozzle (at
the very top edge of the photograph) to a distance of ^1 cm below the nozzle.
The incident green pump is propagating from right to left. The green elastic
scattering has contributions from the vapor, liquid ligaments, and droplets of all
sizes.
Figure 17b shows a gray-scale image of the orange fluorescence and red lasing
from the nozzle spray. The spray is imaged through and orange-red filter to block
the elastically scattered (green) input laser and is recorded on a 1600 ASA color
film. The field of view is the same as in Fig. 17a. The incident pump beam is
propagating from right to left. The fluorescence from the Rhodamine-6G dye-
doped liquid sheets and ligaments remains orange because the irregular shapes of
the liquid sheets and ligaments cannot provide optical feedback for the internally
generated fluorescence and/or amplified fluorescence radiation. Small droplets
also appear orange because small droplets intercept less input-laser radiation and
because the Q values of the MDRs decrease as the droplet radius decreases. In the

b)
c)
Fig. 17 Gray-scale image of the green elastic scattering photograph: a) single laser-
shot of the Rhodamine 6-G dye-doped water spray, b) five laser-shot gray-scale image
of the wavelength-shifted fluorescence and lasing image of the Rhodamine 6-G dye-
doped water spray, and c) ten laser-shot gray-scale image of the wavelength-shifted
SRS image of an ethanol spray.
color .photograph, the orange fluorescence emission is from the liquid ligaments
and smaller droplets. The large droplets appear red in the original photograph
because the large droplets intercept more of the incident input-laser radiation
and possess higher Q values, enabling the lasing threshold to be reached. The
red lasing image from a few large droplets can be distinguished from the orange
fluorescence image of the liquid ligaments and numerous small droplets. Again,
the vapor phase does not contribute to the fluorescence radiation because of the
low-vapor pressure and poor quantum efficiency of the ionic Rhodamine 6-G dye
in the vapor phase.
Figure 17c shows a gray-scale image of the wavelength-shifted SRS image for
a portion of the pure-ethanol spray. The image passes through a red filter and is
recorded on a 1600 ASA color film. The field of view is the same as in Figs. 17a
and 17b. The incident green pump beam is propagating from right to left. Ethanol is
selected because of its low SRS threshold, which is at least three times lower than
the SRS threshold of water. Only SRS emission from the large droplets is detected
because, again, the large droplets in the spray have sufficiently high Q MDRs
and intercept more of the input-laser beam needed for SRS to reach threshold,
which is at least 106 times higher than the lasing threshold. The large droplets
that achieve the SRS threshold are red shifted from the input-laser wavelength,
because a>s = ctfinput — &>vib- The smaller droplets are less likely to achieve the SRS
threshold and only cause elastic scattering in the green. The SRS threshold cannot
be achieved for the liquid sheets and ligaments because their shapes cannot provide
the needed feedback for SRS. The vapor phase will not achieve the SRS threshold
either because of the relatively low-vapor concentration (compared with the liquid
phase) and the lack of optical feedback. Consequently, only the large droplets are
expected to achieve the SRS threshold, emit the red-shifted SRS radiation, and
contribute to the SRS image of the spray.
Table 1 summarizes the scattering and emission properties of the different con-
stituents of the spray. All of the spray constituents contribute to the green elastic
scattering image. The liquid sheets, ligaments, and small droplets appear orange in
the fluorescence image of a Rhodamine 6-G dye-doped water spray. Because the
large droplets achieve the lasing threshold, the red shifted lasing emission from the
large droplets can be distinguished from the orange fluorescence emission from the
small droplets, liquid sheets, and liquid ligaments. We are uncertain of the precise
size below which the MDR Q values and the intercepted input-pump intensities
are too low to support lasing at a fixed input-pump intensity. With a stream of
monodisperse water droplets (a > 20/zm) containing Rhodamine 6-G dye, the
Table 1 Various images from spray constituents
Methods
Spray constituents green orange red

Vapor Yes No No
Sheets and ligaments Yes Yes No
Small droplets, a < 20 jum Yes Yes No
Large droplets, a > 20 n,m Yes Yes Yes
lasing threshold intensity is achieved43 with an input-pump intensity /input ^0.1

kW/cm2. Since the Q values of MDRs decrease with droplet size, the fluorescence
emission from Rhodamine 6-G dye-doped water droplets with a < 20 /xm will
not reach the lasing threshold and will remain orange. We anticipate that at pro-
gressively higher input-laser intensities, the lasing threshold can be reached for
progressively smaller droplets, which would then appear red. The limiting input-
laser intensity is the laser-induced breakdown intensity &l GW/cm2. Similarly, in
an undoped spray, we are able to distinguish the red shifted SRS emission from
large droplets because only these droplets achieve the SRS threshold.
Conclusion
The spherical and slightly deformed spheroidal droplets serve as a high Q value
optical cavity. We have reviewed the modal nature of a sphere and a slightly
deformed prolate and oblate spheroid with an axisymmetric axis along the droplet
flow direction. The thresholds for lasing (for dye containing droplets) and SRS
(when the droplets contain no dye, but can be multicomponent fuel) are greatly
reduced because of the three electromagnetic and quantum electrodynamic effects
associated with spheres and spheroids: 1) intercept more input radiation that is
not on an input resonance and store more input radiation that is on an input
resonance; 2) provide feedback at discrete wavelengths within the fluorescence
and spontaneous Raman profile; and 3) increase the gain coefficient at wavelengths
corresponding to the output resonances.
We have utilized the sensitive nature of the cavity modes (referred to as MDRs)
to determine the droplet size change caused by differences in the evaporation rates
for two cases. First, different droplets in a segmented stream of closely spaced
and, thereby, interacting droplets were examined. Second, droplets in different
streams of a 3 x 3 array of closely spaced streams were examined. Realizing that
the frequency-degenerate modes of a perfect sphere are frequency split when there
is a slight shape deformation, we utilized the degeneracy-splitting feature of the
lasing spectra to determine the shape of the droplets (oblate or prolate) as well
as the deformation amplitude of droplets as a result of inertial forces. In analogy
to precession of a spinning top and of the electronic orbits in a nonspherical
potential, we demonstrated that the precession of the SRS radiation also provides
deformation amplitude information of flowing droplets.
Because the Raman frequency shift is equal to the molecular vibrational fre-
quency, SRS spectra can be used to identify the types of liquids in the droplet.
MDR-related peaks provide information on the droplet size and the temporal pre-
cession of the modes yield the magnitude of shape deformation amplitude. Even
though the droplet size is in the micrometer range, SRS can be reached for the
host liquid and even for minority species components with a concentration above
10%. The study of the pre-resonance Raman effect and dye fluorescence seeding
to improve the detectivity level of minor components is just beginning. The quan-
titative deduction of the species concentration from the SRS intensity is inaccurate
and not precise because the SRS signal is dependent on the many experimental
parameters.
The extension of some of the nonlinear optical techniques developed for single
droplet diagnostics to spray diagnostics is possible. We demonstrated that color
differences from larger lasing droplets can be used to distinguish them from other
components in a hollow cone spray. In addition, we have shown that the red-color
shift of SRS can be used to isolate the region in the hollow cone spray where larger
droplets are found.
Acknowledgments
We gratefully acknowledge the partial support of our research by the U.S. Air
Force Office of Scientific Research (Grant F49620-94-1-0135) and the U.S. Army
Research Office (Grant DAAH04-94-G-0031). We particularly wish to express
our appreciation for the scientific collaborations we had with Yann R. Chemla,
Steven C. Hill, Thomas Jackson, Alfred S. L. Kwok, Md. Mohiuddin Mazumder,
Mel Roquemore, Karl Schaschek, and Kenneth Young.
References
^shkin, A., and Dziedzic, J. M, "Observation of Resonances in the Radiation Pressure
on Dielectric Spheres ," Physical Review Letters, Vol. 38, 1977, pp. 1351-1354.
2
Benner, R. E., Barber, P. W., Owen, J. R, and Chang, R. K., "Observation of Structure
Resonances in the Fluorescence Spectra from Microspheres," Physical Review Letters,
Vol. 44, 1980, pp. 475-478.
3
Chang, R. K., "Micrometer-Size Droplets as Optical Cavities: Lasing and Other Non-
linear Effects,*' Advances in Laser Science II, edited by M. Lapp, W. C. Stwalley, and
G. A. Kenney-Wallace, American Institute of Physics, New York, 1987, pp. 509-515.
4
Ching, S. C., Lai, H. M., and Young, K., "Dielectric Microspheres as Optical Cavities:
Thermal Spectrum and Density of States," Journal of the Optical Society of America, Vol.
B 4, 1987, pp. 1995-2003.
5
Ching, S. C., Lai, H. M., and Young, K., "Dielectric Microspheres as Optical Cavities:
Einstein A and B Coefficients and Level Shifts," Journal of the Optical Society of America,
Vol. B4, 1987, pp. 2004-2009.
6
Chew, H., "Radiation and Lifetimes of Atoms Inside Dielectric Particles," Physical
Review A: General Physics, Vol. 38, 1988, pp. 3410-3416.
7
Chew, H., "Total Fluorescence Scattering Cross Section," Physical Review A: General
Physics, Vol. 37, 1988, pp. 4107-4110.
8
Campillo, A. J., Eversole, J. D., and Lin, H.-B., "Cavity Quantum Electrodynamic
Enhancement of Stimulated Emission in Microdroplets," Physical Review Letters, Vol. 67,
1991, pp. 437-440.
9
Snow, J. B., Qian, S.-X., and Chang, R. K., "Stimulated Raman Scattering from Indi-
vidual Water and Ethanol Droplets at Morphology-Dependent Resonances," Optics Letters,
Vol. 10, 1985, pp. 37-39.
10
Bohren, C. R, and Huffman, D. R., Absorption and Scattering of Light by Small
Particles, Wiley, New York, 1983.
H
Hill, S. C., and Benner, R. E., "Morphology-Dependent Resonances," Optical Effects
Associated with Small Particles, edited by P. W. Barber and R. K. Chang, World Scientific,
Singapore, 1988, pp. 3-61.
12
Hill, S. C., and Benner, R. E., "Morphology-Dependent Resonances Associated with
Stimulated Processes in Microspheres," Journal of the Optical Society of America, Vol. B 3,
1986, pp. 1509-1514.
13
Lai, H. M., Leung, P. T, Young, K., Barber, P. W, and Hill, S. C., "Time-Independent
Perturbation for Leaking Electromagnetic Modes in Open Systems with Application
to Resonances in Microdroplets," Physical Review A: General Physics, Vol. 41, 1990,
pp. 5187-5198.
14
Schafer, F. P., Dye Lasers, Springer Verlag, Berlin, 1973.
15
Acker, W. P., Serpengiizel, A., Chang, R. K., and Hill, S. C, "Stimulated Raman
Scattering of Fuel Droplets: Chemical Concentration and Size Determination," Applied
Physics Letters B, Vol. 51, 1990, pp. 9-16.
16
Richardson, C. B., Lin, H.-B., McGraw, R., and Tang, I. N., "Growth Rate Measurement
for Single Suspended Droplets Using the Optical Resonance Method," Aerosol Science and
Technology, Vol. 5, No. 10, 1986.
17
Richardson, C. B., Hightower, R. L., and Pigg, A. L., "Optical Measurement of the
Evaporation of Sulfuric Acid Droplets," Applied Optics, Vol. 25, 1986, p. 1226.
18
Golombok, M., and Pye, D. B., "Droplet Sizing in Fuel Injections by Stimulated Raman
Scattering," Optics Letters, Vol. 15, 1990, pp. 872-874.

19
Golombok, M., and Pye, D. B., "Droplet Evaporation Measured by Nonlimear Raman
Method," Journal of Physics D: Applied Physics, Vol. 23, 1990, pp. 1103-1108.
20
Golombok, M., and Pye, D. B., "Simulation of Stimulated Raman Scattering in
Droplets," Journal of Physics D: Applied Physics, Vol. 23, 1990, pp. 1109-1113.
21
Fung, K. H., and Tang, I. N., "Composition Analysis of Suspended Aerosol Particles
by Raman Spectroscopy: Sulfates and Nitrates," Journal of Colloid Interfac. Science, Vol.
130, 1989, pp. 219-224.
22
Schweiger, G., "Raman Scattering on Single Aerosol Particles and on Flowing Aerosols:
A Review," Journal of Aerosol Science, Vol. 21, 1990, pp. 483-509.
23
Vehring, R., and Schweiger, G., "Optical Determination of the Temperature of Trans-
parent Microparticles," Applied Spectroscopy, Vol. 46, 1992, pp. 25-27.
24
Kwok, A. S., and Chang, R. K., "Stimulated Resonance Raman Scattering of Rho-
damine 6G" Optics Letters, Vol. 18,1993, pp. 1703-1705.
25
Kwok, A. S., and Chang, R. K., "Supression of Lasing by Stimulated Raman Scattering
in Microdroplets," Optics Letters, Vol. 18, 1993, pp. 1597-1599.
26
Kwok, A. S., and Chang, R. K., "Fluorescence Seeding of Weaker-Gain Raman Modes
in Microdroplets: Enhancement of Stimulated Raman Scattering," Optics Letters, Vol. 17,
1992. pp. 1262-1264.
27
Taylor, T. D., and Acrivos, A., "On the Deformation and Drag of a Falling Vis-
cous Drop at Low Reynolds Number," Journal of Fluid Mechanics, Vol. 18, 1964,
pp. 466-476.
28
Barber, P. W., and Hill, S. C., Optical Particle Sizing: Theory and Practice, edited by
G. Gouesbet and G. Grehan, Plenum, New York, 1988, p. 43.
29
Chylek, P., "Resonance Structure of Mie Scattering: Distance Between Resonances,"
Journal of the Optical Society of America, Vol. A 7, 1990, pp. 1609-1613.
30
Lam, C. C., Leung, P. T., and Young, K., "Explicit Assymptotic Formulas for the
Positions, Widths, and Strengths of Resonances in Mie Scattering," Journal of the Optical
Society of America, Vol. B 9, 1992, pp. 1585-1592.
31
Eversole, J. D., Lin, H.-B., Huston, A. L., Campillo, A. J., Leung, P. T., Liu, S. Y,
and Young, K., "High-Precision Identification of Morphology-Dependent Resonances in
Optical Processes in Microdroplets," Journal of the Optical Society of America, Vol. B I O ,
1993. pp. 1955-1968.
32
Chylek, P., "Partial-Wave Resonances and the Ripple Structure in the Mie Normal-
ized Extinction Cross Section," Journal of the Optical Society of America, Vol. 66, 1976,
pp. 285-287.
33
Fulwyler, M. J., Glascock, R. B., Hiebert, R. D., and Johnson, N. M., "Device Which
Separates Minute Particles According to Electronically Sensed Volume," Review of Scien-
tific Instruments, Vol. 40, 1969, pp. 42-48.
34
Chen, G., Serpenguzel, A., Chang, R. K., and Acker, W. P., "Relative Evaporation
Rates of Droplets in a Segmented Stream Determined by Droplet Cavity Fluorescence
Peak Shifts," SPIE Proceedings Series, Vol. 1862, Society of Photo-Optical Instrumentation
Engineers, 1993, pp. 200-208.
35
Sangiovanni, J. J., and Labowski, M., "Burning Times of Linear Fuel Droplet Arrays:
A Comparison of Experiment and Theory," Combustion and Flame, Vol. 47, 1982, pp.
15-30.
36
Abramzon, B., and Sirignano, W. A., "Droplet Vaporization Model for Spray Combus-
tion Calculations," AIAA Paper 88-0636, 1988.
37
Fillmore, G. L., Buehner, W. L., and West, D. L., "Drop Charging and Deflection in an
Electrostatic Ink Jet Printer," IBM Journal of Research and Development, Vol. 21, 1977,
pp. 37-47.
38
Chen, G., Mazumder, Md. M., Chemla, Y. R., Serpenguzel, A., Chang, R. K., and Hill,
S. C., "Wavelength Variation of Laser Emission Along the Entire Rim of Slightly Deformed
Microdroplets," Optics Letters, Vol. 18, 1993, pp. 1993-1995.
39
Lai, H. M., Lam, C. C., Leung, P. T, and Young, K., "Effect of Perturbations on the
Widths of Narrow Morphology-Dependent Resonances in Mie Scattering," Journal of the
Optical Society of America, Vol. B 8, 1991, pp. 1962-1973.
40
Swindal, J. C., Leach, D. H., Chang, R. K., and Young, K., "Precession of Morphology-
Dependent Resonances in Nonspherical Droplets," Optics Letters, Vol. 18, 1993,
pp. 191-193.
4
Serpenguzel, A., Swindal, J. C., Chang, R. K., and Acker, W. P., "Tsvo-Dimensional
Imaging of Sprays with Fluorescence, Lasing, and Stimulated Raman Scattering," Applied
Optics, Vol. 31, 1992, pp. 3543-3551.
42
Kwok, A. S., Wood, C. F, and Chang, R. K., "Fluorescence imaging of CO2 Laser-
Heated Droplets," Optics Letters, Vol. 15, 1990, pp. 664-666.
43
Tzeng, H.-M., Wall, K. F, Long, M. B., and Chang, R. K., "Laser Emission from In-
dividual Droplets at Wavelengths Corresponding to Morphology-Dependent Resonances,"
Optics Letters, Vol. 9, 1984, pp. 499-501.
Chapter 4
Feasibility of Droplet Size and Concentration

Measurements in a Hypergolic Propellant

Burning Spray
R. Lecourt,* R. Foucaud,* G. Lavergne,1 P. Berthoumieu,f and P. Millan1

Office National d'Etudes et de Recherches Aerospatiales, Chatillon, France
Introduction
I T has been recognized for some time that the design of the injector is one of the
key elements in rocket-motor development.1 Atomization and then vaporization
and combustion of propellants are driven by the injector. The role of the injector in
obtaining the desired performance is very important; a poor choice in the design
of the injector can lead, in extreme cases, to failure of a launcher.
To design injectors, rocket-motor manufacturers use conception criteria. One of
the most famous is Rupe's criterion for impinging jets injectors. The technical skill
of manufacturers has been greatly enriched since World War II. In spite of this,
the development of a new injector is still largely a trial and error task. The design
of an injector is not only a matter of atomization and mixing of the propellants.
The designer has to pay attention to weight, volume of the injector, heat tranfers
towards the injector and chamber walls, machining limitations, etc. Often the first
design of the injector fills all of the objectives but one, and the first experiments
reveal either popping, high-frequency instabilities, hot spots on the walls, or other
undesirable phenomena.2
Classically, the rocket-motor manufacturer has used experimental methods to
solve these kinds of problems. Recent years have seen an increasing use of com-
putational fluid dynamics to cure these problems or even to prevent them at the
first design level. For example, a computer code, PHEDRE,3 was developed in
the 1980s following failure of the second Ariane launch due to high-frequency
instabilities of the Viking engine. Computer codes require entry data for liquid
propellant rocket-motor simulation codes, accurate data on drop size, velocity, and
dispersion near the injector. A great deal of work had been devoted to that subject
rights reserved.
*Research Scientist, Energetics Department, Propulsion Laboratory of the Fauga Mauzac,
31410 Noe, France.
tResearch Scientist, CERT/DERMES, 2 av Edouard Belin, BP 4025, 31055 Toulouse
Cedex, France.
91
92 R. LECOURT ET AL.
in cold flow simulations, even before the development of these computer codes.4"6
However, it soon became apparent that it was impossible to exactly reproduce,
with inert fluids, the physical properties of the true propellants and the dimensional
numbers which rule the atomization processes. Moreover, because of the combus-
tion effects (drop heating, magnification of the turbulence, gas expansion, etc.) a
large uncertainty still exists on the values of the Reynolds and Weber numbers close
to the injector of a rocket motor. Thus, after computations done with the cold-flow
simulations data, a sensitivity study concerning these values must be made.
In parallel with the development of PHEDRE, a considerable amount of ex-
perimental work was done at the Office National d'Etudes et de Recherches

Aerospatiales (ONERA) on the high-frequency instabilities in small bipropellant
rocket motors.7"10 An experimental device was designed and manufactured to mea-
sure the sensitivity of injectors to an acoustic field. The sensitivity of the injectors
was assessed by the natural damping value of the first longitudinal acoustic mode
of a small rocket motors.7 Different damping values were measured for different
injectors10 or different operating conditions. Because it was assumed that the dif-
ferent behavior of each injector was mainly due to its spray characteristics, the
local granulometry and flow rates were measured in cold flow simulations for all
of the injectors; however, no obvious trend appeared from the results.
It is, therefore, clear from the previous three statements that there is a need for
direct observation of the operation of injectors in burning conditions at three levels:
first, at an empirical level, to improve the design criteria of the injectors and assist
the designer if modification of the design is needed; second, to obtain accurate
entry data for computer codes; and, finally, to increase the knowledge of the
atomization and pulverization processes and their coupling with an acoustic field.
This paper describes how a burning spray of hypergolic propellants was studied
by illuminating the liquid phase with a laser sheet. Pictures of the visualized sheets
or* the spray were taken with video and cinematographic cameras. After digitizing,
these pictures were processed in order to obtain quantitative measurements, such
as the granulometry of the spray and the spatial dispersion of the liquid phase.
Previous State of the Art Attempts and Difficulties of Observation

Observation of the combustion in rocket motors has been attempted almost since
the beginning of the development of modern rockets. Reference 1 reports the work
done in the 10 years previous to 1960. It describes combustion chambers equipped
with transparent slits. The slits are cinematographed by a camera without shutter.
In this way, the motion of luminous particles (drops, soot, etc.) along the slit
produces streaks on the film. In a later modification of this device, larger windows
were mounted on the combustion chambers, and the slit was placed on the camera.
Analysis of the streaks allowed measurement of the speed of the particles. This
kind of device was used to study high-frequency combustion instabilities. The
particles were supposed to follow the gases and, thus, the streaks on the film
allowed visualization of the motion of the gases during combustion instability.
Almost at the same time, Ingebo11 managed to measure size and velocity of drops
in a rocket combustor burning ethanol and liquid oxygen. The measurements were
made on photographs obtained with a high-speed camera developed at NASA
Lewis Research Center. He pointed out the great difficulty of direct observation
in classical flames of rocket motors (except H2-O2), that is to say, to provide
sufficient light to penetrate the relatively opaque flames.
MEASUREMENTS IN A BURNING SPRAY 93
In 1965, Levine12 studied the combustion of the same propellants in a two-

dimensional windowed rocket motor, using streak photography and high-speed
cinematography of the emitted light of the flame and of the drops silhouetted
by backlighting. With the backlighting technique, Levine was able to follow the
motion of the drops and the fans during a combustion instability. He attempted
to do the same thing with the LOX-kerosene propellants but the flame was too
opaque, obviously because of the presence of more soot. For this work, Levine
used a very high-visibility combustion chamber which was operated up to 7 MPa.
Unfortunately, neither in this paper12 nor in Ref. 13, which also describes this
work, is there a description of the design of the windowed chamber.

In 1972, McCreath et al.14 described a photographic device for drop size and
velocity measurements in burning sprays. This device was considered the ulti-
mate state-of-the-art photographic technique for that purpose before the advent
of scattering techniques and other recent technological advances. It was used
in dilute kerosene burning sprays. The experiments were made in open air. The
photographic plates were analyzed by an analog system.
Direct observation was also used to study combustion phenomena related to
the use of impinging hypergolic propellants.15'16 These references quote several
other studies of storable liquid propellant sprays by means of a photographic
technique. In both works, the propellants used nitrogen tetroxide as oxidizer
and hydrazine or its derivative as fuel. To study the combustion phenomena, the
authors used color photography because of the color difference between nitrogen
tetroxide (red) and hydrazine and its derivatives (white). In both studies, the
experiments were done with a single element like-unlike injector fired in the
windowed combustion chamber. To see the impinging point of the hypergolic
propellant jets and the resulting spray, Lawver,16 like Zung and White,15 used
powerful white light sources. In Lawver's work, the light was provided from top,
bottom, back, and front of the combustion chamber to obtain a balance between the
reflected and absorbed light. It is difficult to say if this disposition of light sources is
profitable since the quality of the pictures in the copy of Lawver's report was very
poor. Thanks to the short exposure times, the flame was not visible. The pictures
were taken by a single-exposure,15 or high-speed movie camera.15'16 On some of
the Zung and White pictures, the jets and the droplets of the propellants are clearly
visible. At the beginning of his experiments, Lawver used a gaseous film cooling
along the windows of the combustion chamber to protect them from hot gases.
However, because the film cooling created density gradients that were visible on
the photographs and altered the quality of the pictures, Lawver abandoned it.
With this visualization technique, the authors were able to study the parameters
influencing popping, stream separation, or mixing of impinging jets of hypergolic
propellants.
The development of devices, based on Mie or geometrical optics theories of the
light scattering, offers an alternative to the photographic technique for size and
velocity particle measurements. Laser-light scattering devices were first used in
cold flow simulations and then in combustion environments. Reference 17 gives
a detailed description of such a device dedicated to perform measurements in
burning conditions, in laboratory and industrial environments. The authors, Holve
and Meyer, mention particle-size measurements in a furnace at 1600 K. The
well-known Phase Doppler Particle Analyzer (PDPA) was also used in flames.
Presser et al.18 describe velocity and droplet size measurements in spray flames
of methanol and methanol/dodecanol-air. In the conditions of the experiments
(dilute spray, weak emitting flames) it must be concluded that the PDPA worked
well. The authors used a laser sheet to visualize the spray and, consequently, the
regions where there were drops. The results of the PDPA agree qualitatively with
these visualizations. From their paper, it appears that the PDPA was able to make
drop-size and velocity measurements everywhere drops were visualized by the
laser sheet.
Finally, Hulka and Makel19 describe a windowed combustion chamber dedicated
to up-to-date flow visualization and optical diagnostics for the study of a burning
spray of liquid oxygen and gaseous hydrogen. As in previous work of Aerojet,16
the combustion chamber was designed for firing a single element injector. Optical
access to the combustion chamber is through circular windows which can be
placed at several locations along the combustion chamber. This leads to rather
poor visibility, but it must be pointed out that the chamber was designed for an
operating presure of 7 MPa. Moreover, to confine the burning spray, as in the actual
multielement rocket motor, the spray is surrounded by an annular thick layer of hot
gases produced by gaseous H2-O2 combustion. The first combustion experiments
were made at a chamber pressure of 1.2 MPa with injection of only liquid oxygen
and the annular layer of hot gases (no hydrogen was injected around the liquid
oxygen port). In spite of the experience of Lawver,16 a thin barrier of cold gas
(hydrogen) was used to protect the windows. Only a few video and high-speed
camera pictures were taken without external light. The authors reported good
visibility through the hot gases and the cold barrier. They noted that the features
of the liquid oxygen combustion were not distinguishable.
To conclude this short review of direct observation in rocket-motor combustion
chambers, it can be stated that with adequate exposure (short), it is possible to
see through the flame; with powerful external lighting, it is possible to illuminate
the spray and record pictures of it; and suitable hardware is available to manufac-
ture windowed combustion chambers (even with high visibility) with operating
points close to those of actual rocket motors.
Recent Technological Advances

External Lighting of Burning Spray
To light burning sprays, the greatest technological advance since the beginning
of the development of modern rockets is the invention of the laser. In the past 10
years, the classical lasers (argon, yag, etc.) have become more powerful and their
electric suppliers less cumbersome. Moreover, more efficient lasers such as gold
and copper vapor lasers have appeared.
As all of the lasers have become easier to handle and operate, it is possible to
use them in rather hostile environments such as rocket-motor experiment cells.
Now it is possible to make use of the specific properties of the laser light in
comparison with the white light. As the laser light is quasimonochromatic, one
can handle or manipulate it to concentrate in a very small volume in order to obtain
high-volume density of light power. Moreover, these "volumes" which are often
sheets can have different shapes (flat, cylindrical, or conical sheets, for example),
so that the whole light power illuminates only the zones targeted. Then, the laser
sheet can be accurately positioned in space, that is, the space coordinates of the
measurements are also accurately known . Finally, with a flat laser sheet, a short
depth of field is sufficient to take photographs which simplify picture acquisition.
Picture Acquisition
The arrival of the charge-coupled device (CCD) video camera has revolutionized
the range of picture acquisition. With these low-cost cameras, we now obtain
instant results which can be qualitatively analyzed quasi-instantaneously with
video recorders. This allows quick setting of optical parameters, lighting, and
picture acquisition. The electronic shutters of the camera are capable of short
exposures up to l/10,000e s. The CCD transducers also have a high sensitivity
which can be magnified by electro-optic devices.
The drawbacks of the video are still, on the average, poor rate and poor def-
inition. It is possible to improve one of these two properties only by decreasing

the other. For the moment, only cinematography can provide good rates and good
definition.
Image Processing
The production of numerous scientific or technical video pictures has stimulated
the development of image-processing techniques. Thanks to the improvements in
computer science, more and more sophisticated processes [filtering, fast Fourier
transform (FFT)] have been applied to the large amount of information which
represents even a single video image on 256 gray levels (512 x 512 x 8 bits). Now,
with new and improved performance at the workstations in terms of data transfer
and storage capacities, it is possible to join the advantages of cinematographic
pictures and video pictures. Indeed, with a high-definition video camera (2048 x
2048 pixels) mounted on a film analyzer and connected to a workstation, off-line
high-definition digitizing of cinematographic pictures is possible. Thus, as will
be demonstrated later in this paper, all of the information of a 16-mm picture can
be extracted and processed. Moreover, the capacities of the workstations allow
automated and statistical processing of hundreds of such high-definition pictures
on 256 gray levels (4 Mbytes per picture).
Experimental Method
Combustion Chamber
To begin this study of a burning spray of hypergolic propellants, the choice was
made to work at atmospheric pressure. This decision had a pratical dimension: to
avoid difficulties with the windows (breaks, leaks) and, in addition, to be able to
conduct numerous inexpensive experiments a day. It must be pointed out that our
earlier experiments were done with the spray burning in open air, which allowed
us to quickly demonstrate the feasibility of the visualization of the burning spray
with a laser sheet.20
From a scientific point of view, it must be noted that at low pressure all of
the phenomena (pulverization, evaporation, heat transfers, and then combustion)
are spatially stretched. It was found, a posteriori, that this was an advantage
for observing the spray and analyzing and interpreting the pictures. Of course,
this way it is impossible to reproduce the exact burning conditions of a rocket
motor and the work at low pressure must be validated by further experiments in a
windowed combustion chamber simulating more closely the operating point of a
rocket motor.
Therefore, after the experiments in open air, the entire work was repeated in a
nozzleless windowed combustion chamber operating at atmospheric pressure (see
Fig. 1 Windowed combustion chamber.
Fig. 1). To have a high visibility in the chamber and to be able to put windows on
all sides, the design of a chamber with rectangular cross section was chosen. Its
internal dimensions are 160 mm in length and 40 x 50 mm2 for its Cross section.
The visibility through the pyrex windows is 27 mm in height and 143 mm in
length. Because this combustion chamber was designed for short experiments
(2-3 s), it was made of aluminum alloy. There is no film cooling of the windows.
After several experiments, one or two windows were cracked. If the crack was
not on the path of the laser sheet or in the field of view, because the combustion
was operated at atmospheric pressure it was possible to do additional experiments
without leakage through the cracked windows.
Illuminating Technique of the Spray

As was noted by the precursors in this field, the difficulty is to provide more
luminous power to the burning spray than it produces itself. In the past, back-
lighting with a larger power than that the flame was used so that drops appeared
black on a white background. On one side, thanks to the use of lasers, it is easier
to concentrate light in a small volume in order to reach light powers higher than
those of the flames. In laser-sheet visualization, on the other side, the camera
collects only the light scattered by the drops in their direction. It is this small part
of scattered light which has to be more powerful than the flame in order for the
drops to be visible. The first experiments20 showed that either the laser sheet must
be as thin as possible (a few tenths of a millimeter), or it has to be more powerful.
For example, when the experiments began in the windowed chamber (where the
combustion is more efficient than in open air and, therefore, more light is emitted)
we were obliged to abandon the 4-W argon laser in favor of a 15-W copper vapor
laser.
longitudinal M M . transverse
laser sheet \JJ burning LjJ / laser sheet
^^ spray
inject
video or fast movie camera video or fast movie camera

on a binocular or not
LONGITUDINAL TRANSVERSE
VISUALIZATION VISUALIZATION
Fig. 2 Schematic of experimental setups.
The laser sheet was placed in two positions in the burning spray, either perpen-
dicular to the flow direction to obtain a cross section of the spray or parallel to the
flow direction to obtain a longitudinal section (see Fig. 2).
In the first experiments, since the velocity of the drops is about a few tenths of
meters per second before being accelerated by the combustion gases, the steady-
state argon laser was used to obtain the transverse sheet. The drops were illumi-
nated only when they flew across the laser sheet, thus avoiding light streaks on the
pictures.
Copper vapor laser are pulsed lasers whose frequency of pulsation can be easily
adjusted in a large range. Moreover, the lightning time of these lasers is very
short, between 10 and 40 ns. Because of these characteristics, they can be used
as stroboscopes. It seemed convenient to use such a laser to create a longitudinal
(i.e., parallel to the flow direction) stroboscopic laser sheet to illuminate the drops
and freeze their motion.
Because of its greater power, the copper vapor laser was also tried for transverse
visualizations in the combustion chamber. Since the experiment proved to be
successful, all of the following visualizations were made with this laser. Figures 3
and 4 present an example of each kind of visualization, longitudinal and transverse,
in the combustion chamber with the copper vapor laser. The liquid phase in these
photographs appears in yellow. In the longitudinal section, the sheet was vertical
and positioned so that it contains the injection axis of the injector. This way, it was
expected that the sheet would illuminate a high-concentration zone of the spray,
perhaps the highest concentration zone of the spray.
The camera was positioned perpendicular to the laser sheet and, consequently,
to the window. In Fig. 3, the injector is on the left, and the burning spray flows
from left to right. Because of the laser-sheet illumination, desintegration waves,
ligaments, and individual drops are clearly visible from the injector. In the trans-
verse visualization, Fig. 4, the laser sheet was located about 30 mm downstream
of the injector, the camera being positioned obliquely in front of the sheet, with
an angle of 20 deg to the chamber axis.
Numerous drops are visible in the cross section of the chamber, even with the
short duration of the laser pujse (a few tens of nanoseconds). Because this kind
of picture was intended to be used for drop size measurements and the drops
are illuminated for only about 30 ns with a very thin sheet (a few tenths of a
millimeter), we cannot be sure that the drops are illuminated at their greatest
98 R. LECOURTETAL
Fig. 3 Longitudinal visualization of burning spray.
Fig. 4 Transverse visualization of the burning spray.

diameter. For the moment, it is assumed that well-illuminated drops (that is to

say, those showing a high luminance on the photograph) were likely either fully
illuminated in the laser sheet or fully illuminated by internal reflections in the
drop. In addition, a drop which is partially illuminated by the laser sheet receives
less light power per unit volume than a fully illuminated one. Thus, it may have
a lower luminance on the photograph and be eliminated by image processing. In
any case, further experiments are needed to reach a conclusion.
Image Acquisition
When we began our experiments, a video camera was used to test the new
configuration of visualization. Now, it is used only to confirm that the experiment
is successful. Because of its low picture rate (50 ips, i.e., 25 fps) and its rather poor
picture definition (512 x 512 pixels), the video is not capable of doing quantitative
measurements and statistical analysis in sprays. Then it was decided to perform
the pictures acquisition with a cinematographic camera.
Because the experiments are short (about 2 s), the choice was made to use a
high speed 16-mm NAC E10 camera, which can to take up to 10,000 full size
frames per second. In almost all of the experiments, the camera was equipped with
a Nikon 200-mm focal lens.
Tests were made to determine exactly the definition of the cinematographic
pictures. The film used was a color negative Kodak 7292 with a definition of
90 pairs of bars per millimeter, or 180 pixels/mm when translated into image-
processing language. The test was made under average conditions according to
the settings used in the visualization experiments, except that steady lighting was
used instead of copper vapor laser lighting. The test pattern presented in Fig. 5a
was filmed with an aperture of 8 and a rate of 1000 fps. The result of the test is
presented in Fig. 5b. The first point to notice is that a "move" in the direction of
the vertical bars altered the definition between the horizontal bars. This move was
entirely eliminated in the experiments because of the use of stroboscopic lighting.
Therefore, observing only the vertical pairs of bars, the picture shows that they
can be distinguished at least at the spatial rate of 31.5-36 pairs of bars on the
test pattern. By taking the magnification of the lens (0.69) into account leads to
a film definition of 45-52 pairs of bars per millimeter, i.e., about 100 pixels/mm.
In addition, when the bars are closer, not only are they mixed together, but the
resulting spot is less visible on the background.
In the experiments, the greatest magnification was used in the transverse visu-
alizations. Its value was 0.58, less than in the definition test because of the oblique
angle of view. Therefore, the test just described signifies that with a definition of
100 points per millimeter, the apparent size of the smallest visible object is 17 /im
in diameter and that an area accuracy of 5% is obtained for objects larger than
85 /xm in diameter. Since it is possible to slightly increase the magnification to
one, the definition is sufficient for drop size measurements in liquid rocket-motor
sprays if the smallest drops (10-20 /xm) scatter energy enough to be visible on
the film. Finally, to be fully operational, this definition has to be preserved by the
digitizing process of the pictures. At the moment, the most accurate system we
found was able to digitize the pictures at a spatial rate of 100 pixels/mm, that is
to say, about the definition of the film. However, as the film has a smaller grain
than 10 /xm, the spots on the film can be located at less than 10 ^m, relative to
an absolute grid (see Fig. 6). Because the pixels of the digitizing system may not
100 R. LECOURTETAL
25 «U!
IB.
ISC
HI « ii sa
b)
Fig. 5 Views of the test pattern: a) original and b) 16-mm film picture.
10 |im spot -
lOjiim
Fig. 6 Film with two different grains.
be aligned with the spots, a more refined definition is necessary for the digitizing
system (at least twice the definition of the film) in order to preserve the information
of this the film.
Concerning the frame rate, several transverse visualizations of the burning spray
were made under the same conditions, but with different frame rates, 150, 300,
600, and 1200 fps. It was found that the flame light emission became invisible
at the highest frame rate. With the mechanical shutter placed on the camera, that
means an exposure time of 280 /xs. Unfortunately, the aperture is not known
because these experiments were done with a binocular microscope on the camera.
To give an idea of good settings, successful picture acquisition was achieved with
the Nikon 200-mm focal lens in the windowed chamber at 2000 fps and an aperture
of 16.
Finally, use of laser-sheet lighting does not require a large field depth if the lens
is well focused on the sheet. But, as the focus is set before the experiment, it was
observed, that for the smallest field of view (because of differences of refractive
index between cold air and hot gases and, perhaps, because of the fluctuation of
the refractive index during the experiment) a field depth of about a few millimeters
was necessary.
Because of this need for depth of field, even for small fields of view, it appeared
more suitable to use a photographic lens rather than a binocular microscope. In
addition, in our experiments, the smallest fields of view used for drop sizing
required a larger depth of field. As it was aimed to perform measurements at a
well-known abscissa, the image acquisition was performed with an oblique line of
sight (20 deg), Fig. 2 (transverse visualization). For obvious geometrical reasons,
the required depth of field was 6-mm large (field of view: 12.8 x 17.1 mm2). This
value was matched with the Nikon lens (7-8 mm) but may be with small margins to
take the refractive index variations into account. To increase these margins, it is
intended to fit the laser-sheet dimensions to the field of view in order to increase
the power of illumination and, then, to be able to use a smaller aperture.
Image Processing
Regarding image processing, two methods have been developed to conduct tests
of injectors in combustion. The first approach allow one to obtain the size of the
particles seen. This kind of treatment gives accurate results only when it is applied
to images containing particles of size sufficient enough to minimize the error in
the detection. When observed droplets are too small, a density map of droplet
distribution is realized.
102 R. LECOURTETAL
Diameter Measurement
Cinematographic images are digitized with high-resolution systems, high-
definition CCD sensors or line scan camera which allow a definition up to 2000
points, corresponding to the film capacity. Data acquisition made to attain the
following results allowed us to define 1130 points; each point covers a 9-/zm
square on the film and is digitized on 8 bits, that is to say, 256 gray levels.
Thresholding
To observe droplets, a more intense upper light than that provided by the flame
is necessary. Thus, on the positive copy of the film, these droplets appear bright on
a dark background. Droplet detection inside an image is equivalent to finding all
of the bright areas in the dark background. If we represent the variation of lighting
on an image line we obtain the curve shown in Fig. 7 on which three particles are
intersected.
One way to extract the droplets is to give a threshold lighting value which
separates the background from lit droplets. Using this threshold operator, one can
choose a gray level value for which we assume that the value of all of the droplets
is higher than the value of the background. The horizontal line on the curve in
Fig. 7 offers a threshold possibility. It seems obvious than the droplet size will be
influenced by the threshold value. According to the horizontal line position, the
segments representing droplets on the binary image will have different lengths.
Some authors are conducting preprocessing on gray-level images before applying
the threshold. Those processes are used first for eliminating noise and then to
increase the contrast to allow a better threshold choice. Experimental setups offer
the possibility of obtaining very contrasting images on which a threshold operator
can be applied directly to avoid losing important information. It also appears
that noise reduction filtering has a smoothing effect on images. Because this
process decreases black/white transition which makes the threshold level choice
more difficult, it is necessary to apply a contrasting enhancement filter. Thus,
noise suppression is applied to the binary image in which all lit pixels are parts of
the droplets. The main problem is gathering all of the points belonging to the same
drop image. Image scanning is conducted, and only segments included between
two given values are considered. The longest segments cannot be taken for drop
images and the shortest ones can be considered as noise.
Fig. 7 Representation of variation of lighting on an image line.

"-j 48
JB
H44
m 40
• 36
Jggo~
~ H
-£||8g 8 o Threshold level
^ \o !P.
Diameter classes in microns °° ^ a? x;
Fig. 8 Drop size histograms of the spray burning in open air.
In a second step, a sphericity criterion is applied to observed particles in order to

avoid possible stripes on the film. All particles should be inscribed in a rectangle
the sides of which must have rather the same lengths. Furthermore, these particles
should have nearly the same surface as the one of a circle inside the same rectangle.
Regarding drop diameter measurement, the following hypothesis is offered:
all droplets are spherical; the plate diameter is then calculated and its surface is
similar to the preceding one measured.
Threshold influence on particle-sizing results and number of droplets obtained
has also been studied. Figure 8 shows histograms which were obtained with
different threshold values. In the images used, background level was about 30,
and the peaks corresponding to the droplets were as high as 160. The threshold
values were visually chosen from several characteristic pictures. Under 36, the
value is not sufficient to eliminate the background; over 48, the smallest drops
could be lost. It seems that because analyzed images have high contrast (lighting
gradients between image background and particle is very stiff), the influence of
the threshold value is important on the number of drops in each size class but not
on the shape of the distribution. Tests were performed in open air, and more than
1000 images were analyzed (see Fig. 8).
Background Removing
Considering a nonuniform image background, it is necessary to use filtering in
order to avoid mistakes. In Fig. 9, which is an artificial one to support illustration of
the phenomenon, shows a line of an image containing a nonuniform background.
A lighting removal value, equal to the local mean value, allows one to limit the
influence of the background. This technique, possible only when the analyzing
area is close enough to the phenomena size to be observed, was verified in the
present case. Figure 10 presents the results where this method is used on the
previous line.
The windowing size depends on the size and density of the drops present in the
image. In this study, there are a few droplets, the largest of which are less than 150

Fig. 9 Representation of nonuniform image background.
pixels. The windows used are of 30-60 pixels of sides, consequently, the mean is
done on 1000-4000 pixels.
A complete trial run of this processing was made. Sometimes we noticed that the
background lighting varied during the run, with the flame brightening. Removing
the background allowed us to hold a fixed threshold.
The described process was applied on the filtered images. The diameter class
distribution obtained is completely equivalent to have of the unfiltered images.
Only the number of detected droplets increased.
Precision of Sizing Method

A minimum detectable particle size has been defined in order to stay below
the chosen error level. The curve in Fig. 11 denotes this; the possible mistake in
number of pixels on the area measured is a function of the desired accuracy. By
this means, it was possible to evaluate the filming area. To obtain an error of less
than 10% on the diameter measurement assuming that the threshold level allows
particle-size measurement with a possible error of less than 10 pixels, the smallest
droplets have to cover an area of 50 pixels.
For a 50-jnm-particle-diam sizing with a digitizing system giving 1000 pixels
by line, the observed field will be 6.25 mm. The measurement of smallest particle
will be less accurate; for example, the error can increase up to 20% on the diameter
for 35-/zm-diam droplets.
180 j
160..
140..
120..
100--
80--
60..
40..
20..
0-'
Fig. 10 Results when lighting removal value is used.
Influence of possible error on area detection
20%
40
35
30
10%
O 20
5%
0 10 20 30 40 50 60 70 80 90 100
Particle area in pixels
Fig. 11 Influence of possible error on area detection.
Distribution Map
The aim of this process is to show the spatial density distribution of the drops
on a virtual image.
First, a binary acquisition is made (digitizing and thresholding as already de-
scribed) on some thousand images with a threshold which isolates the droplets on
the background. Using an analysis of this sequence, a virtual image is built. In
this image, a pixel value is incremented each time the corresponding pixel level
of the current processing image is above the threshold value. At the end of the
process, the lighting level is equal to the number of droplets which crossed this
pixel during the trial.
The lit areas on this image correspond to high-concentration droplet zones. It's
also possible to establish the distance after which all droplets have burnt. The
upper part of Fig. 12 shows a longitudinal injection axis droplet distribution (the
injector is on the left).
Method of Calibration
The feasibility experiments shown that quantitative measurement of drops size
is possible, but analysis of the pictures yields only apparent sizes. How must the
image be processed to extract the true sizes of the drops?
Even if a scale is recorded before the experiment, such a method of drop
sizing requires calibration to take into account optical aberrations due to the light
diffusion and refractive index variations of the combustion gases.

CM AXIS f\!UM8£r? OF DROPLETS
Fig. 12 Liquid phase distribution in the median plane of the combustion chamber
field of view, 27 x
The principle chosen for the calibration desired was to put into the flame an
object similar to the drops, calibrated in size and illuminated by the laser sheet. It
was decided to use cylindrical silica fibers whose diameters were representative
of those of the drops (that is to say, about 100 and 500 /^m). At that time, the
experiments were conducted with the injector operating in open air. It must be
kept in mind that this may not be fully representative of visualization conditions
in the windowed chamber. The transverse argon laser sheet was used to illuminate
the fibers. The best way to illuminate the cross section of the fibers was to put
only the extremity of the fiber in the laser sheet. The pictures were recorded by a
video camera mounted on the binocular microscope. For the first experiments, the
fibers were placed behind the flame so that they would not be damaged. By doing
so, the optic path, from the fibers to the camera, crossed the entire flame. Second,
the fibers were put into the flame but protected from the flow by their holder. In
this case, a vibration of the fibers by the flow was revealed on the video pictures.
It was also noted that the end of the 500-/xm fiber should have been polished
in order to have a light diffusion close to that of a sphere. Consequently, only
pictures of the smallest fiber (100 /xm) were studied. A video picture of this fiber
is presented before combustion in Fig. 13a, and during combustion in Fig. 13b.
Comparison of these two pictures shows that the diffusion spot at the end of
the fiber was much larger during combustion than without the flame. Several
explanations can be proposed. 1) Because of the refractive index variations, the
combustion gases act as a magnifying lens. 2) Because of the higher light level
during combustion, the CCD transducer is saturated, and the pixels surrounding
the luminous spot are influenced by the saturation. 3) The flow caused the fiber
move or vibrate during the exposure period.
It, therefore, appears that further calibration experiments are required to explain
the phenomena observed. This first attempt of calibration shows that a step-by-
step approach must be used to calibrate the method and that fast cinematography
should help in this work.
Experimental Results
Experimental Study of One Test Injector
To carry out these experiments, we used a monoelement injector designed for
liquid bipropellant rocket motors. It was a four-on-one impinging jets injector. At-
omization is obtained by impingement of four surrounding oxidizer jets (nitrogen
tetroxide) on an axial jet of fuel (monomethylhydrazine). In our experiments, this
injector was used with NTO and MMH propellants with a nominal mixing ratio
of 2:1. The theoretical flame temperature is close to 3000 K.
As explained in the image processing section, two kinds of measurements
were done in the spray produced by the injector, drop sizing and liquid phase
distribution.
Let us recall that the drop sizing was conducted on the apparent size of the drops
on the pictures since the measurement method is not yet calibrated. Therefore, the
size histograms which will be presented are not necessarily representative of the
true sizes. This observation is also true for the repartition measurements, but it is
considered of less consequence regarding quantitative results.
Drop Sizing
The results of two experiments will be presented. For the first one, the spray
was burning in open air and, for the second one, in the windowed combustion
chamber. In both experiments, the spray was illuminated by a transverse laser sheet
and the pictures were taken on the axis of injection. In the open-air experiment,
the laser sheet was created with the argon laser. A field a view of 10 x 13 mm2
was obtained with the binocular microscope. The cinematographic pictures were
acquired at 1200 fps. Because the experiment in the windowed chamber required
higher illumination, the laser sheet was created with the copper vapor laser. The
Nikon lens was used because it afforded a higher depth of field. But, because
the magnification of this lens is less than the one of the binocular microscope, the

b)
Fig. 13 Video pictures of the silica fiber extremity: a) before combustion and b)
during combustion.
190
H 60
• 40
Threshold level
IN in [C g 10 o ir, ~""~'""~ ^^f-''
Diameter classes in microns
Fig. 14 Drop size histograms of the spray burning in the combustion chamber.
field of view was larger: 12.8 x 17.1 mm2. The picture acquisition rate was set
at 2000 fps. The cinematographic pictures were digitized with the most accurate
system available at that time (110 pixels per millimeter).
The resulting size histograms obtained after processing the pictures are pre-
sented in Figs. 8 and 14. Each of these figures contains several histograms. Each
is related to a value of a threshold level of noise subtraction.
The two histograms (Figs. 8 and 14) cannot be compared. The conditions under
which the pictures were acquired in the two experiments (laser, field of view,
optics) make it impossible to conclude anything concerning the influence of the
combustion chamber on drop size.
The Sauter mean diameter was evaluated for the histograms obtained from
the experiment conducted in the combustion chamber. Because the shape of the
histograms is not influenced by the variation of thresholding values, the Sauter
mean diameter is also kept nearly constant, from 254 to 270 /xm for threshold
levels of thresholding from 90 to 40.
Liquid Phase Distribution

In the first experiment in the windowed combustion chamber, the cinemato-
graphic camera was adjusted to record a large field of view (27 x 85 mm2); this
afforded overall observation from the injection face to 50% of the chamber length.
Note that the field is limited by the window height (27 mm). The recording rate of
the camera was set at 8000 fps. The film obtained was very good. At times, droplets
can be followed on several consecutive pictures so that some drop velocities can
be evaluated (from 30 to 150 m/s) and secondary atomization of large droplets be
observed. These pictures were digitized at 50 pixels per millimeter. The result of
the liquid phase distribution measurement process is presented in Fig. 12. The up-
per part of the figure presents a picture built from 3000 cinematographic pictures.
The injector is on the left in this image.
Obviously, the dense zone of the spray is close to the injector. Note that there is
apparently a low density of liquid at the left of this zone because of the presence
Liquid probability of presence 12400picture?.

Axial olane Middle plane External plane
Fig. 15 Liquid phase distribution in several planes of the combustion chamber: axial
plane, middle plane, and external plane; field of view 27 x 45 mm2.
of an opaque cloud of nitrogen tetroxide which diminishes the visibility in this

region. Nevertheless, the size of the dense zone of the spray is not very large. In the
lower part of Fig. 12, the curve represents the liquid density on the horizontal axis
of the picture above. From left to right, the density increases as the visibility of the
liquid phase improves and then decreases because of dispersion and vaporization.
After the first experiment, exploration of the volume of the spray was conducted
by moving the sheet backward and forward from the vertical symmetry plane of
the injector. The field of view was reduced (27 x 45 mm2) to focus attention on the
higher liquid concentration zone and to have enough accuracy to evaluate liquid
Fig. 16 Visualization of the spray just before ignition.

concentration from the pictures. The pictures were recorded at 8000 fps, in five
parallel planes with 5 mm between each plane. One was the symmetry plane of
the injector.
The control recording of the tests with a video camera showed that visualization
was easier (liquid phase more visible) when the optic path in the combustion gases
was shorter, i.e., for the visualization planes close to the forward window.
For this reason, only the pictures of the three nearest planes of the forward
window were digitized. To build each of the three images presented Fig. 15 2400
pictures were needed. As the pictures were digitized at 110 pixels per millimeter
and the field of view was smaller, the description of the liquid phase distribution
of these images is more accurate than the one in Fig. 12. The injector is located at
the top of these pictures. The shape of the burning spray is rather conical, although
in cold flow simulations and transverse visualizations just before ignition, it has
the shape of a cross (see Fig. 16). To have a full three-dimensional view of the
burning spray, it would be interesting to carry out liquid repartition measurements
in transverse visualization.
Conclusion
The triad of laser-sheet visualization, fast cinematography, and image process-
ing appears to be a powerful tool for studying burning sprays in realistic conditions.
Although calibration of the method has not yet been achieved, the experiments
described in this paper demonstrate the feasibility of drop size and liquid phase
distribution measurements in burning conditions at atmospheric pressure. There is
no doubt that the same measurements are possible at higher operating pressures.
Future experiments could conceivably focus on drop-velocity measurements
by particle tracking velocimetry, study of several interacting sprays, and study
of a spray interacting with a film cooling, a wall, or even an acoustic field in a
windowed modulated exhaust motor similar to the blind ones used at ONERA.7
We believe that such experiments, not very expensive if conducted on small
devices, can produce data pertinent to rocket-motor manufacturers and users and
developers of computer codes as well as provide basic knowledge to fundamental
research in burning diphasic media.
References
ârrere, ML, Jaumotte, A., Fraeijs de Veubeke, B., and Vandenkerkhove, J., Rocket
Propulsion, Elsevier, New York, 1960.
2
Anon., "Pressure-fed Thrust Chamber Technology Program," NASA-CR-190666,1992.
3
Habiballah, M., Lourme, D., and Pit, F, "PHEDRE—Numerical Code for Combustion
Stability Studies Applied to the Ariane Viking Engine," Journal of Propulsion and Power,
Vol. 7, No. 3, 1991, pp. 322-329.
4
Dombrowski, N., and Fraser, R. P., "A Photographic Investigation into the Disintegration
of Liquid Sheets," Philosophical Transactions of the Royal Society of London, Series A:
Mathematical and Physical Sciences, Vol. 247, 1954, pp. 101-129.
5
Dombrowski, N., Masson, D., and Ward, D. E., "Some Aspects of Liquid Flow Through
Fan Spray Nozzles," Chemical Engineering Science, Vol. 12, 1960, pp. 35-40.
6
Dombrowski, N., and Hooper, P. C, "A Study of the Sprays Formed by Impinging
Jets in Laminar and Turbulent Flow," Journal of Fluid Mechanics, Vol. 18, pt. 3, 1964,
pp. 392-400.
112 R. LECOURT ET AL
7
Lecourt, R., Foucaud, R., and Kuentzmann, P., "Experimental Study of the Acoustical
Sensibility of LiquidFuel Rocket-Motor Injectors," La Recherche Aerospatiale, 1986-1985,
English edition, pp. 11-22.
8
Lecourt, R., and Foucaud, R., "Experiments on Stability of Liquid Propellant Rocket-
Motors," 23rd Joint Propulsion Conf.t AIAA Paper SI-1112, San Diego, CA, July 1987.
9
Foucaud, R., and Lecourt, R., "Experiments on Stability of Small Storable Liquid
Propellant Rocket-Motors," Propulsion and Energetics Panel 72nd B Specialists' Meeting,
AGARD CP No. 450, Bath, UK, October 1988, pp. 5.1-5.8.
10
Melchior, A., and Lecourt, R., "Combustion Stability Tests of a 6 kN Engine and
Comparison with Subscale Tests," 41st International Aerospace Federation Congress, IAF
90-251, Dresden, Germany, October 1990.
u
lngebo, R. D., "Size Distribution and Velocity of Ethanol Drops in a Rocket Combustor
Burning Ethanol and Liquid Oxygen," ARS Journal, Vol. 31, No. 4, 1961, pp. 540-541.
12
Levine, R.S., "Experimental Status of High Frequency Liquid Rocket Combustion
Instability," Tenth Symposium (International) on Combustion, 1965, pp. 1083-1089.
13
Harrje, D. T, and Reardon, F. H. (eds.), "Liquid Propellant Rocket Combustion Insta-
bility," NASA-SP-194, 1972.
14
McCreath, C. G., Roett, M. F, and Chigier, N. A., "A Technique for Measurement of
Velocities and Size of Particles in Flames," Journal of Physics E: Scientific Instruments,
Vol. 5, 1972.
15
Zung, L. B., and White, J. R., "Combustion Process of Impinging Hypergolic Propel-
lants," NASA-CR-1704,1971.
16
Lawver, B. R., "High Performance N2O4/Amine Elements 'Blowapart'," NASA-CR-
160272, 1979.
17
Holve, D. J., and Meyer, P. L., "In Situ Particle Measurements in Combustion Environ-
ments," Combustion Measurements, edited by N. Chigier, Hemisphere, New York, 1991,
Chap. 8.
18
Presser, C., Semerjian, H. G., Gupta, A. K., and Avedisian, C. T, "Combustion of
Methanol and Methanol/Dodecanol Spray Flames," 26th Joint Propulsion Conf., AIAA
Paper 90-2446, Orlando, FL, July 1990.
19
Hulka, J., and Makel, D., "Liquid Oxygen/Hydrogen Testing of a Single Swirl Coaxial
Injector Element in a Windowed Combustion Chamber," 29th Joint Propulsion Conf. and
Exhibit, AIAA Paper 93-1954, Monterey, CA, July 1993.
20
Lecourt, R., Foucaud, R., Lavergne, G., Berthoumieu, P., and Millan, P., "Hypergolic
Propellant Burning Spray Visualization by Laser-Sheet Method: Application to Droplet Size
and Liquid Concentration Measurements," 3rd Symposium on Experimental and Numer-
ical Flow Visualization, 1993 American Society of Mechanical Engineers Winter Annual
Meeting, New Orleans, LA, December 1993.
Chapter 5
Laser Sheet Visualization of Spray Structure

D. G. Talley* and A. T. S. Thambanf

Phillips Laboratory, Edwards Air Force Base, California 93524-7660
and
V. G. McDonell* and G. S. Samuelsen§
University of California, Irvine California 92717-3550
I. Introduction
T HE design of combustion devices is typically subject to a number of simul-

taneous and usually conflicting constraints. Required energy release rates
and efficiencies have to be achieved over the specified range of flow rates in a
combustion device that is manufacturable, maintainable, reliable, and affordable.
The devices have to be ignitable without producing unacceptable startup tran-
sients, combustion instabilities usually need to be prevented, and heat transfer to
combustion chamber walls cannot exceed material limits. Combustion products
also cannot lead to unacceptable wall deposits, and meeting legislated emission
standards has become increasingly more important. Optimizing these and other
constraints requires careful control of the combustion process, and one of the most
important ways designers have to affect this process is to control the mixing in
the combustion chamber. Prediction and measurement of the mixing distribution
is consequently of first-order importance.
One of the most important factors controlling mixing in spray combustion cham-
bers is the distribution of total liquid mass. This is currently most often measured
using arrays of collection tubes known as patternators. Although sophisticated and
highly refined patternators have been developed for rapid screening of candidate
injectors,1 the inherent intrusive nature of these devices limits their application
to time-averaged, nonreacting flows with simple aerodynamics. Planar laser sheet
imaging techniques, as optical patternators, have the potential to resolve most of
these problems and, therefore, development of these tools have been the subject
of our recent interest. In addition to development efforts, the authors had occasion
rights reserved.
*Research Physical Scientist, OLAC PL/RKFA. Member AIAA.
tU.S. Air Force Palace Knight Student, Pennsylvania State University. Member AIAA.
^Senior Research Scientist. Member AIAA.
^Professor of Mechanical, Aerospace, and Environmental Engineering. Associate Fellow
AIAA.
113
114 D. G. TALLEYETAL
to use these optical tools in a number of programs to provide direct, near-term

support of combustor design efforts. Some examples of these are presented later.
In what follows, optical patternation techniques recently used by the authors
are reviewed, with emphasis on near-term, practical applications. To begin with,
use of a relatively simple planar Mie scattering technique to probe the structure of
a pulsed automotive fuel injector is illustrated. Significant and useful observations
of spray structure are shown to be possible despite considerable uncertainty in
quantitatively relating the measured intensity distributions to mass distributions.
Following that, a more quantitative technique is described based on planar liquid
laser-induced fluorescence (PLLIF). The technique is applied first to a dilute

(optically thin) spray produced by a Research Simplex Atomizer (RSA), and
the results quantitatively compared with point measurements obtained using a
phase Doppler interferometer (PDI). The dilute spray results are also used to
evaluate various approaches that might be used under nonideal conditions that
can occur in practice where optical access may be limited. Finally, the PLLIF
technique is applied to a more dense spray application where useful results are
obtained despite significant optical attenuation. Some additional background is
presented first.
II. Background
Most fundamentally, mixing relates to the relative amounts of chemical species
and energy (temperature) that may be present. Although these can be determined at
a point, most often it is the mixing distribution over an area or throughout a volume
that needs to be known and controlled. For instance, in advanced low-emissions
gas turbine combustors, mixing is instrumental to the success of the concepts.
These concepts rely on combustion occurring at equivalence ratios away from
1.0 in order to minimize reaction temperatures and thereby minimize thermal
NO*. Each concept (e.g., lean direct injection, rich burn-quick mix-lean burn,
premixed/prevaporized) relies on combustion occurring throughout a volume at a
highly uniform equivalence ratio.
For determining mixing distributions, planar imaging techniques have certain
natural advantages over point measurement techniques such as Coherent Anti-
stokes Raman Spectroscopy (CARS). Essentially an array of simultaneous point
measurements, planar techniques not only directly measure two-dimensional mix-
ing distributions, they can also provide information about the mechanisms pro-
ducing the mixing. Planar measurements can be line of sight integrated, as in
conventional schlieren, or they can be spatially localized by interrogating the field
with a laser sheet, thereby providing information only from within the plane of
the sheet. The latter method has the additional advantage that spatially resolved
three-dimensional information can also be obtained if desired by collecting data
from more than one plane.
Partially due to the development of sensitive gated-intensified charge-coupled
device (CCD) cameras, planar laser sheet imaging techniques have now become
both powerful and practical. Signals can be detected from within the relatively
low-intensity levels of a laser sheet, as opposed to requiring all of the laser power
to be focused to a point (the resulting higher signal levels are one of the main
advantages of point measurement techniques). In addition, data rates can be made
high enough that three-dimensional snapshots can be made even in transient flows
by sweeping the sheet across the field much faster than characteristic flow times.
LASER SHEET VISUALIZATION OF SPRAY STRUCTURE 115
Patrie et al.2 characterized a turbulent acetone seeded air jet in this manner at
Reynolds numbers to 3000, and Dahm et al.3 further succeeded in obtaining such
snapshots as a function of time in a Reynolds number 6000 turbulent jet, resulting
in both spatially and temporally resolved four-dimensional information.
Planar laser sheet imaging techniques have been used extensively in gas flows
since approximately the early 1980s,4 but have been much less extensively used so
far in sprays due to the increased complexity of these flows. In spray combustion,
for instance, not only is the total fuel/oxidant distribution of interest, but the
relative proportions of these that exist in the condensed and gaseous states is of
interest well. Optical issues are also more complex. Nevertheless, rapid advances
are being made. PLLIF has been used by a several groups5'6 to visualize liquid fuel
injection in internal combustion engines by dissolving fluorescent dyes in the fuel.
Methods of measuring both gas and liquid concentrations are being developed,7
many of which make use of planar laser-induced exciplex fluorescence (PLIEF),8~1()
where the exciplex compound fluoresces at different wavelengths depending on
the phase. Exciplex methods have the potential to measure droplet temperature
as well.11 Other efforts have gone into exploring nonlinear optical effects such as
stimulated Raman scattering (SRS). These can be used inside droplets to produce
morphology dependent resonances (MDRs), essentially turning droplets into tiny
ring-dye lasers and leading to extremely accurate size information. The latter has
been used in both point (single droplet) and planar (spray) applications.12
When applied to sprays, planar imaging techniques can have several possible
emphases, each of which calls for distinctly different approaches. With sufficient
magnification, single-shot techniques can provide direct measurement of droplet
sizes. Two or more exposures provide measurements of droplet velocity in the
plane of the sheet, i.e., particle imaging velocimetry (PIV). Many exposures tend
to smear out the distinctions between droplets, instead producing more or less
continuous intensity distributions that can, in principle, be related to the total
average liquid mass distribution. This latter approach of averaging many images
is the basis of the methods to be described. We begin with relatively simple
application of the technique to investigate a pulsed automotive fuel injector.
III. Planar Mie Scattering Measurements

of a Pulsed Automotive Fuel Injector
A relatively simple planar Mie scattering technique has recently been used as a
tool to evaluate new concepts for automotive fuel injection. Potential applications
in assembly line quality control were considered as well. Results for a standard
off-the-shelf injector have been reported in the literature.13 The injector had the
geometrical configuration shown in Fig. 1. As the solenoid-actuated poppet is
lifted, back pressure forces the liquid to exit through the annular region between
the 1.07-mm-diam needle and the 1.3-mm-diam orifice. The liquid sheets are then
deflected into a conical shape by a pintle that had a total angle of 85 deg.
In order to obtain time-resolved images of the pulsed spray, the optical setup
illustrated in Fig. 2 was used. A laser beam from a continuous 4-W argon ion laser
was swept by a rotating mirror past an electronic shutter, spherical lens, and glass
rod arrangement. When the beam was in alignment with the rod, a planar sheet
of laser light was created that was approximately 0.5 mm thick at the viewing
location. The exposure time is determined by the time necessary for the beam to
sweep past the rod, which was less than 10 /xs. The injector was triggered by a
116 D. G. TALLEY ET AL.

Poppet
Seat
Orifice Diameter
1.3 mm
Fig. 1 Pulsed hollow cone injector.
photodetector (1) placed in the path of the sweeping laser beam at an angle 9 before
the rod. Ifco is the rotational velocity of the mirror, and re is an operator-adjustable
electronic time delay before the photodetector triggers the injection, then the time
td between the start of injection and the time of exposure is given by
rd = (0/co) - (D
where the start of injection is defined as the time the injector receives the signal
to open. A second photodetector (2) was placed in the laser sheet to allow direct
measurement of the actual exposure delay.
Because of the large mirror rpm, an electronic shutter was added to the beam
path to ensure the spray is imaged only once during each pulse. The shutter
was left open long enough (<!/&> s) to allow the beam to sweep just once from
photodetector 1 past the rod and then shut. The injector control circuit was gated
so that an injection signal was sent only when both the open-shutter and the
photodetector signals are present. As a result, the injector was fired only when an
exposure was to be made.
CW
Laser
Center Line
of Spray
Spherical
Photodetector 2 Lens
Rotating
Mirror
r*
Photodetector 1
Fig. 2 Optical setup for pulsed hollow cone injector tests.

Fig. 3 Evolution of the pulsed spray as a function of time; axial views at 40 psi
and 2-ms pulse duration in a stagnant environment: a) 2.1 ms, b) 2.9 ms, c) 3.9 ms,
and d) 4.7 ms.
Each image was recorded on 35-mm photographic film. Because the light en-
ergy contained in a single sweep of the continuous wave (cw) laser was very
small, multiple exposures were required. This was accomplished by opening the
shutter at a regular frequency low enough to ensure the previous spray was com-
pletely purged from the viewing area before the next injection could begin. The
number of required exposures ranged from several hundred to several thousand.
The images therefore represent ensemble averages at a fixed exposure time delay
from the start of injection. When necessary, the number of exposures was varied
to ensure the spray behavior of interest was adequately revealed. The injected
liquid was a naptha mixture that has similar properties to gasoline but is not as
flammable.
The evolution of the pulsed spray, at an injection pressure of 40 psi and a
pulse duration of 2 ms, is shown in Figs. 3a-3d as a function of time. The firing
frequency was 25 Hz. The fuel is injected downward, and here the laser sheet has
118 D. G. TALLEY ET AL
been oriented to cut the spray parallel to and through its axis, from right to left. A
length scale is provided by the outer sheath of the injector shown in the images,
which is 9 mm. The hollow cone nature of the spray is readily discernible. In many
cases (e.g., Fig. 3c), the images appear to exhibit a three-dimensional structure,
even though the illumination is only two dimensional. This is caused by secondary
reflections of light from the illumination plane to fluid particles outside that plane.
At this pulse duration, the spray volume was measured to be approximately 0.01
ml per pulse, or an average flow rate of 5 ml/s during the pulse. A flow rate
of approximately 7 ml/s was measured in the steady spray at the same injection
pressure.
Four main components of the spray are revealed by Fig. 3. The first component,
shown in Fig. 3a, might be called a leading cone. This is less dense than the main
cone that follows and tends to have a greater diameter. It is created during the
opening motion of the poppet.
The second component of the spray consists of a more dense main cone, as
shown in Fig. 3b. The atomized remnants of the leading cone can dimly be seen to
extend more than twice the distance downstream than the leading cone. The main
cone can also be seen in Fig. 3a as a more intense region emerging just behind the
leading cone. The main cone occurs during the period when the poppet is fully
open, and exhibits a cone angle consistent with the initial cone angle of the steady
spray. Because the bright intensities are confined to narrow regions in the image, it
is likely that a significant portion of the main cone is initially composed of sheets
and ligaments.
The third part of the spray consists of a trailing cone. It is formed during the
closing motion of the poppet and converges to an apex as the fluid flow decreases.
This is illustrated in Fig. 3c, which was taken just after closure of the poppet
and shows the apex after it has convected a small distance downstream. At this
injection pressure, most of the main cone appears to have been atomized by the
time the poppet closes. The three-dimensional light reflections indicate that at
the time of poppet closure, the trailing cone is largely composed of liquid sheets
and ligaments. Waves of large wavelength compared with the thickness of the
sheets can be seen to be formed as the trailing cone is formed, and eventually
contribute to the atomization process. The wave structures are very well defined
despite the fact that this image is an ensemble average of over 600 shots. The
wave structures are not symmetric and rotate with the injector as it is rotated.
Subsequent measurements of poppet motion suggest that these waves are most
probably caused by a side-to-side oscillation of the poppet as it closes.
The last part of the spray consists of leakage, as shown in Fig. 3d. The poppet
does not completely seal immediately after it closes, but allows enough fluid to
seep by to form a pencil stream. Figure 3d also shows the formation of a bulge
in the stream that, though Rayleigh-like, is most probably caused by a bouncing
action of the poppet. Subsequent visualizations reveal that the poppet eventually
seals and the pencil stream stops; it and the Rayleigh-like bulge eventually break
up into smaller droplets.
The three-dimensional structure of the spray was also probed by orienting the
laser sheet perpendicular to the axis and visualizing the spray from behind the
injector. A sample result is given in Fig. 4, which shows cross sections taken at 8,
14, and 32 mm from the injector at an exposure delay of 3.9 ms, corresponding
to the axial view given in Fig. 3c. A segmentation of the illumination pattern
around the circumference of the spray is revealed most clearly in Fig. 4c. Eight

Fig. 4 Cross-sectional view at 3.9 ms: a) 8 mm, b) 14 mm, and c) 32 mm.
or nine lobes can be identified. As with the wave structures discussed earlier, the
segmentation is exceedingly stable over a large number of exposures, and rotates
as the injector is rotated. Further visualizations revealed that this segmentation
emanates directly from the injector orifice, yet a microscopic examination of the
orifice did not reveal any manufacturing or other irregularities that seemed large
enough to explain this effect. It was hypothesized that the irregularities were due
to some fluid dynamic effect internal to the injector, rather than to manufacturing
defects or external instabilities in the spray flow.
In the preceding visualizations, no attempt was made to quantitatively relate
the observed intensity distributions to total liquid mass distributions, yet much of
direct value to the injector designer was learned. This illustrates that planar laser
sheet techniques can be useful even when quantitative sources of error are not
considered at all. In the next section, an alternative approach based on PLLIF is
discussed and quantitative issues are addressed more carefully.
IV. Optical Patternation Using PLLIF

In order to quantitatively relate the preceding Mie scattering intensities to the
total amount of liquid mass present, calculations of the scattering intensity as a
function of droplet size need to be performed. An example of such a calculation
is given in Fig. 5 for a specific refractive index and wavelength and a 90-deg
100.0
MIE SCATTERING
10.0- D CUBED
1.0-
0.1
1 .OE-2
I
1 .OE-3
1 .OE-4
LJ 1 .OE-5 90 Deg. Collection

488 nm wavelength
1.0E-6 1.330 real index
1.0E-7
1.0E-8
10 20 30 40 50 60 70 80 90 100
DROPLET DIAMETER, /xm
Fig. 5 Calculated Mie scattering intensities as a function of drop size compared to
expected values based on droplet volume.
collection angle. Two problems become apparent from this figure. The first is the
highly nonlinear nature of the curve that causes a given intensity to correspond
to more than one possible droplet size. The second can be seen by comparison of
the Mie curve with a second curve representing an expected intensity if scattering
were based solely on droplet volume. Two regimes can be discerned in this case.
Above a diameter of about 50 /xm, the Mie curve roughly follows the volume
based curve. This implies that if the nonlinearities are neglected, the averaged
intensity over many images will be directly proportional to the total volume, i.e.,
mass, present. This is exactly as is desired. Below 50 /zm, however, Mie scattering
becomes relatively more efficient than volume based scattering. This implies that
the recorded intensity distribution will overrepresent the smaller droplets.
Several approaches are possible to minimize these difficulties. The nonlineari-
ties can be reduced by averaging the signal over a range of collection angles, as
from a finite size lens. Alternatively, Hofeldt and Hanson14 have shown that such
averaging can be performed over a range of wavelengths from a broadband laser.
Overrepresentation of the smaller droplets can be resolved by adjusting the mag-
nification and camera sensitivity such that single droplet events can be recorded.
This would allow direct determination of droplet sizes for a true calculation of
total mass. In most cases, however, the magnification required would tend to re-
strict the visualization to only small regions of the spray, and the data handling
requirements would increase significantly.
In contrast, the attraction of PLLIF methods is that, by dissolving a fluorescent
dye into the fluid sprayed, the resulting signal should in principle become volume
based for all sizes, without inherent nonlinearities. To explore the potential of this
approach, the experimental setup illustrated in Fig. 6 was used. Methanol was used
as the spray fluid, into which fluorescein dye (Aldrich F245-6) is dissolved at a con-
centration of 0.1 mmol/1. The illumination source is an argon ion cw laser emitting
at 488 nm. For most of the experiments, the power was set to 1 W. The beam was
expanded and collimated into a parallel sheet through a cylindrical/convex lens

- Methanol doped with Fluorescein
Collimation lens
A Cylindrical lens
Ar+laser(488nm)
Optical filter
Video Camera lens
PC and Frame
grabber board Video monitor
0
•
O O
O O o Video signal
(Frame
1 grabber board)
RGB V
I
Fig. 6 PLLIF experimental setup.
arrangement. The images were acquired using an intensified 8-bit CCD camera
(Xybion Model ISG-250) with f/2.5 zoom lens (Canon Model V6x 18-2.5). The
camera was fitted with a filter (HOYA Y-52) in order to eliminate scattered laser
light while allowing fluorescence emissions to pass through. The camera gating
and gain were controlled through a camera control unit (Xybion Model CCU-01).
The images were acquired through a frame grabber (Data Translation Model DT
2853) installed on an /486-based personal computer. The software used for data
acquisition and some image processing was ImagePro Plus.
To illustrate the difference between scattering and fluorescence, a comparison
between the scattering and fluorescence signals for the same spray is shown in
Fig. 7. The atomizer was the hollow-cone research simplex atomizer discussed in
a following section. The left image in Fig. 7 is scattering and the right image is
fluorescence. Also shown is a comparison of the intensity levels 35 mm below the
injector. Large discrepancies are noted in the center region of the spray. Because
of the overrepresentation of the smaller fines in the spray, the scattering image
appears to show a large amount of mass in the center that appears to come from
nowhere. The fluorescence image shows that most of the mass is, in fact, located
closer to the edge of the spray, which makes more physical sense.
The comparison in Fig. 7 demonstrates the potential of PLLIF techniques;
however, quantitatively relating the recorded intensity to the total liquid mass
present requires careful consideration of several potential sources of error. These
are analyzed in further detail next.
V. Quantitative Issues in PLLIF

To a first approximation, the gray level G registered by a pixel element in
PLLIF can be related to the total liquid mass present as follows. If instrument
noise is neglected, the gray level will be proportional to the total number of
U
Rodiut /mm
Fig. 7 Comparison between Mie scattering and fluorescence images obtained at
90 deg to a laser sheet passing through the centerline of the spray.
photons received from some small solid angle in the optical field. Ideally, these
will originate only from within the region created by the intersection of the finite
thickness laser sheet with the small solid angle. This defines a volume 8 V that
can be approximated as that of a cylinder of length h and cross-sectional area 8 A,
i.e., 8V « h8A. The total amount of photons emitted from within this volume is
assumed to be proportional to the exposure time Af, the laser sheet intensity /,
and the average number of fluorescing molecules present in 8 V at any time during
the exposure. For constant and uniform dye concentration (implying the absence
of any vaporization), the latter will in turn be proportional to the total average
amount of liquid mass present at any time 8m. An average measured liquid mass
concentration can then be defined as pm = 8m/8V. If 8A is sufficiently small and
the statistical sample is sufficiently representative of the spray, this measure will
approximate the true local liquid mass concentration p(jc, y, z) that is related to
droplet statistics through the equation
Pm (2)
where D is the droplet diameter, p/ is the density of the pure liquid, and
n(D | jc, y, z) is the droplet size distribution, where n(D \ x, y, z)dD is the num-
ber of droplets per unit volume having sizes between D and D + dD, as a function
of the spatial position x,y,z. However, since the laser sheet intensity is typically
not uniform across the thickness of the sheet, e.g., having a Gaussian distribution
if the original beam is Gaussian, it is necessary to take into account an inten-
sity distribution /(z), where z is the distance across the sheet. Thus, when all of
the preceding factors are taken into account, the gray level registered by a pixel
element can be estimated as
n oo _
"6
where C is a constant of proportionality. If the laser sheet is thin enough that the
(3)
number density n(D) does not vary significantly over the thickness of the sheet,
then the integral in z can be factored out and replaced by hi, where / is the
average intensity. Gathering all knowns (h, C, p/, 8A, At) into a single constant
and making use of Eq. (2), the gray level registered by a pixel element is found
to be proportional to the average sheet intensity / and the total mass 8m = p8V
inside 8V or, equivalently, by absorbing 8V into the proportionality constant,
directly proportional to the dispersed phase mass density, i.e.,
G = K'Um = Kip (4)

It is evident from the preceding logic that a number of assumptions are required
to quantitatively relate the measured gray level to the total liquid mass. Other as-
sumptions not yet discussed are also involved in applying Eq. (4) to obtain two- (or
three-) dimensional mass distributions in a spray. Correction procedures have been
developed to account for some of these, and others are the subject of continuing
investigation and development. The current status is reviewed subsequently.
A. Secondary Scattering
Probably the single most important factor affecting the quantitative relationship
(4) is secondary scattering by droplets in the spray field. This produces at least
three different significant effects. First, scattering causes extinction of the laser
sheet, causing the average intensity / in Eq. (4) to essentially become unknown.
The basic extinction mechanisms (reflection, refraction, diffraction, absorption)
are illustrated in Fig. 8a; however, absorption is not a significant factor in any
of the present studies due to the small laser dye concentrations used. Second,
the scattered energy leaves the plane of the sheet and can subsequently cause
fluorescence by, and signals to be collected from, other molecules outside this
plane. This phenomenon is responsible for the three-dimensional appearance of
Fig. 3c, for example, and results in a positive bias in recorded gray levels. Third, the
emitted fluorescence itself can be attenuated by scattering from other particles in
the path of the detector, as illustrated in Fig. 8b, causing a potentially compensating
negative bias. These effects become relatively less important in dilute sprays, but
eventually limit the quantitative application of PLLIF, as well as most other optical
techniques, as the spray becomes more dense. At present, quantitative application
of PLLIF is limited to applications where secondary scattering effects are not
significant.
B. Droplet Lensing
Figure 8 also suggests an additional potential source of error. It would ideally be
desired that the droplet interior be uniformly illuminated so that the dye molecules
fluoresce uniformly and isotropically in all directions. In fact, the change in re-
fractive index causes the droplets to act as tiny lenses, concentrating illumination
ABSORPTION
REFRACTION
REFLECT
^"DIFFRACTION
a) O
IMAGE
FLUORESCENCE PLANE
FROM ARBITRARY
VOLUME WITHIN ILLUMINATED
b)
Fig. 8 Scattering phenomena in a spray: a) laser sheet extinction and b) extinction

of fluorescence.
in some regions and reducing it in others. This raises the possibility of an addi-
tional effect of droplet curvature, i.e., size. Nonspherical shapes could contribute
to nonisotropic emissions as well, resulting in an additional effect of shape and
orientation.
To examine these effects for spherical droplets, a study was carried out to quanti-
tatively relate the volume of a drop to the total measured intensity. A piezoelectric
drop generator was used to produce a monodispersed stream of drops, and the
CCD camera was synchronized with the drop generator to effectively freeze the
drop images, allowing multiple frame averages to be obtained. A long-distance
microscope was used to provide high magnification of the individual drops. A
laser intensity of 0.5 W was used with operation in the single-line mode at 488
nm. Approximately 85% of the laser power was maintained in the sheet. A laser
sheet approximately 300 /xm thick and 120 mm high was used, with the droplets
passing through the center of the sheet so that nominally uniform illumination
was used. The fluorescence signal was collected at a 90-deg angle to the sheet,
and the scattered laser light was eliminated with a filter. Methanol doped with 0.1
mmole/1 of fluorescein dye was used for the study.
Figure 9 shows examples of the images obtained. Nonuniformities due to droplet
lensing are quite evident. The illuminating light is focused to a point within the

RETICLE SCALE IMAGE
Fig. 9 Examples of fluorescent images from single droplets.
drop, leading to a local peak in the recorded gray level. Despite this, the correlation
between volume and the summed total intensity GT was examined, where
(5)
and the summation in Eq. (5) is performed over all pixels contained by the droplet
image. The results are shown in Fig. 10. These show that the relationship between
drop volume and integrated intensity remains essentially linear to within the
experimental error of the current system. Thus, it appears possible that droplet
lensing effects can be absorbed into the constant K in Eq. (4). Although these
results are extremely encouraging, it should be noted that only a limited range
of laser intensities, dye concentrations, viewing angles, and drop sizes have been
considered so far. Further investigation will reveal the true extent of the linear
response.
C. Laser Sheet Intensity Variations

If the originating laser beam is Gaussian, then the laser sheet will exhibit a
Gaussian variation not only across its thickness, but also across its width. Thus the
illumination intensity will be lower at the edges of the sheet than near the center. To
account for this effect, a quartz calibration cell filled with a methanol/dye solution
was used. The cell (L = 200 mm, W = 100 mm, H — 10 mm) was placed in
the path of the laser sheet (width approximately 90 mm) and positioned in the
camera's object plane. Multiple images were then acquired and averaged. Using
the image acquisition and analysis software, the averaged image was subject to a
low-pass filter to reduce noise. A histogram of intensity (over a rectangular region
within the boundaries of the test cell) was acquired and saved. This histogram
INTENSITY VS VOLUME
3e+5
2e+5
i 2e+5
1e+5
Oe+0 1e+6 2e+6 3e+6 4e+6 5e+6 6e+6 7e+6

VOLUME ftim3)
Fig. 10 Correlation of droplet volume and total integrated intensity; dashed lines
indicate confidence intervals of 90 and 95%.
Fig. 11 Uncorrected images of a) quartz cell and b) RSA spray vs corrected images
c) and d), respectively.
was then used to correct for the effects of sheet intensity variation. The results of
the correction are illustrated in Fig. 11 that shows images of the quartz cell and
the spray from the research simplex atomizer, both before and after the correction
procedure. The laser sheet profile correction is dependent on the laser power
and the position of the camera relative to the sheet, and hence the calibration
procedure just described has to be repeated whenever one of these parameters
changes. At some point near the edge of the sheet, the illumination will become
small enough that the signal will become comparable to the noise, and then the
given correction will mainly amplify the noise. Therefore, it is recommended that
the correction be applied only over the center part of the sheet where the signal is
acceptable.
D. Corrections for Perspective

In many instances, especially in practical applications, images may need to
be recorded from some direction other than perpendicular to the laser sheet. In
these cases, corrections for perspective will be required. This is accomplished by
performing affine transforms15 on the images. The coefficients for these transforms
were determined by viewing a square grid at the measurement location with the
spray turned off. Figure 12 illustrates correction of the trapezoidal image of the
Fig. 12 Image of the square grid a) before and b) after Affine transform.
grid, in this case viewed at a 23-deg angle, to its original square shape using the
afrme transform. The distortions evident in the images are caused not by optical
effects, but rather by wetting of the paper grid caused by drippage under the spray
nozzle.
E. Variations in Pixel Sensitivity

Corrections are also made for variations in pixel sensitivity by visualizing
a uniform intensity field and storing the resulting image. For the camera used
in these studies, these variations have so far been exceedingly small compared
to other error sources and have not been a factor in any of the visualizations
performed to date.
VI. Application to RSA Atomizer and Comparison with PDI

As a starting point, it was desirable to compare the PLLIF results to an already
well-characterized spray. To this end, a comparison was performed between PLLIF
and extensive phase Doppler interferometer (PDI) measurements already obtained
on an RSA spray.16
Direct comparison between PDI and imaging requires manipulation of the basic
information obtained since the fundamental basis for the two measurements differ.
The PDI measurement is based on the total number of drops that pass through a
region of space per unit of time (i.e., a temporal measurement). Imaging relies on
the fluorescence from a distinct population of drops at a give time (i.e., a spatial
measurement). The intensity of the fluorescence is essentially proportional to the
volume concentration of the spray [see Eq. (4)]. Because the PDI technique al-
lows for the joint measurement of drop size and velocity, however, the temporal
measurement can be converted into a corresponding spatial measurement by nor-
malizing the population of each size by the velocity of each size. The liquid mass
concentration can then be computed from the PDI data as
(6)
where N is the time averaged number density and D|O y is the spatial-weighted
TOP VIEW
RECEIVER
LASER SHEET
I PDI TRANSMITTER
Fig. 13 Experimental setup used for comparative phase Doppler and PLLIF study.
volume mean diameter. The error in this result is considerable given the variation
of TV and DW!S. For example, a 5-/zm variation in D$otS out of 50 /xm results in
33% variation in V. The number density is subject to large errors since it depends
on accurate compensation for probe volume variation with size, total counting
efficiency, and an accurate determination of velocity. In the present spray, an error
of 100% in N is not unreasonable based on interlaboratory comparisons.16
Figure 13 presents a schematic of the optical setups used for the comparative
study. Forward scatter at 30 deg off-axis is used for the PDI measurements. The
PLLIF images were obtained with the laser sheet passing parallel to and through
the centerline of the spray and the fluorescence was collected at 90 deg from the
sheet. It should be noted that although the same atomizer was used with conditions
duplicated as closely as possible, the PLLIF measurements and the earlier PDI
measurements were taken at separate times involving two different experimental
setups. Thus, the following comparisons reflect any variations in the spray itself
caused by the two setups, in addition to systematic errors associated with the
two techniques. Therefore the following comparisons potentially exaggerate the
differences between the two methods.
Figure 14 presents a comparison of the imaging results to PDI results at three
axial locations. No attempt is made here to compare absolute mass distributions,
pending further investigation of droplet lensing effects, that can affect the constant
K in Eq. (4). Instead, the distributions have been normalized to provide the relative
intensity distributions as a function of radial poisition.
The imaging results are based on an average of 50 instantaneous images. At
35 mm, error bars reflecting the variation Ep in p based on the variations £^30,.?
and EM in D-$Q v and N, respectively, taken from Ref. 16 are provided using
Eq. (7),
The agreement between the results obtained using the two techniques is gener-
ally satisfactory. Some discrepancies are observed near the centerline at Z = 50
mm. This is attributed to extinction of the fluorescence signal through the spray
by secondary scattering, causing a blurring effect. This effect worsens with in-
creased distance from the spray and is particularly noticeable at the centerline.
Corrections for this can, in principle, be applied in this case since the PDI results
provide information about the drop size distributions and concentrations through-
out the spray; however, this negates the quick interrogation which the imaging
offers.
Overall, the preceding results tend to support further development of PLLIF
techniques as a nonintrusive patternator, except when secondary scattering might
be important. This latter reservation was known from the outset; however, the con-
siderable time saved over tedious mapping from point measurements may in many
instances make the technique attractive despite some known secondary scattering.
Application of the technique in an even more dense spray will be illustrated later,
where useful information is obtained despite considerable scattering. First, how-
ever, the issue of using PLLIF in practical applications where optical access may
be restricted is addressed, and various techniques for overcoming this problem are
evaluated using the present (nearly dilute) RSA spray.
Z = 25mm
O PDI
— FLUORESCENCE
ct:
2 = 35mm
CO
5
Ld
a:
2 = 50mm
to
z
UJ
10 20 30 40 50
RADIAL POSITION, mm
Fig. 14 Comparison of PLLIF and PDI volume concentration for RSA spray.
VII. Conditions of Restricted Optical Access

In applying laser sheet imaging techniques, the optimum arrangement is typi-
cally one in which the laser sheet is oriented along the plane of maximum interest,
and the image is recorded from a direction perpendicular to the sheet. In practice,
however, optimum optical access may not be available or may be prohibitively
expensive to install. In high-pressure applications, for instance, the number and
size of windows is necessarily restricted, and the windows are often oriented
circumferentially around the combustion or spray chamber at fixed axial dis-
tances. In these instances, direct visualization of distributions perpendicular to the
spray axis, which most directly correspond to mechanical patternation, may be

problematic.
One alternative in such instances is to sweep the laser sheet across the field,
making appropriate Affine transform corrections for nonperpendicular viewing
perspectives as needed, and then constructing three-dimensional maps from the
resulting sequence of images. The desired visualization can then easily be recon-
structed from the three-dimensional map. As mentioned earlier, three-dimensional
reconstruction techniques have been used previously in gas phase PLIF,2 3 but ap-
pear not to have been attempted to date in multiphase PLLIF. To evaluate this ap-
proach, a three-dimensional, time averaged map of the RSA spray was constructed.
Because this spray is steady in the average, rapid data acquisition techniques such
as those described in Refs. 2 and 3 were not required. The laser sheet was ori-
ented parallel to the spray axis with the camera perpendicular to the sheet, and the
spray was traversed through the stationary laser sheet. A total of 32 images at a
9-/zs exposure each were taken at each location, and the average of these images
was stored. A total of 51 images parallel to the spray axis was acquired in this
fashion.
For comparison, images were also taken with the laser sheet perpendicular to the
spray axis at several axial locations. These images were viewed with the camera
at a 23-deg angle to the sheet, the maximum angle that physical limitations of the
spray chamber would permit. The depth of field was large enough that the entire
image remained in focus, and Affine transforms were performed to correct for
perspective as discussed earlier.
Intensity distributions taken from the RSA spray are shown in Fig. 15a and 15b,
in false color. Visualizations with the laser sheet oriented perpendicular to the
spray axis at three axial locations are given in Fig. 15a, corrected for the Gaussian
distribution of the laser sheet, differences in the pixel sensitivity in the CCD array,
3? mm
Fig. 15 a) Visualization of fouled RSA spray. Laser sheet perpendicular to spray axis
at three different axial distances and b) reconstruction of the visualizations in Fig. 15a
from 51 slices parallel to the spray axis.
and the camera viewing angle. A reconstruction of the same visualizations from the
series of images taken parallel to the spray axis, after having performed the same
corrections, are shown in Fig. 15b. Each was constructed from 51 horizontal strips
of intensity data from the parallel images, taken at the appropriate axial distance,
with intensities between the strips filled in by linear interpolation. Circumferential
irregularities reveal the presence of dirt particles in the atomizer. Cleaning later
resulted in a much more uniform circumferential distribution, as would normally
be expected from this atomizer; however, the nonuniform results are presented
here as a better test of the reconstruction technique.
The absolute intensity levels shown in Fig. 15 are not quantitative. Visualization
of the spray parallel to the spray axis encountered a much larger variation in
fluid density, and therefore in corresponding intensity levels, than visualizations
perpendicular to the axis. As a consequence, intensities parallel to the axis tended to
saturate the camera whereas intensities perpendicular to the axis did not, so camera
gain had to be adjusted between the two visualizations to compensate. Although
bothersome in the present case, this problem does not appear to be fundamentally
insurmountable. Use of a larger gray scale resolution camera would reduce it
considerably (the camera used in the present study only had 8-bit resolution), and
this combined with more careful adjustment of laser sheet intensity, camera gain,
and dye concentration in most cases should succeed in eliminating it entirely.
Because this problem is a factor in the present case, however, the following
comparisons will focus only on the geometric structures revealed. Therefore the
false coloring of the images was adjusted more or less arbitrarily to most clearly
bring these out.
Both Figs. 15a and 15b reveal the hollow cone nature of the spray, and both reveal
structures that propagate downstream. Despite some differences, the reconstructed
images approximately show the same structures as the original perpendicular
visualizations, for instance, structure A. Structure B is shown in Fig. 15a as
being a single structure, while in Fig. 15b it shows up as two structures. It is
probable that the reconstructed image is more accurate in this case. This is due
to the loss of resolution caused by correcting the perpendicular visualizations for
such a narrow viewing angle (23 deg) to an equivalent 90-deg viewing angle.
As mentioned earlier, the narrow viewing angle was necessitated by the physical
limitations imposed by the spray chamber. Another reason for believing that the
reconstructed image may be more accurate in this case can be seen by considering
structure C. It appears as a single structure at 37 mm in Fig. 15a, but lower down
at 55 mm and 73 mm in the same figure there are indications that it may actually
be two structures.
It is interesting to note that the three-dimensional reconstruction technique just
described is essentially independent of laser sheet extinction across the spray. This
is because extinction along a ray of light remains essentially the same whether that
ray is part of a horizontal plane or a perpendicular plane. Thus, although the tech-
nique preserves this error, it is unaffected by it. Other secondary scattering effects,
however, such as fluorescence outside the plane of the sheet and extinction of the
fluorescence by other particles, may be somewhat more variable, since this depends
on the sections of the spray through which the emission is viewed, and therefore
would be dependent on viewing angle. Overall, however, even though quantitative
comparisons of intensity levels could not be assessed here, it can nevertheless be
concluded from the preceding results that three-dimensional reconstruction has
considerable potential to provide valuable information about spray structure in

cases of restricted optical access.
VIII. Simplified Approaches Using Local Self-Similarity Assumptions

The three-dimensional reconstruction technique just described requires a large
number of planar images to be taken throughout the spray field, which can be
tedious and time consuming. Simpler alternatives may be possible depending on
the spray and the kind of information required. For instance, if a distribution per-
pendicular to the spray axis is desired, as in mechanical patternation, but optical
access does not allow this to be directly visualized, then one alternative would
be to incline the laser sheet obliquely at some angle to the spray axis, as in
Fig. 18a. This has the advantage that the oblique plane directly contains infor-
mation about the perpendicular distribution, while at the same time remaining
visible with the existing optical access. The resulting visualization can often be
immediately useful even in its raw form, without further processing.
If the approach just described is further combined with modeling, oblique
visualizations can then be corrected to compute the visualization actually desired.
An attractive feature here is that these corrections should become less and less
sensitive to modeling errors as the oblique plane approaches the actual plane
of interest. In particular, for the case of obtaining distributions perpendicular to
the spray axis, even an assumption of self- similarity, which in most cases would
probably not hold over the entire spray, becomes identically true, locally, as
the oblique angle approaches the perpendicular. The use of local self- similarity
assumptions (LSSA) is further explored later to transform oblique visualizations
to equivalent perpendicular distributions.
As was discussed earlier, the gray level registered by each pixel in the CCD
camera is to a first approximation proportional to the dispersed phase density p
that in turn can be related to the droplet number density by the relationship given
in Eq. (2). Local self-similarity implies that distributions perpendicular to the axis
are in some sense geometrically similar except for a stretching factor g(z), as
illustrated in Fig. 16. Mathematically, this implies that when transformed into the
similarity coordinates
z=z (8)
a function /(;c, ;y, z) that is self- similar should reduce to the form
i) (9)
where f ' ( z ) adjusts the magnitude of F(£, 77) as a function of i. Spray fields
can be held to this relationship at various levels of detail, for instance, p(;c, y, z)
or n(D | jc, y, z)- Fortunately, for present purposes of determining total mass
distributions, this relationship need only apply to the least restrictive of these,
namely, p(x, y, z) -> p'(z)P(f , f?).
The function p'(z) will, in general, depend on the amount of vaporization and
droplet deceleration in the spray. In cold-flow applications where vaporization is
negligible, a first estimate of p'(z) can be made by considering the total mass flow
m across a plane perpendicular to the spray axis,
oo /»oo poo /»oo —.
I I I -TP^Vzn(D,Vz\x,y,z)dDdVzdxdy (10)
/ -oo ./-oo .70 .70 °

Fig. 16 Similarity coordinates.
where Vz is the velocity parallel to the spray axis and n(D \ jc, y, z) has been
further expanded by means of the relationship
CO
n(D, Vz \x,y,z)dVz (11)

/.
If n(D, Vz \ x, 3;, z) is also held to be self-similar, then Eq. (10) can be written in
similarity coordinates as
oo /»oo />co roo _
/ / / -PD3VZN(D, Vz l^ (12)
/ -oo ./-oo JO JO &
Since m is independent of z in steady cold flow, it must therefore be the case that
n'(z) = g~2(z), and, correspondingly, through Eqs. (11) and (2), p'(z) = g~2(z).
Thus, to a first approximation Eq. (4) becomes
G(x, y, z) = KIp(x, y, z) = (13)

Equation (13) provides a means of correcting gray levels during a similarity
transformation of a recorded image. The assumption that n(D, Vz \ x, y, z) is
self-similar, which led to Eq. (13), in many cases may not be far from true. In
cold-flow applications where drop sizes do not change, it holds, for instance, if
droplet velocities are assumed to remain relatively constant over the axial distance
spanned by the oblique laser sheet. This may not be so unreasonable if the oblique
angle is narrow enough, especially when it is recalled that most of the mass in the
spray is contained in the larger droplets that carry the most inertia. In any event,
similarity in n(D, Vz \ x, y, z) is only a sufficient condition, not a necessary one,
for similarity in p(x, y, z).
Finally, the function g(z) can be measured by tracking any structure in the
similarity field that would correspond to the same coordinates (£,??), as shown in
Fig. 16. For many sprays, the most obvious choice would be the growth of the
0,0,0
Xf, Yf, Zf
Xi,Yi,Zi
Fig. 17 Simplified LSSA transform for constant cone angles.
boundary of the spray cone. Thus, in addition to the oblique image, the preceding
approach requires one other visualization, namely, one providing the cone angle,
for measurement of g(z).
To evaluate this approach, the preceding RSA spray was also investigated with
the laser sheet offset by 20 deg from the perpendicular to the spray axis. A two-step
approach was taken as illustrated in Fig. 18. The raw image of Fig. 18a was first
corrected for perspective using an affine transform, resulting in the visualization
of Fig. 18b. An LSSA transformation was then performed to reconstruct the
intensity distribution perpendicular to the axis, as in Fig. 18c. For the RSA atomizer
used, the cone angle over the region investigated was nearly straight. Therefore,
g(z) was approximated as a linear function. For this linear case, the similarity
transformation reduces to projecting each point (*/, v/, zi) on the oblique plane to
a point (XQ, yo, £o) on the perpendicular plane along a line that intersects with the
apex of the spray cone on the spray axis, as illustrated in Fig. 17. The projected
coordinates on the perpendicular plane are given by
*o = (14)
where ZQ is some point on the z axis where a perpendicular distribution is to
be reconstructed using LSSA. Following Eq. (13) with g(z) = bz, where b is a
constant, the gray level intensity correction becomes
(15)
The final similarity transformed image of Fig. 18c is compared with the actual
perpendicular distribution reconstructed from slices parallel to the spray axis in
Fig. 18d. Figure 18d corresponds to Fig. 15b discussed in the preceding section,
but unlike the case in Fig. 15, the intensity comparisons between Figs. 18c and
18d are quantitative in this case.
Figure 18 shows that the comparison between the LSSA reconstructed image and
the actual perpendicular distribution is superior to either the original untransformed
image or the affine transformed image, both in terms of intensity magnitudes and
relative geometrical relationships; however, differences exist as well. In the present
case, the magnitude of these differences can be assessed by comparison with an
actual perpendicular distribution. In practice, however, if the actual perpendicular
distribution were available, one would not bother with an LSSA reconstruction
I
\ A
b)
a) d)
Fig. 18 Comparison of LSSA reconstruction to three-dimensional reconstruction:

a) raw image taken at 20-deg viewing angle, b) affine transformed image, c) LSSA
transformed image, and d) reconstrucdon from 51 slices parallel to the spray axis
(Fig. 15b).
at all. The question therefore naturally arises of how the reliability of LSSA
reconstructions can be assessed in the absence of such information. A potential
answer may lie in performing one additional reconstruction, for example from
both a +20-deg and a —20-deg angle. Then confidence in the LSSA reconstruction
might be assessed by comparisons between the two reconstructed images, with
the expectation that these two would probably bracket the true image. Such an
approach has yet to be evaluated. For the present, however, the given results
indicate that oblique visualizations, with or without similarity reconstructions,
offer an important alternative for providing valuable near-term information about
spray structure when optical access is restricted.
IX. An Optically Thick Application

The mass flow rate in the preceding RSA spray examples was approximately 10
Ibm/h, resulting in a relatively dilute, physically small spray where the dye/laser
combination resulted in an optically thin medium with minimal secondary scatter-
ing. In this example, the spray produced by the combustor dome and fuel injector
assembly shown in Fig. 19 was characterized at a flow rate of 80 Ibm/h, where
approximately 68% of the incident light was attenuated by the spray. Anticipating
that this might be sufficient to significantly affect the images obtained, the opti-
cal setup was modified to produce two equal intensity laser sheets which passed
through the spray from opposite directions, as illustrated in Fig. 20.
Fuel Injector
Primary Swirler
Secondary Swirler
Fig. 19 Schematic of GE Aircraft Engines CFM-56 swirl cup.
Flat Mirror
Collimating Lens (Acrylic)
Cylindrical Lens
Beam Steering Device
Fig. 20 Optical setup for two opposed sheets.

Line Profile Data

Image acquired with 1 laser sheet. 0 50 100 150 200 250 300 350 400
Sheet is from right to left. Distance, pixel
Line Profile Data
100 150 200 250 300 350 400

image acquired with 2 laser sheets. Distance, pixel
Sheets are from both left and right.
Fig. 21 Comparison of line profile data from CFM56 swirl cup configuration 1 at
80 Ibm/h for one laser sheet and two laser sheets; images are corrected for laser sheet
intensity variation.
CONF #1 CONF #3
Fig. 22 Images acquired for two CFM56 swirl cup configurations at 80 Ib/h with
two laser sheets from left and right; images are corrected for laser sheet intensity
variation.
To evaluate the effect of attenuation, two images were obtained, one with a single
laser sheet and one with two opposed overlapping laser sheets. These results are
shown in Fig. 21. Immediately to the right of the images are line profiles taken
from the location indicated in the images. The top image clearly shows the impact
of attenuation. When two opposed sheets are utilized, this problem is greatly
diminished, as illustrated in the lower image. Intensity distributions at the tops
and bottoms of these images appear to suggest departures from overall circular
spray patterns. In fact, these are caused by the low illumination intensities at
the edges of the laser sheet that causes the corrections for laser sheet intensity
variations to excessively amplify noise, as discussed earlier.

In this case, the objective was to assess the relative width of the spray and relative
volume at the center of the spray for two hardware configurations. As shown in
Fig. 22, these particular features could be characterized despite the shortcomings
associated with the optically thick nature of this spray.
X. Summary and Conclusions

Laser sheet visualization techniques, for years applied primarily in gaseous
applications, are emerging as potentially powerful tools in multiphase applications
as well. Emphasis has been given here to the application of these tools as optical
patternators to nonintrusively probe the planar and volumetric distribution of
total liquid mass in a spray—one of the most important factors that determine the
performance of spray combustion devices. Despite potentially complex underlying
physics that have yet to be fully addressed, these techniques have demonstrated
the ability to provide much information that is of direct, near-term, practical value.
The use of both Mie scattering and PLLIF techniques has been discussed.
Attention has been given to various issues in quantitatively relating measured
intensity distributions to total liquid mass distributions, along with procedures to
correct for some of the errors. Examples were also given to illustrate that much
information of practical value can be extracted even when quantitative issues have
not been resolved. Some key points are now summarized.
Mie scattering alone can provide significant information regarding the qualita-
tive structure of the spray. Issues that need to be resolved in providing quantitative
information include the nonlinear dependence of scattering intensity on droplet
size, and relatively greater scattering efficiency by small droplets, causing these
to be overrepresented.
The essential attraction of PLLIF techniques is that the signal is in principle
directly proportional to volume for all sizes, without the inherent nonlinearities
of Mie techniques. Initial comparative results with PDI measurements remain
encouraging.
Many complex issues dictate the extent to which PLLIF provides quantitative
results. Steps have been taken to account for some of these, including the effects
of perspective, droplet lensing, and variations in laser sheet intensity.
Multiple scattering effects are probably the single most important limitation
in quantitatively applying PLLIF to practical sprays. Laser sheet attenuation ef-
fects can be reduced to a certain extent by introducing overlapping sheets from
different directions. It may also be possible to use laser power/dye concentration
combinations that saturate the fluorescence inside the droplets, in principle caus-
ing the fluorescent signal to become independent of the laser sheet intensity. These
solutions, however, do not address attenuation of the fluorescent signal itself.
In practical applications where optical access may be restricted, three-dimen-

sional reconstruction techniques can be used to provide the required visualization
that is not directly available. When time or other circumstances allow onjy a few
visualizations, modeling assumptions can be used if the available visualizations
are not too far from the desired visualization. The use of an LSSA was examined
here with encouraging results.
The large range of drop sizes in a polydisperse spray, and even larger volume
range, may produce a very large dynamic range in signal intensity; 8-bit cam-
era resolution may not be sufficient. The use of 12- or even 16-bit cameras is
recommended.
The ability of laser sheet methods to provide valuable information about spray
structure despite the presence of quantitative uncertainties provides a strong im-
petus for continued development of the methods.
Acknowledgments
Financial support was received at various times from Ford Motor Company,
General Electric, Horiba, and the Phillips Laboratory (U.S. Air Force). The mate-
rial in this paper is the accumulation of work performed over a period of years by a
number of different individuals. The authors would like to gratefully acknowledge
the contributions of Yun-Chi Lin, Mary Morris, Mohammad Hassani, Steven Lee,
and T. Iguchi.
References
^cVey, J. B., Russell, S., and Kennedy, J. B., "High Resolution Patternator for the
Characterization of Fuel Sprays," Journal of Propulsion and Power, Vol. 3, 1987, pp. 202-
209.
2
Patrie, B. J., Seitzman, J. M., and Hanson, R. K., "Planar Imaging For 3-D Flow Vi-
sualization," Twentieth International Congress of High Speed Photography and Photonics,
Victoria, British Columbia, Canada, 1992.
3
Dahm, W. J. A, Southerland, K. B., and Buch, K. A., "Four-Dimensional Laser Induced
Fluorescence Measurements of Conserved Scalar Mixing in Turbulent Rows," Fifth Inter-
national Symposium on Application of Laser Techniques to Fluid Mechanics, Paper 1.1,
Lisbon, Portugal, July 1990.
4
Hanson, R. K., "Combustion Diagnostics: Planar Imaging Techniques," Twenty First
Symposium (International) on Combustion, Combustion Inst., Pittsburgh, PA, 1988,
pp. 1677-1691.
5
Lawrenz, W., Koler, F, Meier, F, Stolz, W., Wirth, R., Bloss, W. H., Maly, R. R.,
Wagner, E., and Zahn, M., "Quantitative 2D LIF Measurements of Air/Fuel Ratios During
the Intake Stroke in a Transparent SI Engine " International Fuels and Lubricants Meeting
and Exposition, Society of Automotive Engineers, SAE Paper 922320, Oct. 1992.
6
Bardsley, M. A. E., Felton, P. G., and Bracco, F. V, "2-D Visualization of a Hollow-
Cone Spray in a Cup-in-Head, Ported, I. C. Engine," International Congress and Exposition,
Society of Automotive Engineers, SAE Paper 890315, Detroit, MI, March 1989.
7
Bazile, R., and Stepkowski, D., "Measurements of the Vaporization Dynamics in the
Development Zone of a Burning Spray by Planar Laser Induced Fluorescence and Raman
Scattering," Experiments in Fluids, Vol. 16, 1994, pp. 171-180.
8
Shimizu, R., Matumoto, S., Furuno, S., Murayama, M., and Kojima, S., "Measurement
of Air-Fuel Mixture Distribution in a Gasoline Engine Using LIEF Technique," International
Fuels and Lubricants Meeting and Exposition, Society of Automotive Engineers, SAE Paper
922356, San Francisco, CA, Oct. 1992.
9
Melton, L. A., and Verdieck, J. F., "Vapor/Liquid Visualization in Fuel Sprays," Twen-
tieth Symposium (International) on Combustion, Combustion Inst., Pittsburgh, PA, 1984,
pp. 1283-1290.
10
Senda, J., Fukami, Y., Tanabe, Y., and Fujimoto, H., "Visualization of Evaporative
Diesel Spray Impinging Upon Wall Surface by Exciplex Fluorescence Method," Interna-
tional Congress and Exposition, Society of Automotive Engineers, SAE Paper 920578,
Detroit, MI, Feb. 1992.
1
Murray, A. M, and Melton, L. A.,''Fluorescence Methods for Determination of Tem-
perature in Fuel Sprays " Applied Optics, Vol. 24, No. 17, 1985, pp. 2783-2787.
12
Serpunglizel, A., Swindal, J. C., Chang, R. K., and Acker, W. P., "Two-Dimensional
Imaging of Sprays with Fluorescence, Lasing, and Stimulated Raman Scattering," Applied
Optics, Vol. 31, No. 18, 1992, pp. 3543-3550.
13
Talley, D. G., Lin, Y. C., and Morris, M., "2-D Laser Sheet Visualization of a Pulsed
Hollow Cone Spray: Stagnant and Simulated Two-Stroke Engine Environments," Atomiza-
tions and Sprays, Vol. 1, No. 1, 1991, pp. 89-112.
14
Hofeldt, D. L., and Hanson, R. K., "Instantaneous Imaging of Particle Size and Spatial
Distribution in Two Phase Flows," Applied Optics, Vol. 30, No. 33, 1990, pp. 4936-4948.
15
Russ, J. C., The Image Processing Handbook, CRC Press, Boca Raton, FL, 1992.
16
McDonell, V. G., Samuelsen, G. S., Hong, C. H., Lai, W. A., and Wang, M. R., "Inter-
laboratory Comparison of Phase Doppler Measurements in a Research Simplex Atomizer
Spray," Journal of Propulsion and Power, Vol. 10, 1994, p. 402.

Chapter 6
Use of Fluorescence Methods for Measurement

of Droplet and Vapor Temperatures
Lynn A. Melton*
University of Texas at Dallas, Richardson, Texas 75083-0688
Nomenclature
E* = exciplex, bound only in excited state
G = organic molecule, may be M, in its ground electronic state
M = fluorescent organic molecule in its ground electronic state SQ
unless superscripted
M-G = reaction product in which M and G are chemically bound
(M/GY = E*
S = singlet state
T = triplet state
Subscripts
0,1,2 = numbering of electronic states within an constant spin
manifold beginning with the lowest state; So, S\, 82
are three lowest singlet states
Superscripts
3 = triplet state (precedes symbol)
* = excited electronic state
+ = excited vibrational levels of lowest electronic state
Introduction
S PRAY injection provides a compact means for the rapid introduction of

large quantities of fuel into a combustion chamber. However, the injection of
liquid fuel from one or several nozzle sources brings its own problems: the ideal
precombustion state is to have a homogeneous mixture of fuel vapor and air and
yet the injected fuel enters the combustion chamber as a liquid which is localized
in a spray. The momentum of the spray drives the mixing processes. Some spray
rights reserved.
Trofessor of Chemistry, Department of Chemistry.
143
144 LA. MELTON
droplets become virtually isolated and are surrounded by hot, nearly fuel-free air,
while others nearer to the core of the spray experience an environment in which
thermal diffusion from the hot outer regions and evaporation into a nearly fuel
saturated atmosphere are important. In both cases, partial evaporation must occur
before a combustible fuel/air mixture can be formed. Fuel vapor pressures, key
parameters in determining droplet evaporation rates, depend exponentially on the
liquid temperature. Thus, direct measurement of the temperature of the liquid,
either in isolated droplets or in full sprays, provides important information about
the thermal transport processes within sprays. Such understanding, coupled into
appropriate numerical codes for the prediction of aggregate spray properties, can
lead to more effective design of spray injectors and combustion chambers.
The measurement of temperature within a dynamic spray environment is diffi-
cult. Thermocouples provide point measurements and may distort the local flows
significantly. Laser light scattering techniques, such as those used in the for-
ward scattering1 (Malvern) and phase Doppler2 (Aerometrics) instruments, have
provided much significant information on droplet number densities, size distri-
butions, and velocity distributions but have been virtually insensitive to droplet
temperature. Recent modifications to the phase Doppler techniques (rainbow scat-
tering) may lead to temperature information,3'4 but these techniques will remain
fundamentally point or line-of-sight techniques.
Fluorescence methods, as described in this chapter, offer the possibility of
minimally perturbing, real-time, two-dimensional measurements of temperature
within droplets or sprays. A significant weakness of the fluorescence methods is
their sensitivity to the environment, particularly to quenching by oxygen. This
chapter will describe the photophysics on which these fluorescence methods are
based, assess their strengths and weaknesses, and review a series of droplet,
spray, and vapor jet experiments in which fluorescence methods have been used
to measure the temperature of the liquid and/or vapor phase. OH fluorescence is
widely used to determine the temperature in flames,5 but this paper will concentrate
on methods appropriate for characterization of the precombustion environment and
will not address OH fluorescence methods. It is hoped that this review will lead
to more effective use of these powerful fluorescence methods in fundamental and
applied droplet and spray studies.
Experimental Methods
Photophysics
The photophysical processes which commonly occur in and/or between organic
molecules are illustrated in Fig. 1. The matrix elements responsible for the pho-
ton absorption and emission processes are commonly taken to be temperature
independent. The absorption (or excitation) spectra themselves, however, may
be temperature dependent, since the population of vibronic levels in the ground
and excited states is temperature dependent. The quantum yield for fluorescence,
i.e., the fraction of excited states formed which actually fluoresce, is in general
temperature dependent since the rates of intersystem crossing (to populate triplet
states, which are weakly emitting at best) and internal conversion (to form highly
excited vibrational states of the ground electronic state) are usually temperature
dependent. The collision-induced processes, quenching, chemical reaction, and in
special cases, excited state complex (exciplex) formation, are in general tempera-
ture dependent in both the liquid and gas phases. The temperature dependence may
USE OF FLUORESCENCE METHODS 145
COLLISIONAL QUENCHING OF EXCITED STATES
ENERGY TRANSFER
M* + G INTERSYSTEM CROSSING
INTERNAL CONVERSION
PHOTOREACTION
EXCIPLEX
ABSORPTION
FLUORESCENCE
V
M + G M G
Fig. 1 Photophysical processes commonly occurring in and/or between fluorescent
organic molecules M.
result from changes in the ease of approach of the two molecules, either through
changes in the mean velocity in the gas phase or through temperature-dependent
(viscosity-dependent) diffusion coefficients in the liquid phase. It may also result
from the presence of an activation energy barrier in a reaction channel or, in the
case of exciplexes, thermal equilibrium between the excited organic molecule M*
and the exciplex E*. In the development of effective fluorescent thermometers, it
is not necessary to know the detailed rates for all of these processes but, rather, to
know that they are potentially present and to incorporate appropriate procedures
into the protocols for calibration and measurement.
It is not sufficient to describe the photophysics of a fluorescence thermometry
system in isolation from the droplet and spray experiments for which it is intended.
Heating droplets have transient and inhomogeneous temperature fields; differential
evaporation of the dopants and solvent (fuel) results in spatial- and time-dependent
concentrations of the fluorescent dopants. Droplets act as lenses, and as a result
the fluorescence which reaches the optical collection system does not sample the
droplet interior uniformly. In the discussions which follow, strengths and weak-
nesses of the fluorescent thermometry systems will, when possible, be discussed
in terms of both the fundamental photophysical processes shown in Fig. 1 and in
terms of the adaptation of those photophysical processes to realistic heating and
evaporation processes for isolated droplets and sprays.
For virtually all excited organic molecules, M* or E*9 oxygen is a very effective
quencher, and nitrogen, the inert gases, and alkanes are very ineffective quenchers.
Thus, experiments in which the fluorescent dopants are dissolved in alkane mix-
tures, purged of dissolved oxygen, and used in droplet or spray experiments into
146 LA. MELTON
oxygen-free nitrogen or inert gases may be regarded as unaffected by quenching.

Unless the experimental conditions can be set so that the oxygen concentration in
the liquid remains constant, however, the fluorescent thermometry systems can-
not give reliable results in oxygen-containing environments. For example, in a
fluorescence thermometry experiment in which a droplet falls into heated air, the
concentration of dissolved oxygen in the droplet will vary both in space and in
time, and the observed intensities will not be correctly interpreted as temperatures
by comparing them with calibration results obtained under oxygen-free or under
air-equilibrated conditions. Similar reasoning applies to droplets within sprays.
Unless specifically stated to the contrary, the techniques described subsequently

are subject to this restriction.
Fluorescence Thermometry Systems

Table 1 lists the systems discussed in this chapter and summarizes their photo-
physical bases, strengths/weaknesses, and apparatus requirements. It is intended
as a guide to the diversity of fluorescent systems which have been used for the
measurement of temperature. Potential users should not reject systems solely
upon the information in Table 1, since the particular elements of an engineering
experiment can affect the use of a system.
The temperature dependence of the fluorescence quantum yield can be used
for thermometry. Zhang6 developed and used a fluorescent thermometer which
exploited the decrease in the fluorescence quantum yield of Rhodamine B with
increasing temperature over the range 20-185°C. The preheated 52.5-/X water
droplets fell into quiescent air, and the intensity of the fluorescence, measured as
isolated single droplets traversed the beam of an argon ion laser, was used to infer
the temperature of the droplets within +/—3°C. The use of air apparently did not
bias the results, although the potential for bias is there. In order to make reliable
temperature measurements, it was necessary that the intensity and optical structure
of the argon ion laser beam remain constant, a condition which was difficult to
achieve under these laboratory conditions and which would be virtually impossible
to achieve under full spray conditions. The other systems discussed in this review
depend, in one form or another, on the shapes of the emission spectra and are not
subject to the constraint of constant laser beam intensity and structure.
Melton7 and Murray and Melton8 developed a series of fluorescent thermometry
systems based on the photophysics of exciplexes, particularly those for which the
E* emission is significantly red shifted with respect to the M* emission. In these
exciplex fluorescence thermometry systems the monomer M* and the exciplex £*
both fluoresce, and since the ratio of M * concentration to E* concentration is tem-
perature dependent, either through the viscosity of the solvent (low temperature,
kinetic limit) or through the equilibrium constant (high temperature, thermody-
namic limit), the ratio of the intensity of emission from E* to that from M* can
be calibrated as a fluorescence thermometer. The exciplex systems are currently
limited to use in hydrocarbon solvents because the dopants are more soluble in
nonpolar solvents.
Stufflebeam9 confirmed that, as predicted for bulk non-evaporating liquids,
it was indeed possible to use an exciplex fluorescence thermometer, based on
the intermolecular excited state dimer (excimer) formed between excited state
pyrene and ground state pyrene in decane, to measure the temperature within
+/-1°C over the range 25-9PC. However, pyrene (nbp = 350°C) is much less
Table 1 Fluorescent thermometry systems
Measurement/
System name photophysics/phase Example,1Ref. Strengths/weaknesses
6
One band Intensity/decrease Rhodamine B/water Requires extremely stable
of quantum excitation source
yield/liquid
Two band Ratio of E*/M* Pyrene/alkanes1,9,10,19 +/—1°C sensitivity over
(inter- band intensities/ 20-120°C range, affected

molecular) shift of equili- by oxygen quenching and
brium/liquid concentration changes,
stable excitation source
not required
Two band Ratio of E*/M* PYPYP/alkanes10-20 +/—3°C sensitivity over
(intra- band intensities/ 140-400°C range, affected
molecular) shift of equili- by oxygen quenching but
brium/liquid can be independent of con-
centration changes,
depends on intensity
ratio, stable excitation
source not required
Band shift Ratio of two PYPYP/alkanes13-14 +/—10°C sensitivity over
parts of band/ 20-200°C range, weakly
shift of band/ affected by oxygen quen-
liquid ching (<1 atm air), can be
concentration independent;
stable excitation source
not required
Absorption Ratio of absorp- Eu3+(EDTA)/water15 +/—1°C sensitivity over
tion strengths/ 20-100°C range, not
shift of popu- affected oxygen quenching
lation in ground or concentration changes,
excistate atomic stable excitation source
levels/liquid required, not appropriate
for rapid measurements
Anomalous Ratio of two Pyrene 16,17 +/-50°C sensitivity over
pyrene bands/increase 300-1700°C, use demons-
emission in S2 emission/ trated in flame, stable
vapor excitation source not
required
Lifetime Fluorescence Naphthalene18 +/—15°C sensitivity over
imaging lifetime/decrease 20-450°C range, sensitive
in fluorescence to oxygen quenching,
lifetime/vapour stable excitation source
not required, complex
detection system required
148 L. A. MELTON
Fig. 2 Inter- and intramolecular pyrene excimers.
volatile than decane (nbp = 175°C); were this thermometry system to be used
in falling droplets, evaporation of the more volatile decane would result in an
increase in the pyrene concentration with time and an unknown deviation from
the calibration of E*/M* intensity ratio vs temperature which was obtained under
constant concentration conditions.
In response to this perceived problem, intramolecular excimer systems were ex-
amined as fluorescence thermometers.10 Figure 2 shows inter- and intramolecular
pyrene excimers. Bis-(l-pyrenyl)-! ,3-propane [pyrene-pyrene-propane (PYPYP)],
consists of two pyrene moieties which are tethered by a three carbon alkyl chain,
just the right length to allow the pyrene moieties to fold together into approx-
imately the sandwich geometry of the pyrene excimer; longer or shorter alkyl
chains result in substantial or complete loss of excimer emission.11 Clearly the two
pyrene moieties in PYPYP cannot change their relative concentration, and so long
as the concentration of PYPYP is below the concentration for which there is signif-
icant interaction between an excited pyrene moiety on one PYPYP molecule and a
ground state pyrene moiety on another PYPYP molecule, approximately 5 x 10~4
M, the £*/M* intensity ratio obtained with the PYPYP exciplex fluorescence
thermometer is independent of the concentration of PYPYP. An additional benefit
comes with the intramolecular exciplex fluorescence thermometers: Unlike the
intermolecular exciplex fluorescence thermometers, which may require concen-
trations sufficiently large that questions may be raised about the potential effects
on the fuel evaporation rates, the intramolecular systems are most appropriately
used at very low concentrations.
As shown in Figs. 3 and 4, the spectra of the inter- and intramolecular pyrene
excimers are quite similar at room temperature but differ dramatically as the tem-
perature is raised: the intermolecular pyrene excimer emission virtually disappears
at 120°C, whereas the intramolecular pyrene excimer emission persists to 400°C.
Thus, the intermolecular system yields a higher sensitivity (estimated H-/— 1°C),
shorter range (20-100°C) thermometer whereas the intramolecular system yields
a lesser sensitivity (estimated +/-5°C), longer range (20-400°C) thermometer.10

360 400 440 480 520 560 600

Wavelength (nm)
Fig. 3 Intermolecular pyrene excimer spectra as a function of temperature,19 spectra
normalized to constant monomer intensity; the temperatures, from top down, are 23,
37,47,70, and 124°C. Used with permission of ASME.
The exciplex spectra can also be exploited to develop exciplex shift thermome-
ters. Although the photophysical processes underlying the commonly observed
shift of the exciplex emission band to shorter wavelengths as the temperature is
increased are not well understood,12 this shift, as measured by the ratio of the
intensity in a chosen spectral bandpass on the short wavelength side of the exci-
plex emission band to that in a chosen spectral bandpass on the long wavelength
side, can also be exploited as a thermometer.13 Figure 5 shows the blue shift with
increasing temperature for spectra of PYPYP obtained in hexadecane. Figure 6
shows the intensity ratio EI/E2 as a function of temperature.
It was initially thought exciplex shift might be a homogeneous band shift.
If that were so, then quenching by oxygen would decrease the intensity of the
exciplex band, but it would not change the shape (or apparent position) of the
FLUORESCENCE SPECTRA OF
PY-(CH2)3-PY AS A
FUNCTION OF TEMPERATURE
Normalized
intensity 0.5
400 500 600

Wavelength, nm
Fig. 4 Intramolecular pyrene excimer spectra as a function of temperature,10 spectra

normalized to constant excimer intensity. Used with permission of Applied Optics.
150 L A. MELTON
FLUORESCENCE SPECTRA AT DIFFERENT

TEMPERATURES (2E-5M PYPYP/DECANE)
400425 450 475 5)0 525 50 575 600625 650
Fig. 5 Shift of excimer spectra as a function of temperature.
0.8
0.7-
0.6-
gj
LU
0.5-
0.4-
0.3
~20 40 60 80 100 120 140 160 180
Temperature (C)
nitrogen purged —— air purged ........ OXygen purged
Fig. 6 Intensity ratio EI/E2 as a function of temperature for nitrogen purged, air
saturated, and oxygen saturated solutions; spectral ranges for El and E2 defined in
Fig. 5.
band. The band position could then have been used as a fluorescent thermometer
which would have been independent of PYPYP concentration and of oxygen
quenching. In Fig. 6, the calibration curves are shown for PYPYP solutions which
have been purged with nitrogen, air, and oxygen. The curves for nitrogen and
air are sufficiently close that the errors caused by quenching from the oxygen in
air are minor; such is not the case for the curve for the oxygen purged solution.
Although the PYPYP exciplex shift thermometry system is much less sensitive
to oxygen quenching than many other fluorescent thermometry systems, the goal
of a fluorescent thermometry system which is insensitive to all parameters of the
environment except temperature has not yet been achieved.

Schrum et al.14 have also developed a fluorescence shift thermometer which
makes use of the blue shift in the fluorescence of [N,N'-bis(2,5-di-tert-butylphenyl)-
3,4,9,10-perylenedicarboximide] (BTBP) in methanol or mineral oil as the tem-
perature increases (15-70°C). They achieved temperature resolution of 1-2°C. No
information was given on the sensitivity of the system to oxygen quenching.
Seaver and Peale15 have used a fluorescence thermometer based on solutions
of Eu+3 (EDTA) to measure the temperature of water droplets suspended in an
acoustic field. This thermometry system, which is restricted to aqueous systems
because of solubility problems, made use of excitation spectra; the absorption lines
of two ground electronic state levels of Eu3+ which share a common upper level
were scanned, and the temperature was inferred from the change in fluorescence
intensities in the two lines, which resulted from the changing equilibrium popula-
tions of the ground state levels. Since there was a common upper level, deviations
due to oxygen quenching were not a problem. The scans of the excitation spectra,
however, required minutes and, hence, the technique was applicable only to very
slowly changing (steady-state) systems.
Peterson et al.16'17 measured spectra of vapor phase pyrene as a function of tem-
perature from 650-1300 K. They developed a fluorescent thermometer for these
high-temperature, oxygen-containing systems based on the ratio of the intensity
of the fluorescence from 52 (second excited singlet state) to that from S\ (first
excited singlet state) and made point measurements of the temperature in flame
systems.
Ni and Melton18 have measured the fluorescence lifetime of fluoranthene and
naphthalene as a function of temperature and developed a vapor phase thermometer
based on naphthalene. They found that in the vapor phase in oxygen-free systems,
the fluorescence lifetime of both species decreases significantly with temperature.
In particular the fluorescent lifetime of naphthalene decreases from 180 ns at 25°C
to 35 ns at 450°C. Using a fluorescence lifetime imaging system, based on dual
fast gated image intensifies, they were able to image the temperature field as a
turbulent heated jet of naphthalene-seeded nitrogen exited into a coflow of cold
naphthalene-seeded nitrogen.
Applications
There are three areas of relevance to spray combustion in which fluorescent
thermometers have been used: 1) studies of the heating and evaporation of isolated
droplets, 2) studies of the liquid phase temperature field in sprays, and 3) studies
of the temperature field in the liquid left after a spray has impinged on a heated
plate. This section will review these measurements and tie the development of
152 L A. MELTON
these experiments to the development of the fluorescence thermometry methods

summarized in the Experimental Methods section.
Isolated Droplets
Wells and Melton19 measured the temperature of decane droplets falling into
quiescent hot nitrogen. A nitrogen laser (337 nm) was focused onto the droplet,
and the resulting fluorescence was imaged through a monochromator onto a diode
array detector. The entire fluorescence spectrum could be obtained from a single
droplet. Because they had only a single fixed observation point, they measured
the temperature of the droplets after they had fallen approximately 10 cm through
an oven, whose temperature could be varied. They used an intermolecular pyrene
exciplex fluorescence thermometer and, because it was concentration dependent,
took pains to show that negligible evaporation had occurred during the heating
process. Wells and Melton19 used a pyrene concentration sufficiently high that
90% of the incident laser energy was absorbed within the outer 56 IJL of the 225-
/ji-diam droplet. Because the resulting fluorescence emanated from this layer, they
reported their temperatures as skin temperatures.
Hanlon and Melton20 modified Wells and Melton's experiment significantly.
They constructed a heated fall tube which allowed observation of the droplets at any
point along a 10-cm path. They made use of PYPYP, the intramolecular exciplex
fluorescence thermometer, determined the concentration below which it would be
concentration independent, and used a PYPYP concentration which would allow
evaporation of 80% of the droplet mass before intermolecular effects would perturb
the results. Unfortunately, even though they had significantly higher ambient gas
temperatures than Wells and Melton,19 they observed negligible evaporation of the
hexadecane droplets.
The PYPYP concentration was much lower than the pyrene concentration in
Wells and Melton's experiments, and the droplets (diameter 283 IJL) were now opti-
cally thin. As a result, the fluorescence was collected, with nonuniform weighting,
from a large portion of droplet volume. They20 reported their results as volume-
averaged temperatures. Figure 7 shows the temperature profiles in the fall tube;
Fig. 8 shows the inferred temperatures as a function of fall distance (time). Figure 9
shows the E*/M* intensity ratio calibration curve which Hanlon and Melton20
measured and which they used to convert experimental spectra to droplet temper-
atures. In Fig. 8, the results apparently show that the droplet temperature rises
slowly to about 80°C, jumps by approximately 100°C in a few milliseconds, and
then continues its slow rise. Although the data are reproducible, the interpretation
is flawed because in this transient heating situation; fluorescent emissions from
hot and cold portions of the droplet were unavoidably summed together before the
spectrum was compared with calibration results.
Zhang and Melton21 showed that the inferred droplet temperatures reported by
Hanlon and Melton20 were the result of a nonuniform temperature field within the
droplet combined with a nonlinear (actually concave, see Fig. 9) E*/M* intensity
ratio vs temperature curve. As a result, inferred droplet temperatures below 80°C,
the temperature at which the maximum value of the E*/M* intensity ratio occurs
in the calibration curve, are systematically low, and inferred droplet temperatures
above 80°C are systematically high. If the droplet temperature is uniform or if
the calibration curve is linear, these systematic errors in the inferred temperatures

OVJVJ
d) ,0.^
HEATING '
500 • TUBE D
D
TEMPERATURE 50
° o D D
g - PROFILES
400
A D t
, AA °°° 0
A °0 D D D
b) < A °
A 0 0
§ 300
A A ° o
a) <
'*'"•••• .4.\\°-
UJ
O. c
§ 200
100 a -
O O CD o 0
n
"-6 - 4 - 2 0 2 4 6 8 10cm
DISTANCE (cm) From Entry Port
Fig. 7 Temperature profiles within fall tube apparatus.20 Used with permission of
ASME.
become negligible. They also showed that the temperatures should be described
not as volume averaged but rather as optically weighted temperatures.
Zhang and Melton21 showed that droplet temperature measurements in which
there is 1) a nonuniform temperature field, 2) a nonlinear calibration curve, and
3) illumination by a laser beam large enough to illuminate the whole droplet
are subject to substantial systematic error. If the intention is to make correct
measurements of the transient temperature of a droplet, then either condition 2
or condition 3 must be avoided. The exciplex shift thermometer developed by
Arce,13 which has a virtually linear intensity ratio vs temperature characteristic,
would allow whole droplet illumination with little systematic error in determining
DISTANCE (cm)
TIME(ms)
Fig. 8 Inferred droplet temperatures.20 Used with permission of ASME.

154 L. A. MELTON
CALIBRATION
0.1 mM PYPYP
in Hexadecane
20 60 100 14O 180 220 260 300
TEMPERATURE C
Fig. 9 E*IM* intensity ratio as a function of temperature.20 Used with permission of

ASME.
an optically weighted temperature, but the illumination scheme would preclude

determination of the spatial distribution of temperature within the droplet. Winter22
has carried out preliminary measurements of the (uniform) temperature field within
a droplet using droplet slicing imaging (DSI), a technique in which the exciting
laser beam is focused to a sheet which is narrow compared to the droplet diameter
and passes through the droplet along a midplane. The resulting fluorescence is
focused onto a charge coupled device (CCD) imaging camera located at 90 deg
to the laser sheet. The DSI technique partially eliminates the averaging over hot
and cold regions which hindered the interpretation of Hanlon and Melton's data.20
However, the front hemisphere of the droplet acts as a bad magnifying lens:
the central portion of the DSI image is expanded linearly, but the outer portion is
condensed into a ring at the edge of the droplet. Zhang and Melton23 have developed
algorithms for the restoration of such DSI images and have recommended that the
CCD camera be located at and angle of 60 deg with respect to the laser sheet
in order to capture the temperature field all the way to one edge of the droplet.
Although many of the techniques needed to measure the transient nonuniform
temperature field within a heating droplet appear to be in hand, no such results
have been reported.
Kadota et al.24-25 used exciplex fluorescence thermometry to measure the tem-
perature of a droplet suspended from a fiber in a hot-gas flow. Their steady-state
results showed that the E* / M* intensity ratio did not depend on the presence of
CC>2 and H2O in the vapor but did depend on the oxygen concentration. They
also ignited the droplet and showed that the intensity ratio for a burning droplet,
at a given liquid temperature, was the same as the intensity ratio for a droplet in
nitrogen, a result which implies that the oxygen initially dissolved in the fuel had
become negligible. Thus, they concluded that exciplex fluorescence thermometry
could give accurate results for burning droplets, since oxygen was virtually absent
within the flame sheet surrounding the burning droplet.25 This work provides a
fine example of the way in which prior knowledge of the oxygen distribution in
an experiment can allow use of an otherwise restricted diagnostic technique.
Rubel and Seaver26 have used the Eu3+(EDTA) thermometry system to measure
the steady-state temperature of water droplets which have a monolayer coating of
hexadecyl or octadecyl alcohol.
Sprays
Yoshikazi et al.27 have extended exciplex fluorescence thermometry to the much
more complex problem of thermometry in fuel sprays. They have used a pyrene
intermolecular exciplex fluorescence thermometry system to measure the two-
dimensional distribution of liquid phase temperature in a spray of n-hexadecane

(Ci6H34, nbp = 287°C) or squalane (C30H62, nbp = 350°C) into atmospheric
pressure air. The liquid fuel was preheated to 180°C, and they followed the cooling
of the spray liquid as it mixed with the quiescent air. Because the results did not
differ for the two solvents, they concluded that preferential evaporation of the
fuel (over the pyrene dopant, nbp = 350°C) did not distort the measurements.
The thermometry calibrations were carried out by suspending a droplet of the
doped liquid in heated ambient air (in front of the same excitation and detection
optics), and they argued that such a calibration procedure ipso facto accounted
for the effects of oxygen quenching. In their calibration procedure, however, the
oxygen concentration within the suspended droplet should be approximately the
equilibrium solubility at a given temperature, and it seems likely that the droplets
within a spray will not have time to fully equilibrate with the ambient oxygen.
Impinging Sprays
Ni and Melton28 have used exciplex shift thermometry13 to follow the temper-
ature of the liquid deposited onto a heated steel plate by a spray. As shown in
Fig. 10, when the liquid (decane, nbp = 174°C) is sprayed onto a surface which
is at a temperature higher than the boiling temperature of the liquid, the evapo-
ration of the liquid is rapid but not instantaneous. As shown in Fig. 11, however,
Temperature images of PYPYP/decane

injected onto hot plate (205 C)
€?
0 msec 3 msec 13 msec 73 msec
25 50 75 100 125 150 175

Temperature range (C)
Fig. 10 Temperature of liquid sprayed onto a heated steel plate as a function of time;
decane (nbp = 174° C) onto plate, T= 205° C.28
156 L. A. MELTON
Temperature images of PYPYP/hexadecane

injected onto hot plate (270 C)
•f :-.
0 msec 1.5 msec 13 msec 73 msec
25 50 100 150 200 250 270

Temperature range (C)
Fig. 11 Temperature of liquid sprayed onto a heated steel plate as a function of time;
hexadecane (nbp = 287° C) onto plate, T = 270°C.28
when the liquid (hexadecane, nbp = 287°C) has a boiling point higher than the
surface temperature, the evaporation of the liquid becomes very slow.
Conclusions
This review has provided a summary of fluorescence methods for the measure-
ment of liquid and vapor phase temperatures in systems relevant to spray combus-
tion. It appears that many of the problems associated with the development of the
diagnostic techniques have been overcome, with the possible exception of pertur-
bation by oxygen quenching and, thus, use in realistic engineering applications is
expected to increase.
Acknowledgments
The support of the U.S. Army Research Office through Grants DAAL03-91-G-
0033 and DAAH04-94-G-0020 and of Ford Motor Company through its University
Research Program is gratefully acknowledged.
References
^withenbank, J., Beer, J. M, Taylor, D. S., Abbot, D., and McCreath, G. C., "A Laser
Diagnostic Technique for the Measurement of Droplet and Particle Size Distribution,"
Experimental Diagnostics in Gas Phase Combustion, Edited by B. T. Zinn, Vol. 33, Progress
in Aeronautics and Astronautics, AIAA, Washington, DC, 1977.
2
Bachalo, W. D., and Houser, M. J, "Phase Doppler Spray Analyzer for Simultaneous
Measurements of Drop Size and Velocity Distributions," Optical Engineering, Vol. 23,
1984, p. 583.
3
Sankar, S. V, Ibrahim, K. M., Buermann, D. H., Fidrich, M. J., and Bachalo, W. D.,
"An Integrated Phase Doppler/Rainbow Refractometer System for the Simultaneous Mea-
surement of Droplet Size, Velocity, and Refractive Index," 3rd International Congress on
Optical Particle Sizing, Yokohama, Japan, Aug. 1993.
4
Massoli, P., Beretta, F, D'Alessio, A., andLazzaro, M., 'Temperature and Size of Single
Transparent Droplets by Light Scattering the Forward and Rainbow Regions," Applied
Optics, Vol. 32, No. 18, 1993, p. 3295.
5
Eckbreth, A. C., Laser Diagnostics for Combustion, Temperature and Species, Abacus
Press, Cambridge, MA, 1988.
6
Zhang, J., "Fluorescence Methods for Determination of Temperature in Aerosol Parti-
cles," Ph.D. Dissertation, Mechanical Engineering Dept., Univ. of Nebraska, Lincoln, NE,
1991.
7
Melton, L. A., "Determination of the Temperature of a Liquid," U.S. Patent 4,613,237,
Sept. 23, 1986.

8
Murray, A. M., and Melton, L. A., "Fluorescence Methods for Determination of Tem-
perature in Fuel Sprays," Applied Optics, Vol. 24, No. 17, 1985, p. 2783.
9
Stufflebeam, J. H., "Exciplex Fluorescence Thermometry of Liquid Fuel," Applied
Spectroscopy, Vol. 43, No. 2, 1989, p. 274.
10
Gossage, H. E., and Melton, L. A., "Fluorescence Thermometers Using Intramolecular
Exciplexes," Applied Optics, Vol. 26, No. 11, 1987, p. 2256.
u
Zachariasse, K., and Kuehnle, W., "Intramolecular Excimers with .Alpha.,.Omega.-
diarylakanes," Zeitschrift fur Physikalische Chemie (Frankfurt am Main), Vol. 101, 1976,
p. 267.
12
Martic, P. A., Daly, R. C., Williams, J. L. R., and Farid, S., "Effect of Polymeric
Matrices and Temperatures on Exciplex Formation," Polymer Sci. (Polymer Lett.), Vol. 15,
1977, p. 295.
13
Arce, O., "Temperature and Size Measurements of Free Falling Decane Droplets,"
M.S. Thesis, University of Texas, Dallas, TX, 1993.
14
Schrum, K. F, Williams, A. F., Haether, S. A., and Ben-Amotz, D., "Molecular Fluo-
rescence Thermometry," Analytical Chemistry, Vol. 66, 1994, p. 2788.
15
Seaver, M., and Peale, J. R., "Noncontact Fluorescence Thermometry of Acoustically
Levitated Water Drops," Applied Optics, Vol. 29, 1990, p. 4956.
16
Peterson, D. L., Lytle, F. E., and Laurendeau, N. M., "Determination of Flame Tem-
perature Using the Anomalous Fluorescence of Pyrene," Optics Letters, Vol. 11, 1986,
p. 345.
17
Peterson, D. L., Lytle, F. E., and Laurendeau, N. M., "Flame Temperature Measure-
ments Using the Anomalous Fluorescence of Pyrene," Applied Optics, Vol. 27, 1988,
p. 2768.
18
Ni, T. Q., and Melton, L. A., "2-D Gas Phase Temperature Measurements Using
Fluorescence Lifetime Imaging," submitted for publication in Applied Spectroscopy, 1996.
19
Wells, M. R., and Melton, L. A., Journal of Heat Transfer, Vol. 112, 1990, p. 1008.
20
Hanlon, T. R., and Melton, L. A., "Exciplex Fluorescence Thermometry of Falling
Hexadecane Droplets " Journal of Heat Transfer, Vol. 114, 1992, p. 450.
21
Zhang, J., and Melton, L. A., "Potential Systematic Errors in Droplet Temperatures
Obtained by Fluorescence Methods," Journal of Heat Transfer, Vol. 115, 1993, p. 325.
22
Winter, M., "Measurement of the Temperature Field Inside a Falling Droplet," ILASS-
AMERICAS, 4th Annual Conf., Extended Abstracts, Hartford, CT, 1990.
23
Zhang, J., and Melton, L. A., "Numerical Simulations and Restorations of Laser
Droplet-Slicing Images," Applied Optics, Vol. 33, 1994, p. 192.
24
Kadota, T, Taniguchi, Y., and Kadowski, K., "Exciplex Method for Remote Probing of
Fuel Droplet Temperature," Japan Society of Mechanical Engineers International Journal,
Ser. II, Vol. 34, 1991, p. 242.
158 L A. MELTON
25
Kadota, T., Miyoshi, K., and Tsue, M., "Exciplex Method for Remote Probing of Fuel
Droplet Temperature. Effects of Ambient Conditions on Fluorescence," Japan Society of
Mechanical Engineers International Journal, Sen B., Vol. 36, 1993, p. 371.
26
Rubel, G. O., and Seaver, M., "The Temperature of an Evaporating Water Droplet with
a Monolayer: A Comparison of Experiment and Theory," Atmos. Res., Vol. 31, 1994, p. 59.
27
Yoshizaki, T., Funahashi, T, Nishida, K., and Hiroyasu, H., "2-D Measurements of the
Liquid Phase Temperature in Fuel Sprays," Society of Automotive Engineers International
Congress and Exposition, Paper 950461, Feb. 27-Mar 3, 1995; also Engine Combustion
and Flow Diagnostics, SAE SP-1090, Feb. 1995.
28
Ni, T. Q., and Melton, L. A., 'Temperature Imaging of Liquid Alkanes on Hot Steel
Surface," unpublished, 1993.
Chapter 7
Assessing the Physics of Spray Behavior in

Complex Combustion Systems
V. G. McDonell* and G. S. Samuelsent

UCI Combustion Laboratory, University of California, Irvine,
California 92717-3550
Nomenclature
= projected area of drop
= transfer number
= drag coefficient
= specific heat
= characteristic diameter
= acceleration due to gravity
= latent heat of vaporization
= thermal conductivity
= length scale associated with the flow
= Prandtl number
= Reynolds number
= Stokes number
= environment temperature
= drop surface temperature
= time
= time for evaporation
= time for drop to heat up to saturation temperature
= time for drop to vaporize under steady state conditions
= relative velocity between liquid and gas
= drop velocity
= gas velocity
= gas/drop velocity fluctuation cross correlation
= drop volume
rights reserved.
*Senior Research Engineer. Member AIAA.
^Director, Professor of Mechanical, Aerospace, and Environmental Engineering.
Associate Fellow AIAA.
159
160 V. G. McDONELL AND G. S. SAMUELSEN
We = Weber number
Yps = mass fraction of vapor at drop surface
v = gas kinematic viscosity
pg = gas density
pd = drop density
a = surface tension
Subscripts
d = drop
g =gas
s = surface
I. Spray Combustion Problem
T HE motivation for understanding the dynamic and coupled phenomena asso-

ciated with spray combustion is driven by a need to improve the efficiency of
liquid-fired combustion devices while reducing the amount of pollutants produced.
The complex nature of spray combustion is associated with the large number of
processes occurring. See, for example, Fig. 1 for a typical swirl-stabilized liquid
fueled reaction.
Early physical descriptions of spray combustion in practical devices equated the
process to a diffusion flame.1"3 The importance of drops was demonstrated by the
observation that pressure atomizer spray flames behaved significantly different
than more finely atomized twin-fluid spray flames.2 Khalil and Whitelaw4 also
found significant differences between spray and gas flames. El-Banhawy and
Whitelaw5 pointed out that the variations in the observations from these early
studies are related to the vaporization and residence time scales. This point is
important since the notion of scaling combustion systems continues to be difficult
because of the inability to properly characterize or even select suitable scales.
The early theoretical work of Chiu and Liu6 and Chiu et al.7 attempted to
categorize the early experimental results with application of group combustion
theory, in which the key controlling factors are the nondimensional droplet spacing
and the droplet count. The theory, in principle, applies to a general spray reaction
yLIQUIDFUEL
Complex Aerodynamics
Gas/Drop Interaction
Heat Transfer
AIR——— Pollutant Formation
Combustion/Heat Release
v
Breakup
Vaporization/Cooling
Fig. 1 Description of spray combustion problem in the steady state.
ASSESSING THE PHYSICS OF SPRAY BEHAVIOR 1 61
and, depending on the structure of the spray, can be used to establish the burning
rate.
Despite the intuitive appeal of group combustion theory, the experimental tools
available at the time to examine the various phenomena within the reacting sprays
were inadequate to provide the required input.
Recent advances have been made in the experimental tools. As a result, consid-
erably more information is now available regarding the basic physics occurring
with complex spray systems. In some cases, nonreacting sprays are interrogated in
addition to reacting sprays in an effort to reduce the complexity of interpretation.
Such measurements are required to detail the physics of the phenomena occurring
as well as validate models proposed for comprehensive numerical predictions.
In addition, the theory of spray behavior has also evolved, pushing the need for
detailed measurements further.8
The present chapter addresses these measurements and the information derived
from the data acquired. In-depth discussion of the measurement techniques are
provided throughout the text. Recent reviews are also available that provide more
comprehensive overviews of the main techniques utilized.9
To organize the chapter, the spray combustion process is divided into five
elements: atomization, transport, vaporization, unmixedness, and combustion.
The synergy between the elements cannot be overlooked. For example, in the
continuous combustion process, the transport of the ignited mixture promotes va-
porization that in turn impacts unmixedness. In addition, a synergism between the
elements leads to fundamental boundaries such as lean and rich extinction limits.
In addition to the examination of each element, examples of synergism between
and among the elements are presented.
A. Atomization
The first step in the spray combustion process is the atomization of the liquid.
In practical systems, the atomization occurs in two steps. The first is the atom-
ization that accompanies the injection of the liquid into an ambient flow of air
(liquid injection). The injection of the liquid can be accompanied with a secondary
coflowing fluid (twin-fluid atomizer) or with no accompanying twin fluid and rely
solely on the pressure differential forces to atomize the liquid (pressure-simplex
atomizer). The second step in the atomization is the interaction of droplets with
the ambient air (liquid-air interaction).
7. Fuel Injection
In typical atomizers, a sheet of liquid is injected and breaks up in a three-
dimensional manner (Fig. 2).
In a pressure atomizer, the liquid velocity may exceed the gas velocity, whereas
in a twin-fluid atomizer, a gas commingles with the liquid with a velocity that
typically exceeds that of the liquid. In each case, the velocity differential creates
disruptive aerodynamic forces that result in instabilities and breakup. The basic
mechanism in the breakup process is the competing effect of the disruptive aerody-
namic forces relative to the stabilizing force of surface tension. This competition
is expressed as the Weber number,
(1)
a) 20psig
b) 35psi
c) SOpsig
Fig. 2 Images of breakup process in simplex atomizer methanol spray.

ASSESSING THE PHYSICS OF SPRAY BEHAVIOR 163
Viscous forces may play a role as well, and the Reynolds number Re is often
required for correlations to be successful10 where the Reynolds number of interest
is based on the relative velocity between the gas and droplet and a characteristic
dimension.
It is evident that the measurement of both the characteristic dimension and
relative velocity is critical to establish both the Reynolds number and the Weber
number. Since direct measurement of either has been successfully made, the
approach to date is to assume a gas phase velocity and velocity profile associated
with the liquid surface.
Imaging is especially useful in providing the insight needed to make the re-
quired assumptions.11'13 Examples range from simple photography as shown in
Fig. 2 to more advanced methods such as holography14 and shadowgraphs.13 Such
techniques provide overall visualization of the breakup as well as measurements
of the point at which breakup occurs (breakup length), the wavelength of the sheet
at the point of breakup, and the frequency of breakup. Nonimaging methods can
also be employed with signal analysis to monitor the wave frequency associated
with the sheet breakup process.15
Whereas practical geometries continue, for the most part, to defy complete
physical interpretation, model geometries have been investigated numerically and
empirically with good success. Round liquid jets issued into a stagnant environ-
ment, in particular, have received considerable attention for many years and are
mentioned here primarily for completeness. The basic mechanism for atomization
involves wave formation and subsequent breakup of the liquid. In the simplest
case, surface tension in the liquid will lead to the formation of waves (Rayleigh
mechanism). Next, velocity shearing at the interface between the liquid and the
gas augments or even dominates the surface tension (wind induced16). Recently,
the role of turbulence has received attention.17> 18 In each of these cases, the role of
turbulence in the liquid jet was found to be important in addition to aerodynamics.18
2. Liquid-Air Interaction
Once droplets are formed from the liquid sheet, they interact with the gas phase.
At this point, the atomization process may not even be completed. Depending on
the local conditions, the Weber number and the Reynolds number may reach levels
that result in additional atomization (secondary atomization). This is likely to occur
in elliptic flows with complex aerodynamics where steep gradients in the velocity
field can result in locally high relative velocities. Again, with instruments such as
phase Doppler interferometry (PDI), measurement of the relative velocity can be
accomplished and the likelihood of secondary atomization assessed directly.
3. Practical Design Implications

For the designer of practical equipment, reliance for information is placed on the
database that has evolved over the past 20 years. Two sources of information are
in prominent use: 1) global characteristics (e.g., spray angle, nozzle flow number,
and mean drop size), and 2) empirical that relate fuel type and operating conditions
to global characteristics.19
Lefebvre provides perhaps the single largest collection of empirical relation-
ships for use on atomizers typically in use today.19 These relationships were de-
veloped through comprehensive studies in which properties of the liquid and gas
along with operating conditions and nozzle geometry were systematically varied.
The large majority of the relationships are based on diffraction results. Diffraction
is almost ideally suited for these studies since it provides a direct, nonintrusive
measurement of the mean drop size in the spray along a line of sight and can be
made quickly. Recently studies have appeared that examine the effect of liquid
properties in a more comprehensive fashion by using relative new point-based
measurements.20 Although the extent to which these more time consuming mea-
surements can add to the relations already constructed, the information obtained
(e.g., droplet velocity and gas phase velocity) in addition to droplet size may
give some additional insight into the breakup process. For example, comparisons
between measurements made using these more contemporary techniques suggest

that the basic physics associated with the breakup process in a pressure simplex
nozzle can, for the first time, be accounted for.21'22
4. Summary
Imaging has proven to be the primary means to assess the breakup process
in sprays. In general, the key parameters associated with the breakup process
(e.g., relative velocity) require estimation. Although imaging results have been
utilized to develop correlations and general descriptions of the process, the advent
of advanced diagnostics is allowing details regarding the resulting velocity and
trajectories to be compared to purely theoretical descriptions of the process. The
results are encouraging.
B. Transport
Once the droplets have formed via primary or secondary atomization, they
mix with the surrounding gaseous media, and the droplet motion (velocity and
trajectory) is affected by (and can be dominated by) the momentum exchange
between the droplet and the gas.
1. Theoretical Perspective
In the case of the typical combustion system, the interaction of droplets with
the surrounding gaseous media is highly complex and can be described by
dUd
—7— = —Fd -f (pd - pg)Vg (2a)
where
Fd = ^CDpg(Ud — Ug)2Ap (2b)

Equation (2) is an approximation since additional forces do affect the drop
motion8'23; however, the terms included in Eq. (2) are the most significant in most
cases. What can be seen from Eq. (2) is that the drop motion is described primarily
by the gravity force and by the drag force (which is highly dependent on the
relative velocity of the drop and the gas phase, and the drag coefficient which
itself depends on the local environment).
The Stokes number is used to quantify the droplet response time relative to the
flow time scale. Noting that the droplet response time is directly dependent on the
drag coefficient and, in the simplified case with Stokes drag law (low Reynolds
number) where CD is taken as 24/Re, the response time is proportional to the

diameter squared24
}
18 pg vL ^
Drops with St ^> 1 require much longer times to respond to the gas phase,
whereas drops with St <^ I respond nearly instantaneously. A significant chal-
lenge in determining the Stokes number is the selection of L and the viscosity and
density in a complex reacting flow.

Equation (3) is limited to low Reynolds number regimes. In general, the
Reynolds number can exceed 1 by significant amounts, resulting in non-Stokes be-
havior. In addition, Stokes drag is generally associated with rigid spheres without
internal circulation or vaporization or interaction. As a result, the exact nature of
the drag coefficient in practical vaporizing sprays has proven difficult to account
for. As a result, the accepted practice has been the inclusion of a convective term
for the preceding effects that is easy to integrate25
Using Eq. (4) in combination with the information available from PDI, the drag
coefficient can, in principle, be estimated within the spray.9 In more general cases,
where droplets are vaporizing or interacting,26 imaging can support theoretical
efforts to describe droplet transport.8
2. Detailed Measurements
The preceding analytical descriptions indicate that a critical element of spray
combustion is the dependence of drop motion on drop size. With the development
of phase Doppler interferometry, droplet behavior as a function of size can now
be measured explicitly. Detailed measurements of droplet behavior as a function
of size in spray combustion systems are now being reported.27'36
In particular, cases are now available that characterize the general structure of
the gas phase as well as that of the droplets. As an example, a characterization
of the general structure of the complex spray behavior in a practical combustor
fuel injector (Fig. 3) is presented. Essentially a large air-blast nozzle, this device
injects fuel from a centrally located simplex fuel nozzle. The fuel is filmed upon
the venturi and is subsequently reatomized by the counterswirling air streams
injected on each side of the venturi. Despite the complexity of the device, it has
served as a highly successful practical device used in typical aircraft engines for
over 20 years.
In Fig. 4, the mean gas phase velocity vectors are overlaid on contours of the
turbulent kinetic energy. In this particular case, a 4% pressure drop was utilized
to represent a typical gas turbine engine level. The measurements, conducted at
atmospheric pressure and an inlet temperature of 70°F, illustrate that the gas phase
features a strong on-axis recirculation zone. The turbulent kinetic energy contours
reveal strong mixing at the shear layer between the inner recirculated mass and
the outer high-speed annular flow.
When the droplets are injected into this complex aerodynamic flow, they tend
to segregate according to the Stokes number. Distinct behavior is observed for
PRIMARY SWIRL
COOLING AIR
FLARE AIR
SECONDARY SWIRL
© NOZZLE
© VENTURI Rp = 9.7mm
© RADIAL SWIRLER
® FLARE
© SPLASHPLATE
<G> FLARE AIR PASSAGE
Fig. 3 Schematic of practical combustor dome.
10 A* Fud Nozzle
_ , mA2/sA2
35CH-
280 to 350
-10
20
10 Â
^——————— ^**————————
-10
20
10 AAAAAAAA
-A!-,———,———,—£A*a——
A
-10
20
10 AAAA
0
A. ^ÂA* AAAAAAAA
A '
A
-10
-•m
0 10 20 30 40 50 0 10 20 30 40 50
RADIAL POSmON, mm RADIAL POSITION, mm
Fig. 4 Aerodynamic structure of the gas phase flowfield produced by the combustor
dome operating with a 4% pressure drop.
I
Data Rate, ff/s Data Rate, t/s
80+
64 to 80
48 to 64
32 to 48
16 to 32
Oto16
10 20 30 40 10 20 30 40
RADIAL POSITION, mm RADIAL POSITION, mm
Fig. 5 Velocity characteristics of droplets in the flowfield produced by combustor

dome [cases with St < I (11-20 /mi) and St>l (76-90 /mi)].
different size classes. In this case, the nonreacting spray features droplets with
a wide range of Stokes numbers. For illustration, two examples are shown in
Fig. 5. In this case, Eq. (3) is applied using the measured relative velocity. Since
the response of the drops to the recirculating flow is of interest in this case, L is
taken as the flare exit diameter. For each of the two drop sizes selected (11-20
and 76-90 /zm), a mean Stokes number (0.2 and 5.0) is determined. Figure 5
reveals that the finer drops are, in fact, recirculated quite readily by the gas phase,
whereas the large drops penetrate ballistically as expected intuitively. Note also
that the number of large drops is substantially smaller than the number of small
drops. In addition, an estimate of the concentration of the drops is shown in Fig. 5.
Although not the prime focus of this section, the PDI technique can also provide
some indication of the dispersion of the drops that is important to mixing. It is
noteworthy that a similar basic structure is observed in cases representing more
practical conditions.37
Figure 5 only illustrates the time-averaged behavior.
3. Spray Dynamics
The dynamics of the spray behavior are another critical aspect of the spray
combustion process. To examine this behavior in more detail, two points are
selected in the flowfield produced by the combustor dome. These locations are
situated on the inside and outside of the shear layer separating the recirculated
mass and the high-speed annular flow. Figure 6 illustrates the locations relative to
an image of the spray obtained from light scattered from a sheet of laser light, and
it can be seen that the locations also correspond to points on the inside and outside
of the majority of the liquid.
Figure 7 presents the detailed results at these two points. The results reveal that
at the inner point (r = 18 mm) a substantial number of drops are recirculated. It is

Fig. 6 Image of light scattered by droplet from a sheet of laser light slicing through
the geometric centerline of the flowfield and points at which details are presented in
Fig. 7.
observed, however, that significant numbers of drops of the same size are convected
upstream as well as downstream at the same point. This suggests that the dynamics
of the spray is quite significant. In this particular case, the complex atomization
process and aerodynamics results in the arrival of drops from three sources: directly
from the fuel injector, directly from the venturi, and from downstream.38
4. Reacting Flows
At this point, only the nonreacting case has been examined for the practical
hardware shown in Fig. 3. In the reacting flow, the velocity characteristics of
the spray can be assessed in similar fashion. The same points interrogated under
reacting conditions (Fig. 8) lead to the results shown in Fig. 9. In this case, it can
be seen that the majority of the drops that would have shown as recirculated have
been consumed by the reaction. In addition, the overall velocity values are higher
in the reacting case. This is attributed to acceleration of the gas phase caused by
expansion of the gases as well as the history effect of the drops measured at this

0 40 80
DIAMETER, ^m
Fig. 7a Size and velocity distributions for the nonreacting spray: location 1 indicated
in Fig. 6.
location. Any given drop present at 2.75 Rp must have been larger and hence had
a greater initial velocity at one time.
Similar detailed information is provided in other studies.27'29'33 Focus is placed
on the velocity as a function of size in several studies.29'30'36 In these studies, for
the reacting cases, the velocity behavior of individual drop sizes are compared at
different locations. The ability to track the velocity of individual drops is severely
compromised, however, by the vaporization process in the reacting flow. Even in a
nonreacting vaporizing spray, comparing velocities of a given size drop throughout
the spray can be quite misleading. The reason is that a given size drop at a given
downstream location must have been significantly larger at a given upstream
location in order to end up at the given size at the location of interest. This will be
illustrated with experimental data in the next section.
In addition to the overall distribution, attention has also be recently given to the
temporal nature of the droplet interaction with the gas phase.27-28 The nature of
the arrival time at given points in the spray can be viewed relative to the purely
random behavior in order to establish steadiness. Recent work has developed the
theory to examine sprays from this perspective.39
40
40 80 120
DIAMETER, ^m
Fig. 7b Size and velocity distributions for the nonreacting spray: location 2 indicated
in Fig. 6.
5. Measurement Issues
An issue that remains is the ability to accurately measure the gas phase velocities
in the presence of the drops.
First, the ability to simultaneously measure these quantities is important be-
cause, in general, the coupling between the gas phase and the droplets occurs
due to the instantaneous relative velocity, not the time-averaged mean velocity
difference.
Second, in a practical spray, the size range present typically precludes the
measurement of fine particles and the remaining spray drops at the Same time.
This is because of the limitation of the dynamic range of the instrument.
Third, the optical arrangement optimized for fine particles that are introduced
to track the gas phase velocities will be highly nonoptimum for measuring the
relatively large droplets presented by the spray. The sensitivity of the instrument
to setup inputs exacerbates this problem.33'40 New signal processing technology is
helping reduce the sensitivity of the results, although the dynamic range issue still
exists.41

Fig. 8 Two points selected for detailed assessment shown relative to image of the
reacting spray—same locations as indicated in Fig. 6.
Fourth, the ability to correctly measure the coupling of the fluctuating velocity
components is important to the validation of higher order models. An example of
measuring the gas phase and droplet simultaneously is provided in a recent study
that, although conducted in a highly idealized spray with a narrow dynamic range,
provides a direct measurement of the particle/gas velocity cross correlation, ufg. u'd.
(Ref. 42).
Fifth, the scale and resolution to which the measurement of the gas phase
velocity in the presence of the spray can be made is important. For example,
should it be expected that the velocity characteristics of wakes behind drops can
be measured? In the cases where the gas phase velocity is measured in the presence
of the spray, with the exception of Ref. 42, the gas phase is typically measured
in a totally separate experiment, requiring great care to ensure that the spray is
repeatable.
In addition, questions remain regarding the absolute accuracy of the phase
Doppler measurements of particle mass flux and particle size distribution.43'44
Although it is difficult to place a single value of accuracy on the PDI measure-
ments, an increasing number of studies is creating a consensus. For example, a
comprehensive error assessment of the PDI technique is provided in Ref. 44 and
40
40 80 120
DIAMETER, ^m
Fig. 9a Size and velocity distributions for the nonreacting spray: location 1 indicated
in Fig. 8.
suggests that random errors contribute approximately 10% uncertainty whereas

the miscounting of drops caused by any number of reasons can lead to a 20-25%
underrepresentation of drops. Despite the slow development of a consensus, new
applications, and the huge variety of sprays (e.g., a gas turbine simplex pilot nozzle
injecting a few kilograms per hour of clean fuel compared to a power plant fuel
oil nozzle injecting orders of magnitude more heavy oil) utilized make general-
ization difficult. Perhaps the best approach to assessing the flux issue is the use of
statistical methods.39
6. Summary
The development of the PDI has permitted highly detailed information regarding
the transport of droplets by the gas phase and the interaction between the phases.
Many studies are now available that detail the gas phase velocities and reveal
the complexities of the droplet behavior as a function of size. Recently, efforts
have been directed at improved interpretation of the time-resolved information
produced.
- 40
40 80 120
DIAMETER, ^m
Fig. 9b Size and velocity distributions for the nonreacting spray: location 2 indicated
in Fig. 8.
C. Vaporization
In a spray-fired reaction, it is the vapor that burns. As a result, the step from
atomization to combustion requires vaporization.
1. Theoretical Perspective
From a practical standpoint, work summarized in Ref. 19 provides an effective
starting point from a theoretical basis. The instantaneous rate of vaporization of a
drop in a convective environment is
(5)
P/ g
The transfer number, B, can be formulated based on either thermal or mass

gradients. In the case of the former
BT=cpg(T00-Ts)/Hv (6a)
or in the case of concentration gradients

BM = WO ~ YFs) (6b)
In nonreacting vaporizing sprays, Eq. (6b) must be applied. In this case, the
mass fraction of vapor at the surface of the drop must be determined. In prac-
tice, if the mean vapor concentration within the spray could be determined, the
transfer number would be known since the vapor mass fraction would be constant
throughout a saturated field. (An example is provided later.)
In a spray combustion case, Eq. (6a) can be applied since the thermal gra-
dients tend to dominate. Indeed, it is recommended that Eq. (6a) be used in

high-temperature cases with the boiling temperature substituted for the surface
temperature.
An additional point is that the time required for a given droplet to evaporate is
the sum of the heat up time and the steady vaporization time19:
te = Athu + Atst (7)
2
Whereas the D dependence of the steady-state vaporization time is well known
and established for single isolated drops, the behavior in a convective environment
with interacting drops precludes satisfactory theoretical description. Convection
has been addressed empirically as indicated by the last term in Eq. (5). The
complexity of vaporizing, convective environments with droplet interaction, how-
ever, coupled with the difficulty in acquiring measurements, leads to significant
shortcomings.
Vaporization has proven to be especially difficult to characterize within the
spray. In general, the presence of droplets causes problems in the characterization
of the gas phase scalar fields.
For example, the droplet heating time has posed a challenge experimentally,
although recent advances in optical diagnostic techniques (e.g., rainbow ther-
mometry) has provided the potential for measurement of drop temperature along
with size and velocity within complex sprays.45 The rainbow thermometry tech-
nique provides some interesting potential in terms of identifying saturation regions
within sprays and the extent to which the heat up delay is important. A key issue
of this technique is the extent to which thermal gradients within the drops impact
on the measurement.
Beyond the measurement of drop temperature, the direct measurement of the
vapor concentration within practical sprays has been accomplished recently using
some different techniques. Some promising results were provided early on using
exciplex fluorescence.46 In this technique, the excited state complex (exciplex)
fluorescence is well red shifted in the condensed phase from the excited monomer
in the vapor phase. The result is that, using imaging techniques and suitable
filtering, planar measurements of both the vapor and liquid phases are possible.
Unfortunately, this technique is very difficult to apply in oxygenated environments
due to the high quenching of the monomer.
Alternatively, selective absorption [termed infrared extinction/scattering
(IRES)] has been utilized as well.47"49 For example, the bond between the car-
bon and hydrogen molecule in most hydrocarbon fuels strongly absorbs energy
in wavelengths from 3.3-3.4 /xm. By comparing a line-of-sight measurement of
the extinction at this wavelength to one at a wavelength away from the C=H
absorption band (where only scattering occurs), the relative extinction caused by
the vapor can be determined. The ratio of the extinction caused by scattering and
absorption at one wavelength to the scattering at another gives a line-of-sight
vapor absorption profile. In an axisymmetric spray, a relatively small number of
measurements can be deconvolved into spatially resolved concentration values.
When combined with measurements of the gas phase velocity obtained via PDI,
the results can be used to examine the global vaporization characteristics of the
spray. An example is provided in Fig. 10 from a study conducted on a nonswirling
air-assist atomizer spraying methanol.

Figure lOa presents the vapor concentration along with velocity vectors of the
gas phase as measured using PDI.50 Knowing the velocity of the gas (and therefore
the vapor), the flux of vapor through a given plane can be determined. This gives
the vaporization as a function of distance from the injector. These results are
presented in Fig. lOb. If a suitable scale, can be identified, the vapor production
as a function of distance can be converted into vapor production as a function of
time, which can, in principle, provide the global vaporization rate of the spray.
This approach was explored in detail for a variety of injector types under reacting
as well as nonreacting conditions and illustrated that, for example, the addition of
atomizing air, although reducing droplet size, can delay the location at which a
desired level of vapor occurs.51
A recent study reveals a novel approach to studying the vaporization of a spray.52
In this technique, the fluorescence from a dye-doped into a liquid is compared to
Raman scattering. In that the dye content remains constant as the liquid vaporizes
while the Raman signal remains proportional to the actual droplet mass, the
difference can be used to infer the amount of vapor.
More recently, the direct measurement of the vapor characteristics within a
spray have been obtained using a newly established technique developed for
gas phase mixing studies—acetone fluorescence.53 Although developed for gas
mixing studies, the absorption coefficient for liquid acetone leads to a substantial
difference in the optical thickness of the condensed and vapor phases. Hence the
possibility to utilize acetone fluorescence to study the vapor characteristics while
accounting for the fluorescence from the droplets appears possible.54 Although a
variety of issues and possible limitations are pointed out, this technique appears
promising. The primary drawback is that acetone must be utilized as the fuel.
Worth noting at this point is that the knowledge of the vapor concentration
within the spray permits additional analysis of the spray behavior to be conducted.
This is illustrated in Fig. 11, where the measured concentration and velocity
characteristics are interpolated and used to track the possible path of given drops
within the spray as well as their size evolution using the empirical expression
provided in Eq. (5). This is the same spray for which the gas phase velocities and
vapor concentrations are presented in Fig. 10.
The results demonstrate the point made earlier regarding the comparison of
velocity profiles for a given drop size at different locations in a vaporizing spray.
The mean path is provided by the center track. The tracks on either side are based
on the variation in the droplet velocity measured at each location. In this case, it
is observed that the 15-/zm drop changes size bins as it vaporizes, yet maintains
the velocity characteristics of the larger drop it once was. In this particular case,
the larger drop does not change size bins. In a reacting case, however, the change
in diameter will occur even faster, leading to a situation for the large drops in
[HC] %
P
io?
o
m
o
P
0)
0)
c
m
m
o
c AXIAL POSITION, mm
0.500
CO 0.400 - -
en
0.300--
CO 0.200 - -
0.100--
0.000
25 50 75 100 125 150 175
AXIAL POSITION, mm
Fig. lOb Total vapor flow rate as a function of axial distance determined from the
results shown in Fig. lOa.
15 [iin drops 83 nm drops

(76-90 pirn bin)
(S)
O
-100 0_
-80 -60 -40 -20 0 -80 -60 -40 -20 0
RADIAL POSITION, mm RADIAL POSITION, mm
Fig. 11 Flight paths of 11-20 and 76-90 /xm drops determined from PDI drop size
and velocities and IRES vapor concentration measurements.
particular where they will change bins several times within the field. These types
of results, in particular, approach a Langrangian reference frame for the droplet
measurements, which is something currently not available. In the nonswirling
spray considered in this case, the approach is reasonable. In a highly swirling
combusting spray, the results may be subject to considerable error.
3. Summary
The direct measurement of vapor within a spray remains an experimentally
challenging task. The addition of drop temperature to the well-established phase
Doppler technique appears promising, though issues regarding thermal gradients

within the drop remain. Such a technique could allow better understanding of the
role of droplet heating in combusting sprays. The study of the vapor concentra-
tion within the spray is progressing slowly and is really needed to bridge the gap
between the fuel injection process and the resulting combustion. Deconvolution
of line-of-sight results appears promising although, in the absense of symmetry,
requires considerable constraints from the standpoint of optical access (i.e., mea-
surements from many different perspectives are required). Quicker imaging-based
techniques are currently under development.
D. Unmixedness
The extent to which the fuel vapor and oxidant are not uniformly mixed (un-
mixedness) prior to combustion is receiving increased attention, especially in lean
systems where the goal is to accomplish the combustion process at stoichiometries
low enough to preclude the high temperatures leading to NOX formation. In these
cases, variations in local mixture ratios can lead to local hot zones (that form NOX)
as well as locally cold zones (that can lead to inefficiencies resulting in CO and
hydrocarbon emissions).
Mixing is an inherently spatial and temporal issue. As such, planar techniques
appear best suited for studying mixing. In gaseous systems, planar laser induced
fluorescence (PLIF) techniques are established tools for assessing mixing quality
and uniformity.
As mentioned earlier, the potential to utilize acetone fluorescence to directly
image the fuel vapor concentration in a reacting spray provides one avenue to
deduce the degree of mixing present.54
The potential to examine the mixing of the spray itself has been the subject of
study in recent years. If it is assumed that the uniformity of the spray is an indi-
cator of fuel uniformity, then the mixing characteristics of the liquid phase can be
utilized. An extensive evaluation and summary of a planar liquid laser-induced flu-
orescence (PLLIF) technique for liquid sprays is provided elsewhere in this book.55
An example of the application of PLLIF is presented in Fig. 12 from a study
in which the effects of atomizer air-to-liquid (ALR) ratio on the uniformity of the
liquid fuel distribution was evaluated.56 The goal was to identify the minimum air-
to-liquid ratio required to provide adequate coverage. In this case, a jet was injected
with air-assist perpendicularly to a high-speed crossflow. The injector was located
at a radial distance of 0 mm and at 12.5 mm laterally relative to the coordinate
system indicated. Results are presented for four injector air-to-liquid ratios at a
plane perpendicular to the crossflow 10 mm below the point of injection. In addition
Injector Hole -20 -20 Fuel Mass

12.5mm - 15 ( - 15 Fraction
- 10 ' 10
10.95
-5 -5
10.85
ALR-O ALR-1 10.75

125-
i 0.65
i 0.55
i 0.45
0 5 10 15 20 25 30 0 5 10 15 20 25 30
10.35
Radial Distance, mm Radial Distance, nan
- 20 1075
- 15 lo.15
100
10 Jo.05
- 5
75
50 -
ALR-4.67
5 10 15 20 25 30 0 5 10 15 20 25 30
Radial Position, mm Radial Position, mm
Fig. 12 Spatial uniformity and SMD along jet centerline 10 mm below the injection
point of a liquid fuel jet injected into a high-speed crossflow—effect of air to liquid
ratio.
to the fuel distribution, drop size results are presented along a line perpendicular
to the crossflow 10 mm downstream of the jet. These results illustrate the coupling
between the drop size and the uniformity. For example, locally larger drop sizes
are measured in the region of the jet farthest from the wall—a result of their larger
momentum. Also, the increased ALR not only increases penetration, but reduces
drop size.
Figure 13 summarizes the results and reveals that a jet-to-crossflow momentum
ratio of 2000 is adequate to provide reasonable coverage of the spray jet by 10 mm
downstream of the injection point. In this particular case, the liquid coverage
appears adequate spatially. The temporal unmixedness of the liquid could be,
in principle, assessed using the same technique. The same technique, applied to
combusting sprays, offers the potential to describe the liquid distribution. In this
case, the effects of temperature on the fluorescence spectra must be established.
In one case,54 a temperature dependency of 35% was established for acetone, but
these effects remain a critical issue to understand in order to apply these techniques
to combusting flows.
2. Summary
The study of fuel-air mixing using advanced diagnostics in liquid fueled com-
bustion problems is still in the infancy stages. The highest promise appears to be
Fuel Mass Flow = 0.1463 g/sec

Crossflow Velocity = 38 in/sec
0.45 400
0.40 - -350
-300
Q) 0.35 -
•9 - 250
0.30 -
-200
0.25 -
- 150
0.20 -
- 100
0.15 - -50
0.10
0 500 1000 1500 2000 2500 3000 3500
(pU2)jet/(pU2)Cross
Fig. 13 Summary of mixing effectiveness for constant fuel flow.
in fluorescence or Raman scattering. Despite this, the desired quantity to assess

in combustion systems is the local fuel/air ratio. Toward this end, techniques that
simultaneously measure liquid and vapor phase fuel are desired. Exciplex appears
at this point to be the only technique to do this. However, alternative approaches
are beginning to appear.
E. Combustion
Intimately associated with mixing is the combustion process. Information de-
sired includes the identification of the reaction front and the temperature and
species compositions and concentrations within the reacting flow. In addition,
the behavior of the liquid droplets in terms of size and velocity is of paramount
interest.
In the case of reaction, an index is required of the location at which combustion is
initiated. A classic tool for probing the reaction zone in non-premixed combustion
is OH fluorescence.57 In sprays, this technique has not been entirely successful
because of the presence of droplets. It has proven useful, however, in marking the
qualitative aspects of the combustion zone.29 An issue is the Raman background
produced by droplets which requires considerable effort to suppress and finally
allow interpretation of the OH.58
Alternatively, the temperature can be measured to indicate the region of heat
release. Application of techniques such as coherent anti-Stokes Raman scattering
(CARS) is possible in the presence of drops. One example is the measurement of
gas temperature in a model gas turbine combustor.59 If the dielectric breakdown
induced by the drops focusing the high-energy laser beams is properly accounted
for, reasonable temperatures can be measured.60
With caution, temperature measurements using thermocouple probes can pro-

vide significant insight into the general nature of the reaction.32'33
Recent work seeking to investigate group combustion phenomena utilized si-
multaneous scattering of a laser sheet by the droplets and reaction luminosity
(CH, OH, and €2 emission) to correlate the presence of droplets and reaction.61
In this case, an effort to directly evaluate some of the theoretical ideas put forth
previously was conducted. The experimental work found that some of the basic
structure were not adequately described by the pure form of group theory, sug-
gesting that application of group combustion theory requires some modification
to be completely successful. In this example, an experiment was derived directly

from a theoretical description of the spray process. That some modification to the
theory might be required illustrates the importance of maintaining close dialog
between the modeling and experimental efforts.
2. Summary
The combustion process is largely a gas phase problem. The presence of drops
tends to exacerbate the challenges of obtaining measurements of the characteris-
tics species and thermal fields associated with the combustion process. The ability
of the PDI technique to probe the droplet size and velocity characteristics adds re-
markably to the information available via typical gas phase reaction measurements
(e.g., PLIF, CARS, laser anemometry).
II. Summary and Conclusions—Putting the Steps Together

Although the preceding step-by-step sequence reveals that considerable insight
can be provided regarding the breakup, transport, vaporization, unmixedness, and
combustion processes, it is apparent that the elements must be considered together
as a whole. Complementary results from a variety of state of the art diagnostics
provide by far the greatest insight into the structure and physics of the processes
occurring within complex spray combustion systems.
From a design perspective, the most practical information is available in the area
of atomization, where engineers can find correlations based on injector geometry
and operating conditions. Design at this point does not, however, generally incor-
porate the role of transport, vaporization, unmixedness, or combustion of the spray.
Only recently have tools been applied to study these phenomena, and how these
results fit into the design process is still being established. Such measurements are
necessary to unravel the physics occurring.
At this point, detailed assessment of the droplet size and velocity conducted via
PDI appears most widely in the literature. However, only a few of those studies
include gas phase structure or information regarding scalar fields. In general, the
luxury of the wide array of diagnostics required to completely characterize the
spray combustion process is not available and, for the most part, such characteri-
zation requires a significant amount of time to complete. One example is the study
where PDI and OH PLIF are utilized to validate a physical model of stabilization.58
Probably the most challenging area from an experimental point of view is
our ability to deal with the instantaneous structure and behavior. In this regard,
information and protocol is only beginning to emerge. To address the mechanisms
behind pollutant formation and reaction stability, time-resolved information is
required. Finally, questions regarding the resolution scales of the measurement
techniques remain.
Additional efforts are required to close the loop on experiments and numerical
predictions. As computing power increases, so does the ability to solve com-
prehensive codes with increasingly sophisticated models. In parallel, diagnostic
techniques continue to evolve. It is essential to maintain a bridge between the two.
In most cases where a combined numerical/experimental effort takes place, the
greatest insights are attained. At a minimum, the experimental data needs to be
sufficiently documented and archived to prove of value to those developing and
applying theoretical and numerical analyses.62
References
a, Y., and Ogasaward, M., "Studies on the Structure of a Spray Combus-
tion Flame," Fifteenth Symposium (International) on Combustion, Combustion Institute,
2
Komiyama, K., Flagan, R. C, and Heywood, J. B., "The Influence of Droplet Evap-
oration on Fuel-Air Mixing Rate in a Burner," Sixteenth Symposium (International) on
Combustion, Combustion Institute, Pittsburgh, PA, 1977, pp. 549-560.
3
Owen, F. K., Spadaccini, L. J., Kennedy, J. B., and Bowman, C. T., "Effects of Inlet
Air Swirl and Fuel Volatility on the Structure of Confined Spray Flames," Seventeenth
Symposium (International) on Combustion, Combustion Institute, Pittsburgh, PA, 1979,
pp. 467-473.
4
Khalil, E. E., and Whitelaw, J. H., "Aerodynamic and Thermodynamic Characteristics
of Kerosene-Spray Flames," Sixteenth Symposium (International) on Combustion, Com-
bustion Institute, Pittsburgh, PA, 1977, pp. 569-576.
5
El-Banhawy, Y., and Whitelaw, J. H., "Experimental Study of the Interaction Between
a Fuel Spray and the Surrounding Combustion Air," Combustion and Flame, Vol. 42, 1981,
p. 253-275.
6
Chiu, H. H., and Liu, T. M., "Group Combustion of Liquid Droplets," Combustion
Science and Technology, Vol. 17, 1977, p. 127.
7
Chiu, H. H., Kim, H. Y, and Croke, E. J., "Internal Group Combustion of Liquid
Droplets," Nineteenth Symposium (International) on Combustion, Combustion Institute,
8
Sirignano, W. A., "Fluid Dynamics of Sprays—1992 Freeman Scholar Lecture," Journal
of Fluids Engineering, Vol. 115, Sept. 1993, pp. 345-378.
9
Bachalo, W. D., "Experimental Methods in Multiphase Flows," International Journal
of Multiphase Flows, Vol. 20, Supplement, 1994, pp. 261-295.
10
Arai, T., and Hashimoto, H., "Behavior of Gas-Liquid Interface of a Liquid Film Jet,"
Bulletin ofJSME, Vol. 28, No. 245, pp. 2652-2659.
H
Mansour, A., and Chigier, N., "Dynamic Behavior of Liquid Sheets," Physics of Fluids
A, Vol. 2, No. 5, 1990, pp. 2971-2980.
12
Lavergne, G., Trichet, P., Hebrard, P., and Briscos, Y, "Liquid Sheet Disintegration
and Atomization Process on a Simplified Airblast Atomizer," Journal of Engineering for
Gas Turbines and Power, Vol. 115, July 1993, pp. 461-466.
13
Stapper, B. E., and Samuelsen, G. S., "An Experimental Study of the Breakup of a
Two-Dimensional Liquid Sheet in the Presence of Coflow Air Shear," AIAA Paper 90-0461,
Jan. 1990.
I4
Santangelo, P. J., and Sojka, P. E., "A Holographic Investigation of the Near Nozzle
Structure of an Effervescent Atomizer-Produced Spray," Atomization and Sprays, Vol. 5,
No. 2, 1995, pp. 137-156.
15
Eroglu, H., and Chigier, N., "Wave Characteristics of Liquid Sheets with Impinging
Air Jets," Atomization and Sprays, Vol. 2, No. 2, 1992, pp. 121-135.
l6
Reitz, R. D., "Atomization and Other Breakup Regimes of a Liquid Jet," Ph.D. Thesis,
Princeton Univ., Princeton, NJ, 1978.
17
Wu, P.-K., Miranda, R. F, and Faeth, G. M, "Effects of Initial Flow Conditions on
Primary Breakup of Nonturbulent and Turbulent Round Liquid Jets," Atomization and
Sprays, Vol. 5, No. 2, 1995, pp. 175-196.
18
Mansour, A., and Chigier, N. A., "Effect of Turbulence on the Stability of Liquid Jets
and the Resulting Droplet Size Distributions," Atomization and Sprays, Vol. 4, No. 5,1994,
pp. 583-604.
19
Lefebvre, A. H., Atomization and Sprays, Hemisphere, New York, 1989.
20
Dorfner, V, Domnick, J., Durst, F, and Kohler, R., "Viscosity and Surface Tension
Effects in Pressure Swirl Atomization," Atomization and Sprays, Vol. 5, No. 3, 1995,
pp. 261-285.
21
Mao, C-.P, Chuech, S. G., and Przekwas, A. J., "Analysis of Pressure Swirl and Pure
Airblast Atomization," Atomization and Sprays, Vol. 1, No. 2, 1991, pp. 215-235.
22
Li, X., Chin, L. P., Tankin, R. S., Jackson, T., Stutrud, J., and Switzer, G., "Comparison
Between Experiments and Predictions Based on Maximum Entropy for Sprays from a
Pressure Atomizer," Combustion and Flame, Vol. 86, 1991, pp. 73-89.
23
Faeth, G. M., "Mixing, Transport, and Combustion in Sprays," Progress in Energy and
Combustion Science, Vol. 13, 1987, pp. 293-345.
24
Crowe, C. T., Chung, J. N., and Troutt, T. R., "Particle Mixing in Free Shear Flows,"
Progress in Energy and Combustion Science, Vol. 14, 1988, pp. 171-194.
25
Putnam, A., "Integratable Form of Droplet Drag Coefficient," Journal of the American
Rocket Society, Vol. 31, 1961, pp. 1467-1468.
26
Nguyen, Q. V, and Dunn-Rankin, D., "Experiments Examining Drag in Linear Droplet
Packets " Experiments in Fluids, Vol. 12, No. 3, 1992, pp. 157-165.
27
Presser, C., Gupta, A. K., Semerjian, H. G., and Avedisian, C. T., "Droplet Transport
in a Swirl Stabilized Spray Flame," Journal of Propulsion and Power, Vol. 10, No. 5, 1994,
pp. 631-638.
28
Presser, C., Gupta, A. K., and Semerjian, H. G., "Aerodynamic Characteristics of
Swirling Spray Flames: Pressure Jet Atomizer," Combustion and Flame, Vol. 92, 1993,
pp. 25-44.
29
Goix, P. J., Edwards, C. F, Cessou, A., Dunsky, C., and Stepowski, D., "Structure of
a Methanol/Air Coaxial Reacting Spray Near the Stabilization Region," Combustion and
Flame, Vol. 98, 1994, pp. 205-219.
30
McDonell, V G., and Samuelsen, G. S., "Gas and Drop Behavior in Reacting and
Nonreacting Air-Blast Atomizer Sprays," AIAA Journal of Propulsion and Power, Vol. 7,
No. 5, 1991, pp. 684-691.
31
McDonell, V G., Wood, C. P., and Samuelsen, G. S., "A Comparison of Spatially-
Resolved Drop Size and Drop Velocity Measurements in an Isothermal Chamber and
a Swirl-Stabilized Combustor," Twenty-First Symposium (International) on Combustion,
Combustion Institute, Pittsburgh, PA, 1986, pp. 685-694.
32
McDonell, V. G., Adachi, M., and Samuelsen, G. S., "Structure of Reacting and Non-
Reacting Swirling Air-Assisted Sprays," Combustion Science and Technology, Vol. 82,
1992, pp. 225-248.
33
Ghaffarpour, M., and Chehroudi, B., "Experiments on Spray Combustion in a Gas Tur-
bine Model Combustor," Combustion Science and Technology, Vol. 92, 1993, pp. 173-200.
34
Edwards, C. F, and Rudoff, R. C., "Structure of a Swirl-Stabilized Spray Flame
by Imaging, Laser Doppler Velocimetry, and Phase Doppler Anemometry," Twenty-Third

pp. 1353-1359.
35
Mao, C. P., Wang, G., and Chigier, N., "An Experimental Study of Air-Assist Atom-
izer Spray Flames," Twenty-First Symposium (International) on Combustion, Combustion
Institute, Pittsburgh, PA, 1986, p. 665.
36
Hardalupus, Y, Taylor, A. M. K. P, and Whitelaw, J. H., "Velocity and Size Charac-
teristics of Liquid-Fuelled Flames Stabilised by a Swirl Burner," Proceedings of the Royal
Society of London, Vol. A428, 1992, pp. 159-172.
37
McDonell, V. G., Seay, J. E., and Samuelsen, G. S., "Characterization of the Non-
Reacting Two-Phase Flow Downstream of an Aero-Engine Combustor Dome Operating

at Realistic Conditions," American Society of Mechanical Engineers International Gas
Turbine Institute Meeting, Paper 94-GT-263, The Hague, The Netherlands, June 1994.
38
Wang, H. Y., McDonell, V G., Sowa, W. A., and Samuelsen, G. S., "Experimental
Study of a Model Gas TVirbine Combustor Swirl Cup, Part II: Droplet Dynamics," Journal
of Propulsion and Power, Vol. 10, No. 4, 1994, p. 446.
39
Edwards, C. F., and Marx, K. D., "Theory and Measurement of the Multipoint Statistics
of Sprays," Recent Advances in Spray Combustion, edited by K. K. Kuo, Progress in
Astronautics and Aeronautics, AIAA, Washington, DC, 1995.
40
McDonell, V. G., and Samuelsen, G. S., "Sensitivity Assessment of a Phase Doppler
Interferometer to User Controlled Settings," Liquid Particle Size Measurement Techniques:
2nd Volume, ASTM STP 1083, edited by E. D. Hirleman, W. D. Bachalo, and P. G. Felton,
American Society for Testing and Materials, Philadelphia, PA, 1990, pp. 170-189.
41
Zhu, J. Y, Bachalo, E. J., Rudoff, R. C., Bachalo, W. D., and McDonell, V. G.,
"Assessments of a Fourier Transform Doppler Signal Analyzer and Comparisons with a
Time-Domain Counter Processor," Atomization and Sprays, Vol. 5, No. 6, Nov. 1995, pp.
585-602. ___
42
Wang, M. R., and Huang, D. Y, "Measurement of u'g.ufd. in Mixing Layer Flow with
Droplet Loading," Atomization and Sprays, Vol. 5, No. 3, 1995, pp. 305-328.
43
McDonell, V. G., and Samuelsen, G. S., "Intra- and Interlaboratory Experiments to
Evaluate the Performance of Phase Doppler Interferometry," Recent Advances in Spray
Combustion, Vol. I, edited by K. K. Kuo, Progress in Astronautics and Aeronautics, AIAA,
Washington, DC, 1995.
44
Hardalupus, Y, Taylor, A. M. K. R, and Whitelaw, J. H., "Mass Flux, Mass Fraction
and Concentration of Liquid Fuel in a Swirl-Stabilized Flame," International Journal of
Multiphase Flow, Vol. 20, Supplement, 1994, pp. 233-259.
45
Sankar, S. V, Ibrahim, K. M., Buermann, D. H., Fidrich, M. J., and Bachalo, W. D.,
"An Integrated Phase Doppler/Rainbow Refractometer System of the Simultaneous Mea-
surement of Droplet Size, Velocity, and Refractive Index," Third International Congress
on Optical Particle Sizing, Yokohama, Japan, Aug. 1993.
46
Melton, L. A., and Vierdieck, J. F, "Vapor/Liquid Visualization for Fuel Spray,"
Combustion Science and Technology, Vol. 42, 1985, pp. 217-222.
47
Chraplyvy, A, R., "Nonintrusive Measurement of Vapor Concentrations Inside Sprays,"
Applied Optics, Vol. 20, No. 15, 1981, pp. 2620-2624.
48
Adachi, M., McDonell, V. G., and Samuelsen, G. S., "Nonintrusive Measurement of
Gaseous Species in Reacting and Non-Reacting Sprays," Combustion Science and Tech-
nology, Vol. 79, 1991, pp. 179-194.
49
Drallmeier, J. A., and Peters, J. E., "Experimental Investigation of Fuel Spray Vapor
Phase Characterization," Atomization and Sprays, Vol. 1, No. 1, 1991, pp. 63—68.
5()
McDonell, V. G., Adachi, ML, and Samuelsen, G. S., "Structure of Non-Swirling Air-
Assisted Sprays Under Reacting and Non-Reacting Conditions, Part I: Gas Phase Behavior,
Part II: Drop Behavior" Atomization and Sprays, Vol. 3, 1993, pp. 389-486.
51
McDonell, V. G., and Samuelsen, G. S., "Effect of Fuel Injection Mode on Fuel
Vapor in Reacting and Non-Reacting Methanol Sprays," 24th Symposium (International)
on Combustion, Combustion Institute, Pittsburgh, PA, 1992, pp. 1557-1564.
52
Bazile, R., and Stepowski, D., "Measurements of the Vaporization Dynamics in the
Development Zone of a Burning Spray by Planar Laser Induced Fluorescence and Raman
Scattering," Experiments in Fluids, Vol. 16, 1994, pp. 171-180.
53
Lozano, A., Yip, B., and Hanson, R. K., "Acetone: A Tracer for Concentration Measure-
ments in Gaseous Flows by Planar Laser-Induced Fluorescence," Experiments in Fluids,
Vol. 13, 1992, pp. 369-373.
54
Bazile, R., and Stepowski, D., "Measurements of Vaporized and Liquid Fuel Concen-
tration Fields in a Burning Spray Jet of Acetone Using Planar Laser-Induced Fluorescence,"
Experiments in Fluids, Vol. 20, No. 1, Nov. 1995, pp. 1-9.
55
Talley, D. G., Thanban, A. T. S., McDonell, V. G., and Samuelsen, G. S., "Laser Sheet
Visualization of Spray Structure," Recent Advances in Spray Combustion, edited by K. K.
Kuo, Progress in Astronautics and Aeronautics, AIAA, Washington, DC, 1995.
56
Seay, J. E., McDonell, V. G., Lee, S. W., and Samuelsen, G. S., "Atomization and
Evaporation Performance of an Airblast Injector in a Subsonic Crossflow," Journal for
Propulsion and Power (submitted for publication).
57
Hanson, R. K., "Combustion Diagnostics: Planar Imaging Techniques," Twenty-First
pp. 1677-1691.
58
Stepowski, D., Cessou, A., and Goix, P., "Flame Stabilization and OH Fluorescence
Mapping of the Combustion Structures in the Near Field of a Spray Jet," Combustion and
Flame, Vol. 99, 1994, pp. 516-522.
59
Zhu, J. Y., Tsuruda, T, Sowa, W. A., and Samuelsen, G. S., "Coherent Anti-Stokes
Raman Scattering (CARS) Thermometry in a Model Gas Turbine Can Combustor," Journal
of Engineering for Propulsion and Power, Vol. 115, 1993, pp. 515-521.
60
Dunn-Rankin, D., Switzer, G. L., Obringer, C. A., and Jackson, T. A., "Effect of Droplet
Induced Breakdown on CARS Temperature Measurements," Applied Optics, Vol. 29,1990,
pp. 3150-3159.
61
Nakabe, K., Mizutani, Y, Akamatsu, F, and Fujioka, H., "Observation of Droplet
Group Combustion in Terms of Simultaneous Measurement of Mie Scattering and Spectral
Luminosity from Spray Flames," Atomization and Sprays, Vol. 4, 1994, pp. 485-500.
62
McDonell, V. G., and Samuelsen, G. S., "An Experimental Data Base for the Com-
putational Fluid Dynamics of Reacting and Non-Reacting Methanol Sprays," Journal of
Fluids Engineering, Vol. 117, Mar. 1995, pp. 145-153.

Chapter 8
Structure of a Coflow Laminar Spray Diffusion Flame

G. Chen* and A. Gomezf

Yale University, New Haven, Connecticut 06520-8286
I. Introduction
I N most practical spray combustion systems droplets are known to burn col-
lectively as a cloud, rather than individually.1 Accordingly, it is desirable to
extend the present understanding of single droplet combustion to include droplet
interactive behavior. Theoretical studies have identified different regimes of cloud
burning in simplified cloud models with three-dimensional interactions.2"4 Exper-
imental verification of group burning regimes, on the other hand, has been limited
to two types: 1) systems with a small number of droplets or a single droplet stream,
or two-dimensional droplet streams, in which reasonably quantitative observations
were made5'8; and 2) turbulent spray combustion systems with three-dimensional
droplet interactions in which largely qualitative observations on group combustion
were reported.1 In the first case, to achieve better control, the combustion envi-
ronment was clearly oversimplified, whereas in the second class of experiments,
turbulence obscured the interpretation of the results. To bridge the gap between
these two extremes, three-dimensional laminar spray flames are considered here,
as well-defined and well-controlled environments to study the evaporation, com-
bustion, and interaction of droplets.
An electrostatic spray (ES) of charged droplets was shown to be a useful tool
for this type of combustion experiment.9 Among its many advantages, it can
produce quasimonodisperse charged droplets that are self-dispersed by coulombic
repulsion. In previous studies the combustion of such sprays in laminar counterftow
diffusion flames was investigated.9'11 Their behavior in laminar axisymmetric
coflow diffusion flames is examined in this chapter.
An analytical model for a Burke-Schumann diffusion flame with fuel spray
injection was recently formulated in a configuration similar to the present one.12
No experiments, however, were ever reported on strictly laminar, stable, coflow
diffusion spray flames, the flame of Moore and Moore being no exception since,
although labeled as laminar, it was actually transitional, with possibly just a
laminar base.13 The lack of experimental evidence on laminar spray flames is
Copyright ©1995 by the American Institute of Aeronautics and Astronautics, Inc. All
rights reserved.
Tostdoctoral Research Associate, Department of Mechanical Engineering.
Associate Professor, Department of Mechanical Engineering.
187
188 G. CHEN AND A. GOMEZ
explained by a recent study in which investigators reported that buoyancy-induced

instabilities prevented the establishment of a stable laminar spray flame at normal
gravity, under conditions in which the slip between the two phases was small.14
In contrast, a self-sustained, stable, laminar, diffusion flame, with fuel injected
as a monodisperse electrostatic spray, was successfully established in this study.
Two features of practical sprays are retained in this environment: slip between the
droplets and the host gas and three-dimensional droplet-droplet interactions. The
electrostatically generated droplets travel, in fact, at relative velocities of several
meters per second with respect to the gas phase. The spray in the present case
is dense enough so that interactions between droplets in terms of mass and heat
transfer set in and oxidizer penetration in the innermost region of the spray is
impeded. Consequently, a fuel-rich, relatively cold, and nonflammable mixture is
formed, where droplets do not burn individually but rather as a group or cloud
surrounded by a common diffusion flame. The structure of such a flame is the
topic of this chapter. Preliminary experiments were presented elsewhere.15
II. Experimental Arrangement

A schematic of the burner is shown in Fig. 1. The typical system to produce
a monodisperse electrostatic spray has been described in detail elsewhere.16 It
comprises a metal capillary maintained at a high potential and a suitable ground
electrode to close the electric circuit, with the liquid fuel being fed through the
Soot
Blue Flame
n-Heptane
Fig. 1 Schematic of the laminar coflow spray burner.
STRUCTURE OF A COFLOW LAMINAR SPRAY DIFFUSION FLAME 189
capillary. Its adaptation to combustion environments requires a careful selection

of the ground electrode for two reasons, both related to the need to shield the high
electric field region where the spray is generated from the flame: first, electric
breakdown of the relatively conductive combustion gases may otherwise occur
that would adversely affect the stability of the spray; second, it is desirable to
minimize electric effects on the combustion processes, so that observations made
in these experiments are generally applicable to laminar spray flames, regardless
of the particular spray generation method. The ground electrode satisfying this
prerequisite is a porous bronze plate, as depicted at the top of the burner in Fig. 1.
It has a 1.9-mm central opening and is positioned 0.9 mm downstream of the

tip of the capillary where the electrospray is anchored. An electrospray in the
so-called cone-jet mode was established when a voltage of 1492 V was applied
to the metal capillary.16 In this mode the liquid meniscus at the outlet of the
capillary had a conical appearance with a thin jet issuing from the cone tip. The jet
broke up farther downstream, near the plate opening, by varicose wave instabilities
into a stream of charged droplets of two sizes: larger, primary ones and smaller
satellites. The primary droplets accounted for more than 98% of the total flow rate
whereas the satellites for less than 2%.17 The generated droplets passed through the
opening in the porous plate because their large inertia overcame any electrostatic
attraction of the ground plate. About 1 mm above the plate the stream opened
up into a conical fan and ultimately dispersed downstream because of the droplet
mutual charge repulsion. Some droplets, the satellites, reversed their flight and fell
back to the plate. The bulk of the flow rate, consisting of monodisperse droplets,
continued to move away from the plate and fed a self-sustained, stable, continuous,
axisymmetric, candlelike flame. In the present experiments a coflow of air at 2.6
1/min was supplied through the burner housing. After passing through a perforated
plate, a passage filled with glass beads and the porous plate, it exited with a
uniform velocity of about 5 cm/s. This shroud flow improved the flame stability
and ensured overventilated conditions, the overall equivalence ratio being equal to
0.35. A small teflon tube (4-mm inner diameter) was mounted coaxially with the
capillary, to keep fuel and oxidizer separated all of the way to the burner mouth
and also to insulate electrically the capillary from the burner housing. The metal
capillary and the teflon tube were kept coaxial with the burner by a perforated
teflon plate. The liquid fuel was heptane that was doped with 0.3% by weight of
a proprietary antistatic additive, Stadis 450 (Du Pont), whose sole effect was to
increase the fuel electric conductivity and ensure proper electrospray dispersion.
The liquid fuel flow rate was kept constant at 8.4 cc/h. In some exploratory
experiments, elimination of the shroud air flow caused no changes in the flame
appearance, which implies that the satellite droplets that had fallen back onto the
burner played no significant role in the stabilization and burning of the flame.
Droplet size, velocity, and concentration were measured by a commercial two-
component phase Doppler anemometer (PDA) (Dantec Electronik). The use of a
nearly monodisperse spray significantly facilitates the size measurements, over-
coming limitations in the instrument dynamic range. Further details on the diag-
nostic system are given elsewhere.16
The gas-phase temperature was determined by cylindrical, silica-coated Pt-
6%Rh/Pt-30%Rh thermocouples, with a bead diameter of 140 jum. The presence
of soot or droplets in some regions of the flame prevented a complete mapping
of the temperature field. No corrections for convective and radiative effects were
made for reasons to be explained.
10mm
Fig. 2 Photograph of the flame.

A. Flame Appearance
The monodisperse electrospray of heptane, once ignited, burned as a stable,
axisymmetric flame. The flame, shown in Fig. 2, had in all respects the appearance
of a classic candle flame, except that here the fuel originated from a point source of
droplets rather than from a candlewick. Five regions could be readily identified by
visual inspection: 1) a bright blue, horizontal, flame ring, that acted as stabilization
region of the combustion process and was positioned 3.5 mm above the porous
plate; 2) a relatively dark inner region, in the lower part of the flame, right above
the flame ring; 3) a bright yellow outer shell whose luminosity was due to black-
body radiation from the soot at high temperature; 4) a diffuse, deep blue flame that
was the outermost luminous region in the lower part of the flame and appeared
to envelope the combustion region; and 5) an orange region at the very top of the
flame that was presumably the soot oxidation region. The tip of the flame flickered
slightly at a.frequency of about 12 Hz, similarly to what is observed in gaseous
flames and attributed to buoyancy-induced instability.18 The luminous height of
the flame was 34.5 mm.
B. Evidence of Group Combustion Under Very Dilute Conditions

Figure 3 shows radial scans of temperature at four axial locations. The two lower
scans show a toroidal pattern with the peak temperature region near the flame lo-
cation, similarly to gas diffusion flames.19 Farther up into the flame, the trends are
the same, albeit the measurements are incomplete because of soot deposition on
the thermocouples bead. Eventually, as the flame closes in on the centerline, the
location of the peak temperature should reach the axis (see Fig. 6) and the profile
would show a monotonically decreasing temperature as a function of radial coordi-
nate. Peak temperatures at different heights above the burner are all approximately
1700 K. A similarity of thermal structures between gaseous and spray diffusion
flames was already established in transitional/turbulent conditions.20 The present
measurements confirm it under strictly laminar diffusion flames.
If individual droplet burning were occurring in the core of the spray, streaks
of 40-/xm droplets surrounded by a flame should have been discerned, since the
droplets followed diverging trajectories. No observations of this sort were made.
Also, under individual droplet burning, a much more uniform temperature profile
in the radial direction would be expected, in contrast with the data in Fig. 3.
Thus, the present experiments offer no evidence of individual droplet burning
1loUU
ftnn
1600
°P: '. 'Q '.

1400
1200 i s >° Q
\ y..\
' . •
;
1000 - o - atz=8.99mm • '. .'•
• - a - - at z=10.99mm • • '. •
800
- - • • - - atz=12.99mm \ • '. -a -
600 7 •• - at z=14.99mm %
' to "
o
b
/inn , , i , , -, i , , , , i , , , , i , , , , i , , , , i . , , . i , , , ,
0 1 3 4 5 6 7
r(mm)
Fig. 3 Radial scans of the gas-phase temperature at selected axial locations.
500
400 -
300 -
200 -
z=9.99mm
z=l 1.99mm
z=13.99mm
100 -
-4
Fig. 4 Radial scan of a) the ratio of interdroplet spacing over droplet average diam-
eter (left ordinate) and b) the average droplet diameter (right ordinate) at selected
axial locations.
in the core. Rather, conditions are more akin to what is often labeled internal
group combustion, with a continuous flame enveloping a relatively cooler core in
which droplets are evaporating. Only a few droplets were found to continue to burn
outside the flame. Individual droplet burning in the core of the flame was prevented
because the outer flame impeded oxygen penetration and air entrainment from the
bottom of the flame was also very limited. Further evidence in support of this
conclusion will be given later in the discussion on droplet evaporation.
Since the treatment of group combustion is often cast in terms of the ratio of
interdroplet spacing to droplet diameter, l/D, this ratio is plotted in Fig. 4 at
selected axial locations, as computed from the inverse cubic root of the measured
droplet number density and the size. The vertical bars on each curve of l/D indicate
where the peak temperature is at each axial location. Also plotted in Fig. 4 is the
droplet diameter variation along the radial direction at z = 11.99 mm, with the
length of the vertical bars being equal to two standard deviations of the diameter at
every location. Notice that in the core of the spray, the Relative Standard Deviation
(RSD), that is the ratio of the standard deviation to the average diameter, is fairly
small, on the order of 10-15%. Therefore, close to the centerline, the droplet
size distribution is sufficiently narrow for l/D to be a meaningful quantity. At
the outermost locations, on the other hand, the RSD increases significantly. For
example, the droplet diameter data in Fig. 4 have RSDs between 0.25 and 0.4 at
the locations outside of the flame, which implies that the size distributions are
too broad for l/D to be computed reliably. Furthermore, on the oxidizer side
nothing prevents the droplets from burning individually and the concept of group
combustion is not particularly useful. In conclusion, if we conservatively focus just
on the region in the core of the spray, the values of l/D range between 20 and 60.
Especially those in the upper portion of the evaporation region, are significantly
70
60
50
i 40
Q
30
20
10
10 15 20 25
z (mm)
Fig. 5 Average droplet diameter and average droplet velocity vs axial distance from
the burner.
larger than the typical value of 10-25, that experiments on droplet streams and
theoretical models of multiple droplet combustion report as upper limit for the
droplets to start experiencing significant interactions.6'21 Yet, the lack of evidence
of individual droplet burning implies that interactions must be present even in this
study, which suggests the inadequacy of systems of a small number of droplets or
droplet streams to represent properly three-dimensional group combustion.
C. Droplet Life Histories

The evolution of the droplets along a particular trajectory, the flame centerline,
is considered in Fig. 5, showing the mean droplet diameter and axial velocity
measured on the flame axis by the PDA and plotted vs the axial distance from
the porous plate. The droplets were injected in the combustion region with an
initial mean diameter of approximately 50 /zm and an initial average velocity of
5 m/s. As the droplets entered the flame environment they continued to decelerate
under the combined action of the drag and of a small electric force, that is due to
the coulombic repulsion from the droplets and to some penetration effects from
the high field region within the burner housing where the spray is formed. The
coulombic repulsion is crucial in causing spray dispersion at the base of the flame.
Thanks to this effect, the spray originates from a point source of droplets that are
spread into a conical fan, a three-dimensional configuration that is more realistic
in simulating spray combustion than the single droplet stream that the breakup
of a jet by axisymmetric instabilities produces. The combined effect of the two
forces of electrical nature is estimated at less than 15% of the drag force at z = 6
mm. Farther up into the flame, however, electric effects become less and less
significant since the hypodermic needle is farther away and the spray is even more
diluted. Consequently, the present results should be of general applicability and

not specific to only electrostatic sprays.22
The droplet size increased slightly in the first 2 mm. This phenomenon is at-
tributed to the droplet initial heat up. A change in temperature of 70 K of liquid
heptane can, in fact, account for a 1.6-/^m increase in droplet diameter. Further-
more, estimates of the droplet heat up time using a lumped approach yielded char-
acteristic times that were comparable to the droplet residence time in this region.
Farther downstream, the droplet diameter decreased because of evaporation. It is
important to remark that the droplets were virtually monodisperse locally, as indi-
cated by the small value of the RSD that was 0.09 at the first measurement location
2 mm above the burner. The RSD remained small up to z = 12 mm. At z = 16 mm
the RSD was 0.26. Beyond that point, the size distribution continued to broaden
until no droplet signal was detected past z = 22 mm. At this height, approximately
corresponding to the height of the dark zone described in the preceding section,
evaporation was complete. The data in Fig. 5 show that in their journey along the
flame axis the droplets continued to decelerate nearly linearly with axial distance.
This quasilinear dependency can be predicted if the Stokes drag law and the d-
square evaporation law are obeyed, as detailed in Ref. 22. Our findings also confirm
analytical predictions obtained in laminar sprays under similar conditions.12
Figure 6 presents the thermocouple temperature measurements along the flame
axis. These measurements, since they will be used quantitatively, require a discus-
sion of errors due to conduction gain/loss, radiation loss, and the impingement of
droplets on the bead. The latter is estimated to decrease the temperature reading
on the thermocouples by no more than 40 K at z > 10 mm where the measurement
of the temperature on the axis became sufficiently steady.22 Standard corrections
for radiative losses resulted in a 55-K temperature increase. Conduction gain was
estimated at less than 70 K based on the results of Sato et al.23 Since the effect of the
1800
1700
1600
1500
1400
1300
1200
1100
0 5 10 15 20 25 30 35 40
z (mm)
Fig. 6 Gas-phase temperature measured along the burner axis vs axial distance from
the burner.
conduction gain is partially counterbalanced by those due to droplet impingement

and radiative loss, the actual temperature should be about 25 K higher than what
was measured, an inconsequential error. Because of these partial cancellations,
the temperature measurements were left uncorrected.
In contrast with gas diffusion flames,24 the temperature along the axis decreased
in the first few measurements. This trend is a consequence of the flame geometry
and the close proximity of the anchoring ring flame at the lowest measurement
positions (see Fig. 2) that results in more efficient heat transfer to the axis and
relatively high temperatures. Farther downstream, as the flame opened up, the
temperature in the core decreased at first, and then it became fairly constant, in part
as a consequence of the ongoing evaporation. Subsequently, it increased, peaking at
z — 33 mm a few millimeters below the tip of the flame. No temperature is reported
in the region between 17 and 31 mm because of the high soot concentration that
prevented any reliable measurements.
It is instructive to compare the centerline droplet history to that along two
other trajectories, which are labeled T-l and T-2 in Fig. 7a (r-z plane). These
^trajectories were determined using the simultaneous measurements of axial and
radial velocity components of the droplets. The two additional trajectories cover
paths closer to the flame location. The latter, also shown in the figure, is given by
the radial location of the temperature peaks at various axial positions. Figure 7b
shows the mean droplet diameter along the three trajectories. The length of the
error bars at selected locations is equal to two standard deviations of the measured
droplet distributions. The closer the trajectory was to the flame zone, the smaller
the initial size of the droplets was because of evaporation effects, but the droplets
still retained remarkably monodisperse size distributions through the bulk of their
evaporation history in both T-l and T-2. The size distribution began to broaden at
z = 14 mm, where the RSD values were 0.24 at z = 14.5 mm and 0.21 at z = 14
mm on T-l and T-2, respectively. In the remaining few millimeters, the percentage
of the initial mass flux that had yet to evaporate was negligibly small (<1%). The
nonmonotonic behavior of the droplet diameter for z > 16 mm may be attributed to
the combined effect of the broadening of the size distribution and of the preferential
evaporation of the smallest droplets. A similar phenomenon was also observed in a
counterflow configuration.9 From the examination of the life history of the droplets
along these three distinct trajectories and the visual inspection of the flame, it is
concluded that the dark region in the lower part of the flame is coincident with the
bulk of the droplet vaporization region. This region occupies a sizable portion of
the flame, with a height reaching 2/3 of the total flame height.
The measurements of the droplet velocity field along the three trajectories en-
ables us to convert the spatial abscissa in Fig. 7b into a temporal one by computing
= Jof *V (1)
where s is the streamwise coordinate and V is the velocity vector. The results are
replotted in Fig. 8 as the square of the droplet diameter vs time. The goal is to
examine whether there were regions of quasisteady evaporation at least through
part of the droplet lifetime, despite the fact that the droplets underwent group
combustion. The data clearly show that after an initial transient, the droplets follow
the J-square law through a significant portion of their lives, under conditions in
which their size distributions are either monodisperse or sufficiently narrow to
justify the use of mean diameters. The continuous lines in the figure are linear fits
5 — o- - Axis
: - o- - r (T-2) I
4 r o r(Tpk) -
/-^ 3 — ^^_ o -
j=
r
2
Î /
,r 0-°"°
^^ •*-*"**'*-- ^;
1 - / . f f P ^9^*'* ~
0 - O— O —0 — O — 0— -O —0 —— O — 0- -0 —O -
() 5 10 15 20 2
a) z (mm)
60
] 0 - - o - - on Axis ]
50
Ojg ^ ' • ,Q - - • - - on T- 1 .
|QQ» o - - a - - on T-2
40 ~ > • »
'p a '
^ b'ru
Q
30 b
^; -f-
,1'.°.
^;
1 • •
°- • '. *>.
•.
20 o -
° .-...'.
b-a • o
10 " , , , , 1 i , , , 1 , , , 1 , , , , 1 , , , , "
10 15 20 25
b) z (mm)
Fig. 7 Comparison of droplet life histories along three trajectories: a) trajectories in
the r-z plane; b) average droplet diameter vs axial coordinate as measured from the
burner mouth.
3000
2500
2000
1500
1000
500
L_L
- 1 0 1 2 3 4 5 6 7
t (ms)
Fig. 8 Square of the average droplet diameter along selected trajectories vs residence
time, as calculated from the initial measurement location.
to the three profiles yielding average evaporation coefficients of 0.577, 0.58, and
0.66 mm2/s for the centerline, T-l and T-2 trajectories, respectively. They differ
only by 13% despite the different temperature along each trajectory, as indicated
in Fig. 3. The approximate constancy of the evaporation coefficients confirms one
of the assumptions in group combustion theory2 and is due to the insensitivity of
evaporation coefficient to the gaseous temperature for the range of temperatures
studied here.
Since at least for the centerline measurements the temperature field is character-
ized for the first five points for which the d-square law applies, the experimentally
inferred evaporation coefficient can be compared with that calculated with the
assumption of quasisteady evaporation of an isolated droplet in a convective envi-
ronment. Accordingly, the evaporation coefficient K is
where p\ is the liquid density, A and Cp are the average gas thermal conductivity
and specific heat capacity, evaluated according to Williams,25 and B is the transfer
number given by
= Cp(T00~Tb)/L (3)
where L is the fuel latent heat of vaporization, Tb is assumed to be the boiling point
for heptane and 7^ is the average temperature in the surrounding environment. In
the evaluation of 5, the temperature measured by the thermocouple on centerline
was used for 7^; it was approximately constant at T = 1200 K in the range
of interest, that is z > 8 mm. The expression for the transfer number in Eq. (3)
applies to the evaporation of an isolated droplet in the absence of combustion.
The heat released from the flame contributes indirectly to the value of T^ via heat
transfer from the flame. The evaporation coefficient in Eq. (2) was also corrected
for forced convection effects, using the Frossling formula25 based on the relative
velocity between the droplets and gas. The computed evaporation coefficient was
0.66 mm2/s, a value that is 13% larger than the experimentally determined result.
This discrepancy can be attributed to limitations in the accuracy of Eq. (2), which
is estimated at 15% (Ref. 26), and to inevitable errors in the measurement and
processing of the data that are estimated at about 4.4%. In conclusion, droplet
interactions are sufficiently intense to prevent individual droplet burning even
in this highly diluted spray, but not intense enough to cause a departure from a
d-square law evaporation behavior.
IV. Conclusions
The following are the principal conclusions of this experimental study on the
structure of an axisymmetric laminar spray flame.
1) A self-sustained, laminar diffusion flame was successfully established with
the fuel injected as a monodisperse spray under conditions of significant slip
between the droplets and the gas.
2) The flame appearance and temperature field showed little or no evidence of
individual droplet burning. Rather, internal group combustion was experienced,
in which the bulk of droplet evaporation occurred in an inner core that provided
vapor diffusing into a flame enveloping the bulk of the droplet cloud.
3) Droplet life histories showed that the droplets follow the ^-square law during
a significant part of their lives with evaporation coefficients on the order of 0.6
mm2/s. The experimentally inferred evaporation coefficient on the centerline is
within the margin of error of that calculated in the case of isolated droplet evapo-
ration in a convective environment. Thus, although the interactive effect between
the droplets results into a flame enveloping a droplet cloud, individual droplet
evaporation still applies within the cloud; that is, the droplet interaction is not
sufficiently intense to cause a departure from the d-square law.
Acknowledgments
The support of NASA, under the Microgravity Science and Applications Pro-
gram, Grant NAG3-1259, and the National Science Foundation (NSF) through an
NSF Young Investigator Award, and Equipment Grant CTS-9112601, is gratefully
acknowledged.
References
^higier, N. A., and McCreath, C. G., "Combustion of Droplets in Sprays," Acta Astro-
nautica, Vol. 1, 1974, pp. 687-710.
2
Labowsky, M, and Rosner, D. E., "Group Combustion of Droplets in Fuel Clouds.
I. Quasi-steady Predictions," Evaporation-Combustion of Fuels, edited by J. T. Zung,
American Chemical Society, 1978, pp. 63-79.
3
Chiu, H. H., Kim, H. Y., and Croke, E. J., "Internal Group Combustion of Liq-
uid Droplets," Nineteenth Symposium (International) on Combustion, Combustion Inst.,
Pittsburgh, PA, 1982.
4
Correa, S. M., and Sichel, M., "The Group Combustion of a Spherical Cloud of Monodis-
perse Fuel Droplets," Nineteenth Symposium (International) on Combustion, Combustion
Inst., Pittsburgh, PA, 1982.
5
Brzustowski, T. A., et al., "Interaction of Two Burning Fuel Droplets of Arbitrary Size,"
AIAA Journal, Vol. 17, 1979, pp. 1234-1242.
STRUCTURE OF A CO FLOW LAMINAR SPRAY DIFFUSION FLAME 199
6
Sangiovanni, J. J., and Labowsky, M., "Burning Times of Linear Fuel Droplet
Arrays: A Comparison of Experiment and Theory," Combustion and Flame, Vol. 47, 1982,
pp. 15-30.
7
Zhu, J. Y, and Dunn-Rankin, D., "Temperature Characteristics of a Combusting Droplet
Stream," Twenty-Fourth Symposium (International) on Combustion, Combustion Inst.,
Pittsburgh, PA, 1992.
8
Queiroz, M., and Yao, S., "A Parametric Exploration of the Dynamic Behavior of Flame
Propagation in Planar Sprays," Combustion and Flame, Vol. 76, 1989, pp. 351-368.
9
Chen, G., and Gomez, A., "Counterflow Diffusion Flames of Quasi-Monodisperse Elec-
trostatic Sprays," Twenty-Fourth Symposium (International) on Combustion, Combustion

Inst., Pittsburgh, PA, 1992.
10
Chen, G., and Gomez, A., 'The Individual Effect of Droplet Size and Velocity on the
Structure of Counterflow Spray Diffusion Flames " Eastern States Section Fall Technical
Meeting, Combustion Inst., Pittsburgh, PA, 1993.
H
Gomez, A., and Chen, G., "Charge-Induced Secondary Atomization in Diffusion
Flames of Electrostatic Sprays," Combustion Science and Technology, Vol. 96, 1994,
pp. 47-59.
12
Greenberg, J. B., "Spray Diffusion Flames with Arbitrary Initial Droplet Velocity Dis-
tributions," Combustion Science and Technology, Vol. 75, 1991, pp. 11-30.
13
Moore, J. G., and Moore, J., "The Distributions of Temperature and Major Species
in Laminar Diffusion Flames," Sixteenth Symposium (International) on Combustion, Com-
bustion Inst., Pittsburgh, PA, 1976.
14
Levy, Y, and Bulzan, D., "On the Combustion of a Laminar Spray," NASA TM-106210,
1993.
15
Gomez, A., and Chen, G., "Experimental Investigation on the Combustion of Monodis-
perse Sprays in Co-flow Laminar Diffusion Flames," Sixth International Conference on
Liquid Atomization and Spray Systems, Rouen, France, 1994.
16
Gomez, A., and Tang, K., "Charge and Fission of Droplets in Electrostatic Sprays,"
Physics of Fluids, Vol. 6, 1994, pp. 404-414.
17
Tang, K., and Gomez, A., "On the Structure of an Electrostatic Spray of Monodisperse
Droplets," Physics of Fluids, Vol. 6, No. 7, 1994, pp. 2317-2332.
18
Ellzey, J. L., Laskey, K. J., and Oran, E. S., "A Study of Confined Diffusion Flames,"
Combustion and Flame, Vol. 84, 1991, pp. 249-264.
19
Santoro, R. J., et al., "The Transport and Growth of Soot Particles in Laminar Diffusion
Flames," Combustion Science and Technology, Vol. 53, 1987, pp. 89-115.
20
Onuma, Y, and Ogasawara, M., "Studies on the Structure of a Spray Combustion
Flame," Fifteenth Symposium (International) on Combustion, Combustion Inst., Pittsburgh,
PA, 1975.
21
Marberry, M., Ray, A. K., and Leung, K., "Effect of Multiple Particle Interactions on
Burning Droplets," Combustion and Flame, Vol. 57, 1984, pp. 237-245.
22
Chen, G., "An Experimental Investigation on Laminar Diffusion Flames of Monodis-
perse Sprays," Ph.D. Thesis, Yale Univ., New Haven, CT, 1995.
23
Sato, A., et al., "A Correctional Calculation Method for Thermocouple Measurements
of Temperatures in Flames," Combustion and Flame, Vol. 24, 1975, pp. 35-41.
24
Gomez, A., Littman, M. G., and Glassman, I., "Comparative Study of Soot Formation
on the Centerline of Axisymmetric Laminar Diffusion Flames: Fuel and Temperature
Effects " Combustion and Flame, Vol. 70, 1987, pp. 225-241.
25
Williams, F. A., Combustion Theory, Addison-Wesley, Reading, MA, 1985.
26
Law, C. K., and Williams, F. A., "Kinetics and Convection in the Combustion of Alkane
Droplets," Combustion and Flame, Vol. 19, 1972, pp. 393-405.

Chapter 9
Coaxial Airblast Atomizers with Swirling Air Stream

Y. Hardalupas* and J. H. Whitelaw1

Imperial College of Science, Technology and Medicine,
London SW7 2BX, United Kingdom
Nomenclature
DG , KG = inner diameter and radius of the gaseous jet
DL = inner diameter of liquid jet
Z)0, RQ = outer diameter and radius of the liquid jet
d = droplet diameter
G = volume flux of liquid
MFR = gas-to-liquid mass flow rate ratio, Eq. (3)
MMD = mass median diameter, Eq. (9)
MR = gas-to-liquid momentum ratio, Eq. (4)
R05 = radius defined by Eq. (11)
/?oo = radius of measurement location at the largest distance from the
nozzle axis
5, S' = swirl numbers according to Eqs. (6) and (7)
SMD = Sauter mean diameter, Eq. (8)
SMD — spatially averaged SMD at one axial station, Eq. (10)
St = Stokes number
U, V, W = axial, radial, and tangential velocity components
UG = gaseous velocity averaged over the area of the annulus
UL = liquid velocity averaged over the area of the liquid jet
u', i/, w' = rms of fluctuations of axial, radial, and tangential velocity
components
wTJ, uw = cross correlations
UG,UL = local gaseous and liquid velocities
VR = gas-to-liquid velocity ratio, Eq. (5)
WeCT\t = critical Weber number for droplet breakup
rights reserved.
*EPSRC Advanced Research Fellow, Mechanical Engineering Department. Member
AIAA.
^Professor, Head of Thermofluids Section, Mechanical Engineering Department.
201
202 Y. HARDALUPAS AND J. H. WHITELAW
Weexit, We\oc = Weber number at nozzle exit, Eq. (1), and local value, Eq. (2)
fjiG,vG — dynamic and kinematic viscosity of the gas
PG = density of gas
a = surface tension
r = droplet response time, Eq. (12)
Introduction
I T is important to be able to control and predict the size characteristics of the

sprays produced by coaxial injectors for the performance of rocket engines.

Early work on sprays from coaxial air-blast injectors was performed by droplet
capture and imaging techniques1 and hot-wax freezing2'3 and accuracy was limited.
The characteristics of sprays from coaxial air-blast atomizers with axial streams
have been measured with nonintrusive laser interferometric techniques,4"9 which
allowed more accurate and detailed size measurements. Hardalupas and Whitelaw,9
for example, found maximum mean diameters only close to the axis of symmetry,
where most of the liquid remains and limits the rate of spread of the sprays and,
thus, the mixing between fuel and oxidizer over a wide range of liquid and gas
flow rates similar to those of rocket engines. The development of the sprays with
distance from the nozzle confirmed that the rate of spread is limited even for
distances comparable to the length of the combustion chamber of the preburners
of the Space Shuttle main engine (SSME), and this is a disadvantage of the
performance of coaxial air-blast atomizers with axial airstreams.
Although such coaxial atomizers with axial external gas streams have been
widely used in rocket engines, problems have been reported and can be traced back
to the atomization of liquid oxygen during the startup of the engine; for example,
combustion instabilities are related to the injector performance10 and cracks on
sheet metal, the turbopump dome, and on the blades of the first and second stage
of the turbopump of the preburner of the SSME.11 Efforts for improved designs
of coaxial air-blast atomizers include coaxial arrangements with swirling liquid
streams and axial external gas flow.12"14 However, no study of sprays produced
by coaxial atomizers with axial liquid streams and swirling external gas flow is
available that could be an alternative design of coaxial injectors for rocket engines.
Studies of sprays produced by air-assisted swirl atomizers15"17 have shown that
introduction of swirl in the external air stream of air-assisted atomizers increases
the rate of spread of the sprays and improves atomization. However, the complexity
of the nozzle geometry of commercial air-assisted atomizers and the flow has not
allowed general conclusions. The simple coaxial atomizer geometry of the current
study, although relevant to the operation of rocket engines, allows better under-
standing of the physics involved in the atomization process, and its conclusions can
be extended to the more complex geometries of commercial air-assisted atomizers.
The main spray characteristics of importance to combustion applications are the
mean droplet size, which influences the evaporation rate and the droplet response to
the gaseous flowfield, and the rate of spread, which influences the mixing between
fuel and oxidizer.18"20 The mean diameters and rate of spread of sprays have been
represented by empirical correlations in terms of velocity, density, viscosity, and
surface tension of the gas and the liquid and the geometry of the nozzle21 with
limited success. These two characteristics of sprays from coaxial air-blast nozzles
with axial liquid streams and swirling external gas flow are examined here as a
function of swirl number of the airstream, gas and liquid flow conditions, and the
COAXIAL AIRBLAST ATOMIZERS WITH SWIRLING AIR STREAM 203
dimensions of the nozzle and are compared to those of sprays produced by coaxial
atomizers with axial streams. It is common for the important parameters to be
expressed in terms of nondimensional numbers and the nondimensional groups
used in the presentation of the results are defined as follows.
The exit Weber number is
We^t = pG(UG-UL)2DL/a (1)

where UG — UL is the relative velocity between the gaseous and the liquid jets at
the exit and DI is the diameter of the liquid jet exit. The local Weber number of
the droplets is quantified by the local slip velocity, UG — UL, between the gas and
droplets with diameter d, so that
uL)d/a (2)
The gas-to-liquid mass flow rate ratio is
MFR = pcUc(D2G - Dl)/pLULDl (3)

whereDo is the external diameter of the liquid tube and DG is the gaseous jet
diameter. Two additional scaling parameters are the gas-to-liquid momentum and
velocity ratio,
MR = PGU2(D2G- D2} I PL U2 D2L (4)
VR = UG/UL (5)
The swirl number S is defined as the ratio of the axial flux of the angular momentum
to the axial flux of axial momentum and will be used to quantify the swirl intensity
in the gaseous stream; that is,
rf° UWr2dr
S = -^—R ———— (6)
where RQ is the outside radius of the liquid tube, U and W are the local mean axial
and tangential velocity at the exit of the gaseous jet, respectively, and r is the local
radius.
The flow configurations and instrumentation are considered in the next section,
which is followed by the presentation of the results. The mechanisms affecting
atomization and rate of spread are discussed in the section following the results
and the consequences of nozzle dimensions and swirl number are correlated to
the nozzle operating conditions. The paper ends with a summary of the more
important conclusions.
Experimental Arrangement and Instrumentation

The air-blast atomizer of Fig. 1 was constructed and operated at atmospheric
pressure with air in the annulus and water in the central tube which comprised a 10-
mm-diam tube reduced to an external diameter of 2.95 mm with internal diameter
DI of 2.3 mm (0.090 in.). The air flow rate was supplied through four inlets with
their axes normal to that of the nozzle with flow straighteners to remove residual
swirl and a conical contraction to further reduce flow asymmetries. Nozzles with
120-
4 inlets of axial gas

4 tangential inlets of swirl gas
OP 02.95 ID 02.3
V z, U
Fig. 1 Geometry and main dimensions of the coaxial atomizer.
straight exits, as shown in Fig. 2a, could be attached at the exit of the air jet with
diameters of 8.95, 14.95, and 22.95 mm resulting in annular widths of 3, 6, and
10 mm; the length of the straight part of the nozzle was 18, 28, and 38 mm for the
three diameters, respectively.
The flow conditions for the examined sprays are summarized in Table 1 and
cover a range of Weber numbers at the exit of the nozzle from 230 to 630, of air-
to-liquid momentum ratio from 21 to 65, velocity ratios from 14 to 38, mass flow
rate ratio from 0.5 to 2.6, constant liquid jet Reynolds number around 10,000, and
air jet Reynolds numbers from 76,000 to 140,000. The Reynolds number of the air
flow was based on the velocity averaged over the area of the annulus at the exit, UG,
and the exit diameter of the air jet DG. Other parameters were based on the air and
liquid velocities averaged over the area of the annulus and the area of the liquid jet,
respectively. The operating conditions of the injectors of the preburner of the SSME
are also presented for comparison purposes. Although the exit Weber number of
the current atomizers was two orders of magnitude lower than that of the SSME
injectors, it was not as important for the atomization process of these injectors22
as the remaining parameters which are comparable to those of the SSME.
Two methods were used to generate swirl of the airstream in the annulus. The
first used four tangential inlets, in additional to the four axial inlets of Fig. 1,
and generated a maximum swirl number at the exit of the nozzle with 10-mm
annular width of around 0.3. Table 1 lists the parameters of the swirling flows
considered. The second method used triple-start helical swirlers to generate higher
swirl numbers and the helices had a 6.35-mm pitch, with their starts shifted by 120
deg, an axial width of each groove of 1.6 mm, wall thickness of 0.6 mm resulting in
helix angles of 5, 7.5, and 13 deg for the 10-, 6-, and 3-mm annulus, respectively.
The rest of the dimensions and the position of the swirlers in the air jet with
014.95
Fig. 2 Nozzle geometries exchanged at the exit of the injector of Fig. 1 during the
experiments: a) low-swirl number and b) high-swirl number. (Continued)
helical swirler
helical swirler
helical swirler
b)
Fig. 2 (Continued).
3-, 6-, and 10-mm annulus are shown in Fig. 2b. The flow conditions at the exit
of the nozzles with helical swirlers and the angle of the velocity vector at the exit
of the swirler blades relative to the axis of symmetry of the sprays are given in
Table 1. The swirl number could not be determined from the definition provided
by Eq. (6) because of airflow recirculation at the nozzle exit, and it was evaluated
according to the inclination of the velocity vector relative to the axis of symmetry
at the exit of the swirler and its dimensions, as suggested for a guided-vane cascade
in an axial tube,23 that is,
(7)
where a is the angle of the vanes relative to the axis of symmetry at the exit of the
swirler, D0 is the external diameter of the liquid tube, and DG is the diameter of
COAXIAL AIF
Table 1 Operating conditions of the examined sprays
Velocity
JU
Annular Air Liquid vector Reynolds Reynolds Exit Momentum Velocity Mass 03
velocity,3 velocity,13 angle,c
IIAIOIV 1SNT
Case width, Swirl number number Weber ratio ratio flow rate ratio,
no. mm m/s m/s deg number airjetd liquid jete numberf air/liquid8 air/liquidh air/liquid1
preburner 0.6 366 30.5 - 0.0 2.1 x 106 480,000 745,000 10.6 12 0.84
SSME
N
1 10 85 3.6 0.0 1.29 x 105 9740 230 60.6 23.6 2.60 rn
2 6 123 3.6 - 0.0 1.21 x 105 9740 494 50.8 34.2 1.49 3D
0)
3 3 135 3.6 0.0 0.8 x 105 9740 598 21 37.5 0.56
SI 10 85 3.6 0.3* 1.29 x 105 9740 230 60.6 23.6 2.60 H
S2
S3
S4
S5
10
6
10
6
85
138
50.2
80
3.6
3.6
3.6
3.6
-
85C
82.2C
0.2J
0.1J
7.9k
5.3k
1.29 x
1.36 x
0.76 x
0.8 x
105
105
105
105
9740
9740
9740
9740
230
630
75
202
60.6
63.6
21.4
21.5
23.6
38.3
13.9
22.2
2.60
1.66
1.54
0.97
ii—
z
0
S6 3 127 3.6 7?c 3.2k 0.76 x 105 9740 528 18.7 35.3 0.53
a
UG , averaged over the area of the annulus of the air jet. b UL , averaged over the area of the liquid jet. c Angle between axis of symmetry of the swirler and vanes. d ReG =
UcDc/vc- QR*L = ULDL/VL- f According to Eq. (1). g According to Eq. (4). h According to Eq. (5). According to Eq. (3). ] According to Eq. (6) using the measured
13
radial profiles of V and W at the nozzle exit. kSwirl number according to Eq. (7). m
ro
o
the air jet. The swirl number Sf is given in Table 1 and is larger than the critical
value of 0.67 required for a recirculation at the nozzle exit.23 The values of S'
can be regarded only as an approximation, since the characteristics of the helices
remain for some distance downstream of the exit plane and the nonuniformity of
the flow leads to overestimations, which could be as high as a factor of two when
compared with the measurements.24
The air was supplied to the nozzle by a compressor and, when the tangential
inlets were used to generate swirl, the axial and swirling airflow rates were me-
tered separately by rotameters before passing to separate settling chambers. The
rotameters operated at gauge pressures between 30 and 300 KPa and the flow rates
were corrected to normal temperature and pressure (NTP) from calibration charts
provided by the manufacturer. The liquid was pumped from a tank and the flow
rate was kept constant at 0.015 x 10~3 m3/s by adjusting a valve in the return
line of excess liquid to the tank and metered by a rotameter, which operated at
a gauge pressure of 30 KPa. The atomizer was positioned in the vertical direc-
tion, and the resulting spray exhausted vertically downward toward a collection
tank which collected most of the liquid content of the spray. An exhaust system
attached at the side of the collection tank removed the air with the mist of the
small droplets generated by the spray. Flow straighteners were positioned at the
entrance of the collection tank to ensure that the spray remained undisturbed by
the exhaust system. The air and water flow rates supplied to the nozzle during
the measurements were kept constant within 5%. The airflow was occasionally
seeded with TiO2 powder, which was nominally micron sized and small enough
to trace the airflow, when the velocity characteristics of the airflow in the annulus
at the exit of the nozzle were measured. The powder was dispensed by two reverse
cyclone feeders,25 which were connected at two of the axial inlets of the atomizer.
The airflow rates calculated by the integration of the axial velocity profiles at the
exit of the nozzle were within 15% of those measured by the rotameters.
The velocity, diameter, and liquid flux of the fuel droplets were measured by
a phase-Doppler velocimeter,26'27 which comprised transmitting optics based on a
rotating grating as a beam splitter and frequency shifter and integrated receiving
optics, which collected the light scattered from the measuring volume in the
forward direction at an off-axis scattering angle of 30 deg on the bisector plane
of the two laser beams to ensure that refraction through the droplets dominated
the scattered light. The collected light was focused to the center of a 100-^m slit
and passed through a mask with three evenly spaced rectangular apertures before
reaching the three photodetectors. The beam intersection angle of the anemometer
was adjusted to allow the measurement of droplet diameters up to 360 /xm. The
optical characteristics of the instrument are given in Table 2.
The measured size distributions and the mean diameters at each point were
based on 20,000 measurements resulting in statistical uncertainties of less than
2% (Ref. 28) and the sizing accuracy of the instrument was less than 2 /zm for
droplets larger than 20 /xm. The uncertainty is larger for the smaller droplets due
to the tolerance of the phase-measuring electronic circuit26 and the oscillations
of the phase shift remaining on the calibration curve of the instrument.29 Droplet
velocities were obtained in 60 size ranges of 6 />tm with uncertainties less than 1 %
and 4% for the mean and rms values, respectively, based on the average sample
size of at least 1000 in each range for the smaller sizes and around 2% and 6% for
the larger droplets due to the smaller sample sizes.
The representative diameters of the sprays were estimated from the temporal
size distribution (number/in2s), which is related to the flux of liquid droplets,

Table 2 Optical characteristics of the phase-Doppler instrument
Transmitting optics
Laser: He-Ne laser
operating power, mW 35
wavelength, nm 632.8
Beam intersection angle, deg 3.024
Measurement volume length at 1/e2 intensity, mm 4.88
Measurement volume diameter at 1/e2 intensity, /zm 129
Fringe spacing, /zm 1 1 .991

Number of fringes 11
Frequency shift, MHz 0-3
Receiving optics
Focal length of collimating lens, mm 500
Location of receiving optics
from forward scatter angle, deg 30
Equivalent aperture at collimating lens:
dimension of rectangular aperture, mm 67 x 10.6
separation between aperture 1 and 2, mm 13.3
separation between aperture 1 and 3, mm 26.6
Magnification 1/2
Spatial filter slit width, //m 100
Effective length of measuring volume, /xm 3 1 2.5
Phase angle-to-diameter conversion factor
for channel 1 and 3, without rotation of the plane of the beams, //,m/deg 0.973
Phase angle-to-diameter conversion factor
for channel 1 and 3, after rotation of the plane of the beams by ±45 deg,
jitm/deg 1.163
rather than the spatial (number/m3), which is related to the number density of
the droplets, as explained by Refs. 30 and 31. The reason for this choice is that
the liquid flux is a conserved quantity and is used by current prediction models
for the calculations of local droplet characteristics.32 The representative diameters
used to characterize the sprays were the Sauter mean diameter (SMD) and the
mass median diameter (MMD) defined as
id? (8)
MMD = diameter carrying 50% of the mass flux (9)
where «/ is the number of measurements in the size range / which corresponds to

a droplet diameter di and N is the total number of ranges. The overall statistical
and measurement uncertainty of the mean diameters is expected to be around 5%.
The mean and rms of the fluctuations of the radial, V and t/, and tangential,
W and w', velocity components, as well as the time-averaged cross correlation
terms TTv and MID, were measured as a function of droplet diameter by rotating the
plane of the laser beams by ±45 deg around the direction parallel to the axis of
the flow.33 The uncertainty of the mean and the rms of the fluctuations of the radial
and tangential velocity components as a function of droplet size was 3% and 10%,
respectively, when the number of measurements in the considered size ranges was
around 1000. The corresponding uncertainty in the correlation coefficient ~uv/uf-v'
or uw/uf - w' was around 20%. The number of individual droplet realizations at
each point for each of the required three measurements was at least 30,000.
The volume flux (m3 of Iiquid/m2s) of the liquid droplets was measured accor-
ding to Ref. 31 and did not involve droplets larger than 360 /zm. Uncertainties in
the measurement of volume flux in sprays have been discussed,30"32'34 although not
in a conclusive way. The size distributions indicated that the number of droplets
larger than 300 jj,m was at least two orders of magnitude smaller than the maximum
probability for most of the measurements, and so the error to the flux measurement
due to the contribution of droplet sizes larger than 360 /xm was negligible. The
rejection of measurements by the validation procedure of the instrument was
larger in the dense region of the spray due to attenuation of the laser beams and
occurrence of multiple droplets in the probe volume which resulted in systematic
reduction of the flux measurements. The increased rejection of Doppler signals in
the near nozzle region was due to the 5/8 frequency validation procedure of the
zero crossings counter processor27 caused by reduced signal to noise ratio and also
due to the phase ratio sizing validation. Some phase ratio rejections may have been
caused by the presence of ligaments, although the number of such signals was low,
as suggested by photographic studies and by the value of the local Weber number,
which was evaluated from our measurements to be lower than unity. Thus, the
estimated values of the liquid flow rate after integration of the measured liquid
flux radial profiles at z/DL = 26 was around 50% lower than the liquid flow rate
measured by the rotameter, but the difference decreased with the increase of the
axial distance from the nozzle. In dilute regions of the sprays, errors of the order
of ±20% were present. For this reason the radial profiles of flux are presented as
relative values after normalization by the local maximum value measured by the
instrument, and the relative flux is likely to be precise to within 15%.
The size characteristics of sprays produced by nozzles with different geometry
and flow conditions were compared by estimating average SMD values over all
of the spray instead of the centerline values to avoid uncertainties caused by
differences in the rate of spread of the sprays. The average SMD was obtained
by integrating the measured radial profiles after weighting with the local liquid
volume flux and the area of a representative ring at each radial location,
SMD = I °° SMD(r)G(r)27trdr / f °° G(r)2xrdr (10)

Jr=0 / Jr=Q
where ROQ is the measurement location at the largest distance from the axis of
symmetry for each axial station and SMD(r) and G(r) are the values of the Sauter
mean diameter and volume flux at radial distance r from the axis, respectively.
Equation (10) assumes symmetry of the spray, and in sprays where asymmetries
were observed the SMD was evaluated as an average of the four resulting values
from Eq. (10) after integration of the radial profiles along directions forming a
cross. The values of SMD at each axial station were constant to within ±20%,
which indicates that the effects of droplet coalescence and evaporation were neg-
ligible.
The rate of spread of the spray was evaluated using the radial position RQ.S,
where the estimated liquid flow rate from integration of the radial profiles of
COAXIAL AIRBLAST ATOMIZERS WITH SWIRLING AIR STREAM 21 1
liquid flux was half the estimated total value at each axial station,
o.5 f # oo
2nG(r)rdr = 0.5 / 2nG(r)rdr (11)

=0
The flux half- width, namely, the radial position where the flux was half the local
centerline value,9 is appropriate only when the maximum value is on the centerline,
and this does not apply to the hollow-cone sprays of the high-swirl-number cases.
The uncertainty of this measurement is expected to be around 15%, as is for the
relative flux.
Results
Velocity characteristics of droplet sizes in the ranges 6-12, 48-54, and 102-
108 /zm, referred to as 9, 50, and 105 /xm, are presented; the smallest droplets
followed faithfully the mean and partially the turbulent flow characteristics of the
airflow, the 50-/xm droplets corresponded to a value close to the most probable
diameter in a large part of the spray, and the 105-^m droplets indicated the motion
of the droplets, which carry most of the liquid volume flux. The response of the
droplet sizes to the airflow was evaluated by the droplet response time r, the time
needed for a droplet to accelerate to 66% of the air velocity and is
18/LtG PG
where d is the droplet diameter under consideration, pL and pc are the densities of
the liquid and the air, respectively, and [LQ and VG are the dynamic and kinematic
viscosity of the air. Ratios between a characteristic flow time scale and the response
time of the droplet are defined as Stokes numbers and when larger than or around
unity indicate droplet response to the considered gas flow quantity. Thus, the mean
and turbulent flow Stokes numbers, defined as
Stm = Tm/r (13)
Sttllrb = Tturb/T (14)
characterize the droplet response to the mean and turbulent motion of the airflow.
The mean flow time scale at the nozzle exit is Tm = DC / UG > and the turbulent
time scale of the energy containing eddies of the airflow is rturb = V w '» where
A, = 1/8 DC is the length scale associated with the energy containing eddies and
u' is the rms fluctuations of the axial velocity of the airflow. The values of the
cited Stokes numbers together with the droplet relaxation time for the 9-, 50-, and
105-/xm droplets are given in Table 3 for nozzles with different annular widths of
the airstream and indicate that 9-/xm droplets followed the airflow fairly well, the
50 /zm followed the mean flow partially only, and the 105 /zm did not respond
at all justifying the choice of the droplet sizes for the presentation of the results.
It should be noted that the Reynolds number of the large droplets, based on the
droplet diameter and the mean slip velocity, may be higher than unity, for example,
for 100-/zm droplets it was about 100 close to the nozzle, and thus the flow around
the large droplets was not Stokesian, and the use of Eq. (12) for the droplet
response time was an approximation. However, the modification to the formula
ro
—L
ro
r<
T
J)
Table 3 Stokes numbers for sprays produced by coaxial airblast atomizers with swirling airstreams Sg
Coaxial atomizer Coaxial atomizer Coaxial atomizer -g

withl 0-mm annulus, with 6-mm annulus, with 3 -mm annulus, c/)
Droplet Response ^
Size, time, Stm Stm Stm Stm Stm St0 g
ms Stm 5^ case SI caseS2 case S4 Stm Stiurb case S3 case S5 Stm Stturb caseS6 f-
9 0.3 1 0.9 2.7 6.4 1.4 0.7 0.6 4.2 4.2 0.3 0.3 1.9 ^
50 9.3 0.03 0.03 0.2 0.5 0.06 0.02 0.018 0.2 0.1 0.008 0.009 0.1 ±
105 40 0.007 0.007 0.04 0.1 0.01 0.005 0.004 0.05 0.04 0.002 0.002 0.02 H rn
1
for x will still result to a small value of the Stokes number, which corresponded
to unresponsive flow and, for convenience, the use of Eq. (12) was retained.
The mean velocity and the rms of the fluctuations were normalized by the liquid
velocity averaged over the area at the exit of the liquid tube, UL. The radial distance
r from the axis of the spray, and the axial distance z from the exit of the nozzle
were normalized by the diameter of the liquid jet exit DL. The radial profiles of
liquid volume flux were normalized by the maximum value of the flux at each
axial station, G w , as measured by the instrument.
The characteristics of sprays from coaxial air-blast atomizers with axial air and
liquid streams have been reported elsewhere9 for a range of gas and liquid flow
rates and annulus widths of 3, 6, and 10 mm and provide a basis for comparison
with the present findings. The current results are presented separately for low- and
high-swirl numbers.
Low-Swirl Number
The characteristics of the sprays produced by a nozzle with an annular width of
the air jet of 10 mm and swirl number of 0.3, case SI of Table 1, at z/DL = 26,
52, and 91 from the nozzle are presented. Figure 3 shows that the Sauter and
median mean diameters were largest at the center, where the liquid jet breaks up
and decreased to the edge of the spray where the small droplets were generated in
the initial region due to the high shear at the interface with the fast moving air jet
and later due to breakup of the larger droplets which remained close to the center.
•5
e
0 10 20 30
r/DL
Fig.;. 3 Radial profiles of Sauter mean and median diameter at z/DL = 26,52, and 91
for case SI.
214 Y HARDALUPAS AND J. H. WHITELAW
Since small droplets dispersed faster than the larger droplets, the number of large
droplets relative to small droplets close to the centerline increased with the distance
from the nozzle exit and caused an increase of the mean diameters at z/DL = 52,
whereas it decreased again downstream at z/DL = 91, as the larger droplets also
dispersed from the centerline. Figure 3 and the values listed in Table 4 suggest that
the ratio of the MMD to the SMD was around 1.2, and the size distributions of the
sprays close to the nozzle followed the root-normal distribution for this ratio.4'35'36
Thus, the SMD is sufficient to describe the size distribution35 and will be used in
the rest of the text to quantify the quality of atomization.
The volume flux of the liquid droplets, Fig. 4, had a maximum on the axis of
symmetry and was nearly uniform over a larger area of the central region than for
the axial airstream sprays,9 suggesting that this was due to the presence of swirl
in the airstream and is further discussed in the Discussion section. The smaller
droplets away from the centerline carry only a small amount of the liquid flux,
and this indicates a narrow central dense region in the spray, which is mainly
responsible for the mixing of the fuel and the oxidizer.
The mean velocity vectors, Fig. 5, show a minimum on the axis and a maximum
at the high-shear region between the air and the liquid jet stream close to the
nozzle exit. The 9-/zm droplets moved faster than the larger droplets, at least in
the central part of the spray, and the rate of acceleration of the larger droplets
in the central part was higher than at the edge with no axial velocity minimum
in the central region downstream of z/DL = 52. The larger droplets moved faster
than the airflow at the edge of the spray and even at the center after z/DL = 52,
since they could not follow the air motion and maintained their upstream velocity
for a larger distance. The flow of the 9-^m droplets, which indicates the airflow,
expanded fast at Z/DL = 26 in the region of the high-speed airstream, and the
angle of the mean velocity vectors relative to the vertical direction was larger
for the airflow than for the large droplets due to the delayed response of the
larger droplets to the airflow characteristics. The tangential velocity component
increased only toward the edge of the spray, and all droplet sizes had similar mean
tangential velocity component after z/DL = 52. However, the small tangential
velocity component of the large droplets could cause centrifuging and increase
their radial velocity component away from the axis, in agreement with observations
in swirling kerosene burners.37 The centrifuging of the large droplets is evaluated
in the Discussion section and was responsible for the faster spread of the sprays
with swirling airstreams relative to those with axial airstreams and the uniform
liquid flux distribution in the central region.
r/Di
Fig. 4 Radial profiles of liquid flux at z/DL = 26, 52, and 91 for case SI, normalized
by the centerline value at each axial station.
o
o
Table 4 Parameters affecting atomization in sprays produced by coaxial atomizers with low-swirl airstreams
Annular SMDat MMDat MMDI Average Max H

Case width, z/DL = 26, z/DL = 26, SMD SMD* f/Gm, b i c
T•âcceler? (UGC - "Lc),d (UG ~ UL)S ^
no. mm /xm lim ratio Mm m/s ms m/s m/s
N
1 10 140 173 1.24 132 56.2 2.4 16 81.4 m
2 6 118 146 1.23 100 59.7 1.4 17 119.4 V)
3 3 127 158 1.24 97 41.9 1.4 13 131.4 $
H
41.4
i
SI 10 140 168 1.20 110 3.1 10.6 81.4
S2 10 144 178 1.24 135 47.6 2.6 13.7 81.4
S3 6 121 148 1.22 100 55.8 1.8 14 134.4
a
r-
According to Eq. ( 1 0). bMaximum air velocity on the centerline. c
According to Eq. ( 1 5). d
Maximum measured slip velocity on the centerline. e
Velocity
difference at nozzle exit.
ro
01

Fig. 5 Three-dimensional mean velocity vectors of 9-, 50-, and 105-/xm droplets at
zlDL = 26,52, and 91 for case SI; the tangential velocity component has been multiplied
by a factor of 3.
The motion of the large droplets was deterministic, as indicated by the increase
of the anisotropy between the rms of the fluctuations of the axial velocity u' and
those of the radial v' and tangential wr velocity components with droplet size,
Fig. 6, and by values of the correlation coefficient uv/u' • v' close to unity, Fig. 7,
which were larger than those of the air, in accord with the fan spreading effect.38
Since the large droplets cannot respond to the airflow turbulence, they moved with
straight trajectories and maintained their upstream velocities over larger distances,
which justifies the observed increase of the axial velocity fluctuations. The values
of ITuJ/V • w' were much smaller than those of ~uv/u' • v1', indicating low correla-
tion between the fluctuations of the axial and tangential velocity components after
z/DL = 52. The velocity characteristics of the 9-/zm droplets agree qualitatively
with the airflow measurements39 and suggest a delay in the development of the air
jet close to the axis of the symmetry, because of the delayed momentum transfer
from the air to the liquid droplets.
case SI
z/D L =26
20 30
r/D,
Fig. 6 Radial profiles of the rms of the axial, radial, and tangential velocity fluctua-
tions of 9-, 50-, and 105-/im droplets at z/DL = 26.
High-Swirl Number
The characteristics of the sprays produced by the nozzle with annular width of
10 mm and conditions according to case S4 of Table 1 at z/DL = 1.3 and 26 are
presented. The Sauter mean diameters of Fig. 8 had a maximum in the central
region and close to the nozzle exit, z/DL == 1.3, due to large droplets breaking up
from the liquid jet surface and being reduced at the edge of the spray. Some of the
large droplets reversed their flow and moved toward the nozzle exit, where they
reached the edge of the air jet nozzle and generated a liquid film on the wall of the
pipe, which was disintegrated by the shear of the airflow and reatomized causing
the local maxima of the SMD atr/DL = 6. Farther downstream, the droplets broke
up due to shear by the high-velocity airstream, and the spray had a minimum SMD
at the center and a maximum value of around 100 /xm away from the center.
The liquid flux profiles, Fig. 9, indicated a hollow-cone type spray with a net
transfer of liquid toward the nozzle at the center. At z/DL = 1.3, the liquid jet still
existed on the axis and carried most of the fuel away from the nozzle, while being
surrounded by droplets moving toward the nozzle. At z/DL = 26, there was a
small transfer of liquid toward the nozzle at the center, although most of the liquid
had dispersed from the axis of the atomizer and moved away from the nozzle,
because of the rapid spreading of the droplets after the breakup of the liquid jet.

-l.O
r/Di
Fig. 7 Radial profiles of correlation coefficients uv/u'v' and uw/u'w' of 9- and 105-^m
droplets at z/DL = 26,52, and 91 for case SI.
-30 -20 -10 0 10 20

r/DL
Fig. 8 Radial profiles of Sauter mean diameter at z/DL = 1.3 and 26 for case S4.

r/D
Fig. 9 Radial profiles of liquid flux at z/DL = 1.3 and 26 for case S4, normalized by
the maximum value at each axial station.
6m/s
50 jim
Fig. 10 Three-dimensional mean velocity vectors of 9-, 50-, and 105-^m droplets at
Z/DL = 1.3 and 26 for case S4.
The main difference of the airflow between the low- and high-swirl-number
flows is the presence of a recirculation zone close to the exit of the nozzle, as
indicated by the mean velocity vectors, Fig. 10. The reduced slip velocity inside
the recirculation zone delayed the initial breakup of the liquid jet which occurred
at around IQDL with the liquid jet breaking into ligaments and spreading out
outside the recirculation zone, where fine droplets were generated as a consequence
of the shear caused by the high-velocity airstream. It should be noted that no
measurements were obtained close to the axis of symmetry atz/DL = 1.3, because
the liquid jet was still present in this region. At z/DL = 26, the air recirculation
zone was limited to a small central region and all of the droplets moved with
similar velocities, suggesting that the development of the flow was faster than
for the previously examined sprays. The tangential velocity component has the
opposite sign than that for the low-swirl-number case because the direction of the
swirl was opposite. The values of the radial and tangential velocity components
at z/DL = 1.3 were similar to that of the axial velocity in the freestream outside
the recirculation zone, which agrees with the expectations for high-swirl-number
flows. The angle of the mean velocity vectors of the large droplets relative to the
axis of the spray was larger than for the airflow due to strong centrifuging of the
large droplets to the outer edge of the air jet by the swirl. This explains the large
width and the hollow-cone nature of the sprays at z/DL = 26.
r/Di
Fig. 11 Radial profiles of the rms of the axial, radial, and tangential velocity fluctu-
ations of 9-, 50-, and 105-/Lon droplets at z/DL = 26 for case S4.

120
r/DL
Fig. 12 Radial profiles at z/DL = 26 for cases S5 and S6 of a) Sauter mean diameter,
b) liquid flux, and c) mean axial velocity.
The rms fluctuations of the axial velocity, Fig. 11, had large values at the shear
layer between the recirculation zone and the freestream at z/DL = 26. Although
this is expected for the air and so for the 9-/xm droplets, it is not clear why the large
droplets had similar characteristics, since their turbulent Stokes number suggests
that they cannot respond to the air turbulence. However, the probability density
functions of the axial velocity of the large droplets in the shear layer were bimodal,
with distinct maxima at velocities around zero and large positive values, causing
the large values of the rms fluctuations. This was a result of the partial interaction
of droplets with the larger eddies in the region of the separation streamline of the
airflow.40 This phenomenon becomes important in reacting flows because large
eddies can cause large temporal variations of the local liquid concentration, see,
for example, the theoretical work,41'42 and so of the local mixture fraction, which
can limit the ability of the mixture to react and affect the stability of the flame. The
anisotropy between the rms of the fluctuations of the radial and tangential velocity
components, v' and w'', and the axial velocity component uf was lower than for
the low-swirl-number case. It appears that the airflow recirculation zone in the
high-swirl cases and the strong swirling airstream close to the nozzle deflected
the droplets from the straight trajectories in the axial direction, as suggested by
the fan-spreading effect and provided a similar mechanism in the tangential and
radial direction which increased the rms of the fluctuations of the droplet velocity.
The effect of the annulus of the nozzle was also examined, and pig. 12 presents
the Sauter mean diameter, the liquid flux, and the mean axial velocity profiles at
z/DL = 26 for the 6- and 3-mm annulus nozzle and flow conditions according
to cases S5 and S6 of Table 1, respectively, and the results are compared with
those for the 10-mm annulus. The values of the SMD were higher for the small-
annulus nozzles, but the rate of spread of all sprays was similar. The swirl number
decreased with nozzle annulus, however, and the importance of each parameter on
spray characteristics is discussed in the following section. The sprays were still
hollow-cone type, but the net liquid flux remained positive indicating no strong
effect of an air recirculating region at the central part of the spray, which became
less important for the smaller annulus nozzle, since the net liquid flux on the
centerline was higher for the small-annulus nozzle. The mean axial velocity for
the nozzles with annular widths of 6 and 3 mm at Z/DL = 26 was close to zero
but with the direction away from the nozzle and suggested that an air recirculation
zone existed upstream of this axial station but that its strength was limited, since
few droplets were transported toward the nozzle in contrast to the 10-mm annulus
nozzle (Fig. 9). The breakup length of the liquid jet was around 10DL and similar
to that of the 10-mm annulus. Increase of the liquid flow rate by around 70%, 30%,
and 10% for the nozzles with 10-, 6-, and 3-mm annular widths, respectively, led
to dramatic increase of the breakup length of the liquid jet and worse atomization
resulting in a large number droplets with diameter larger than the size limit of
the instrument. This occurred because the momentum of the reverse airflow at the
central part of the spray was higher for the 10-mm annulus nozzle and the liquid
jet momentum could be increased further before complete penetration of the air
recirculating region occurred without breakup of the liquid jet.
Discussion
This section compares the spray characteristics from coaxial air-blast atomizers
with swirling and axial airstreams to establish the effect of swirl on atomization
and rate of spread. The mechanisms influencing the atomization and the rate of
spread are examined, and the mean droplet diameters are correlated to the nozzle
geometry, the swirl number, and the flow conditions, and preferably expressed as
non-dimensional numbers.
Figure 13 compares the spray characteristics of cases SI and S2 of Table 1 with
low-swirl number with those without a swirling airstream for a 10-mm annulus
(case 1). It should be noted that the air and liquid flow rates were kept constant, so
that the observed variations on the centerline development of the SMD downstream
of Z/DL = 60 for each case can only be due to swirl. Also, comparison between the
radial profiles of SMD and liquid flux at Z/DL = 91 for cases 1 and SI, Figs. 13b
and 13c, shows that a large number of droplets with SMD around 70 ^m existed
in the swirling spray outside the boundary of the spray with axial airstream, at
around r/DL = 16, and these droplets carried a large amount of liquid flow rate.
Thus, direct comparison of the atomization between sprays with large differences
in the rate of spread can only be successful by using the spatially averaged SMD,
estimated according to Eq. (10). Table 4 compares the centerline value of the SMD
at Z/DL = 26 with the average SMD at this axial station and, for example, for case
SI, confirms that the centerline SMD values can be misleading.
The average SMD values decreased for swirl numbers higher than 0.2, Fig. 15a,
suggesting improved atomization. For low-swirl numbers, however, the effect on

200
10 mm annulus
180 centreline
160
140
120
100
80
0 20 40 60
E
O
«•».
o
Fig. 13 Profiles: a) centerline Sauter mean diameter, cases 1, SI, and S2; b) radial
Sauter mean diameter; and c) liquid flux atz/DL = 91 for cases 1 and SI.
atomization was small and, indeed, for a swirl number of 0.2 resulted in a reduction.
The breakup of the liquid jet in coaxial airstreams without swirl generated liquid
clusters close to the axis of symmetry, which carried most of the liquid flow rate,
and the final droplet sizes in the spray were determined by the breakup of the
liquid clusters due to local shear with the air.22 Photographs of low-swirl-number
sprays indicated that the breakup mechanism was similar to that of sprays without
a swirling airstream. Thus, the droplet sizes were determined from the value of
the local Weber number defined according to Eq. (2). The acceleration time of the
air velocity along the centerline is defined as
= (zro - Lc)/(UGm - UL) (15)

eeler =
where it is assumed that droplets at the edge of the liquid core with length Lc
had velocities equal to that of the liquid jet, UL, and zm is the axial distance from
the nozzle exit where the centerline axial velocity of the air flow is maximum
and equal to UGm, as defined in Fig. 14a. The corresponding acceleration Stokes
number
Stacceler == (16)
dU/dz
m
18
15
12
j
3 9
6
A case SI
O caseS2
3 • case 2
A
case S3 centreline
0
0 20 40 60 80 100 120 140
b)
Fig. 14 a) Parameters affecting atomization in coaxial atomizers with air swirl num-
ber up to 0.2 and b) centerline development of the mean axial velocity of the air flow
in the sprays for cases 1, SI, and S2 and cases 2 and S3.
characterizes the droplet response to the centerline flow acceleration. Values of

Stacceier less than unity imply lack of response of the liquid to the airflow and so
imply increase of the slip velocity and of the local Weber number We\oc. Values
of Wfeioc larger than a critical value, WfeCnt» of 12 imPty breakup.43'44 Then the
maximum stable diameter in a spray is defined as
The values of racceier were calculated from Eq. (15) assuming that Lc was constant
and equal to around 4DL and UGm was the maximum velocity of the 9-/xm droplets
on the centerline as measured by the phase Doppler and are given in Table 4
together with the velocity differences (UG - UL) and the maximum measured
slip velocity along the centerline between the airflow and the 105-/xm droplets
(uGc — WLC)- racceier was mainly affected by the air-to-liquid momentum ratio and
the diameter of the air jet for coaxial atomizers with axial airstream,22 but both
these parameters were kept constant for the 10-mm annulus nozzles. Increase
of swirl number, however, resulted in lower values of UGm and longer racceier>
Fig. 14b and Table 4, which stemmed from the faster spreading of the swirling
airflow. Thus, the increase in racceier with the swirl number explained the increase
of the average SMD for the 10-mm annulus nozzle for swirl numbers up to
0.2, Fig. 15a. For sprays from the 6-mm annulus nozzle the average SMD was
lower than for the 10-mm annulus nozzle, Fig. 15a, but Facceier was also lower,
Fig. 14b, and thus the suggested mechanism still explained the atomization. Thus,
the specified mechanism explained the atomization for different nozzle geometries.
The main parameters affecting this atomization mechanism are the air-to-liquid
jet momentum ratio, the diameter of the air jet and the swirl number influencing
âcceier.tne liquid-to-air density and air viscosity influencing the droplet response
time and thus the value of Stacceien and the surface tension and air density affecting
the local Weber number. For swirl number of 0.3, however, the atomization was
improved even though Tacce\er increased. Thus, a different mechanism affecting
atomization is introduced by the presence of swirl for values of swirl number
larger than 0.2.
An alternative way to improve atomization is to cause the large droplets to

spread from the centerline to the high-velocity airstream, close to the nozzle and
before the airstream decelerates with jet expansion. In this way, the slip velocity
between droplets and air and the local Weber number increase close to the nozzle
and, consequently, the maximum stable diameter according to Eq. (17) is reduced
and the atomization improved. Experiments with a converging air jet exit and axial
airstream9 attempted to use this mechanism by increasing the radial velocity of the
droplets close to the nozzle using the inclined direction of the air jet relative to the
axis of the spray, but met with limited success because the high-axial velocities
made the droplets move along slightly inclined trajectories relative to the vertical
direction without allowing them to spread quickly in the high-speed airstream. The
increase of swirl number above 0.3 resulted in more droplets being centrifuged in
the high-velocity airstream closer to the nozzle either due to swirl or due to the
air recirculation zone at the nozzle exit for high-swirl-number airstreams, where
they broke up due to the increased shear resulting in improved atomization.
The disintegration of the liquid jet before the end of the air recirculation zone is
essential for the latter atomization mechanism, since the air velocity downstream
of the recirculation zone was much lower than close to the nozzle, and the local
Weber number decreased resulting in worse atomization. The disintegration of the
liquid jet depends on the momentum ratio of the reversed airflow to the liquid
flow, which decreased with the liquid flow rate, and after a critical value the liquid
jet penetrated the recirculation zone before breaking up. The airflow recirculation
zone length and strength varies with the diameter of the air jet and the swirl number
S, so that it is important to evaluate their influence on the disintegration of the
liquid jet. The length of the air recirculation zone, LRZ, normalized by the air jet
diameter is linearly related to the swirl number45
LRZ/DG=kS (18)
where k is a constant. Assuming that the liquid jet and the reverse airflow behave
like two opposed flows which must have the same stagnation pressure, the dynamic
pressure on the centerline of the liquid jet, pL(ULDL/z)2/2, and the reverse
airflow, poU^z/2, must balance at the forward stagnation point occurring at an
axial distance, ZFSP, from the nozzle exit given by
ZFSP = E±_[PL_
DL URZi PG ^
where URZ is the reverse flow velocity and is URZ = cUc, where c is around 0.3,
whereas the influence of the swirl number on URZ is small.45 Thus, the maximum
liquid velocity £/Lmax for which the liquid jet disintegrates till the end of the
3 4 5
swirl number
35
10 mm annulus
30 O liquid; case 1
25 A liquid; case SI
D liquid; case S2
20
liquid; case S4
15 air; case 1
10
air; case SI
air; case S2
5 air; case S4
20 40 60 80 100
b) z/Di
35
3 mm annulus
30
O liquid; case 3
25 A liquid; case S6
__
20
•A air; case 3
air, case S6
15
10
5
0
20 40 60 80 100
c) L
Fig. 15 Influence of swirl number on atomization and rate of spread of sprays and
air flows: a) spatially averaged Sauter mean diameter [Eq. (10)] for the three nozzles,
b) rate of spread U0.5 tEcl* (H)l for nozzles with 10-mm annular width, and c) 6-mm
annular width.
recirculation zone corresponds to the case where LRZ = Z?SP and considering
Eqs.(18)and(19)
(20)
where k\ — ck. In the present study the liquid jet diameter and the air and liquid
densities were kept constant. Thus, the maximum liquid velocity for which the
liquid jet breaks up upstream of the end of the recirculation zone increases with the
swirl number, the air velocity at the nozzle exit, and the diameter of the air jet. For
nozzles with 10-, 6-, and 3-mm annular widths, increase of the liquid velocity by
a factor of 1.7, 1.3, and 1.1, respectively, resulted in the liquid jet penetrating the
recirculation zone and worse atomization than for the cases with axial airstreams.
Equation (20) suggests that the ratios of UL max for nozzles with 10- and 6- and 10-
and 3-mm annular widths are 1.4 and 2.5, respectively, which are larger than those
of 1 .25 and 1.5 suggested from the experiments; this difference is probably caused
by some dependence of the assumed constant parameters c and k on the diameter
of the air jet. The average SMD of Fig. 15a shows that the effect of swirl is reduced
for nozzles with smaller annular widths and the preceding analysis justifies this
result, since the liquid jet velocity for the nozzles with annular widths 6 and 3 mm
was closer to the maximum value for penetration of the recirculation zone by the
liquid jet than for the nozzle with 10-mm annular width.
The rate of spread of the sprays is discussed below. The droplet response to
air turbulence in the vicinity of the nozzle exit was determined by the turbulent
Stokes number, defined in Eq. (14), and evaluated in Table 3 for each nozzle
geometry. Table 3 suggests that the turbulent Stokes number of droplets larger
than around 20 /xm was lower than 0.1 and, thus, these droplets could not respond
to air turbulence. Since droplets larger than 20 fim carried more than 95% of
the liquid flow rate, the air turbulence was not responsible for droplet dispersion
and, thus, the rate of spread of the sprays. It should be noted that the turbulent
Stokes numbers increased with the distance from the nozzle,38 and large droplets
may respond to the airflow turbulence and disperse away from the centerline
farther downstream from the nozzle. The low-turbulent Stokes numbers close to
the nozzle suggest that the initial droplet trajectories, which are affected by the
breakup of the liquid jet, define the rate of spread in this region for nozzles with
axial airstreams. This observation has implications for the computer modeling of
such sprays, since large droplets do not respond to the gas-phase turbulence and
there is no need to simulate the interaction with the gas-phase turbulence in order
to calculate the rate the spread close to the nozzle.
An additional mechanism exists in nozzles with swirling airstreams and is re-
lated to droplet centrifuging by the swirling vortex and the corresponding timescale
rswiri associated with the swirling vortex is
rswir, = l/a> = RG/W (21)
where CD is the angular velocity of the swirling vortex, which has a radius equal to
that of the air jet, RG, and tangential velocity W, as measured for the 9-, 50-, and
105-/xm droplets. The corresponding centrifuge Stokes number after considering
Eqs. (12) and (21) is then defined as
Sto, = rswiri/T = l^G/pLa)d (22)
which quantifies the centrifuging effect due to the swirling airstream of the sprays46
and droplets are centrifuged when Stw is less than one. Table 3 suggests that St^
was less than 1 for droplets larger than around 50 /xm, suggesting that these
droplets centrifuged once they acquired tangential velocity. Droplet centrifuging
increased the rate of spread of the sprays, but the tangential velocity of the air jet
was reduced with the distance from the nozzle due to the expansion of the flow,
so that the ability of the droplets to acquire tangential velocity sufficient to cause
centrifuging depends on their transit time along an axial distance z from the nozzle
exit which is
= Z/UL (23)
where it was assumed that the initial velocity of the large droplets was UL. The
equivalent transit Stokes number
Sttransit = ^transit/7 (24)

should be around unity for droplets to acquire significant tangential velocity and
so, for example, 100-/xm droplets which have a response time around 40 ms will
travel a distance of 120 mm, thus up to z/Di — 52, before acquiring sufficient
tangential velocity to centrifuge. The initial value of the tangential velocity of
the air should also be high enough to be able to influence the droplets, and this
suggests that a minimum swirl number is required for centrifuging to occur.

The rate of spread of the sprays was evaluated by the values of #0.5 at different
axial stations estimated according to Eq. (11) for all of the cases of Table 1 and
is presented in Figs. 15b and 15c for nozzles with 10- and 3-mm annular widths,
respectively. The rate of spread of the air jet was also evaluated by estimating /?o.5
after integrating the velocity profiles of the 9-^m droplets and is also presented
in the figures. The results show that the rate of spread increased with the swirl
number, but a value of at least 0.2 was required. This observation agrees with the
analysis of the preceding paragraph and the values of the centrifuge Stokes number
of Table 3. Figure 15b indicates even a small reduction in the rate of spread for
swirl number 0.2 relative to the spray without a swirling airstream and may be
explained by the swirl-induced decrease of the air pressure in the central region;
this may delay the disintegration of the liquid jet and cause some of the droplets
to move initially toward the center and in this way reduce the rate of spread. This
mechanism is similar to the stabilizing effect of swirl on atomization observed
by the theoretical work.47 Thus, there are two counteracting mechanisms affecting
the rate of spread, the centrifuging of the droplets due to swirl which increases the
rate of spread and the air pressure gradient in the radial direction which decreases
it. For a swirl number lower than 0.2 and nozzles with 10-mm annular width, the
latter mechanism is dominant. For the sprays generated by nozzles with smaller air
jet diameters, the minimum value of swirl number required for droplet centrifuging
is expected to increase, because the distance z over which the tangential velocity
of the air decreases below a value required for centrifuging becomes shorter, since
the airflow development scales with the air jet diameter. Thus, for constant liquid
flow rates at the small-diameter nozzles, the velocity of the large droplets remained
nearly constant, resulting in shorter transit time, according to Eq. (23), and reduced
probability for droplets to acquire sufficient tangential velocity to centrifuge. This
expectation was not confirmed experimentally, since for the nozzle with 6-mm
annulus the maximum swirl number generated by the tangential inlets was only
0.15. For this value of swirl number no centrifuging occurred.
The sprays with high-swirl-number airstream offer the best combination of
atomization and spreading rate. Figure 15a shows that the SMD of these sprays at
Z/DL = 26 was much smaller than that of the rest of the sprays and the rate of
spread was much larger. The flow rate of air with high swirl was less than that in
the low-swirl sprays to achieve better performance, but the breakup length of the
liquid jet was greater with high swirl. An important observation for calculation
models is that the rate of spread of the spray for high-swirl numbers is larger than
that of the airflow as shown for the 10- and 3-mm annulus nozzles, cases S4 and
S6, in Fig. 15. Since air turbulence could not have caused this effect, this finding
supports theoretical suggestions41'42 that droplet centrifuging by the large eddies
of the airflow can disperse droplets more than the continuous phase.
Summary and Conclusions

Phase-Doppler measurements of size, velocity, and liquid flux in sprays pro-
duced by air-blast coaxial nozzles were obtained over a wide range of air and liquid
flow rate conditions and different nozzle geometries. The sprays encompassed a
range of Weber numbers at the nozzle exit from 230 to 630, of air-to-liquid mo-
mentum ratio from 21 to 60, air-to-liquid velocity ratio from 14 to 38, air-to-liquid
mass flow rate ratio from 0.5 to 2.6, constant liquid jet Reynolds numbers around
10,000 and air jet Reynolds numbers from 76,000 to 140,000, and swirl numbers
from 0 to 7.9. The main findings were as follows.
1) The size characteristics were determined by the local Weber number based
on the local slip velocity and the droplet diameter rather than the exit Weber
number based on the slip velocity at the exit and the diameter of the liquid tube.
A physical atomization model was suggested to predict the size characteristics of
sprays produced by coaxial air-blast atomizers with swirling external airstream
with swirl number up to around 0.2, which showed that the atomization process
was affected by the air swirl and the air-to-liquid momentum ratio at the nozzle
exit, the air jet diameter, the liquid-to-air density ratio and the surface tension, and
the kinematic viscosity and density of the air.
2) For swirl numbers higher than 0.3, atomization improved due to droplet
centrifuging away from the centerline in the high-velocity annular airstream,
where shear was higher, and this effect was enhanced when a recirculation zone
was established at the nozzle exit. The introduction of the high-swirl number
airstream resulted in hollow-cone type sprays and is the most effective mechanism
to improve the atomization and the rate of spread of coaxial air-blast atomizer
sprays. This mechanism can be maintained for sufficiently low-liquid flow rates,
and this limit is defined by the swirl number of the air stream, the ratio of the
diameters of the air and liquid jets, the air velocity at the nozzle exit, and the
square root of the air-to-liquid density ratio.
3) The effects of air mean and turbulent flow on the droplet motion were
quantified in terms of Stokes numbers, defined as the ratio of the characteristic
time scale to the droplet response time. The air turbulence close to the nozzle
could not influence droplet dispersion and, thus, the rate of spread of the spray,
which was determined by the initial droplet trajectories after the breakup of the
liquid jet and the droplet centrifuging effect due to air swirl.
4) The centrifuging mechanism affecting the rate of spread of the sprays was
quantified by a centrifuge Stokes number, defined as the ratio of the time scale
of the swirling vortex to the response time of the droplet. However, the transit
Stokes number, which is associated to the transit time of the droplets through the
region of the airflow where the air tangential velocity remains high, should be
close to unity to allow droplets to acquire tangential velocity and centrifuge. This
mechanism was responsible for the increased rate of spread of the sprays, which
became larger than the rate of spread of the corresponding airflow for the higher
swirl numbers.
Acknowledgments
The authors wish to acknowledge financial support from NASA Marshall
Space Flight Center, under Grant NAS8-38872. J. R. Laker designed and con-
structed the electronics of the phase-Doppler instrument. P. Trowellconstructed
the experimental facility. We are grateful to H. Struck and H. McDonald for
many useful discussions during the conduct of this research. Y. Hardalupas was
supported by a 1994 EPSRC Advanced Research Fellowship.
References
êiss, M. A., and Worsham, C. H., "Atomization in High Velocity Airstreams," ARS
Journal, Vol. 29, April 1959, pp. 252-259.
2
Burick, R. J., "Atomization and Mixing Characteristics of Gas/Liquid Coaxial Injector
Elements," Journal of Spacecraft and Rockets, Vol. 9, May 1972, pp. 326-331.
3
Falk, A. Y, "Coaxial Spray Atomization in Accelerating Gas Stream," NASA CR-
134825, June 1975.
4
Ferrenberg, A., Hunt, K., and Duesberg, J., "Atomization and Mixing Study," NASA
CR-178751, Dec. 1985.
5
Eroglu, H., and Chigier, N. A., "Initial Drop Size and Velocity Distributions for Airblast
Coaxial Atomizers," Journal of Fluids Engineering, Vol. 113, Sept. 1991, pp. 453-459.
6
Sankar, S. V, Brefia de la Rosa, A., Isakovic, A., and Bachalo, W. D., "Liquid Atom-
ization by Coaxial Rocket Injectors," AIAA Paper 91-0691, 1991.
7
Sankar, S. V, Wang, G., Brefia de la Rosa, A., Rudoff, R. C., Isakovic, A., and Bachalo,
W. D., "Characterisation of Coaxial Rocket Injector Sprays Under High Pressure Environ-
ments," AIAA Paper 92-0228, 1992.
8
Zaller, M. M., and Klem, M. D., "Coaxial Injector Spray Characterization Using Wa-
ter/Air as Simulants," Proceedings of the 28th JANNAF Combustion Meeting (San Antonio,
TX), 1991, pp. 1-9, NASATM 105322, Oct. 1991.
9
Hardalupas, Y, and Whitelaw, J. H., "The Characteristics of Sprays Produced by Coaxial
Airblast Atomizers," Journal of Propulsion and Power, Vol. 10, July 1994, pp. 453^-60.
10
Wanhainen, J. P., Parish, H. C., and Conrad, E. W., "Effect of Propellant Injection
Velocity on Screech in 20,000-Pound Hydrogen-Oxygen Rocket Engine," NASA TN
D-3373, April 1966.
11
Wang, T.-S., "Computational Analysis of the Three-Dimensional Steady and Transient
SSME Fuel Preburner Combustor," Proceedings of International Union of Theoretical and
Applied Mechanics Symposium on Aerothermodynamics in Combustors, Taiwan University,
Taipei, 1991, pp. VI.93-VI.95.
12
Petersen, E. L., Rozelle, R., and Borgel, P. J., "Characterization and Wall Compatibility
Testing of a 40 K Pound Thrust Class Swirl-Coaxial Injector and Calorimeter Combustion
Chamber," AIAA Paper 91-1873, 1991.
I3
Elam, S. K., "Subscale LOX/Hydrogen Testing with a Modular Chamber and a Swirl
Coaxial Injector," AIAA Paper 91-1874, 1991.
14
Hulka, J., Schneider, J. A., and Dexter, C. E., "Performance and Stability of a Booster
Class LOX/Hydrogen Swirl Coaxial Element Injector," AIAA Paper 91-1877, 1991.
15
Mao, C.-P, Wang, G., and Chigier, N., "The Structure and Characterization of Air-
Assisted Swirl Atomizer Sprays," Atomization and Spray Technology, Vol. 2, No. 2, 1986,
pp. 151-169.
16
Mao, C.-P, Oechsle, V, and Chigier, N., "Drop Size Distribution and Air Veloc-
ity Measurements in Air Assist Swirl Atomizer Sprays," Journal of Fluids Engineering,
Vol. 109, March 1987, pp. 64-69.
17
McDonell, V. G., Adachi, M., and Samuelsen, G. S., "Structure of Reacting and
Nonreacting, Swirling Air-Assisted Sprays," Combustion Science and Technology, Vol. 82,
No. 4-6, 1992, pp. 225-248.
18
Faeth, G. M., "Evaporation and Combustion of Sprays," Progress in Energy and
I9
Faeth, G. M., "Mixing, Transport and Combustion in Sprays," Progress in Energy and
20
Energy and Combustion Science, Vol. 8, No. 3, 1982, pp. 171-201.
21
Lefebvre, A. H., Atomization and Sprays, Hemisphere, New York, Ch. 6, 1989, p. 238.
22
Engelbert, C., Hardalupas, Y., and Whitelaw, J. H., "Break-up Phenomena in Coaxial
Airblast Atomizers," Proceedings of the Royal Society of London, Vol. A451, pp. 189-229,
1995.
23
Wall, T. F, "The Combustion of Coal as Pulverized Fuel Through Swirl Burners,"
Principles of Combustion Engineering for Boilers, edited by C. J. Lawn, Academic, London,

England, UK, 1987, pp. 197-335.
24
Joseph, V, Ganesan, V, and Shet, U. S. P., "Characteristics of Open LPG Jet Flames
in Helical and Tangential-Slot Swirl Burners," Journal of the Institute of Energy, Vol. 60,
Dec. 1987, pp. 193-198.
25
Glass, M., and Kennedy, I. M., "An Improved Seeding Method for High Temperature
Laser Doppler Velocimetry," Combustion and Flame, Vol. 29, 1977, pp. 333-335.
26
Hardalupas, Y, "Experiments with Isothermal Two Phase Flows," Ph.D. Dissertation,
Univ. of London, Imperial College, Mech. Eng. Dept., London, England, UK, 1989.
27
Hardalupas, Y, "Description of the Fluids Section 'Model 2' Phase Doppler Counter,"
Imperial College of Science, Technology, and Medicine, Mechanical Engineering Dept.,
Rept. FS/90/29, London, England, UK, Aug. 1990.
28
Tate, R. W., "Some Problems Associated with the Accurate Representation of Drop-
Size Distributions," Proceedings of the 2nd International Conference on Liquid Atomization
and Spraying Systems (ICLASS) (Madison, WI), 1982, pp. 341-351.
29
Hardalupas, Y, and Taylor, A. M. K. P., "The Identification of LDA Seeding Particles
by the Phase Doppler Technique," Experiments in Fluids, Vol. 6, 1988, pp. 137-140.
3()
Bachalo, W. D., Rudoff, R. C., and Brefia de la Rosa, A., "Mass Flux Measurements of
a High Number Density Spray System Using the Phase-Doppler Particle Analyzer," AIAA
Paper 88-0236, 1988.
31
Hardalupas, Y, and Taylor, A. M. K. P., "On the Measurement of Particle Concentration
near a Stagnation Point," Experiments in Fluids, Vol. 8, Oct. 1989, pp. 113-118.
32
Dodge, L. G., Rhodes, D. J., and Reitz, R. D., "Drop-Size Measurement Techniques
for Sprays: Comparison of Malvern Laser-Diffraction and Aerometrics Phase/Doppler,"
Applied Optics, Vol. 26, June 1987, pp. 2144-2154.
33
Hardalupas, Y, and Liu, C. H., "Size-Discriminated Velocity Cross-Correlation Mea-
sured by a Single Channel Phase Doppler Velocimeter," Laser Techniques and Applications
in Fluid Mechanics, edited by R. J. Adrian, D. F. G. Durao, F. Durst, M. Maeda, and J. H.
Whitelaw, Springer-Verlag, Heidelberg, 1993, pp. 145-165.
34
Hardalupas, Y, Taylor, A. M. K. P., and Whitelaw, J. H., "Mass Flux, Mass Fraction
and Concentration of Liquid Fuel in a Swirl-Stablised Flame," International Journal of
Multiphase Flow, Vol. 20, Suppl., Aug. 1994, pp. 233-259.
35
Simmons, H. C., "The Correlation of Drop-Size Distributions in Fuel Nozzle Sprays.
Part I: The Drop-Size/Volume-Fraction Distribution," Journal of Engineering for Power,
Vol. 99, July 1977, pp. 309-314.
36
Faeth, G. M., "Structure and Atomization Properties of Dense Turbulent Sprays,"
Proceedings of the 23rd Symposium (International) on Combustion, Combustion Inst.,
Pittsburgh, PA, 1990, pp. 1345-1352.
37
Hardalupas, Y, Taylor, A. M. K. P., and Whitelaw, J. H., "Velocity and Size Charac-
teristics of Liquid-Fuelled Flames Stabilised by a Swirl Burner," Proceedings of the Royal
Society of London, Vol. A428, No. 1874, 1990, pp. 129-155.
232 Y HARDALUPAS AND J. H. WHITELAW
38
Hardalupas, Y., Taylor, A. M. K. P., and Whitelaw, J. H., "Velocity and Particle Flux
Characteristics of Turbulent Particle-Laden Jets," Proceedings of the Royal Society of
London, Vol. A426, No. 1870, 1989, pp. 31-78.
39
Ribeiro, M. M., and Whitelaw, J. H., "Coaxial Jets With and Without Swirl," Journal
of Fluid Mechanics, Vol. 96, Pt. 4, 1980, pp. 769-795.
40
Hardalupas, Y, Taylor, A. M. K. P., and Whitelaw, J. H., "Particle Dispersion in a
Vertical Round Sudden Expansion Flow," Philosophical Transactions of the Royal Society
of London, Vol. A341, No. 1662, 1992, pp. 411-442.
41
Crowe, C. T., Gore, R. A., and Troutt, T. R., "Particle Dispersion by Coherent Structures
in Free Shear Rows," Paniculate Science and Technology, Vol. 3, 1985, pp. 149-158.
42
Squires, K. D., and Eaton, J. K., "Particle Response and Turbulence Modification in
Isotropic Turbulence," Physics of Fluids, Vol. A2, No. 7, 1990, pp. 1191-1203.
43
Clift, R., Grace, J. R., and Weber, M. E., Bubbles, Drops and Particles, Academic,
London, England, UK, 1978.
44
Pilch, M., and Erdman, C. A., "Use of Breakup Time Data and Velocity History Data
to Predict the Maximum Size of Stable Fragments for Acceleration-Induced Breakup of a
Liquid Drop," International Journal of Multiphase Flow, Vol. 13, No. 6,1987, pp. 741-757.
45
Chen, R.-H., Driscoll, J. F., Kelly, J., Namazian, M., and Schefer, R. W., "A Comparison
of Bluff-Body and Swirl-Stabilised Flames," Combustion Science and Technology, Vol. 71,
No. 1-3, 1990, pp. 197-217.
46
Dring, R. P., and Suo, M., "Particle Trajectories in Swirling Flows," Journal of Energy,
Vol. 2, July-Aug. 1978, pp. 232-237.
47
Lian, Z. W., and Lin, S. P., "Breakup of a Liquid Jet in a Swirling Gas," Physics of
Fluids, Vol. A2, No. 12, 1990, pp. 2134-2139.
Chapter 10
Pressure Oscillation Effects on Jet Breakup

Sibtosh Pal,* Harry Ryan,1 David Hoover,* and Robert J. Santoro§

Pennsylvania State University, University Park, Pennsylvania 16802
Nomenclature
a = speed of sound
D = water jet diameter
D0 = injector diameter
d = drop size diameter
/ = frequency
f(d) = probability density function of drop size
L = injector length
LQ = distance over which transverse gas flow is effective
pa = acoustic pressure amplitude
Pc = chamber pressure
PI = internal liquid pressure
Ps(0) = external pressure on liquid surface
Pa(0) = surface tension pressure
R(9) = radius of curvature
Re = Reynolds number of the liquid jet
r(0) = perimeter
U = velocity
W = width of a deformed liquid jet
Weg = Weber number based on gas density and gas velocity, pgU^D0/a
Wegi = Weber number based on gas density and liquid velocity, pgU^D0/a
Wei = Weber number based on liquid density and liquid velocity, piUfD0/<j
z — axial location from injector face
9 = angle
Copyright © 1995 by Robert J. Santoro. Published by the American Institute of

Aeronautics and Astronautics with permission.
*Research Associate, Mechanical Engineering. Member AIAA.
^Graduate Student, Mechanical Engineering. Student Member AIAA.
^Graduate Student, Mechanical Engineering; currently at Boeing Commercial Airplane
Group, Everett, WA.
^Professor, Mechanical Engineering. Member AIAA.
233
234 S. PAL ET AL.
X = wavelength
p = density
a = surface tension
Subscripts
g = gas
/ = liquid
10 = arithmetic mean
32 = Sauter mean
Superscripts
, — first derivative with respect to 0
rf = second derivative with respect to 0
Introduction
F OR liquid rocket engines, combustion instability has been an area of consider-

able interest because of its potential impact on engine operation. Instabilities
in rocket chambers can be triggered and amplified by a coupling between the fluid
dynamic and/or combustion processes with the acoustic resonance modes of the
chamber.1 The lack of fundamental understanding of combustion instability has
resulted in the use of passive control devices like baffles and acoustic cavities
within the rocket chamber to damp chamber response to particular frequencies.
However, this approach often involves costly design and fabrication procedures.
There is, therefore, a need for understanding the fundamental processes which
initiate and sustain combustion instability in order to design low cost and stable
rocket combustors.
High-frequency instabilities which involve transverse modes of the chamber are
particularly severe for rocket engine operation. Other frequency phenomena such
as chug (low frequency) and buzz (intermediate frequency) have been amenable
to tractable solutions.1 Consequently, the priority of current research is to under-
stand the mechanisms which lead to the occurrence of high-frequency instability
phenomena.
A commonly proposed mechanism for the initiation of high-frequency insta-
bilities is the coupling between oscillating pressure and velocity fields and pri-
mary and secondary atomization processes.1'2 Previous researchers have adopted
a global approach for investigating combustion instability by triggering pressure
perturbations in an inherently stable motor.1 For these experiments, varying the
injection system operating parameters to influence the atomization characteristics
was observed to have a strong influence on combustion stability.3-4
Newman5 conducted a fundamental experimental study of liquid jets subjected
to oscillatory pressure and velocity fields corresponding to the first tangential mode
in a circular cross-sectional chamber. His results showed that for the laminar flow
regime, the intact jet length was reduced for both oscillating pressure and velocity
fields. Debler and Yu6 studied the effects of applying a sinusoidal disturbance with
a loudspeaker on an unconfined laminar liquid jet and observed similar results.
Heidmann and Groeneweg7 proposed an analytical model of liquid jet dynamic
response to velocity and pressure oscillations by correlating the product of jet
breakup time and frequency to the response characteristics of the atomization rate.
These workers7 extended an expression developed by Clark8 that relates the de-
formation, and subsequent breakup, of a liquid jet subjected to a steady transverse
PRESSURE OSCILLATION EFFECTS ON JET BREAKUP 235
flow of gas. Their7 analytical results predicted that velocity perturbations in-phase
with the pressure oscillations would tend to destabilize a combustion system,
whereas out-of-phase velocity oscillations would create a stabilizing influence.7
Furthermore, they7 predicted that by significantly increasing either the jet diam-
eter or relative velocity between the liquid and surrounding gas, the combustion
process would be stable.
In the present paper, the effects of oscillatory pressure and velocity fields are re-
examined over a wider parametric range. The current work focuses on identifying
whether the pressure or velocity field affects the liquid jet breakup phenomena
more effectively. An experiment was developed to simulate the injection conditions

in a liquid rocket engine in the absence of combustion. An acoustic driver was used
to excite the resonant modes in a cylindrical chamber. Two injection schemes
were studied in the cylindrical chamber; one consisting of unielement injectors
that produce intact liquid jets and the other an impinging jet injector in which two
liquid jets impinge at a point to form a spray fan.
For the single liquid jet experiments, imaging studies showed a dramatic fanlike
flattening of the liquid jet for transverse resonance modes of the chamber. The
occurrence of the fanlike structure is explained as a consequence of the oscillat-
ing velocity field in the gas phase. The similarity in appearance of the fanlike
atomization to previous results presented by Clark8 motivated an experimental
comparison between steady vs oscillating velocity fields. The dynamic pressure
force (Weber number) was related to the liquid jet deformation for both veloc-
ity fields. Additionally, a two-dimensional potential flow model that resolves the
equilibrium shape of a liquid column subjected to a steady transverse flow of gas
by balancing the dynamic and surface tension forces is presented. This model
predicts a deformed shape that qualitatively explains the experimentally observed
fanlike geometry.
For the impinging liquid jet experiments, imaging studies showed that for the
same transverse resonance modes of the chamber the breakup of the sheet formed
downstream of the impingement point was significantly enhanced. These imaging
studies were complemented with drop size measurements made using phase-
Doppler interferometry9'1() and liquid sheet atomization frequency measurements11
for both nonexcited and oscillatory flow conditions.
Experimental
Chamber and Excitation Apparatus
The steel cylindrical chamber depicted in Fig. 1 was used to investigate the
effects of acoustic pressure oscillations on single liquid jet and impinging liquid
jet atomization. The inner diameter and length of the cylindrical chamber are
152 mm and 610 mm, respectively. Two diametrically opposed 76.2-mm-diam
quartz windows, centered 152 mm from the top of the chamber (section B-B in
Fig. 1) allow visual access into the chamber. Additionally, the chamber has two
diametrically opposed 12.7-mm quartz windows (section C-C in Fig. 1), centered
216 mm from the top of the chamber. These two smaller windows are oriented
90 deg with respect to the larger windows. The chamber has two speaker arms,
located 152 mm from the top of the chamber (section B-B in Fig. 1). The diameter
and length of the speaker arms are 38.1 mm and 156 mm, respectively. For the
current set of experiments, a 120-W Altec Lansing speaker was attached to the
end of one speaker arm, whereas a Plexiglas® window was placed at the other
236 S. PAL ET AL.

MOVABLE INJECTOR
61Qm>
SECTION B-B
r MICROPHONE PORT
SECTION C-C SECTION D-D
Fig. 1 Schematic of the chamber.
for additional visual access. The chamber features a modular, traversable injector
system that allows for both easy replacement of injector type and injector face
positioning capability from the top of the chamber to the center of the large quartz
windows (0-152 mm of travel). The chamber can be pressurized up to 2.0 MPa
(20 atm), with the 76.2-mm-diam quartz windows being the limiting factor. For
pressurized studies, the speaker was housed in a pressure vessel (not shown in
Fig. 1) connected to the main chamber.
The acoustic resonance modes of the chamber were first calculated by solving
the three-dimensional wave equation for this cylindrical geometry. The standing

Table 1 Calculated chamber resonant modes
Frequency,
Hz Longitudinal Radial Tangential
283 1 0 0
566 2 0 0
849 3 0 0
1132 4 0 0
1326 0 0 1
1356 1 0 1
1415 5 0 0
1442 2 0 1
1575 3 0 1
1698 6 0 0
1744 4 0 1
1939 5 0 1
1981 7 0 0
2155 6 0 1
2201 0 0 2
2219 1 0 2
2264 8 0 0
2273 2 0 2
2359 3 0 2
2384 7 0 1
2475 4 0 2
2547 9 0 0
2617 5 0 2
2624 8 0 1
2760 0 1 0
wave frequencies and the corresponding modes are listed in Table 1 for frequencies
up to the first radial mode (2760 Hz). The calculations indicated that the frequency
response of the chamber was extremely rich and that certain higher order combined
modes were closely spaced in frequency.
The resonant modes of the chamber along with the corresponding acoustic
pressure amplitudes were then experimentally determined for conditions in which
no liquid flow was present. The chamber has six microphone ports, two located
diametrically opposite each other, 83 mm from the top of the chamber (section
A-A in Fig. 1), and the remaining four located at 90-deg intervals along the cham-
ber circumference, 254 mm from the top of the chamber (section D-D in Fig. 1).
The initial chamber response was characterized by introducing white noise as
the input to the speaker and measuring the chamber response at the various mi-
crophone ports with PCB piezotronics model 106B microphones. The observed
resonant mode frequencies agreed with the calculated values for the individual
longitudinal and tangential modes. The mixed modes were difficult to verify ex-
perimentally, however, because of the complex and closely spaced mode structure.
238 S. PAL ET AL.
Subsequently, the speaker was tuned individually to each of the resonant frequen-
cies up to the chamber's first radial mode, and the response was measured at the
various microphone ports. These results showed that the pure longitudinal modes
were more energetic than the pure tangential and tangential/longitudinal mixed
modes.
Single Jet Experiments

For the single jet experiments, three different single tube orifice geometries
were used, as shown in Fig. 2, to introduce a water jet into the chamber. The
orifice diameter of the first injector was 1.6 mm, and the diameters of the other
two injectors were both 2.54 mm. The L/D0 (length/orifice diameter) ratios for
these tubes were 16, 10, and 35, respectively. The 2.5-mm-diam, 35L/D0 single
If II
E1 6
E |
in in
CM
o = 1 ,6mm C\> =2.54mm
NOZZLE 1 NOZZLE 2
%
6.3mm 1 mm
ORIFICE
E
E
o»
odi
00|
DO =2.54mm
NOZZLE 3
Fig. 2 Single jet injectors used for this investigation.
tube injector had an upstream orifice insert (diameter of 1 mm) to prevent induced
pressure oscillations from coupling with the liquid feed system. This injector was
used for the majority of the studies reported here. To achieve higher flow velocities,
injectors not employing this insert orifice were required. Because of the higher
flow velocities, these injector conditions were considered to be less susceptible to
low-frequency pressure coupling.
The Altec Lansing speaker, driven with an amplified signal from a function
generator (see Fig. 1), was used to induce the acoustic field in the chamber. Both
the frequency and amplitude of the induced acoustic field were varied until the
liquid jet was perturbed. With optical access provided by the window attached to
the opposite speaker arm, cinematographic techniques [35-mm, charge injection
device (CID) solid-state and Spin Physics cameras] were used to record the effect
of the applied acoustic field on the liquid jet. A spot lamp or strobe light typically
oriented perpendicular to the camera axis was used to illuminate the liquid jet.
The results presented here were all obtained using CID solid-state camera (512 x
512 pixels) for imaging of the atomization process.
The single jet acoustic excitation experiments were performed at pressurized
conditions ranging from 0.1 to 2.0 MPa (1 to 20 atm). Specific operating conditions
are summarized in Table 2. The liquid jet parameters noted correspond closely to
liquid oxygen (LOX) injected flow conditions in the Space Shuttle main engine
(SSME) oxidizer preburner. The injector element used in the SSME oxidizer
preburner, however, is a shear coaxial element which uses coflowing gas to help
atomize the LOX jet.
The single jet acoustic excitation experiments were complemented by an exper-
iment in the same chamber to study the effects of a steady transverse gas flow on
jet breakup. This experiment closely followed an experiment conducted by Clark.8
The setup just described was modified to cinematographically investigate trans-
verse gas stream effects on single liquid jets. The 76.2-mm-diam quartz windows
and speaker of the test section were removed from the chamber. Two tubes with
inside diameters of 10 mm were placed 180 deg opposed, separated by 11 mm and
normal to the jet axis at a distance of 26 mm from the injector face, and were used
to subject the liquid jet to a transverse gas flow. These tests were only performed
at atmospheric pressure.
Table 2 Parameter ranges for jet breakup
Parameter Range
Injector diameter, D*, mm 1.6-2.5
Frequency, /, Hz 1000-5000
Acoustic pressure amplitude, Pa, kPa 28.8 (maximum)
Chamber pressure, Pc, MPa 0.1-2.0
Jet Reynolds number, Re* 1250-50,000
Gas Weber number, Wegi 0.01 -200
Liquid Weber number, Wfef 10-14,000
Jet velocity, U?, m/s 0.5-20
a
Parameter values that are comparable to Space Shuttle main
engine oxidizer preburner conditions for liquid oxygen (LOX)
flow.
240 S. PAL ET AL.
H—————— ^ 762mm •
~H 21.6mm I*—
.FLOW iFLOW
i !
T i i ! i 1
I i ! 1 1
40.6mm
A-*-
l i ! 1 1
Ai>r
> '
1 20.1mni
!~M
A
8.9mm
Fig. 3 Impinging jet injector used for this investigation.
Impinging Jet Experiments

For the impinging jet experiments, a single injector design was used, as shown
in Fig. 3. The twist-drilled injector consisted of two 1.27-mm-diam orifices, which
were designed for the jets to impinge at a point 8.5 mm from the injector face
with an included impingement angle of 60 deg. The L/D0 of the orifice passages
was 18.3.
An illustration representing the spray from a typical impinging jet injector is
shown in Fig. 4. The flowfield from this type of injector is characterized by a sheet
formed at the impingement point in the plane bisecting the two impinging liquid
Entrance Effecta
Fig. 4 Schematic of a typical impinging jet spray.

jets. The liquid sheet subsequently breaks up in a periodic fashion into ligaments
and drops. Details of the breakup phenomena of impinging jet sprays can be found
in many references.10"14
In the current study, the effect of acoustic pressure oscillations on the sheet
breakup phenomena was experimentally addressed. Specifically, instantaneous
images of the impinging jet flowfield for both nonexcited and oscillating flow con-
ditions were obtained. These visualizations were complemented by phase-Doppler
interferometry measurements9'1() of drop size, and sheet atomization frequency
measurements,11 the details of which are discussed next.
Drop size measurements in the impinging jet spray field were made in the
chamber using an argon-ion (X = 514.5 nm) laser-based, fast Fourier transform
(FFT) version of the phase-Doppler particle analyzer (PDPA).9'15 The theory be-
hind the operation of the PDPA has been reviewed thoroughly in the literature.9'15
The collection optics were oriented 30 deg off axis from the forward propagation
direction of the laser beam, which is the optimum angle for measuring transparent
spherical drops.9 With this optical configuration, the drop size measurement range
spanned 40-1500 /zm.
The measurement of the sheet atomization frequency of the impinging jet spray
field closely follow the experiments by Heidmann et al.,11 who observed that
the sheet formed downstream of the impingement of two liquid jets broke up
into strands of liquid in a periodic fashion. In an attempt to characterize this
sheet atomization frequency, they conducted a simple experiment which involved
traversing a small (relative to the sheet surface area) spatial area beam from a white
light source through the sheet surface at an axial location along the centerline in the
vicinity of the sheet breakup length, and collecting the line-of-sight light signal on
the other side of the sheet with a photoelectric cell. They recorded and analyzed the
measured interrupted light signal with both an oscilloscope and a pulse meter. Their
results showed that for a given impingement angle, the sheet atomization frequency
scaled with the jet velocity. Similar results were also obtained by Ryan et al.,10 who
calculated the sheet atomization frequency from wave separation measurements
and the sheet velocity. In the present study, the experiment conducted by Heidmann
et al.11 was repeated with some minor changes, for both nonexcited and oscillatory
flow conditions. A helium-neon laser and a photodiode were used instead of the
white light source and the photoelectric cell. The photodiode signal was sampled
at 41 kHz with an analog-to-digital (A/D) converter.
Results and Discussion

Single Jet Breakup Results by Acoustic Forcing
The single jet experiments showed that for certain chamber resonant frequencies
and over a wide range of flow conditions, the liquid jets emanating from all tested
injectors demonstrated enhanced breakup. Images obtained for low-velocity jets
(in the Rayleigh breakup regime16) under both nonexcited and acoustically excited
flow environments at atmospheric conditions are presented in Fig. 5. These images,
which show both acoustically perturbed and unperturbed jets, were taken through
the Plexiglas window along the axis of the speaker arms. The CID camera with a
shutter speed of l/30th of a second was used, therefore yielding a time-averaged
view of the phenomenon. The orifice diameter of the injector was 2.54 mm (injector
3, Fig. 2) and the velocity was 1.6 m/s, resulting in a jet Reynolds number of 4040.
242 S. PAL ET AL.

Fig. 5 Water jet at the chamber centerline subjected to a 1-T/3-L resonant mode
(1560 Hz), acoustic field causes the water jet emanating from injector 3 (D0 = 2.54
mm) to flatten out into a thin sheet, image time duration is 1/30 s, velocity of 1.6 m/s.
The first image in this sequence, Fig. 5a, shows the undisturbed jet. For this
particular injector, the field of view is 15 jet diameters. At a frequency of 1560 Hz,
corresponding to the first tangential/third longitudinal (1-T/3-L) mode (discussed
later), the jet deformed into a two-dimensional fan perpendicular to the speaker
axis, with the extent of deformation increasing with the amplitude of the acoustic
pressure oscillation as Figs. 5b-5g show. The measured peak-to-peak acoustic
pressure oscillation corresponding to each plate is tabulated in the figure. These
acoustic pressure amplitude measurements were taken with the microphone placed
in the port directly underneath the speaker arm (section D-D in Fig. 1). Figure 5b
shows that at a peak-to-peak acoustic pressure amplitude of 1100 Pa (155 dB) the
jet begins to bulge in the middle. The image appears similar to the axisymmetric
disturbance observed for Rayleigh breakup,17 and findings presented by Debler
and Yu.6 This comparison is misleading, however, because the cross section of
the present jet becomes elliptical in the presence of the acoustic wave rather than
remaining circular. With increasing acoustic pressure, Fig. 5c, the jet deforms into
a two-dimensional fan, with drops breaking off the bottom of the sheet under the
action of shear and surface tension forces. This phenomenon is visually similar
to the sheet produced by impinging liquid jets.10'11'14 High-speed cinematography
employing a Spin Physics camera with imaging rates of 4000 frames/s showed
evidence of wave patterns on the sheet. Consequently, the mechanism for disinte-
gration of the two-dimensional fan into drops may be analogous to that postulated
for the breakup of a liquid sheet.17-18 The next two images for higher acoustic pres-
sure amplitudes, Figs. 5d and 5e, show a progressive increase in the extent of fan
broadening and a decrease in the intact fan length. Finally, at the highest acoustic
pressure amplitudes, Figs. 5f and 5g, the intact fan area reduces considerably with
the jet almost instantaneously atomizing into drops.
A similar series of images is shown in Fig. 6 for a higher jet velocity of 3.6
m/s (Re = 9100) with the same injector. By contrasting jet images for the same
acoustic power level, but for different jet velocities, the importance of jet Reynolds
number on the jet fanning phenomenon was ascertained. The qualitative obser-
vations made on the variation of width and length of the intact fan as a function
of the amplitude of the acoustic pressure oscillation for the lower velocity case
are also similar for this case. Comparison between jet fanning images for identi-
cal acoustic pressure oscillation amplitudes, but different jet Reynolds numbers,
yields additional information. For example, comparison of Fig. 5d with Fig. 6b
Fig. 6 Water jet at the chamber centerline subjected to a 1-T/3-L resonant mode
(1560 Hz), acoustic field causes the water jet emanating from injector 3 (D0 = 2.54
mm) to flatten out into a thin sheet, image time duration is 1/30 s, velocity of 3.6 m/s.
244 S. PAL ET AL.
PRESSURE
CHAMBER
WALL
Fig. 7 Measured pressure distribution at one phase angle within the chamber along
a horizontal plane 254 mm from the top of the chamber (see Fig. 1) corresponding to
the 1-T/3-L resonant mode (1560 Hz).
or Fig. 5g with Fig. 6d, shows that the broadening of the jet decreases with an
increase in jet Reynolds number for the same acoustic pressure oscillation ampli-
tude. The observations can be quantitatively explained after a discussion of the
acoustic pressure oscillation pattern corresponding to this frequency (1560 Hz).
The oscillating pressure amplitude pattern within the chamber for the 1560-Hz
frequency resonance at atmospheric conditions was measured with two traversable
microphones. The microphones were traversed through the chamber in 12.7-mm
intervals at section D-D shown in Fig. 1. From measurements of the acoustic
pressure amplitude along the orthogonal diameters (12 locations) and the phase
angle between the two pressure traces, the oscillating pressure pattern within the
chamber was constructed. The resulting acoustic pressure amplitude profile at
one phase angle, as shown in Fig. 7, has tangential-like characteristics which
are evident by comparing this figure with the classical representation of the first
tangential (1-T) acoustic mode shown in Fig. 8. The frequency (1560 Hz) is
closest to the calculated value for the 1-T/3-L mode. Note that for the longitudinal
component of the 1-T/3-L mode, the window location is close to the theoretically
placed pressure antinode. Additional pressure measurements made by traversing a
microphone through the speaker arms into the chamber verified the aforementioned
observations. The peak-to-peak acoustic pressure amplitudes tabulated in Figs. 5
and 6 were measured at the junction point between the speaker arm and the main
chamber corresponding to a location close to the theoretically placed pressure
antinode for the longitudinal component of the 1-T/3-L mode. Note that the single
liquid jet flattens into the observed two-dimensional sheet along the pressure nodal
line corresponding to the tangential component of the 1-T/3-L mode.
The jet flattening phenomenon under the same oscillating flowfield conditions
(1560 Hz, 1-T/3-L mode) was also investigated for cases where the jet was located
off axis to the center of the chamber. In Fig. 9, the three pairs of image plates
(Figs. 9a and 9b, 9c and 9d, and 9e and 9f) for a jet velocity of 2.7 m/s (Re =
6940) correspond to three jet locations, viz., at the chamber center, 19.1-mm off
center along the pressure nodal line, and 19.1-mm off center along the diameter
perpendicular to the pressure nodal line. The images presented in Fig. 9 are for a
PRESSURE
SPEAKER!
CHAMBER
WALL
Fig. 8 Pressure field for a first tangential mode as calculated by solving the three-
dimensional wave equation.
Fig. 9 Water jet subjected to a 1-T/3-L resonant mode (1560 Hz), chamber pressure
2.0 MPa and Re 6940: a) at centerline, nonexcited conditions, b) at centerline, 28.8
kPa acoustic pressure amplitude, c) 19.1 mm away from centerline along the pressure
nodal line, nonexcited conditions, d) 19.1 mm away from centerline along the pressure
nodal line, 28.8 kPa acoustic pressure amplitude, e) 19.1 mm away from centerline
along diameter perpendicular to the pressure nodal line, nonexcited conditions, and
f) 19.1 mm away from centerline along diameter perpendicular to the pressure nodal
line, 28.8 kPa acoustic pressure amplitude.
246 S. PAL ET AL
chamber pressure of 2.0 MPa (20 atm). Figures 9a, 9c, and 9e are for nonexcited
conditions, whereas Figs. 9b, 9b and 9f are for a peak-to-peak acoustic pressure
amplitude of 28.8 kPa (183 dB). The amplitude level of the pressure oscillations
realized for the elevated chamber pressure was an order of magnitude greater than
that for atmospheric pressure. For the same oscillating flowfield conditions, the jet
exhibits the greatest enhancement of breakup at the centerline. Also, for the same
off-center separation, the jet is flattened more along the pressure nodal line than
away from the pressure nodal line. For the conditions of Figs. 9b and 9d, the jet
experiences the local effects of only an oscillating velocity field, whereas for the
conditions of Fig. 9f, the jet is subjected to both oscillating velocity and pressure
fields. For a tangential mode, the magnitude of the oscillating velocity field is
highest at the chamber center and decreases with distance from the center. At any
fixed radius, however, the velocity amplitude would be highest at the two points
bisecting the pressure nodal line. These arguments suggest that the jet flattening
phenomenon is due to the oscillating velocity field.
To quantify the visually observed trends, the ratio of the width of the fan W to
the undisturbed jet diameter D was measured from Figs. 5 and 6 (and other images
not shown) as a function of downstream distance. The ratio W/D vs the Weber
number based on the maximum gas cross velocity and density (Weg) are plotted
in Fig. 10 for various jet Reynolds numbers at two downstream locations. The W
and D measurements were made from time-averaged images and hence, represent
average quantities. The variation in the Weber number Weg plotted in this figure
reflects changes in the maximum gas velocity due to varying the acoustic pressure
amplitude. The maximum gas velocity was obtained from the one-dimensional
approximation19'20
Ug = Pa/Pga (1)
where Pa is the maximum measured pressure amplitude and a the speed of sound.
The Weber number Weg defined in this way does not exceed unity for the conditions
plotted in Fig. 10. For a constant Weg, the ratio W/D increases exponentially with
downstream distance. Furthermore, the maximum W/D never exceeds six for the
0.000 0.200 0.400 0.600 0.800 1.000

Weg
Fig. 10 W/D vs Weg as a function of Re for a water jet subjected to a 1-T/3-L acoustic
mode.
cases tested. In fact, the fan invariably disintegrates into drops for values of W/D
greater than this threshold.
At atmospheric pressure, the two-dimensional fanning phenomenon was also
observed at frequencies of 1330 Hz and 1720 Hz, which are close to the calculated
values of the 1-T/l-L and 1-T/4-L modes, respectively. The pure 1-T mode was
measured to be a weak mode at a slightly lower frequency and did not induce the
two-dimensional fanning phenomenon. The plane of jet brreakup also coincides
with the pressure nodal line for the tangential component of the modes, with the
breakup occurring at similar flow conditions as the 1-T/3-L (1560 Hz) mode.
The importance of the tangential character of these modes is underscored by the

observation that excitation for pure longitudinal modes of the chamber were not
found to perturb the jet. At elevated chamber pressures up to 2.0 MPa (20 atm),
the sheet formation phenomenon was observed at other distinct frequencies (1920,
2140, and 2560 Hz). Note the proximity of these frequencies to the calculated
1-T/5-L, 1-T/6-L, and 1-T/8-L modes listed in Table 1.
The physical mechanism that causes the jet to deform into a sheet under the
oscillating flow conditions corresponding to the mixed tangential modes of the
chamber is of considerable interest. For example, it could arise as the result of pres-
sure coupling or velocity coupling mechanisms. The results show, however, that
the maximum deformation is observed when the jet is on the pressure nodal line,
where the jet is subjected to minimal pressure oscillations. Also, the jet deforms
into a fan over a wide range of flow conditions for only frequencies that closely
correspond to mixed tangential resonant modes of the chamber. The preferred
frequency for the breakup of a round jet has been shown to be proportional to the
ratio of the jet velocity U\ to the injector orifice diameter D0 (Refs. 21 and 22).
In the present experiments, the jet deformation phenomenon has been observed
for jet velocities between 0.5 and 20 m/s, with the magnitude of jet flattening
decreasing significantly with jet velocity, and for two injectors with different ori-
fice diameters, 1.6 and 2.54 mm. Over the tested range of these two parmeters,
the preferred breakup frequencies of the jets span roughly two orders of magni-
tude. However, the jet deformed only at resonant frequencies characteristic of the
chamber. Acoustically enhanced breakup of liquid jets in open (nonchambered)
environments is known to occur at the preferred frequencies through a pressure
coupling mechanism.6 Consequently, the present results, which are independent of
the jets' preferred frequency, argue against a pressure coupling mechanism. In the
absence of pressure coupling as a viable mechanism, velocity coupling appears to
be responsible for the liquid jet deformation.
Single Jet Breakup in Steady Transverse Gas Flow

To investigate the velocity coupling mechanism, a study of the effects of steady
transverse gas flow on jet breakup in an unconfined environment, which closely
follows an earlier study by Clark,8 was undertaken. The experiment was initially
conducted using one impinging airstream with an exit velocity of 10 m/s. The
liquid jet was observed to flatten into a sheet that curved away from the direction
of airflow, as illustrated in Fig. lib. Consequently, the width of the curved sheet
was extremely difficult to measure. The introduction of a diametrically opposed
impinging gas stream reoriented the curved sheet to a normal sheet (Fig. lie)
without markedly changing the shape of the sheet, thereby making measurement
of the width of the sheet easier. A typical phtograph of the phenomenon is shown
248 S. PALETAL
NOZZLE
(a)
NOZZLE
Fig. 11 Schematic for the transverse steady air flow investigation: a) undisturbed
water jet issuing from injector 2, D0 = 2.54 mm, b) water jet subjected to transverse
airflow from one 10-mm-i.d. pipe, and c) water jet subjected to transverse airflow
from two opposed 10-mm-i.d. tubes.
in Fig. 12 [injector mounted in chamber (Fig. 1) with no windows]. The observed

fanning of the jet and the subsequent atomization appear similar to the images
of the deformed jet under the influence of an acoustic field (Figs. 5 and 6). This
simple experiment suggests that velocity coupling is the mechanism that causes
this phenomenon.
The width W of the liquid sheet was mesured 38.7 mm below the tubes supplying
the transverse gas flow. The ratio of the measured width W of the liquid sheet to
the undisturbed diameter D of the liquid jet plotted against the Weber number Weg
of the impinging airstream is shown in Fig. 13 for injector 3. The jet Reynolds

Fig. 12 Sidelit photograph of a water jet emanating from injector 2 being deformed
by opposed transverse airflow, Fig. lid, airflow is through two 5-mm-i.d. copper
tubes, Ui = 3 m/s.
number Re ranges from 4,000 to 10,000 and the Weber number Weg ranges from
less than 1 to about 40. Note that the Re range is comparable to the earlier results on
jet deformation under acoustic forcing (Fig. 10); however, the Weber-number Weg
range is more than an order of magnitude larger for the steady transverse gas flow
results. These results show that the W/D ratio increases with transverse air velocity
(Weg) and decreases with liquid jet velocity. For Weber numbers less than 10, the
W/D ratio is roughly 1.5 and then increases to a maximum of about 4 at higher
Weber numbers. For any Reynolds number, there exists a maximum Weber number
beyond which the two-dimensional sheeting process does not occur. Increasing
the Weber number above this critical value causes the jet to atomize instantly into
drops. The results shown in this figure are viable because the trends in the W/D
ratios for this experiment and for the acoustic experiment are qualitatively similar,
despite the fact that the Weber numbers Weg for the two cases differ by more than
an order of magnitude.
250 S. PAL ET AL.
£—— A Re-3990
Q —— a Re-6740
V —— V Re-7860
O —— O Re-9230
—I—
10 20 30 40
We c
Fig. 13 W/D vs Weg as a function Re for a water jet subjected to opposed transverse
air flow, width measurements were taken approximately 39 mm from the injector face.
The reason for this discrepancy in the Weber-number range could be partly
due to the physical differences between the steady vs oscillating velocity fields
and partly due to estimating, instead of measuring, the amplitude of Ug for the
oscillating velocity field. The amplitude of the velocity field experienced by the
jet in the acoustic forcing experiments may be significantly higher than that pre-
dicted by the one-dimensional analysis [Eq. (1)] based on the maximum measured
value of the pressure amplitude. Clearly, the velocity field within the chamber,
as opposed to the pressure field, has to be measured to resolve this discrepancy.
This discrepancy, however, does not diminish the observed similarities between
the two sets of experiments. Rather, the appropriateness of the use of the Weber
number based on gas properties for comparing the two phenomena may need to
be re-examined.
As already mentioned, experiments on the breakup of a liquid jet in a transverse
flow of gas were carried out by Clark8 for a wide range of flow conditions.
Specifically, he investigated the effects of gas-velocity-based Weber numbers Weg
in the range form 30 to 11,000 on the deformation of liquid jets emanating from
various injectors within a Reynolds number range of 7,600 to 145,000. dark's
results8 showed that the liquid jet deformed into a fanlike structure that bent away
from the direction of gas flow.
In the Clark study,8 a model of the breakup process is proposed based on an
integration of the momentum equation. The model neglects the effects of surface
tension forces as well as viscous and gravitational forces. An expression for the
ratio of the width of the broadened liquid jet, W, to the initial diameter of the jet,
D, is derived8
W pg\(U? + Ul)/L0
j i ) & \ui (2)
where L0 is the distance over which the transverse gas flow is assumed to be
effective.
The deformation ratios predicted by dark's model8 [Eq. (2)] for the current ex-
perimental conditions on transverse airflow effects on a liquid jet shown in Fig. 13,
6- /A A— A Re-3990
/ m— • Re-6740
5- A T—T Re-7860
/ *—* Re-9230
A
Q 4-
A
£
3- / .-^"T^~' ,__.
2-
/* m'^^=:::'-~~'*~~~^'
1-
0- ———————— i,———————— «,———————— i,———————— ,

0 10 20 30 4(
We g
Fig. 14 W/D vs Weg as a function of Re for a liquid jet subjected to transverse gas
flow predicted by Clark's analytical model* for the experimental results shown in
Fig. 10.
are presented in Fig. 14. The model predicts that the W/D ratio increases linearly
with Weber number Weg, with a higher jet deformation predicted for decreasing
Reynolds number. This trend agrees reasonably well with the experimental results
shown in Fig. 13. The agreement between the model and experiment is better for
high-Reynolds numbers than low-Reynolds numbers. Finally, note that the Weber
numbers Weg for the current experimental conditions do not exceed 40 and, there-
fore, surface tension forces are important. The model, therefore, overpredicts the
deformation of the jet in the low-Weber number Weg range because it neglects the
effects of surface tension forces.
Potential Flow Modeling of Single Jet Deformation

To evaluate the effects of a transverse gas flow on the geometry of a liquid jet, a
simple model based on combining potential flow theory with surface tension effects
was developed. The liquid jet is modeled as a column of fluid, thereby reducing
the problem to two dimensions. The spirit of the model was to qualitatively predict
the manner in which a circular liquid jet would deform under crossflow and to
determine the important parameters. The simplifying assumptions of potential flow
and two dimensionality make the model tractable because liquid inertial effects,
viscous effects, and separation and wake effects are neglected. The calculation of
the pressure distribution on a cylinder in crossflow based on potential flow theory
is a classical problem23 and can be carried out by numerous methods. In the current
problem, the cylinder is composed of a fluid and, therefore, its shape will deform
under crossflow in a manner that equates the internal pressure P, to the sum of the
external pressure Pv(<9) and the surface tension pressure Pa(0) along every point
on the perimeter r(9) (Ref. 24) as
Pi(0) = Ps(0) (3)

For a crossflow velocity, the internal pressure P/ is constant. The external static
pressure Pv(0) can be calculated by using the panel method25 for solving potential
252 S. PAL ET AL.
flow problems. The surface tension pressure Pff(Q) is related to the radius of
curvature R(0) and the surface tension cr as
Pff(0)=cr/R(0) (4)
The radius of curvature is in turn related to the perimeter r (9) ?~>
[r2(0) + r'2(<9)]3/2
&(£)} __ _____L *- ' \^/J_________ ,~^
V
' r2(<9) + 2r'2(6>) - r(6>)r"(6>) V
Finally, by including Eqs. (4) and (5) in Eq. (3), the following second-order
differential equation relating the perimeter of the shape to the pressures Pi (0) and
Ps(0) is obtained:
/ 2 2 2 3 2
~ ' /
r(0) a
Since the potential flow solution to the problem is desired, symmetry requires
only solving Eq. (6) for 9 ranging from 0 to n/2 with the following boundary
conditions:
r\rffi\
(7)
de
dr(0)
(8)
In addition, to conserve mass, the area of the deformed shape is equal to the area
of the initial circular cross section as
•I f 2, D2
r (0)dO=7t—
Equation (9) indicates that the internal pressure is constant for a given gas velocity.
(9)
The interative procedure for numerically solving Eq. (6) with the three condi-
tions [Eqs. (7-9)] involved assuming, for a chosen gas velocity, a geometry for the
deformed shape and the internal pressure PI. The external static pressure distribu-
tion as a function of the angle 9 is calculated using a panel method potential flow
code. For the present set of calculations, a panel size corresponding to an angular
increment A# of 0.5 deg was chosen as a compromise between accuracy and com-
putational time. Equation (6) was then solved with a fourth-order Runge-Kutta
numerical procedure using Eq. (7) as an initial condition. Since the numerical
procedure is a marching technique, a Newton-Raphson technique25 was employed
to evaluate the parameters PI and r(0). The external static pressure distribution for
the new geometry was again calculated and the iterative procedure was repeated
until convergence (less than 10~10) was attained.
The results of these calculations for a sequence of air velocities for Weber
numbers Weg up to one are shown in Fig. 15. Under crossflow, the initial circular
cross section flattens into an elliptical-like shape normal to the direction of the air
flow. The W/D ratio of the deformed shape is also seen to increase rapidly with
increasing Weber number Weg.

3
We g U g (m/s)
2 1 o.oo o.o
2 0.35 3.0
3 0.98 5.0
1
-1
_2
-3
-4 - 3 - 2 - 1 0 1 2
X (mm)
Fig. 15 Calculated deformation of a water jet when subjected to uniform transverse

air flow of velocity Ug.
As mentioned previously, the model does not include the velocity characteristics
of the liquid jet, meaning that the jet Reynolds number is zero. The model, there-
fore, predicts the maximum distortion that a liquid jet can achieve under a steady
transverse flow of gas for a given Weber number. Hence, it is not appropriate to
compare the jet deformation ratio W/D predicted by this model with that predicted
by dark's model.8 In fact, dark's model8 [Eq. (2)] predicts, for a stationary water
column, that W/D should be infinite irrespective of Weber number Weg.
Finally, a comparison of the W/D ratios predicted by the model for the two
Weber-number conditions (in Fig. 15) with the experimental results of the acoustic
forcing experiment (Fig. 10) shows the same Weber number order of magnitude
discrepancy described in the last section. Quantitative agreement should not be
expected from such a simple modeling approach, rather the qualitative mechanistic
agreement is emphasized.
Additional insight into the deformation of a liquid jet in the direction normal
to the crossflow of gas can be found in the literature pertaining to the motion of
liquid drops in a gas.26 Liquid drops falling under the action of gravity are ellip-
soidal in shape with the major axis normal to the drop trajectory. As an example,
falling raindrops are elliptical and not teardrop shaped (a common misconception
a pointed out by Clift et al.26). Using the experimental data and dimensional analy-
sis presented by Clift et al.26 for drops moving in a gravitational field, the liquid jet
in a transverse gas flow corresponds to a region where an elliptical shape should
be expected.
The preceding analysis, along with the complementary observation regarding
moving drops, and the results from the off-center single jet experiments, support
a velocity coupling mechanism as being responsible for the fanning phenomenon
observed in the present experiments.
Acoustic Forcing of Impinging Jet Spray

The impinging jet spray experiments focused on characterizing the effects on
the sheet atomization phenomenon of the oscillating pressure and velocity fields
corresponding to the energetic 1-T/3-L (1560-Hz) resonant mode of the chamber.
254 S. PAL ET AL.
•i•-.*....- »-...-•
>-.,i*£
Fig. 16 Instantaneous images of an impinging jet water spray subjected to both a

nonexcited environment and 1-T/3-L resonant mode (1560 Hz), 1.27 mm diameter and
18.3 L/D0 of the impinging injector for the acoustic condition, the impinging jet spray
was oriented along the pressure nodal line: a) Re = 5600, t// = 4.4 m/s, nonexcited
condition, b) Re = 5600, £// = 4.4 m/s, 1290 Pa acoustic peak-to-peak pressure, c) Re
= 10,200, Ui = 8.0 m/s, nonexcited condition, d) Re = 10,200, Ut = 8.0 m/s, 1290 Pa
acoustic peak-to-peak pressure, e) Re = 17,100, t// = 13.5 m/s, nonexcited condition,
and f)Re = 17,100, £// = 13.5 m/s, 1290 Pa acoustic peak-to-peak pressure.
Imaging studies of the sheet breakup phenomenon were complemented by phase-

Doppler interferometry drop size measurements and sheet atomization frequency
measurements for both nonexcited and acoustic oscillating flow conditions. For
all of the results presented here, the chamber pressure was atmospheric, and the
impinging jet injector was oriented so that the plane of the liquid sheet coincided
with the pressure nodal line of the 1-T/3-L resonant mode.
The imaging studies showed that the atomization of the liquid sheet formed at the
jet impingement point was enhanced for the acoustic oscillating flow conditions as
demonstrated by the sequence of instantaneous images shown in Fig. 16. Here, the
three sets of images, Figs. 16a and 16b, 16c and 16d, and 16e and 16f, correspond
to increasing liquid jet velocities. The first impinging jet spray image in each
set corresponds to nonexcited conditions, whereas the second image corresponds
to a peak-to-peak oscillatory pressure level of 1290 Pa. For the nonexcited flow
conditions, the change in spray characteristics with increasing liquid jet velocity
is similar to those reported by Ryan el al.10 and the earlier study of Heidmann
et al.11 Figures 16a and 16b for the lowest impinging jet velocity case ([// = 4.4
m/s, Re = 5600) clearly show an enhancement in the breakup of the impinging
jet sheet with acoustic oscillations. The contiguous liquid sheet observed in Fig.
16a is replaced by a sheet of drops in Fig. 16b. As the liquid sheet was oriented to
coincide with the pressure nodal line, the local gas velocity oscillates normal to
the thin-liquid sheet, thereby enhancing atomization. Also, in view of the single jet
flattening phenomenon results presented earlier, the two liquid jets likely undergo
some deformation prior to impingement and, therefore, possibly contribute to the

enhanced atomization of the liquid sheet. For Figs. 16c and 16d at the intermediate
jet velocity (£// = 8 m/s, Re = 10,200), the atomization is enhanced to a lesser
degree. A close inspection of these two figures (and others) shows that the sheet
vestiges that are evident for the nonexcited case are replaced by drops for the
acoustic oscillation case. Finally, the spray flowfield for both cases look nearly
identical for the highest jet velocity case (£// = 13.5 m/s, Re = 17,100) as shown
in Figs. 16e and 16f. These results show that chamber resonant modes can affect
the atomization characteristics of impinging jet injectors.
Measurements of drop size with phase-Doppler interferometry in the spray field

provided additional insight into the effect of an oscillating velocity field on the
atomization characteristics of an impinging jet injector. For these measurements,
both the transmitting and receiving optics of the phase-Doppler particle analyzer
were placed at a 15-deg angle with respect to the pressure nodal line on each side
of the 76.2-mm-diam quartz windows (section D-D in Fig. 1). The resulting net
30-deg off-axis alignment represents the best possible configuration for the sizing
of spherical transparent drops.9'15 With this orientation, the probe volume was
positioned 50.8 mm downstream of the injector face at the chamber centerline.
The optics were chosen to provide a measurement range of up to 1500 /xm. The
impinging jet velocity (Ut = 8 m/s, Re = 10,200) and the peak-to-peak oscillating
pressure level (2Pa = 1290 Pa) were identical to the conditions corresponding to
Figs. 16c and 16d.
The measured probability density functions (pdf) f ( d ) of drop size for both
the nonexcited and oscillatory flow conditions are presented in Fig. 17. For each
measurement, a sample size of 10,000 drops was used for calculating the statis-
tics. The validation rate for these measurements was about 25%, suggesting the
0.0030
0.0020 -
0.0010 J
0.0000
250 500 750 1000 1250 1500
Fig. 17 Drop size number distribution/(d) as a function of drop diameter d in an

impinging jet spray for both non excited and oscillatory flow conditions, 1-T/3-L mode
at 1560 Hz, 1290 Hz peak-to-peak pressure, obtained from 10,000 individual drop size
measurements, 50.8 mm from the injector face, along the chamber centerline; for the
oscillatory pressure condition, 1290 Pa, peak-to-peak pressure D0 = 1.27 mm, L/D0 =
18.3, Ui = 8.0 m/s.
256 S. PAL ET AL.
presence of nonspherical ligaments at the measurement location. Both pdfs are

positively skewed and peak at about 400 /urn. In contrast to the nonexcited case,
the pdf for the oscillatory flow condition shows a statistically significant increase
in drop count between 250 and 750 /xm and a decrease at the large-size range. The
arithmetic mean diameter d\Q decreases from 523 /zm to 488 /zm, and the Sauter
mean diameter d32 decreases from 966 /zm to 831 /zm. This measured decrease in
drop size agrees qualitatively with the difference in spray structure noted earlier.
For both nonexcited and oscillatory flow conditions, the sheet atomization fre-
quency of the impinging jet injector was also measured. Similar to the experiment
of Heidmann et al,11 these measurements were made to identify and isolate any
possible mechanism for coupling between the imposed oscillatory acoustic fre-
quency and the sheet atomization frequency. For the experiments, the helium-neon
laser beam was introduced into the chamber normal to one of the 12.7-mm-diam
quartz windows (section C-C in Fig. 1), and the transmitted beam collected through
the other 12.7-mm-diam quartz window. The injector was positioned such that the
laser beam was normal to the impinging jet sheet surface 67.4 mm downstream
from the impingement point.
The photodiode signal captured the interruption of the laser beam by the liga-
ments and drops tearing off from the bottom of the sheet. The sheet atomization
frequency analysis involved rescaling the signal from zero to one, corresponding
to beam interruption and no interruption, respectively; measuring the time period
between successive crossings of a given threshold level; and then binning the
number of threshold crossings in frequency space. Whereas this analysis proce-
dure provided a distribution, the Heidmann et al.11 measurements of the same
phenomenon with a pulse meter only yielded a mean frequency.
The sheet atomization frequency measurements for two threshold levels are
presented in Fig. 18. These results for a jet velocity of 7.1 m/s (Re = 8900) con-
trast the sheet atomization frequency content between non-excited and oscillatory
flow (2Pa = 1290 Pa) conditions. The choice of a threshold level is somewhat
arbitrary; here, values of 0.5 and 0.85 (0 and 1 correspond to beam fully inter-
rupted and uninterrupted, respectively) were chosen to contrast the effect of this
parameter on the results. For either of the two flow conditions, these measurements
do not highlight a single frequency for impinging jet atomization, as suggested by
Heidmann et al.11 but rather a mean frequency with a large standard deviation.
Note, however, that the instrumentation used by Heidmann et al.11 only provided a
mean frequency. Results of sheet atomization frequency by Ryan et al.10 calculated
by relating the sheet velocity to measurements of successive detached ligament
structures also showed a mean frequency with a large standard deviation. Further-
more, since the distributions are skewed, the mean frequencies do not coincide
with the peak frequencies. The threshold acts as a filter for the smaller liquid struc-
tures and drops and, therefore, the high-frequency content of the distributions is
higher for the larger threshold. For the 0.5 threshold level, the distributions do not
exhibit any noticeable change between nonexcited and oscillatory flow conditions,
whereas at the higher 0.85 threshold level, there is a measurable change in the dis-
tributions in the frequency range up to about 2000 Hz. However, the distributions
do not indicate any energetic levels around the imposed frequency of 1560 Hz.
This suggests that the accelerated breakup and finer atomization, observed by both
imaging studies and drop size measurements, is a result of the oscillating velocity
field around the liquid sheet periphery and not due to the liquid sheet responding
to the imposed frequency.

0.30
°~° Quiescent, 0.50 threshold level

1-T/3-L Mode, 0.50 threshold level
Quiescent, 0.85 threshold level
1-T/3-L Mode, 0.85 threshold level
0.00
1000 2000 3000 4000
Sheet Atomization Frequency (Hz)

Fig. 18 Ratio of the number of threshold crossings plotted vs the sheet atomization
frequency and threshold level in an impinging jet spray, Results for nonexcited and
oscillatory flow conditions, 1-T/3-L mode at 1560 Hz, 1290 Hz peak-to-peak pressure,
D0 = 1.27 mm, L/D0 = 18.3, Ut = 8.0 m/s.
Summary
A series of experiments on single round jets subjected to acoustic pressure
oscillations in a round chamber were conducted. Oscillating pressure and veloc-
ity fields were demonstrated to dramatically affect jet breakup. The single jets
were observed to deform into two-dimensional fans at specific frequencies corre-
sponding to mixed tangential mode chamber resonances. The phenomenon was
explained on the basis of velocity coupling. An examination, involving several
types of evidence, suggested that the velocity field can induce the observed jet
deformations. The jet fanning phenomenon was observed over a wide range of jet
Reynolds Re and Weber Wegi numbers. For the jet experiments along the chamber
centerline, the plane of the two-dimensional fan coincided with the pressure nodal
lines of several mixed tangential/longitudinal modes. Also, the off-center single
jet experiments showed that for the same liquid and acoustic flow conditions, the
two-dimensional fanning phenomenon was more pronounced at locations where
the magnitude of the oscillating velocity field was greater. The pure longitudinal
modes were never observed to distort the liquid jet.
Comparisons of the liquid jet deformation induced by a transverse steady gas
flow showed visually similar geometric effects. The Weber number Weg for the
oscillating velocity field was an order of magnitude lower than that for steady
flow at a given deformation ratio (W/D) and Reynolds number. Explanation of
this discrepancy may require consideration of the appropriateness of the Weber
number Weg as a basis for comparing the two phenomena. Additionally, effects
due to differences in the oscillating and steady velocity fields or inaccuracy in
calculating Wes from pressure amplitude measurements need to be considered.
An analytical effort that predicts the equilibrium jet cross section by balancing
potential flow and surface tension illustrated that a transverse velocity field is
258 S. PAL ET AL.
capable of explaining the observed jet broadening. The results of this model
can be interpreted as yielding the maximum distortion achieved by a liquid jet
subjected to a transverse gas flow for a given gas Weber number. Additionally,
by drawing on the literature pertaining to the deformation of drops moving under
the action of gravity, similar elliptical deformation of the jet cross section was
predicted.
The flowfield from an impinging round jet injector was affected by the same
chamber resonant modes that enhanced the atomization characteristics of single
jets. Imaging studies and drop size measurements showed that the atomization of
the sheet formed by the impinging jets was dramatically enhanced for conditions
when the sheet was oriented to coincide with the pressure nodal line. Sheet atom-
ization frequency measurements indicated no coupling between the imposed and
impinging jet frequencies. Again, the velocity coupling mechanism was used to
explain the augmennted atomization results.
Acknowledgments
The authors would like to acknowledge the support of this work by NASA
through the Propulsion Engineering Research Center at the Pennsylvania State
University under Grant NAGW 1356 Supplement 2, and the Air Force Office
of Scientific Research, Air Force Systems Command, USAF, under M. Birkan.
H. M. Ryan and D. V. Hoover would like to acknowledge support under the
NASA Space Grant College Fellowship Program and the Air Force of Scientific
Research through the AFRAPT Program, respectively. The authors also thank G.
H. Koopmann from the Mechanical Engineering Department at the Pennsylvania
State University for his valuable and insightful suggestions. The assistance of
D. Boone and L. Schaaf in the assembly of the experimental apparatus is also
gratefully acknowledged.
References
ârrje, D. T., and Reardon, F. H., eds., "Liquid Propellant Rocket Combustion Insta-
bility," NASA SP-194, 1972.
2
Anderson, W. E., Ryan, H. M., and Santoro, R. J., "Combustion Instability Phenomena of
Importance to Liquid Bi-Propellant Rocket Engines," 28th JANNAF Combustion Meeting,
San Antonio, TX, Oct. 28-Nov. 1, 1991.
3
Lecourt, R., and Foucaud, R., "Expreiments on Stability of Liquid Propellant Rocket
Motors," AIAA/SAE/ASME/ASEE 23rd Joint Propulsion Conf., AIAA Paper 87-1772,
San Diego, CA, June 29-July 2, 1987.
4
Wanhainen, J. P., and Morgan, C. J., "Effect of Injection Element Radial Distribution
and Chamber Geometry on Acoustic-Mode Instability in a Hydrogen Oxygen Rocket,"
NASA TN D-5375, Cleveland, OH, August 1969.
5
Newman, J. A., "A Preliminary Study of the Effects of Vaporization and Transverse
Oscillations on Liquid Jet Breakup," Masters Thesis, Department of Aerospace and Me-
chanical Sciences, Princeton Univ., Princeton, NJ, July 1967.
6
Debler, W., and Yu, D., "The Breakup of Laminar Liquid Jets," Proceedings of the
Royal Society of London, Vol. A 415, 1988, pp. 107-119.
7
Heidmann, M. F., and Groeneweg, J. R, "Analysis of the Dynamic Response of Liquid
Jet Atomization to Acoustic Oscillations," NASA TN D-5339, Cleveland, OH, July 1969.
8
Clark, B. J., "Breakup of a Liquid Jet in a Transverse Flow of Gas," NASA TN D-2424,
Cleveland, OH, 1964.
9
Bachalo, W. D., and Houser, M. J., "Phase/Doppler Spray Analyzer for Simultaneous
Measurements of Drop Size and Velocity Distribution," Optical Engineering, Vol. 23, No.
5, 1984, pp. 583-590.
1()
Ryan, H. M., Anderson, W. E., Pal, S., and Santoro, R. J., "Atomization characteristics
of Impinging Liquid Jets," 31st Aerospace Sciences Meeting and Exhibit, AIAA Paper
93-0230, Reno, NV, Jan. 11-14, 1993.
H
Heidmann, M. R, Priem, R. J., and Humphrey, J. C., "A Study of Sprays Formed by
Two Impinging Jets " NACA TN 3835, March 1957.
12
Vassallo, P., Ashgriz, N., and Boorady, F. A., "Effect of Flow Rate on the Spray
Characteristics of Impinging Water Jets," Journal of Propulsion and Power, Vol. 8, No. 5,
1992, pp. 980-986.
13
Hautman, D. J., "Spray Characterization of Like-On-Like Doublet Impinging Rocket
Injectors," 29th Aerospace Sciences Meeting, AIAA Paper 91-0687, Reno, NV, Jan. 7-10,
1991.
14
Dombrowski, N., and Hooper, P. C., "A Study of the Sprays Formed by Impinging Jets
in Laminar and Turbulent Flow," Journal of Fluid Mechanics, Vol. 18, 1963, pp. 392-406.
15
Ibrahim, K. M., Werthimer, G. D., and Bachalo, W. D., "Signal Processing Considera-
tions for Laser Doppler and Phase Doppler Applications," Fifth International Symposium
on the Application of Laser Techniques of Fluid Mechanics, Lisbon, Portugal, July 9-12,
1990.
16
Reitz, R. D., "Atomization and Other Breakup Regimes of a Liquid Jet," Ph.D. Thesis,
Princeton Univ., Princeton, NJ, 1978.
17
Lefebvre, A. H., Atomization and Sprays, Hemisphere, Philadelphia, PA, 1989,
pp. 28-78.
18
York, J. L., Stubbs, H. E., and Tek, M. R., "The Mechanism of Disintegration of Liquid
Sheets," Transactions of the American Society of Mechanical Engineers, Vol. 75, 1953,
pp. 1279-1286.
19
Morse, P. M., Vibration and Sound, 4th ed., Vol. 1, Published by the American Institute
of Physics for the Acoustical Society of America, 1976, pp. 225-265.
2()
Kinsler, L. E., Frey, A. R., Coppens, A. B., and Sanders, J.V., Fundamentals of Acous-
tics, 3rd ed., Vol. 1, Wiley, New York, 1982, pp. 98-117.
21
Petersen, R. A., and Samet, M. M., "On the Preferred Mode of Jet Instability," Journal
of Fluid Mechanics, Vol. 194, 1988, pp. 153-173.
22
Faeth, G. M., "Structure and Atomization Properties of Dense Turbulent Sprays,"
Twenty-Third Symposium (International) on Combustion, Combustion Inst., Pittsburg, PA,
1990, pp. 1345-1352.
23
Kirchhoff, R. K., "Computer Graphic Solutions," Potential Flows, 1st ed., Vol. 1,1985,
pp. 25-28.
24
Goedde, E. R, and Yuen, M. C., "Experiments on Liquid Jet Instability," Journal of
Fluid Mechanics, Vol. 40, 1970, pp. 495-511.
25
Anderson, D. A., Tannehill, J. C., and Pletcher, R. H., Computational Fluid Mechanics
and Heat Transfer, 1st ed., Vol. 1, Hemisphere, Washington, DC, 1984, pp. 302-318.
26
Clift, R., Grace, J. R., and Weber, M. E., Bubbles, Drops, and Particles, 1st ed., Vol. 1,
Academic, New York, 1978, pp. 26-27.



Chapter 11
Simulation of High-Pressure Spray Field Dynamics

Joseph C. Oefelein* and Vigor Yangf

Nomenclature
BT — thermal transfer number
CD = drag coefficient
Cp = constant-pressure specific heat
T>im = effective diffusion coefficient
d = diameter of equivalent sphere
e = specific internal energy
et = specific total internal energy
Fs = interphase source term associated with momentum
/ = droplet distribution function
g, G = generic weighting functions
h = heat transfer coefficient
hi = specific enthalpy of species /
ftO . = heat of formation of species /
A/ZU = enthalpy of vaporization
J = coordinate transformation Jacobian
m = mass
m = mass flow rate
N = total number of species
nd k = number of identical droplets associated with £th parcel
p = pressure
Q = turbulent energy flux
Qs = interphase source term associated with energy
qe = energy diffusion flux
qi = mass diffusion flux of species /
Re = Reynolds number
Ru = universal gas constant
Copyright © 1995 by Joseph C. Oefelein. Published by the American Institute of

Aeronautics and Astronautics, Inc. with permission.
*Research Assistant, Department of Mechanical Engineering. Member AIAA.
^Professor, Department of Mechanical Engineering. Associate Fellow AIAA.
263
264 J. C. OEFELEIN AND V. YANG
Si = turbulent species flux

T = temperature
T — turbulent momentum flux
t = time
u = velocity vector
u,v — velocity components in the axial and transverse coordinates, respectively
V — volume
ygas = volume associated with gas phase
Vtotai = total system volume
W = molecular weight
w = average interface velocity
Xi = mole fraction of species /
jc, y = axial and transverse coordinates, respectively
x,y = position vectors
Yi = mass fraction of species i
Z = compressibility factor
Greek Symbols
a = thermal diffusivity
A = characteristic spatial interval
8r = characteristic temporal interval
0 = void fraction
A. = thermal conductivity
jj, = viscosity
v = kinematic viscosity
p = density
ps = interphase source term associated with mass
r = viscous stress tensor
TI = lifetime of equivalent spherical fluid particle
co = vorticity
a)Sti = interphase source term associated with species
Subscripts
c = critical property
d = equivalent spherical droplet
/ = species /
m = mixture
o = reference state
p = parcel
r = reduced property
s = surface property
t — turbulent quantity
oo = ambient flow condition
Superscripts
— — Reynolds-averaged quantity
= Favre-averaged quantity
' — Reynolds fluctuation
" = Favre fluctuation
SIMULATION OF HIGH-PRESSURE SPRAY FIELD DYNAMICS 265
I. Introduction
T HIS chapter focuses on the simulation of high-pressure, multiphase, com-

bustion dynamics typically observed in contemporary propulsion and power-
generation systems. Emphasis is placed on supercritical processes. Modeling these
dynamics in a fully coupled manner poses a variety of challenges. In addition to
the classical closure problems associated with incompressible turbulent flows,
reacting, multiphase flows introduce the complicating factors of chemical ki-
netics, highly nonlinear source terms, and a variety of multispecies, multiphase
interactions. Flowfield evolution is governed by both compressibility effects (i.e.,
volumetric changes induced by changes in pressure) and variable inertia effects

(i.e., volumetric changes due to variable composition and/or heat addition). The
situation is compounded with increasing pressure due to the introduction of ther-
modynamic nonidealities and transport anomalies.
Inherently, the flowfields associated with state-of-the-art combustors are com-
posed of an intricate array of mixing layers between fuel, oxidizer, and product
species. Liquid fuel and/or oxidizer, initially in subcritical states, are injected into a
high-pressure environment that often exceeds the respective thermodynamic criti-
cal points of the propellants. As liquid surfaces approach the critical condition, the
difference between gas and liquid densities becomes smaller and transient effects
in both phases begin to predominate. Depending on the hardware configuration,
fluid properties, and flow characteristics, two possible scenarios exist with respect
to the injection process. If heating rates are low, dynamic forces and surface ten-
sion effects promote an array of atomization and secondary breakup mechanisms.
For this situation, a heterogeneous spray field is formed which evolves continu-
ously over a wide range of thermodynamic regimes in response to a cascade of
vaporization, mixing, combustion, and expansion processes. If heating rates are
high, diffusion processes can predominate prior to atomization. For this situa-
tion, injected liquid jets vaporize forming a continuous fluid in the presence of
extremely steep interfacial density gradients. Regardless of the physical mecha-
nisms present, the resultant inter- and intraphase coupling dynamics yield an array
of highly nonlinear, turbulent, chemical, gas-gas, and gas-liquid interactions for
which established theories are lacking and that are extremely difficult to treat nu-
merically. These interactions are dominated by widely disparate time and length
scales, many being smaller than can be resolved numerically given the current
state of art in computer architecture. As a consequence, the theoretical basis for
the study of such systems must be formulated in terms of numerically resolvable
flow dynamics, with unresolvable or subgrid-scale (sgs) dynamics modeled using
validated analytical or empirical techniques.
The complexity of the system just outlined is numerically demanding and
several tradeoffs are required with respect to the hierarchy of processes and
characteristic scales resolved. A variety of uncertainties exist with regard to the
closure problem, and in many cases computational capacity is in direct conflict
with the accuracy of the sgs models employed. As part of a systematic effort
to address these difficulties, this chapter presents a generalized theoretical for-
mulation that accommodates sgs turbulence, interphase transport, and chemical
kinetics in a numerically compatible manner. The governing system for a hetero-
geneous, two-phase, reacting flow is derived in its most general form through the
use of convolution integrals. These integrals filter out small-scale dynamics from
resolved scales over a defined set of spatial and temporal intervals. The goal is
to provide a unified and numerically compatible framework which facilitates the

closure problem in a well-posed manner.
On establishing the theoretical framework, a representative set of calculations
are presented to demonstrate key concepts. Here, the current state of the art as-
sociated with inert spray field dynamics at supercritical pressures is presented.
A detailed property evaluation scheme and consistent closure methodologies are
specified that handle unsteady sgs turbulence and spray field interactions in a
fully coupled manner over the full range of thermodynamic regimes. Turbulence
is modeled by means of the large-eddy-simulation technique (LES) using the sgs
models developed by Erlebacher et al.1"3 Spray field evolution is modeled us-

ing the separated-flow (SF) methodologies described by Faeth4 coupled with a
set of recently obtained correlations5'7 that describe high-pressure vaporization
dynamics associated with a dilute oxygen spray vaporizing in a hot hydrogen
stream. Future calculations will incorporate a detailed combustion model using
the multiple scalar mixing methodologies outlined by Girimaji.8 A companion set
of calculations are also being conducted to systematically assess the interrelation
between fluid dynamic and physicochemical processes over a wide range of pres-
sures in hydrogen/oxygen mixing layers.9 Overall objectives are to gain insight
with respect to 1) the many uncertainties associated with modeling turbulent, re-
acting, multiphase flows at high-pressure, 2) the sensitivity of prevailing dynamics
to various modeling assumptions, and 3) the fundamental coupling mechanisms
associated with multiphase combustion processes at high pressures. In all sub-
sequent discussions, the term "droplet" will be used in both the classical sense,
when referring to subcritical behavior, and also to denote dense fluid particles
when referring to supercritical behavior.
II. Approach
Numerous spray combustion models have been proposed over the past three
decades. Reviews by Faeth4'1() have systematically organized relevant literature up
to 1987. With subtle differences in specific detail, these models can be generally
classified as either locally homogeneous-flow (LHF) models, where the condensed
phase is assumed to be in local thermodynamic equilibria with the gas phase, or
separated-flow (SF) models, where finite-rate interphase transport is considered.
Of these two classes, SF models have received the widest acceptance.
Lagrangian tracking schemes have proven to be the most efficient way to model
interphase exchange rates, with one of two methodologies employed to account
for the effects of turbulence on droplet dispersion: 1) gradient diffusion models10
or 2) stochastic-separated-flow (SSF) models.4 Gradient diffusion models employ
empirically derived proportionality constants that are tuned to achieve observed
experimental trends. SSF models employ ensemble droplet dynamics through the
specification of a characteristic turbulent eddy lifetime. This lifetime is obtained
through scaling arguments with respect to the turbulence model employed. A
particle is assumed to interact with an eddy for a time defined as the smaller of the
eddy lifetime or the characteristic transit time as given by the Lagrangian equations
of motion. Generally, SSF methods have found wider acceptance due to difficulties
associated with obtaining a gas-particle exchange coefficient coupled with the
inconvenience of incorporating gradient diffusion expressions into Lagrangian
parcel tracking routines. Dukowicz11 and Gosman and loannides12 were the first
to adopt stochastic methods to study droplet dispersion due to turbulence. Shuen13
has extended the analysis by Gosman and loannides to include the effects of
turbulence on interphase heat and mass transport. This model has been evaluated
in a wide variety of flows with encouraging results.
Typical closure schemes employed in most contemporary spray combustion
algorithms, for example those developed by Liang and Chan,14 Shuen et al.,15 and
Chen and Shuen,16 use the conventional time-averaged Reynolds stress approach
(i.e., Reynolds averaging) coupled with the SSF methodology outlined earlier.
The two-equation k-€ model fashioned after the work of Jones and Launder17 is
used almost exclusively. There are two major deficiencies associated with this
framework when used to simulate unsteady dynamics. First, the use of Reynolds
averaging for unsteady flow analysis implies that turbulent time scales are small in
comparison to convective time scales. Such an assumption is usually adequate for
slowly accelerating flows or flows with low-frequency periodic behavior. It is not
adequate for short-duration transient flow or flow with high-frequency periodic
behavior. For such flows, the time scales of the large-scale turbulent structure and
convective processes are comparable. Second, the k-e equations do not describe
all of the scales of importance in an unsteady flow. Thus, for multiphase flows it
becomes necessary to infer additional terms to account for interphase modulation
effects on the turbulence field. This entails the incorporation of a momentum
exchange function into the derivation of the kinetic energy equation coupled with
the a priori specification of modulation effects on the turbulent dissipation rate.
The modeling of such terms presents many challenges. Three models have been
employed to simulate modulation effects using the k-€ framework. These are
the works of Mostafa and Mongia,18 Amsden et al.,19 and Fashola and Chen.20
Although each has demonstrated encouraging results for steady-state conditions,
no definitive conclusions may be drawn with respect to unsteady flowfields because
of uncertainties associated with model constants.
To eliminate the deficiencies just outlined, a large-eddy-simulation procedure
is employed wherein the large-scale turbulent structure is directly simulated and
finer scale dissipative structure is modeled. This technique has evolved over sev-
eral decades largely on the foundations laid by meteorologists at the National
Center for Atmospheric Research. Underlying physical concepts and comprehen-
sive historical data have been presented in articles by Fox and Lilly,21 Schumann,22
Yoke and Collins,23 and Rogallo and Moin.24 Pioneering work in this area was
conducted by Deardorff,25 with theoretical support by Lilly.26 In these simulations,
the numerical grid is refined to a point where small-scale turbulence is presum-
ably isotropic, and time-dependent large-scale structures are calculated directly.
Conceptually, the small scales carry only a small portion of the total turbulent
energy and thus less complicated sgs models are required to resolve small-scale
interactions.
From a mathematical perspective, the simulation of turbulence using the LES
methodology begins with the formal filtering of small-scale effects from large-
scale motions in the full conservation equations. This filtering operation is per-
formed through the use of convolution integrals.27 The premise behind this concept
is based on the fact that values at discrete mesh points represent flow variables
only in some average sense. In general, the filtering operation is a mathematically
formal generalization of the Reynolds-averaged approach. The fundamental dif-
ference between solutions obtained using LES vs Reynolds averaging is in the
definition of the small scale. In LES, small scales are defined as those less than the
grid size A. For Reynolds averaging, all dynamical degrees of freedom smaller
than the size of the largest energy containing eddies are averaged out. Thus, no
dynamical information exists with respect to the smaller scales.
III. Concept of Filtering

From an operational point of view, the filtering approach applies a linear oper-
ator that is commutative with the space and time derivatives in the conservation
equations. Different filtering operators at different levels of resolution can be hi-
erarchically organized in terms of characteristic scales, such as grid size and/or
time increments. In doing so, both the coherent and the stochastic attributes of the
flowfield are resolved.

Flows of practical interest are nonstationary, nonisotropic, and inhomogeneous.
Thus, it is necessary to use a filter with a spatially dependent width to encompass
the variation in size of the turbulent eddies that are present in various parts of the
flow domain. Accordingly, filtering intervals must be bounded in both time and
space. Such a representation of the field variable 0f- can be formally expressed by
convolution integrals of the form proposed by Leonard27
4>i(y, r)G(x -y, t - r; A, 8r)dy dr (1)

Jo
where x and y represent three-dimensional position vectors of common origin,
with 0 < r < t and Vgas represents the total system volume associated with the
gas phase. These integrals are normalized such that
G(x -y, t - r- A,8r)dy dr = 1 (2)
The quantity G represents a weighting function that quantifies the influence of

sgs dynamics at remote points (y, r) on filtered values at (or, /). Here, A and Sr
represent the characteristic spatial and temporal intervals over which filtering is
performed. These intervals must be sufficiently large compared to characteristic
sgs interactions, but small with respect to the large-scale changes in the flow.
Throughout this study, overbars will denote resolvable scales, with primes denoting
sgs quantities.
To account for variable density effects, the filtering operation given by Eq. (1)
is augmented with Favre decompositions28 of the form
<t>t = 4>i + <$>" 4>i=î/p (3)

These are applied to velocity, energy, and species concentration. Classical
Reynolds decompositions,29 i.e.,
0. = 4>i + <t>i (4)

are applied to pressure and density. Favre decompositions yield a mass-weighted
average based on the function G whereas Reynolds decompositions are based on
a centered operator. The relation between Favre-averaged and Reynolds-averaged
quantities are as follows:
(5)
Equations (1) and (2) provide a formal definition of the averaging process. This
formalism effectively separates numerically resolvable interactions from sgs pro-
cesses. When the average is uniform over an unbounded homogeneous dimension
in space or time, the postulates of Reynolds29 lead to
pfauj = pfauj + p4>"u" (6)

27 30
These postulates do not apply, however, over bounded domains. ' For this case
the convective fluxes are given by
pfauj = pfaiij + LIJ + Ctj + Rij (7)

where
In the momentum equation, the terms L/y, C//, and /?/; are referred to as the
Leonard stresses, cross-term stresses and sgs Reynolds stresses, respectively. Anal-
ogous terms appear in the energy and species equations. By comparing Eqs. (6)
and (7), it is apparent that the Reynolds-averaged approach is a special case of the
general situation described by Eq. (7).
During the past decade, considerable progress has been made in the LES of in-
ert, single component, incompressible, turbulent flows. Early works relied heavily
on the use of Reynolds averaging to eliminate the Leonard stresses and cross-term
stresses, with Reynolds stresses computed using the Smagorinsky model.25'27'31'32
To enhance numerical accuracy, recent simulations have been based on the direct
calculation of Leonard stresses, with scale-similarity and eddy-viscosity models
employed to model sgs cross-term stresses and Reynolds stresses, respectively.
Among these models, only the Bardina et al. model33 satisfies the important phys-
ical constraint of Galilean invariance.34
Reliable sgs models for inhomogeneous, turbulent, compressible flows are still
in the developmental stage. To account for such effects, the current work adopts the
pioneering work of Erlebacher et al.1-3 These researchers propose a compressible
generalization of the Smagorinsky model coupled with a scale-similarity model
to obtain the cross stresses. The sgs heat and mass fluxes are obtained in a similar
manner using the gradient transport hypothesis. Leonard stresses are calculated
directly.
The Smagorinsky model employs an eddy viscosity vt which is obtained by
assuming that small scales are in equilibrium at the sgs level. The resultant balance
between energy production and dissipation yields an expression of the form
vf = (C,A) 2 |S|
where A is a filter width typically proportional to the grid size, Cs is the Smagorin-
sky constant, and |S| = J25/y S// the magnitude of the large-scale strain rate tensor
T, 1 (**i . 9i
Here, HI represents the large-scale velocity.

In its original form, there are six limitations associated with the Smagorinsky
model.
1) It does not account for compressibility effects.

2) It does not yield the correct limiting behavior near walls.
3) It does not account for back scatter of energy from small scales to large
scales.
4) It does not vanish in laminar flow.
5) It is too dissipative within transition regions.
6) The optimal model constant must be changed as a function of flow charac-
teristics.
Item 1 represents the most critical limitation with respect to the current analysis.
Erlebacher et al.1"3 have demonstrated that deficiencies associated with compress-

ibility effects are easily corrected through the introduction of Favre filters and the
inclusion of Leonard stresses and cross-term stresses in the calculation procedure.
Relevant details are outlined in the next section.
IV. Theoretical Formulation

This section presents an analysis of the governing equations that accommodate
inhomogeneous turbulence, heterogeneous spray dynamics, and finite-rate chem-
ical kinetics over a wide range of thermodynamic states and spray field regimes.
These equations are derived in their most general form through the use of convolu-
tion integrals that filter out the small-scale dynamics from numerically resolvable
flow properties on well-defined spatial and temporal intervals. Beginning with the
instantaneous system, filtering is performed in two stages. First, the influences
of sgs gas-gas turbulence and chemical interactions are incorporated yielding the
well-known closure problem for turbulent reacting gases. The second filtering
process incorporates sgs gas-liquid interactions and is performed as a superposi-
tion on the first to eliminate the introduction of new correlations. Conceptually,
the resultant system of integrodifferential equations naturally incorporate dense
spray effects, with closure dependent on the accurate specification of coupled sgs
gas-gas and gas-liquid interactions.
A. Gas-Gas System
Neglecting body forces, radiation, and Dufour and Soret effects, the filtered
conservation equations of mass, momentum, total energy, and species concentra-
tion for a compressible, chemically reacting gas of N species can be expressed,
respectively, in conservative form as
^ + V • (pu) = 0 (8)
ot
p/] = V - f - V - r (9)
ot
where
u)l + pv[Vu
o
— (pet) + V - [(pet + p)u] = V - qe + V • (f • «) - V - Q - V - (T • «)

ot
- V - [ltr(r«")] + V - (r~u") (10)
where
e, = e + i« • u + {tt(T)/p
p p^l
i=h°fi+ r I en p rT
J Pta •'Tref
i=\
) = V.qi-V-Si+vi i = !,...,# (11)

ot
where
In the preceding equations, r represents the viscous stress tensor, et the total
internal energy, qe the energy flux due to heat conduction and mass diffusion, and
qi the mass diffusion flux of the fth species. The viscous stress tensor is assumed
to follow Stokes' hypothesis, and Pick's and Fourier's laws have been used to
approximate species and thermal diffusion processes, respectively.
The equation of state is expressed in general form as the product of the com-
pressibility factor Z and the ideal gas equation of state for a mixture; i.e.,
p = pRmT + pR'^T" (12)

where
N N
Y Y"
Rm = Z
Such a representation accounts for thermodynamic nonidealities without introduc-

ing significant numerical complexities.
Equations (9) and (11) differ from their instantaneous counterparts by the ap-
pearance of the turbulent momentum and species fluxes, T and S, , and mean chem-
ical source terms &>/, i = 1, . . . , N. Quantities T and Si represent the Leonard,
cross, and Reynolds terms depicted in Eq. (7)
T = p(5ii - uu) + p(£u» + uû) + pu"u" (13)
» + l^fi) + pY~u» (14)
The mean chemical source terms represent the average rate of production of species
/ within the spatial and temporal intervals A and ST.
Equation (10) represents the rate of change of the sum of the mean internal
energy, mean kinetic energy, and turbulent kinetic energy. Numerous additional
terms appear as a result of the filtering process. Interpretation of the left-hand side
and first two terms on the right-hand side of Eq. (10) are analogous to that of their
instantaneous counterparts. The last four terms on the right-hand side represent
the collective effect of turbulence interactions on the filtered total energy field.
Quantity Q represents the Leonard, cross term, and Reynolds enthalpy fluxes
Q = pCp(fu - fu) + pCp(fu" + fû) + pCpfî" (15)

Thus V • Q represents transport of mean enthalpy due to the presence of sgs
fluctuations. The last three terms
V • (T • u) V - [±tr(7VO] V • (r~M")
represent, respectively, 1) transport of mean kinetic energy due to the production
of sgs turbulent kinetic energy and sgs interactions with the mean flow, 2) transport
of turbulent kinetic energy due to the presence of sgs turbulent fluctuations, and
3) transport of mean internal energy due to dissipation and diffusion of turbulent
kinetic energy and mean shear stress interactions with the sgs turbulence field.
B. Gas—Liquid System
Subgrid-scale spray field interactions are incorporated into Eqs. (8-11) through
the introduction of a second filter of the same form as Eq. (1). This filter is
analogous to the weighting function first proposed by Gough35 and Gough and
Zwarts36 and is applied in a manner similar to Anderson and Jackson,37 Whitaker,38
andO'Rourke. 39
Two fundamental assumptions are connected with this approach. First, it must
be assumed that the effect of the spray field on the filtered system can be described
in terms of a unique set of properties averaged over the intervals A and ST. Second,
it must be assumed that the unsteady spray field distribution can be characterized
by a weighting function g. This implies that the dynamic behavior of g can be
described in terms of a characteristic set of parameters that are coupled locally to
the dynamic behavior of the gas phase.
By definition, the weighting function is a joint probability-density function
(pdf) in both time and space and, therefore, is nonnegative and approaches zero
sufficiently rapidly with distance from the point (or, £) to ensure the existence of
its integral. In contrast to Eq. (2), however, the normalization condition for this
function is
where
/' 'III
JO J J JVto,
g(x —y, t — r; A, 8r)dy dr = 1 (16)
K
Vtotal = ^gas + ]T V-
/= !
Here, V, represents the volume of the z'th droplet, with K representing the total
number of droplets in the system at time t.
With the preceding definition, the gas-phase volume fraction or void fraction is
given in one of two ways,
= f g(x-y,t - dr
Jo
(17)
The expected value of an arbitrary tensor, \fs associated with the flowfield is given
by
g(x -y, t - r\ A, , r)dy dr (18)

where (^(jc, 0) represents the average effect of the spray field on the quantity fy
over the intervals A and 8r.
Incorporation of spray field effects into Eqs. (8-11) is performed by means
of two commutation relationships that describe the effect of the order in which
differentiation and averaging are performed. These relationships are derived by
differentiating Eq. (18) with respect to time and space and applying multiple
applications of Leibnitz's theorem and Gauss' theorem to the resultant right-hand
side expression. On performing these operations, the commutation equations take
the form
g(x-y,t - r; A, 8r)\lr(y, T)H> - ndS dr (19)
and
=V
dr
/
(l
- dr (20)
JQ
The vector w that appears in Eq. (19) represents the average velocity of the interface
at the point (jc, t) in response to sgs fluctuations induced at remote points (y, r).
The vector n represents the unit normal to this interface, taken as positive in an
outward direction from the gas phase.
Equations (19) and (20) form the fundamental basis on which formal averaging
of sgs spray field effects are incorporated into the filtered gas-gas system. The
physical interpretation of the resultant surface integrals provides insight toward

the appropriate characterization of the spray field and its evolution.
Consider the following surface integral
dr
This expression represents the interfacial area of influence at points (x, t) in re-
sponse to sgs fluctuations induced at remote points (y, r). Similarly, the expression
/'
Jo
g(x - j, t - r; A, Sr)\lr(y, r) • ndSdr (21)
represents the average of \lt(x, t) over this area.

The significance of Eq. (21) becomes apparent after incorporating the preceding
framework into Eqs. (8—11). Before doing so, however, it is necessary to obtain
one additional relation. Assuming that the quantity \/f(y, r) is common to both
phases, the following identity may be deduced:
/'
Jo
g(x -j, t - r; A, 8r)\/s(y, • ndS
J
dr
g(x -y, t - r; A, , r) • nddS dr (22)
Here, n^ represents a unit normal with respect to the interface taken as positive in
an outward direction from the /th droplet. Naturally, g is nonzero for only a small
fraction of droplets in the vicinity of points (#, t).
Using the developments just given, spray field effects are incorporated into
Eqs. (8-11) by multiplying each by the weighting function g, integrating over
the region occupied by the gas, then applying the commutation laws given by
Eqs. (19) and (20). In each case, resultant surface integrals are replaced using the
identity depicted in Eq. (22). To illustrate the procedure, consider the equation for
conservation of mass. Performing the outlined sequence of operations yields
/'
JQ
g(x-y,t-r-A,Sr)-dy dr
g(x -y, t - r; A, 5r)V • (pu)dy dr = 0 (23)

where
g(x -y, t - r; A, <5r)p(y, T)H> - nddS dr

S(y,r)
t I K
_
£
[_ /=1 S(y,i
g(x -y, t - r; A, 6r)p(y, r)« • nddS dr
Summing terms (i) and (ii) yields
3_
(24)
Tt
where
/?,(*, 0= / dr
JO
(25)
If sgs turbulence and spray field interactions are averaged over the unique intervals
A and <5r, then averages taken with respect to the weighting function G are
equivalent to those taken with respect to g and no new correlations are generated.
Thus,
(p) = p (pu) = pu
and Eq. (24) can be written
_3_
(26)
~di
In a similar manner, Eqs. (9-11) take the form
— (Opu) + V • (8(puu + pi)] = V • (6f) - V • («D + F, (27)

Ot
— (6 pet) + V • [8(pe, + p)u] = V - (Oqt) + V - ( 8 f - u )

ot
- V • (6>(2) - V • (0r • fi) - V • [0 + V - [#Cr~^)] + g, (28)
-(QpYi) + V - (epYtu) = V • (0qi) - V - {9 Si) + On, + d>,,,

ot
1 = 1, ...,N (29)
where
K
F*(x,t)= I/" f y J° \i^ S(y,T)
-(-pI+T)}.nddS\dr (30)
6.C..,)-/t/01
-(-/7/+T) M -^ e } w,/d5 dr (31)
Wv,i(Jt,0=/ T]^^ ^(^-J'^ - r; A,

Jo ^JJs(y^
dr (32)
Two benefits emerge as a result of the mathematical formalism. First, the void
fraction, given by Eq. (17), and source terms (25) and (30-32) are incorporated
naturally into the governing system without the introduction of new correlations.
Such a condition is the minimum requirement for the incorporation of physically
harmonious closure schemes. Second, the coupling between phases is based on
the a priori specification of an interface condition. For subcritical conditions a
distinct interface exists due to the presence of surface tension, thus, the choice is
obvious. At supercritical conditions the interface must be defined with respect to
one of the critical isolines. In the current work, the critical mixing temperature is
employed for this purpose.
Assuming no net accumulation of respective quantities, the instantaneous bal-
ances of mass, momentum, energy, and species at the interface of an arbitrary
droplet are
[p(u - H>)} • nd = [pd(ud - w)} - nd (33)
{pu(u - w) - (—pi -h r)} • nd = {pdud(ud - w) - (-pdl + rd)} - nd (34)
(pet(u -w)- (-pi +r)-u-qe}'nd

(ud - w) - (-pdl + rd) • ud - qdê] - nd (35)
[pYi(u - w) - qi] • nd = [pYdti(ud - w) - qdj] • nd (36)

Substitution of right-hand side quantities into Eqs. (25) and (30-32) yield general
analytical expressions that describe gas-liquid coupling dynamics. In practice,
several simplifying assumptions are required to fully characterize the state of the
spray field. The next section outlines relevant assumptions generically.
C. Spray Field Characterization

The modeling of sgs gas-liquid interactions is dependent on three fundamental
tasks: 1) obtaining an analytical description of spray field behavior in terms of a
globally applicable set of parameters, 2) obtaining an analytical description of the

weighting function g, and 3) obtaining an analytical description of the dynamic
evolution of g in terms of the quantities prescribed in 1 and 2.
In this section, characteristic parameters will be denoted generically by an M
dimensional vector s. The void fraction given by Eq. (17) and interphase source
terms given by Eqs. (25) and (30-32) will then be reformulated in terms of this
vector and an identity that equates the weighting function g to a joint droplet
distribution function f(s\x, t). The joint droplet distribution function describes
the probable number of droplets per unit system volume per unit time that exist in
a given state (s\ , $2, . . . , SM).
The source terms employed in this study are derived based on the assumptions
that the spray field exists as discrete droplets of arbitrary geometry, with the
behavior of individual droplets governed by three mutually exclusive events in the
time interval t —>• t + dt .
1) They may assume new states s1/ + dsi, i = 1, . . . , Af , and move to new
coordinates x + dx.
2) They may coalesce with adjacent droplets at given points in space forming a
single entity.
3) They may breakup at given points in space forming an array of droplets.
Conceptually, these assumptions accommodate dense spray dynamics just beyond
the liquid-core or spray-fan regions of the injector through the specification of
time-dependent droplet distributions as an inlet boundary condition.
With the preceding assumptions, the volume of an arbitrary droplet can be
fully characterized by various components of s. For example, the radius would
characterize the volume of spherical droplets and/or deformed fluid particles that
are redefined in terms of equivalent diameters. Thus, the droplet volume will be
denoted Vd(s). Given this definition, the relation
represents the probable volume per unit system volume occupied by all droplets
existing in a particular state (s\, si, . . . , SM) in the vicinity of the point (jc, t).
Comparing this definition with Eq. (22) yields the following identity:
g(x — y, t — r; A, 8r)\l/(y, t) • nddS
(37)
*S(s) J
Using Eq. (37), the void fraction may be written in terms of the droplet distribution
278 J. C, OEFELEIN AND V. YANG
Table 1 Surface averaged interphase source terms

represented by Eq. (39)
t) (Pd(ud -w)}-nd
, t) (Pdud(ud - w) - (-pdl + rd)} - nd
,0 [ped,t(Ud - w) - (~PdI + *d) ' ud - qd<e] • nd
,0 {pYd.i(ud - H>) - qdti] - nd
function as
0(x,t) = l- f ••• I j Vd(s)f(s; x, t)dsids2, ...,dsM (38)

** SM J $2 " s\
Similarly, the source terms (25) and (30—32) may be expressed in the generic form
V(x,t)= [ • ' • / f f(s;x,t)\<£<( 1r(s).nddS\dsids2,...,dsM (39)

JsM Jsi Jsi IV J Js(s) I
J
where *!>(*, t) and \/r(s) - n^ represent respective arguments listed in Table 1.

Equations (38) and (39) provide a formal link to the framework established by
Williams40 and O'Rourke.39 Equation (39) represents local interphase transport
processes through integration over the surface of all classes of droplets times the
probable number of drops in each class, as dictated by the shape of the droplet
distribution function. Terms denoted by \fr(s) • nd in Table 1 represent the rate of
exchange of mass, momentum, energy, and species per droplet in a particular state.
Mass transfer occurs through vaporization and dilatation processes. Momentum,
energy, and species transfers occur as a result of these mass transfer mechanisms
coupled with the net effects of pressure forces, viscous forces, and molecular
diffusion processes.
A further simplification of Eq. (39) is obtained if respective surface integrals
in Table 1 are replaced with volumetric averages. This form is given in Table 2.
These expressions are employed in all subsequent calculations with a Lagrangian
tracking scheme to model the dynamic evolution of /.
Table 2 Volume- averaged interphase source terms

represented by Eq. (39)
________________________s^
/
P,(x,t)
F,(*,/)
a.e.,0 "" dv< dt

T/ HV, dn,\ d^n
d
»s,i(X,t)
dt j
V. System Closure
Equations (26-29) coupled with the interphase source terms (25) and (30-32),
the void fraction given by Eq. (17), and an appropriate equation of state represent
the most general form of the governing system. This system is fully characterized
by the primitive variables p, /?, u, f, and Y\, ..., YN on the specification of em-
pirical and/or analytical models that adequately describe: 1) thermodynamic and
transport properties Z, Cp, //,, A, and 2^m; 2) subgrid-scale gas-gas interactions as
characterized by J, Q, and S/; 3) subgrid-scale gas-liquid interactions as charac-
terized by 0, ps, Fs, QS9 d)Sti'9 and 4) chemical kinetics as characterized by <y,-.
To illustrate key issues related to the closure problem, a series of recently

obtained correlations5'7 are employed to analyze the effect of pressure on a dilute
oxygen spray vaporizing in a turbulent hydrogen stream. Only a representative set
of calculations are presented here in the interest of demonstrating key concepts.
More detailed analyses focused on turbulent interactions and predominate reaction
mechanisms at high pressure are currently being conducted by Oefelein.9
Because mixing layers are fundamental to all combustion devices, the two-
dimensional domain illustrated in Fig. 1 has been selected as the baseline config-
uration. This flowfield has been selected for three primary reasons.
1) It embodies several of the phenomena of interest in full scale combustors,
with minimal complications introduced through boundary effects.
2) It is sufficiently confined to accommodate the inherently small spatial and
temporal intervals required to accurately resolve key processes.
3) A large database exists for validation purposes.
Case studies selected are designed to isolate phenomena of interest while mini-
mizing the uncertainties associated with various modeling assumptions. Specifics
regarding the cases selected and relevant boundary conditions are discussed in
subsequent sections. Details associated with the closure scheme are outlined here-
after.
A, Thermodynamic and Transport Properties

To account for thermodynamic nonidealities and transport anomalies over a
wide range of pressures and temperatures, an extended corresponding-states prin-
ciple similar to that developed by Rowlinson and Watson41 is employed with two
different equations of state. The 32-term Benedict-Webb-Rubin (BWR) equation
of state proposed by Jacobsen and Stewart42 is used to predict the PVT behavior of
the liquid phase, saturated vapor mixtures, and gas-phase properties in the vicinity
of the critical point. The Soave-Redlich-Kwong (SRK) equation of state43 is used
"T"'"
5 w
-t......
U2l...
Fig. 1 Computational domain employed for analysis of high-pressure spray field

dynamics.
elsewhere. Combined, this methodology provides a highly accurate evaluation

scheme for thermodynamic and transport properties in a variety of multiphase
mixtures. The following is a description of the corresponding-states methodology
along with details regarding the evaluation of relevant properties.
The law of corresponding states expresses the generalization that equilibrium
properties that depend on intermolecular forces are related to the critical properties
in a universal way. In 1873, van der Waals showed that this law is theoretically
valid for all pure substances whose PVT properties can be expressed in terms of
a two-constant equation of state. In 1939, Pitzer showed that this law is similarly
valid for substances which can be described by a two parameter intermolecular

potential function. The corresponding-states principle holds well for fluids con-
taining simple molecules and, on semiempirical extension, also holds for many
other substances where molecular orientation is not important. Additional correc-
tion factors are required for polar substances.
The extended corresponding-states model used in this study is based on three
assumptions: 1) the configurational properties of a single phase mixture rjm can
be equated to those of a hypothetical pure fluid, i.e.,
rjm(p, T,Xi,...,XN) = rix(p, T) (40)

2) the properties of the hypothetical pure fluid obey classical two-parameter
corresponding-states formalism
ri*(p, T) = ri0Fn(W0, Wx,*x, fx) (41)

where TJO corresponds to a reference fluid; and 3) the reference fluid density
and temperature p0 and T0 obey an extended equilibrium corresponding-states
principle given by
Po = pTix T0 = T/fx (42)

The terms Tix, f x , and Wx in Eqs. (40-^12) are, respectively, the equivalent sub-
stance volume reducing ratio, the equivalent substance temperature reducing ratio,
and molecular weight for the multicomponent mixture. The equivalent substance
volume reducing ratio accounts for the distribution of energy with respect to the
reference fluid. The temperature reducing ratio accounts for molecular size differ-
ences. Fn in Eq. (41) is a dimensional scaling factor. The functional form of these
parameters are described later.
Implementation of the corresponding-states methodology requires the selection
of a reference fluid. In this study methane is employed for two primary reasons.
First, a reliable database exists44 with sufficient data correlated for the equation of
state, relevant thermodynamic properties, and transport properties. Second, it is
similar in structure to the chemical systems of interest.
To apply the model to mixtures, analytical expressions for F^ must be specified
along with a set of mixing and combining rules for fix, fX9 and Wx, a reference
fluid equation of state, and relevant property data for the reference fluid. Following
Leland and Chappelear,45 the mixing rules employed are as follows:
N N
iXjhij (43)
'Xjfijn*J (44)
3
W / (45)
where subscript ij corresponds to binary pair parameters. Combining rules for

these terms are given by
)i-tij) (46)
/y = (/*/;)* d-*y) (47)
Wij=2WiWj/(Wi + Wj) (48)
In Eqs. (46) and (47), the quantities /,-;- and ky represent binary interaction parame-
ters that account for molecular energy and volumetric effects in the binary system.
The quantities ft, and ft are the equivalent substance reducing ratios for compound
/ in the mixture. These quantities are obtained by a two-parameter methodology
as follows:
Ki^Wcj/V^WiW^TrÛi) (49)
fi = (Tc,i/TCt0)Oi(Vrti, Trti,Q)i) (50)
Functions 0/ and 0t- are shape factors46'47 that account for nonsphericity with
respect to molecular structure. Subscript c denotes a critical value and r a reduced
value.
The functional form of the B WR equation of state is
9 15
2
P(P> ) = I>»(7V + E an(7V"-17exP{ -yp2}
r
(51)
n=l n=10
where y is an empirically fitted parameter termed the strain rate. Coefficients an

are functions of temperature and the universal gas constant Ru . These quantities
are given in Ref. 44. The SRK equation is of the form
(52)
1 — pb \ + pb
where
and
fco = 0.48 + 1.57ey - 0.176o>2
Here, a) represents the accentric factor and £la and Qt> are empirically derived
constants. When evaluated with respect to bar, mol/1, and K, these constants take
on the values of 0.42748 and 0.086640, respectively.

Temperature, K
Fig. 2 Density predictions of pure oxygen using the BWR equation of state (T < 200
K) and the SRK equation of state (T > 200 K) in comparison to experimental data
points obtained by Vargaftik.48
Equation (42) coupled with Eq. (51) for T < 200 K, Eq. (52) for T > 200 K,
and the mixing rules given by Eqs. (43-45) were used to obtain the PVT behav-
ior for the multicomponent system considered in this study. Figure 2 illustrates
the effectiveness of this combination in predicting the behavior of oxygen within
the liquid, saturated vapor, and gas-phase regimes. In this plot, density vs tem-
perature are compared with experimentally derived values obtained by Vargaftik48
over the interval 40 < T < 1000 K and for pressures of 1, 10, 50, 100, 200, and
400 atm. The average deviation between experimental and calculated densities
obtained with this procedure was less than 0.2%.
In this study, an explicit expression for the constant pressure specific heat as a
function of temperature and pressure is required. Having established an analytical
representation for real mixture PVT behavior, this property is obtained through a
two-step process. First, respective reference properties are transformed to those
for the mixture at a given pressure using the corresponding-states methodology
outlined earlier. A pressure correction is then obtained using a departure function
of the form43'49
AC, f} -R (53)
T
= J oo
This function describes the deviation of a known reference value with respect to
pressure at a given temperature and composition. It is derived by means of the
Maxwell relations.50
Viscosity JJL and thermal conductivity A are obtained using the methodologies
developed by Ely and Hanley.51'52 Equations (40-48) are employed with scaling
factors of the form
*fhx* (54)
(55)
Values of n = 1 and n = — 1, respectively, are assigned to the exponent in Eq. (45).

For mixtures of molecules of substantially different size; i.e., when the dif-
ference between two binary species approaches volumetric ratios on the order
V c ,i/ Vc,2 ~ 6, the mean density approximation given by Eq. (40) fails. This fail-
ure has been explained by the nonequilibrium molecular dynamics studies of
Evans and Hanley.53 Since most thermophysical properties are determined from
relatively short-range forces, the properties of the larger component dominate.
To correct for this effect in the prediction of mixture viscosity, Eq. (54) is used
together with an Enskog correction of the form given by Ely and Hanley54
li = ô(p0, T0)F» + A/zEnsk°s (56)

This correction is based on the exact solution of the Enskog model for a multicom-
ponent mixture of hard spheres.55 The proposed correction is shown to improve
predictions for mixtures which exhibit large size and mass differences for both
dense and dilute gas states.
In a similar manner, Ely and Hanley52 propose an expression of the form
A = I0(p0, T0)FX + A/'(r) + AA crit (p, D (57)

for thermal conductivity. The first term on the right-hand side accounts for purely
collisional and transitional effects. The second term accounts for transfer of energy
due to internal degrees of freedom. This term is modeled by means of a modified
Eucken correlation with an empirical mixing rule for polyatomic gases.43 The last
term in Eq. (57) accounts for near-critical effects. In mixtures, however, the critical
anomaly is typically diminished due to differences with respect to critical criteria
for different components. Because this effect is relatively small, it is neglected in
the current work.
The effective diffusion coefficient Vim for each species / is related to binary
diffusion coefficients Pt; through the formula given by Bird et al.56
(58)
A theory describing diffusion in binary gas mixtures at low to moderate pressures

has been well developed.43 At low pressures, these coefficients vary inversely with
pressure or density and are essentially independent of composition. At high pres-
sure, however, the product T>ijp (or T>tjp) is no longer constant, but decreases
with an increase in p (or p) and is dependent on composition. To account for these
trends, a two-step approach is adopted. First, low-pressure theory is employed to
obtain the binary mass diffusivities in a dilute gas. Calculations are performed
analytically using Chapman-Enskog theory coupled with the Lennard-Jones in-
termolecular potential energy function given by Wilke and Lee.57 A high-pressure
correction is then applied using the corresponding- states methodology proposed
by Takahashi.58 The final expression takes the form
T>ijP = (VijP}0Fv (59)

where (T>ijp)0 represents the low-pressure value, and FX> = (T>ijp)r the reduced
value. Takahashi has obtained a tabulated representation of Fp as a function of
reduced mixture temperature and pressure. These data are employed in this study.
284 J.C.OEFELEIN AND V.YANG
B. Turbulence
The works of Erlebacher et al.1'3 are employed to model the sgs Leonard terms,
cross terms, and Reynolds terms associated with Eqs. (13-15). The Leonard-
stress terms and related counterparts in the energy and species equations require
no modeling. They can be computed directly once a filter function is specified.
Cross-stress terms are modeled using scale-similarity models of the form proposed
by Bardina et al.33 and Speziale34
Tc = p(M - M) Qc = pCp(fu - ft) Sf = p(YiU - 7/fi)

These models employ coefficients of unity to ensure Galilean invariance with

respect to the overall model. The Reynolds-stress terms and respective counterparts
in the energy and species equations are modeled using a generalization of the
Smagorinsky eddy-viscosity model for compressible flows25'27-31-32 coupled with
a gradient transport model to simulate sgs energy and species transport.
The sgs Reynolds-stress tensor is separated into deviatoric and isotropic parts
as follows:
R = pu"u" =
where
RD = p(u"u" - \u~u"I)
RI = \u~u"I
The deviatoric part is modeled using the Smagorinsky model
RD = ~~^
where
Here, the dimensionless quantity CR represents the compressible Smagorinsky

constant, with A being the filter width. The isotropic part is modeled through a
turbulence-production-equals-dissipation hypothesis. This term is given by
where C/ is also a dimensionless constant.

The temperature— velocity and species— velocity correlations are modeled as
follows59:
S =-
where the constants Prt and Sct represent the turbulent Prandtl and Schmidt
numbers, respectively.
Combined, the overall sgs models employed in Eqs. (9) and (10) are
s 3 3
Q = pCp(fu - ffi) - pC p (C*/Pr,)A 2 n| Vf
Through direct numerical simulations of compressible isotropic turbulence,

Erlebacher et al.2 found that the constants CR and C/ assume the values of 0.012
and 0.0066, respectively. Thus, these values are adopted in the current work.
Values of 0.5 are employed for both Sct and Prt.
In principle, the sgs fields depend on the precise definition of the filtering oper-
ation and the parameters characterizing it. Through analysis of three-dimensional
channel flows, Piomelli et al.60 have shown that model consistency, for the combi-
nation just outlined, is achieved using a Gaussian filter in each of the three spatial
dimensions. For a given direction i = 1, 2, 3 this filter is of the form
= ( ^- exp - —^ fe - y/)
where */ corresponds to the point of interest, yi is the dummy variable, and A/

the filter width. This filter provides a global character with respect to the range of
scales that contribute to sgs behavior. Conceptually, the inclusion of these scales
yield a more accurate simulation of first-order and second-order moments. Because
of this potential benefit, the current work employs a Gaussian filter. Historically,
temporal filtering has not been employed and is thus neglected.
In dividing the turbulent flowfield into large and small scales, one presumes
that a cutoff length A/ can be sensibly chosen in each spatial direction such that
all fluctuations on a larger scale are large eddies and the remainder constitute the
small-scale fluctuations. Local grid spacing must be of the same order as the local
filter width to ensure that the grid is fine enough to resolve the filtered fields.
In practice, these widths are assumed to coincide. For flows of practical interest
the sensible choice for A, may vary significantly over the flow domain. Such a
situation demands the use of stretched grids to minimize the overall computational
burden. Thus, nonuniform filter widths are inherent in the filtering process.
Performing the filtering operation with nonuniform filter widths violates the
basic assumptions associated with the commutation laws given by Eqs. (19) and
(20). To correct this situation it is necessary to transform the physical system to
a uniform coordinate system with a spatially homogeneous filter width A. This
specific problem has been addressed by Ghosal and Moin.61 Such a transformation
coincides with the mapping procedure employed to obtain the numerically compat-
ible body-fitted coordinate system. For two-dimensional calculations, coordinates
(;c, y) are transformed to (£, 77) space through the transformations £ = %(x, y)
and r] — r)(x, y). Typically A£ = Ar? = 1. On performing this transformation,
the two-dimensional weighting function employed in this study takes the form
- r/; A) = (6/jr A)7($, »j) exp{-(6/A2)[(£ - £')2 + 0? - r,')2]}

(60)
where J(%,rj) is the Jacobian of the transformation. Care must be taken with
regard to the order of accuracy of the transformation process to maintain equivalent
accuracy between filtered and sgs quantities.
C. Spray Field and Droplet Models

Spray field behavior is modeled by solving the Lagrangian equations of motion
and transport for the life histories of a statistically significant sample of droplets.
The droplet field is assumed to be dispersed such that particle collisions are
infrequent. It is also assumed that the Kolmogorov microscale is of the same order
or larger than the largest droplets in the spray field. For such a situation, interactions
between the gas phase and spray field are dominated by instantaneous laminar
fluid dynamics. Given these assumptions, the Lagrangian equations governing
instantaneous droplet motion can be expressed as follows10:
*Xd dUd 3
- - " («-«,) (61)
dt " dt 4 " "Pddl
rlvj HDJ ^ u.
- -- ud) (62)
dt dt 4 pdd2d
where CD represents the drag coefficient and Red the droplet Reynolds number
Red = —— L[(u - udf + (v- vd)2]J * (63)

p,
In the preceding equations virtual mass, Bassett, and gravitational forces are
neglected. Quantities u = u + u" and v = v + v" represent the instantaneous ve-
locity components associated with the gas phase. Subscript d refers to liquid-phase
quantities. Unsubscripted quantities correspond to the gas phase. In all calcula-
tions, the instantaneous velocity field is simulated using stochastic techniques
similar to the SSF methodologies outlined by Faeth.4 Fluctuating quantities are
specified stochastically, with a normal distribution, at intervals coincident with
the local characteristic eddy lifetime. The motion of droplets are tracked as they
interact with a succession of turbulent eddies. A droplet is assumed to interact
with an eddy for a time taken as the smaller of either the eddy lifetime or the
transit time required to traverse the eddy.
Droplet mass and heat transfer are governed by equations of the form
= -m• d ,<-^
(64)
dt
J rri
™>dCp,d-T- = hdnd2d(T — Td) — mdAhv (65)
where md represents the vaporization rate, hd the heat transfer coefficient, and A/z,,
the enthalpy of vaporization. In contrast to the latent heat of vaporization, which
is the energy required to vaporize one mole of pure liquid in its own vapor at a
given temperature Td and saturation pressure p s at(^)» the enthalpy of vaporization
represents the energy required to vaporize one mole of the liquid component into
a gaseous mixture at a given temperature and pressure.
Equations (61-65) are evaluated using a fourth-order Runge-Kutta scheme with

semiempirical models employed to quantify the drag coefficient, vaporization
rate, and heat transfer coefficient. The drag coefficient accounts for the dynamic
influence of pressure and viscous forces acting on droplet surfaces. Putnam62 has
proposed the following simple expression for the standard drag coefficient:
_ Red < 103

D
~ 0.424 Red > 103 (66)
This correlation has proven to be quite accurate in describing the dynamic forces
acting on spherical droplets.
At subcritical pressures, the assumption of sphericity with respect to the droplet
field appears to be adequate due to the presence of surface tension. This assumption
is violated, however, as pressure approaches and exceeds the thermodynamic
critical point. Deviation from the reference drag curve given by Eq. (66) is mainly
attributed to an increase in form drag which results from increasing droplet aspect
ratio. To compensate for asymmetries associated with droplet deformation at high
pressure, Hsiao et al.6 and Hsiao7 have obtained a correction factor of the form
CD = C°DRe°-3/(l + aBT)b (67)

where CQD corresponds to Eq. (66). In contrast to the low-pressure case, where the
enthalpy of vaporization is rate controlling, thermal diffusion processes prevail
at high pressure. Thus, Eq. (67) is scaled with respect to a high-pressure transfer
number denoted BT, where
BT = (Foe - TC)/(TC ~ To) (68)

Here, Tc represents the critical mixing temperature defined as the temperature
for which both phases exist in equilibrium at a given pressure. This quantity is
calculated as a function of pressure and composition using phase equilibrium
theory and the property evaluation scheme outlined earlier. Such calculations are
based on the minimization of the Gibbs function. Through systematic analysis and
data reduction, coefficients of a = 0.05 and b — 1.592/pJ?Q2 have been deduced,
where pr represents the reduced pressure. A comparison of this correlation with
direct numerical simulations of an oxygen droplet vaporizing in a convective
hydrogen environment is given in Fig. 3. Details can be found in Refs. 6 and 7.
Major assumptions and detailed formulations associated with droplet transport
models are given by Law,63 Sirignano,64 and Faeth.4' I0 These works provide an in-
depth analysis of droplet vaporization and combustion starting with the classical
works of Ranz and Marshall66 and Spalding.67'68 As described by Faeth,10 the
following classical correlations are used to model subcritical droplet dynamics
= 2irpsVsmdd k(l + BM) (69)
B M ) L e ~ l / [ ( l + BM)Le~l - 1] (70)
Quantities Le = ^.s/ PsCptS'Dsm represent the Lewis number and BM the Spalding
transfer number; i.e.,
n,o2) (71)
100
25 50 75 100
24(1 + 1/6Re2/3)/Re°'7
Fig. 3 Comparison of the high-pressure drag correlation given by Eq. (67) with data
obtained through direct numerical simulations; from Hsiao et al.6 and Hsiao7; used
with permission.
Subscript s refers to the vapor mixture at the droplet surface. The 1/3 rule is
employed to evaluate these quantities,65 with YXio2 evaluated by means of the
relation
where
To account for convective effects at subcritical pressures, the multiplicative

correction factors proposed by Faeth and Lazar69 are employed. These factors take
the form
md
= 1+ (72)
hd
(73)
where Sc == ^s/ps^sm and Pr = fjisCptS/Xs are Schmidt and Lewis numbers,

respectively.
At pressures above the critical point, droplets, initially in a subcritical state,
are exposed to a supercritical environment. As liquid surfaces approach the crit-
ical mixing condition, the difference between gas and liquid densities becomes
smaller, and characteristic times associated with transport processes in both phases
approach the same order of magnitude. Concurrently, the enthalpy of vaporization
and surface tension approach zero. These conditions markedly alter droplet vapor-
ization, deformation, and breakup characteristics. For the situation just described,
droplet regression rates proceed at such high rates that the change in droplet tem-
perature, as described by Eq. (65), can be neglected. Thus, only Eqs. (61-64) are
evaluated for the high-pressure cases.

1.2,—————,————......——......
4 6 10
Droplet Lifetime, ms
Fig. 4 Droplet diameter vs lifetime for a spherical oxygen droplet vaporizing in a
supercritical, nonconvective, hydrogen environment; from Lafon et al.5; used with
permission.
To account for high-pressure effects, correlations recently obtained by Lafon

et al.,5 Hsiao et al.,6 and Hsiao7 were employed. High-pressure vaporization rates
were modeled by means of two correlations. The first quantifies the lifetime asso-
ciated with spherical oxygen droplets vaporizing in a supercritical, nonconvective,
hydrogen environment. The second provides a correction factor to account for
deformation and breakup processes imposed by convective effects. These correla-
tions use the critical-mixing temperature to define a reference interface between
the gas and liquid phases.
Figure 4, which is taken from Ref. 5, shows representative curves of droplet
diameter vs lifetime for a spherical oxygen droplet vaporizing in a supercritical,
nonconvective, hydrogen environment. These data were obtained by means of
direct numerical simulations at the conditions indicated in the figure. Through
extensive analysis over a wide range of ambient temperature and pressure, two
rate-limiting parameters were identified with respect to the droplet lifetime. The
first was the transfer number given by Eq. (68). This parameter quantifies the
characteristic time associated with thermal diffusion processes. The second is
the ratio of thermal diffusivities a.^/oid, where subscript oo represents the gas
phase mixture and d the droplet. This parameter quantifies the deviation in the
characteristic time due to variations in mixture diffusivity. The final correlation,
as presented in Ref. 5, is given in Fig. 5. Linear least-squares data reduction yield
the following equation:
,/^=o = \(dl/ad}f (aooM*) [0.0115 + 0.542(1 - 7C*)] (74)
where
= 1 + 3.9 fl - exp j-0.035 (— - l}\]

L I \«d /JJ
Here, d0 represents the initial droplet diameter. Quantity 7C* is related to Eq. (68)

0.20
H2-O2 System
70.15 o Supercritical
Near-Critical
J
cfO.10
s°
0.05
0.85 0,90 0.95 1.00
Fig. 5 Dimensionless corrected droplet lifetime associated with a spherical oxygen

droplet vaporizing in a supercritical, nonconvective, hydrogen environment; from
Lafon et al.5; used with permission.
by means of the following relation

7C* = BT/(l + BT)
The least square error associated with Eq. (74) is 5.2%.
To account for deformation and breakup processes which arise in the presence
of a convective stream, Hsiao et al.,6 and Hsiao7 have obtained a correction factor
based on the droplet Reynolds number Red and reduced pressure p/pc- This
correlation is given as
1-1
(75)
A comparison of this correlation with direct numerical simulations is given in
Fig. 6. Details can be found in Refs. 6 and 7.
Modeling droplet vaporization rates using Eqs. (74) and (75) requires a priori
knowledge of the path dependency associated with the vaporization process. To
illustrate this point, the path dependency associated with a liquid oxygen droplet
vaporizing in a supercritical, convective, hydrogen environment at 400 atm is given
in Fig. 7. These data were acquired by means of direct numerical simulations. Here,
data is plotted in terms of an equivalent spherical diameter vs the reduced droplet
lifetime and droplet Reynolds number. At a Reynolds number of zero, a nonlinear
power-law dependence of the form
(76)
exists where ft represents a proportionality constant and n a real number. As the

Reynolds number increases to approximately 20, this dependence approaches a
somewhat linear condition where ft = 1 and n = 0. For this situation, droplet
vaporization rates are essentially constant with time and fully characterized by
the droplet lifetime and respective initial conditions. Beyond a value of 20, non-
linearities again begin to predominate with a functional dependence close to, but
1.21 . . . . . . . , . . • , • • • , . . . . . .
n p - 100 atm
1.0 A 200
O 400
0.8
O2~ 0.4 0.6 0.8 1.0 1.2

M+o.ieeRe/V0'88)"1
Fig. 6 Convective correction vs Reynolds number and pressure for the lifetime of
an oxygen droplet vaporizing in a supercritical, hydrogen environment; from Hsiao
et al.6 and Hsiao7; used with permission.
not exactly, the form given by Eq. (76). The data presented represent the most
extreme conditions typically encountered and, in general, the Reynolds number
effect on the path dependency is small. Thus, in all cases considered a linear path
dependence is assumed with respect to the vaporization process. This condition is
denoted by the dashed line in Fig. 7.
VI. Numerical Implementation and Computational Domain

The governing system just outlined is evaluated by means of an optimal numer-
ical framework that takes full account of variable properties over the entire range
).0 0.2 0.4 0.6 0.8 1.2

Reduced Droplet Lifetime, t/T,jfe
Fig. 7 Path dependency associated with a liquid oxygen droplet vaporizing in a
supercritical, convective environment at 400 atm; from Hsiao et al.6 and Hsiao7; used
with permission.
of thermodynamic regimes. Eulerian gas-phase dynamics are treated using vec-

torized, finite-volume, alternating-direction-implicit (ADI) factorization70-71 with
viscous preconditioning72 and the convergence acceleration techniques developed
by Buelow et al.73 and Venkateswaran and Merkle.74 Temporal discretization is
obtained using second-order dual-time stepping integration. Spatial discretiza-
tion is obtained using the fourth-order flux differencing methodologies outlined
by Rai and Chakravarthy.75 The physical time step employed for numerical in-
tegration of the gas-phase system rg is selected such that, on the average, local
Courant-Friedrichs-Lewy (CFL) numbers of order one are obtained with respect
to time- accurate flow quantities. This criteria dictated a time step of rg = 1 /zs for
all calculations presented hereafter.
The Lagrangian liquid-phase system is handled using a fourth-order Runge-
Kutta scheme. The time step associated with this system is selected as
TR-K = I0~l{rr, r/, rc, r,, rg]

Here rr represents the characteristic droplet relaxation time, r/ the droplet lifetime,
rc the time associated with droplet heating, and rt the time associated with the
local sgs velocity fluctuation. Using the solution from the preceding step as initial
conditions, the Lagrangian system is advanced over an interval corresponding to rg
for each pseudotime iteration performed with respect to the gas phase. Performing
the integration in this manner insures that both systems converge simultaneously in
a fully coupled manner. Combined, the theoretical/numerical framework offers a
tractable means of assessing and analyzing detailed interactions between processes
in a manner which makes optimal use of current computational capacities. A
complete description of the algorithm employed is given by Oefelein.9
The computational domain, as depicted in Fig. 1 , is composed of two coflowing
hydrogen streams, initially at 1000 K, with a dilute 100 K oxygen spray injected
at the trailing edge of a splitter plate. The domain length is 200 mm and width
is 50 mm. The spray is injected at pressures of 1, 100, and 400 atm such that
an overall oxygen to hydrogen mass flow ratio of 0.5 defined with respect to the
shear layer thickness. In the present investigation, the initial velocity distribution
is defined by a hyperbolic tangent profile of the form
In all cases, Ul = 200 m/s (M} = 0.084) and U2 = 50 m/s (M2 = 0.021), with
droplets injected at the local gas-phase velocity. The initial shear layer thickness
is specified to be 3 = 1 mm. To provide consistency between cases, the initial
profile is perturbed at the fundamental and first subharmonic modes with respect
to shear layer growth at an rms amplitude of 0.25%. The characteristic frequencies
associated with these modes are 8658 Hz and 4329 Hz, respectively.
Free-slip boundary conditions are specified at the upper and lower boundaries.
The method of characteristics is applied at the inlet and exit. In this study, the
entire domain is subsonic. Thus, there is one outgoing characteristic associated
with the inlet, and four outgoing characteristics associated with the exit boundary.
To supplement these conditions, inflow mass fluxes, temperature, and composition
are specified at the inlet along with a far-field reference pressure at the exit. A
buffer zone is attached to the exit to provide adequate damping of pressure waves
in the far field. This technique effectively eliminates reflections associated with
the far-field boundary condition, producing a realistic pressure distribution at the
exit boundary.
Because of the large number of droplets present in an actual spray, a sampling
technique is employed whereby characteristic groups of droplets are represented
by computational parcels. The spray source terms given by Eq. (39) and Table 2
are evaluated by means of a droplet distribution function of the form
/( W )=— (78)
k=l
Here tidtk represents the number of identical droplets in the &th parcel, with sitk
denoting the /th property associated with this parcel, and P the total number of
parcels. Thus, the initial conditions associated with the Lagrangian system involve
the specification of initial parcel positions, velocity, mass, temperature, and the
number of droplets represented by the parcel.
The quantity nd^ is determined at the inlet boundary through joint specification
of a number distribution function and the total particle mass Md. These quantities
are related by the relation
(79)
In this study, an upper limit distribution of the form depicted in Fig. 8 is employed.
The Sauter mean diameter associated with this distribution, D32, is 100 /xm. At
each physical time step injected parcels are assigned randomly selected diameters
over the interval 0 < ddtk < 200 /xm in distributions proportional to the trends
dictated by Fig. 8.
The grid used in all calculations is given in Fig. 9. The overall mesh size is
120 x 75. An optimal combination of stretching is employed to minimize the
overall grid density required to resolve predominate processes. This combination
14 100
12
80 *
10 0
8 60 -§
E 6 40
I 03
"3
Q
4 I
20 d
2
V
0 50 100 150 200 258
Drop Diameter, urn
Fig. 8 Number distribution and cumulative volume of droplets injected at the inlet
boundary.
25 50 75 100 125 150 175 200

x, mm
Fig. 9 Grid distribution used for the simulation of two-dimensional mixing layer
dynamics.
was arrived at in two stages. First, linear stability analysis was employed using
the Orr-Sommerfeld equation to obtain analytical predictions of the fundamental
modes associated with mixing layer growth. The axial and transverse grid spacing
required to match predicted values were then obtained to insure that spatial growth
and dispersion characteristics in the linear regime were resolved accurately. Using
the preceding spatially fourth-order accurate scheme outlined, an axial distribution
of 20 cells per wavelength and transverse distribution of 8 cells per shear layer
thickness were found to fall within 2% of analytical predictions. Upon establishing
this criteria, parametric studies were conducted at 1,100, and 400 atm to determine
the optimal combination for which grid-independent solutions were obtained at
each of the pressures of interest. Figure 9 represents the configuration that produces
grid-independent solutions for the entire range of pressures considered.
VII. Results and Discussion

In this section, a sequence of representative calculations are presented which
highlight predominate processes associated with high-pressure spray field dynam-
ics in turbulent mixing layers. First the baseline flow characteristics at 1, 100, and
400 atm are analyzed with no spray injected to establish a point of reference with
respect to the gas-gas dynamics. On establishing these trends, the spray field is
introduced and a similar analysis is conducted to assess various multiphase inter-
actions. Three cases are considered. The first is a simulation at 1 atm with classical
low-pressure correlations employed to model droplet vaporization dynamics. This
provides a point of reference with respect to established theories at low pressure.
The latter two cases are conducted at 100 and 400 atm, respectively, to assess
the impact of pressure on the fully coupled system. The cases selected represent
the first logical step toward modeling high-pressure, multiphase, combustion pro-
cesses and are presented here only do demonstrate key concepts and predominate
physical phenomena. More detailed analyses which focus on turbulent interac-
tions and predominate reaction mechanisms at high pressure are currently being
conducted by Oefelein.9
A. Analysis of Gas-Gas Dynamics

Key characteristics associated with the hydrogen stream with no spray injected
are given in Figs. 10-12. Here, the instantaneous pressure, vorticity, velocity,
and normalized turbulent viscosity fields are compared at 1, 100, and 400 atm to
establish the baseline trends. The pressure and vorticity fields are given in Fig. 10.
The corresponding velocity fields are given in Fig. 11. When normalized with
respect to the ambient pressure, /?ref = 1, 100, and 400 atm, and the centerline
Pressure, -0.2336 <> (p-pref)/Pref * 0.1415%, A = 0.02%
Vorticity, 0 < co/coref z 100%, A = 10%, coref = -74x103 s"1

25 50 75 100 125 150 175 200

x, mm
Fig. 10 Pressure and vorticity fields associated with the hydrogen mixing layer at 1,
100, and 400 atm with no spray injected.
vorticity at the inlet, a>ref = —74000 s"1, trends exhibited by respective fields are
essentially independent of pressure. Peak-to-peak differences of approximately
2% are exhibited with respect to the normalized pressure fields. Vorticity, initially
concentrated in the basic velocity profile, is constantly redistributed into larger
and larger vortices, with wavelength doubled and thus frequency halved after
each interaction. Two such interactions occur within the 200-mm domain length
employed in this study. In each case, the vorticity diffuses such that the relative
core strength is decreased to approximately 60% of the initial reference value on
reaching an axial location of approximately 160 mm downstream from the inlet.
u-Velocity, -20 ^ u ^ 250 m/s, A * 10 m/s
v-Velocity, -75 ^ v ^ 75 m/s, A = 10 m/s
Velocity Vectors
E 25
"0 25 50 75 100 125 150 175 200

x.mm
Fig. 11 Velocity fields associated with the hydrogen mixing layer at 1,100, and 400
atm with no spray injected.

Viscosity Ratio
p - 1 atm, 0 £ v/v, £ 0.2704
p = 100 atm, 0 £ v/v, <, 37.80, A - 5

p - 400 atm, 0 < V/V, < 156.3, A = 5

50
§25
25 50 75 100 125 150 175 200

x, mm
Fig. 12 Normalized turbulent viscosity fields associated with the hydrogen mixing
layer at 1,100, and 400 atm with no spray injected.
At each of the pressures considered, extrema associated with the axial velocity
component are —20 and 250 m/s. Those associated with the transverse velocity
component are —75 and 75 m/s.
The normalized turbulent viscosity fields associated with each case are given in
Fig. 12. This quantity is given by the ratio vt/v/, where vt represents the turbulent
eddy viscosity and v/ the laminar kinematic viscosity. This ratio is directly propor-
tional to the sgs Reynolds number and characterizes the predominance of turbulent
vs laminar diffusion of momentum. In contrast to the trends exhibited earlier, a
strong pressure dependence is exhibited with respect to this quantity. This depen-
dence is attributed to the marked decrease in the laminar kinematic viscosity which
occurs with increasing pressure. At 1 atm, this ratio ranges from 0 in the freestream
to a maximum of only 0.2704. Thus, laminar diffusion processes dominate for this
case and the contours that bracket maximum values are shown to highlight regions
where high shear stresses exist. In contrast to the case at 1 atm, ranges of 0-37.80
and 0-156.3 are observed at 100 and 400 atm, respectively. Thus, turbulent diffu-
sion processes dominate at these pressures, and increase with increasing pressure.
The corresponding turbulence intensities, qsgs/AU where qsgs = A/(2/3)A:SgS and
A(7 = U\ — C/2, however, are relatively insensitive to pressure, with ranges of
0-1.012, 0-1.163, and 0-1.169% exhibited, respectively.
B. Analysis of Multiphase Dynamics

The effects of multiphase interactions on overall system dynamics at 1, 100,
and 400 atm, respectively, are presented in Figs. 13-15. In all cases considered,
little change was observed with respect to the normalized peak-to-peak pressure
fluctuations. Changes were observed, however, with respect to the velocity and
Spray Distribution, 2583 Parcels, 6.3x104 Particles
Mass Fraction, 0.9764 <> YH2 < 1

Viscosity Ratio, 0 < v/v, < 0.3000
Vorticity, 0 < to/o^ ^ 100%, A - 10%, coref--74x103 s'1

50
§25
25 50 75 100 125 150 175 200

x, mm
Fig. 13 Spray distribution, mass fraction, turbulent viscosity, and vorticity fields
associated with the hydrogen-oxygen system at 1 atm.
turbulent viscosity profiles. Thus, in each figure, plots of the instantaneous spray
field distributions, gas-phase mass fraction, normalized turbulent viscosity ratios,
and normalized vorticity fields are given for comparison.
The instantaneous spray field distributions are represented by symbols that indi-
cate respective parcel locations. These symbols represent the equivalent diameter
of each parcel and are plotted to scale with respect to the grid axis. The number
of parcels present in the flow domain is indicated at the top of each plot along
with the total number of droplets that the parcels represent. At 1 atm, spray field
evolution is only slightly influenced by gas-phase fluctuations. Within the 200-mm
domain, trajectories follow an essentially straight path along the centerline, with
only slight dispersion exhibited by the smallest droplets in the flow. At 100 atm,
the situation is markedly different. For this case, droplet mixing and dispersion
is dominated by the prevalent large-scale structures. Significant coupling is ex-
hibited at a location approximately 60 mm downstream of the inlet, with most of
the droplets completely vaporized at a downstream location of 150 mm. Similar
trends are exhibited at 400 atm. For this case, however, the trends are more pro-
nounced, with significant coupling beginning approximately 45 mm downstream
and most of the droplets vaporized on reaching a location approximately 80 mm
downstream. The average droplet Reynolds number associated with the entire field
at the given instant in time are 1.804, 90.39, and 158.3, respectively.
Mass Fraction, 0.8893 ^ YH2 ^ 1, A * 0.01

Viscosity Ratio, 0 <> v/v, < 33.83, A - 5
Vorticity, 0 < a>/a>ref <, 100%, A - 10%, a>ref = -74x103 s"1

50
§26
25 50 75 100 125 150 175 200

x.mm
The instantaneous mass-fraction contours illustrate how the vaporized oxygen

striates and evolves within the large-scale structure. These striations produce
local changes in mixture composition that in turn alter the large-scale attributes
and small-scale dynamics of the flow. At 1 atm, an almost negligible amount of
mass is exchanged, with a maximum of yO2 = 2.360% accumulating at 160 mm
downstream of the inlet. At 100 and 400 atm, maximums of 11.07 and 22.02%
are observed at the same approximate location. In all cases, local maxima are
coincident with regions of maximum vorticity.
The introduction of the spray field into the gas-phase system alters the nor-
malized turbulent viscosity ratio relative to the values obtained with no spray.
At equivalent instants in time, values of 0.3000, 33.83, and 124.2 were obtained
yielding deviations of 10.95, —10.50, and —20.54% with respect to the baseline
cases. In a similar manner, values of 45, 50, and 55% of the initial reference value
were exhibited with respect to the vorticity fields. This is in contrast to a value of
60% that was exhibited in the absence of the spray. Thus, deviations of —25.00,
— 16.67, and —8.333 in relative core strength occurred at the 160-mm location.
Corresponding extrema associated with the velocity field intensified and were
again relatively insensitive to pressure. Here, extrema of —30 and 250 m/s where
exhibited with respect to the axial component and —90 and 90 ms exhibited with
respect to the transverse component.
Mass Fraction, 0.7798 < YH2 < 1, A = 0.01

Viscosity Ratio, 0 < v/v, < 124.2, A = 5
Vorticity, 0 < a>/corBf < 100%, A = 10%, a>ref * -74x103 s'1

50
I 25
25 50 75 100 125 150 175 200

x, mm
VIII. Conclusions
Fundamentals associated with the simulation of high-pressure spray combustion
processes for the general case of a turbulent, reacting, multiphase flow have
been presented through the development of a generalized theoretical framework.
This framework accommodates unsteady flow dynamics over a wide range of
thermody namic and spray field regimes in a numerically compatible manner. Using
this framework, representative cases associated with high-pressure spray field
dynamics in turbulent, two-dimensional, mixing layers have been systematically
analyzed. The basic flow consisted of two coflowing hydrogen streams with a
dilute oxygen spray injected at the trailing edge of a splitter plate. Pressures of
100 and 400 atm have been considered along with the 1 atm case that was included
to provide a point of reference with respect to classical low-pressure models.
Results have highlighted the dramatic effect of pressure on fully coupled spray
field dynamics. In contrast to the low-pressure case, high-pressure mixing and
dispersion characteristics are much more sensitive to both sgs fluctuations and the
large-scale coherency associated with the gas phase. Gas-phase dynamics are also
markedly affected by the spray field as a result of both transport between phases
and induced variations in mixture properties. Such changes alter the evolutionary
structure of the flowfield. Significant gas-liquid coupling is attributed to two basic
mechanisms: 1) pressure-induced enhancement of turbulent diffusion processes
and 2) mixture-induced variations of gas-phase structures.
Turbulent diffusion processes increase significantly at high pressures. Over the

ranges considered, the change in density that accompanies increasing pressure
causes the molecular kinematic viscosity to decrease significantly, producing an
overall increase in local Reynolds number. This effect produced a three-order-of-
magnitude increase in the normalized turbulent viscosity ratio, with peak values
ranging from 0.3000 for the 1 atm case to 124.2 at 400 atm. Such effects amplify
the dispersion attributes associated with the spray field.
In each of the high-pressure cases considered the motion of large-scale dy-
namics dominated the mixing and dispersion of droplets. This behavior produced
striations of vaporized oxygen in the flowfield, with corresponding variations in

thermophysical properties. The effect of these striations on turbulent diffusion
processes is dependent on the local values of density and dynamic viscosity in
the mixture. At the instant in time shown, the cases presented exhibited an overall
decrease with respect to turbulent diffusion processes. At other instants in time,
however, the opposite trend has been observed. Regardless of the trend, striations
associated with the mixture can produce significant changes in the coherent flow
structures and thus markedly alter the time-averaged history of the prevailing
dynamics.
Acknowledgments
This work was sponsored under the NASA Marshall Space Flight Center Grad-
uate Student Researchers Program, Grant NGT-50953. The support and encour-
agement of Charles F. Schafer is greatly appreciated.
References
^rlebacher, G., Hussaini, M. Y., Speciale, C. G., and Zang, T. A., 'Toward the Large
Eddy Simulations of Compressible Tiirbulent Flows," Inst. for Computer Applications in
Science and Engineering, ICASE 87-20, NASA Langley Research Center, Hampton, VA,
1987.
2
Erlebacher, G., Hussaini, M. Y., Speziale, C. G., and Zang, T. A., "Toward the Large
Eddy Simulation of Compressible Turbulent Flows," Inst. for Computer Applications in
Science and Engineering, ICASE 90-76, 1990.
3
Erlebacher, G., Hussaini, M. Y, Speziale, C. G., and Zang, T. A., "Toward the Large
Eddy Simulation of Compressible Turbulent Flows," Journal of Fluid Mechanics, Vol. 238,
1992, pp. 155-185.
4
Faeth, G. M., "Mixing, Transport, and Combustion in Sprays," Progress in Energy and
5
Lafon, P., Yang, V, and Habiballah, M., "Supercritical Vaporization of Liquid Oxygen
Droplet in Hydrogen and Water Environments," unpublished.
6
Hsiao, C. C., Yang, V, and Shuen, J. S., "Vaporization of Liquid Oxygen Droplets in
a Supercritical Hydrogen Stream," unpublished.
7
Hsiao, C. C., "Droplet Vaporization and Combustion in Quiescent/Forced-Convective
Environments," Department of Mechanical Engineering, Ph.D. Thesis, Pennsylvania State
Univ., University Park, PA, 1995.
8
Girimaji, S. S., "Assumed /?-pdf Model for Turbulent Mixing: Validation and Extension
to Multiple Scalar Mixing," Combustion Science and Technology, Vol. 78, 1991, pp. 177-
196.
9
Oefelein, J. C, "Simulation and Analysis of High-Pressure Spray Combustion Pro-
cesses in Turbulent Mixing Layers," Department of Mechanical Engineering, Ph.D. Thesis,
Pennsylvania State Univ., University Park, PA, 1996.
10
Faeth, G. M., "Evaporation and Combustion of Sprays," Progress in Energy and
H
Dukowicz, J. K., "A Particle-Fluid Numerical Model for Liquid Sprays," Journal of
Computational Physics, Vol. 35, 1980, pp. 229-253.
12
Gosman, A. D., and loannides, E., "Aspects of Computer Simulation of Liquid-Fueled
Combustors," AIAA Paper 81-0323, 1981.
13
Shuen, J. S., "A Theoretical and Experimental Investigation of Turbulent Sprays,"
Department of Mechanical Engineering, Ph.D. Thesis, Pennsylvania State Univ., University
Park, PA, 1984.
14
Liang, P. Y., and Chan, D. C., "Development of a Robust Pressure Based Numerical
Scheme for Spray Combustion Applications," AIAA Paper 93-0902, 1993.
15
Shuen, J. S., Chen, K. H., and Choi, Y, "A Time-Accurate Algorithm for Chemical
Non-Equilibrium Viscous Rows at All Speeds," AIAA Paper 92-3639, 1992.
16
Chen, K. H., and Shuen, J. S., "A Coupled Multi-Block Solution Procedure for Spray
Combustion in Complex Geometries," AIAA Paper 93-0108, 1993.
17
Jones, W. P., and Launder, B. E., "The Prediction of Laminarization with a Two-
Equation Model of Turbulence," International Journal of Heat and Mass Transfer, Vol. 15,
1972, pp. 301-314.
l8
Mostafa, A. A., and Mongia, H. C., "On the Interaction of Particles and Turbulent
Fluid Flow," International Journal of Heat and Mass Transfer, Vol. 31, No. 10, 1988, pp.
2063-2073.
19
Amsden, A. A., O'Rourke, P. J., and Butler, T. D., "KIVAII: A Computer Program for
Chemically Reactive Flows with Sprays," Los Alamos National Lab., LA-11560-MS, Los
Alamos, NM, 1989.
20
Fashola, A., and Chen, C. P., "Modeling of Confined Turbulent Fluid-Particle Flows
Using Eulerian and Lagrangian Schemes," International Journal of Heat and Mass Transfer,
Vol.33, 1990, pp. 691-700.
21
Fox, D. G., and Lilly, D. K., "Numerical Simulation of Turbulent Flows," Review of
Geophysics and Space Physics, Vol. 10, 1972, pp. 51-72.
22
Schumann, U., "Subgrid Scale Model for Finite Difference Simulations of Turbulent
Flows in Plane Channels and Annuli," Journal of Computational Physics, Vol. 18, 1975,
pp. 376-404.
23
Voke, P. R., and Collins, M. W., "Large-Eddy Simulation: Retrospect and Prospect,"
Physico Chemical Hydrodynamics, Vol. 4, 1983, pp. 119-161.
24
Rogallo, R. S., and Moin, P., "Numerical Simulation of Turbulent Flows," Annual
Review of Fluid Mechanics, Vol. 16, 1984, pp. 99-137.
25
Deardorff, J. W., "A Numerical Study of Three-Dimensional Turbulent Channel Flow
at Large Reynolds Numbers," Journal of Fluid Mechanics, Vol. 41, 1970, pp. 453-480.
26
Lilly, D. K., "On the Application of the Eddy Viscosity Concept in the Inertial Subrange
of Turbulence," National Center for Atmospheric Research, NCAR-123, Boulder, CO,
1966.
27
Leonard, A., "Energy Cascade in Large-Eddy Simulations of Turbulent Fluid Rows,"
Advances in Geophysics, Vol. 18, 1974, pp. 237-248.
28
Favre, A., "Equations des gaz turbulents compressibles," Journal de Mecanique, Vol.
4, No. 3, 1965, pp. 361-390.
29
Reynolds, O., "On the Dynamical Theory of Incompressible Viscous Fluids and the
Determination of the Criterion," Philosophical Transactions of the Royal Society of London,
Series A, Vol. 186, 1894, pp. 123-164.
30
Monin, A. S., and Yaglom, A. M., Statistical Fluid Mechanics, MIT Press, Cambridge,
MA, 1971.
31
Smagorinsky, J., "General Circulation Experiments with the Primitive Equations. I.
The Basic Experiment," Monthly Weather Review, Vol. 91, 1963, pp. 99-164.
32
Reynolds, W. C., "Computation of Turbulent Flows," Annual Review of Fluid Mechan-
ics, Vol. 8, 1976, pp. 183-208.
33
Bardina, J., Ferziger, J. H., and Reynolds, W. C., "Improved Subgrid Scale Models
Based on Large Eddy Simulation of Homogeneous, Incompressible, Turbulent Flows,"
Stanford Univ., TF-19, Stanford, CA, 1983.
34
Speziale, C. G., "Galilean Invariance of Subgrid-Scale Stress Models in the Large
Eddy Simulation of Turbulence," Journal of Fluid Mechanics, Vol. 156, 1985, pp. 52-62.
35
Gough, P. S., "The Flow of a Compressible Gas Through an Aggregate of Mobile,
Reacting Particles," Ph.D. Thesis, McGill Univ., Montreal, Quebec, Canada, 1974.
36
Gough, P. S., and Zwarts, F. J., "Modeling Heterogeneous Two- Phase Reacting Flows,"
AlAA Journal, Vol. 17, No. 1, 1979, pp. 17-25.
37
Anderson, T. B., and Jackson, T, "A Fluid Mechanical Description of Fluidized Beds,"
Industrial and Engineering Chemistry Fundamentals, Vol. 6, 1967, pp. 527-539.
38
Whitaker, S., "The Transport Equations for Multi-Phase Systems," Chemistry and
Engineering Science, Vol. 28, 1973, pp. 139-147.
39
O'Rourke, P. J., "Collective Drop Effects on Vaporizing Liquid Sprays," Department
of Mechanical and Aerospace Engineering, Ph.D. Thesis, Princeton Univ., Princeton, NJ,
1981.
40
Williams, F. A., "The Transport Equations for Multi-Phase Systems," Physics of Fluids,
Vol. 1, No. 6, 1958, pp. 541-545.
41
Rowlinson, J. S., and Watson, I. D., "The Prediction of the Thermodynamic Properties
of Fluids and Fluid Mixtures -1. The Principle of Corresponding States and its Extensions,"
Chemistry and Engineering Science, Vol. 24, No. 8, 1969, pp. 1565-1574.
42
Jacobsen, R. T, and Stewart, R. B., "Thermodynamic Properties of Nitrogen Including
Liquid and Vapor Phases from 63K to 2000K with Pressure to 10,000 bar," Journal of
Physical and Chemical Reference Data, Vol. 2, No. 4, 1973, pp. 757-922.
43
Reid, R. C., Prausnitz, J. M., and Poling, B. E., The Properties of Gases and Liquids,
4th ed., McGraw-Hill, New York, 1987.
44
Ely, J. F, and Huber, M. L., "NIST Thermophysical Properties of Hydrocarbon Mix-
tures," National Inst. of Standards and Technology, Gaithersburg, MD, July 1992.
45
Leland, T. W., and Chappelear, P. S., "The Corresponding States Principle. A Review
of Current Theory and Practice," Industrial and Engineering Chemistry Fundamentals, Vol.
60, 1968, pp. 15-43.
46
Leach, J. W., Chappelear, P. S., and Leland, T. W., "Use of Molecular Shape Factors in
Vapor-Liquid Equilibrium Calculations with the Corresponding States Principle," American
Institute of Chemical Engineers, Vol. 14, 1968, pp. 568-576.
47
Fisher, G. D., and Leland, T. W., "Corresponding States Principle Using Shape Factors,"
Industrial and Engineering Chemistry Fundamentals, Vol. 9, No. 4, 1970, pp. 537-544.
48
Vargaftik, N. B., Tables on the Thermophysical Properties of Liquids and Gases, 2nd
ed., Wiley, New York, 1975.
49
Modell, M., and Reid, R. C., Thermodynamics and Its Applications in Chemical Engi-
neering, 2nd ed., Prentice-Hall, Englewood Cliffs, NJ, 1983.
50
VanWylen, G. J., and Sonntag, R. E., Fundamentals of Classical Thermodynamics,
Wiley, New York, 1986.
51
Ely, J. F, and Hanley, H. J. M, "Predictions of Transport Properties. 1. Viscosity of
Fluids and Mixtures," Industrial and Engineering Chemistry Fundamentals, Vol. 20, No. 4,
1981, pp. 323-332.
52
Ely, J. F, and Hanley, H. J. M., "Predictions of Transport Properties. 2. Thermal
Conductivity of Pure Fluids and Mixtures," Industrial and Engineering Chemistry Funda-
mentals, Vol. 22, No. 4, 1981, pp. 90-97.
53
Evans, D. J., and Hanley, H. J. M., "Viscosity of a Mixture of Soft Spheres," Physics
Reviews A, Vol. 20, No. 4, 1979, pp. 1648-1654.

54
Ely, J. F, and Hanley, H. J. M., "An Enskog Correction for Size and Mass Difference
Effects in Mixture Viscosity Predictions," Journal of Research of the National Bureau of
Standards, Vol. 86, No. 6, 1981, pp. 597-604.
55
Tham, M. K., and Gubbins, K. E., "Kinetic Theory of Multicomponent Dense Fluid
Mixtures of Rigid Spheres," Journal of Chemical Physics, Vol. 55, No. 1, 1971, pp. 1-19.
56
Bird, R. B., Stewart, W. E., and Lightfoot, E. N., Transport Phenomena, Wiley, New
York, NY, 1960.
57
Wilke, C. R., and Lee, C. Y., "Estimation of Diffusion Coefficients for Gases and
Vapors," Industrial Engineering Chemistry, Vol. 47, 1955, pp. 1253-1257.
58
Takahashi, S., "Preparation of a Generalized Chart for the Diffusion Coefficients of
Gases at High Pressures," Journal of Chemical Engineering (Japan), Vol. 7, No. 6, 1974,
pp. 417-420.
59
Eidson, T. M., "Numerical Simulation of the Turbulent Rayleigh-Benard Problem using
Numerical Subgrid Modeling," Journal of Fluid Mechanics, Vol. 158, 1985, pp. 245-268.
60
Piomelli, U., Moin, P., and Ferziger, J. H., "Model Consistency in Large Eddy Simula-
tion of Turbulent Channel Rows," Physics of Fluids, Vol. 31, No. 7, 1988, pp. 1884-1891.
61
Ghosal, S., and Moin, P., "The Basic Equations for the Large Eddy Simulation of
Turbulent Flows in Complex Geometry," Journal of Computational Physics, Vol. 118,
1995, pp. 24-37.
62
Putnam, A., "Integratable Form of Droplet Drag Coefficient," ARS Journal, Vol. 31,
1961, pp. 1467-1468.
63
Energy and Combustion Science, Vol. 8, 1982, pp. 171-201.
64
Sirignano, W. A., "Fuel Droplet Vaporization and Spray Combustion," Progress in
Energy and Combustion Science, Vol. 9, 1983, pp. 291-322.
65
Faeth, G. M., "Current Status of Droplet and Liquid Combustion," Progress in Energy
and Combustion Science, Vol. 3, 1977, pp. 191-224.
66
Ranz, W. E., and Marshall, W. R., "Evaporation From Drops," Chemical Engineering
Progress, Vol. 48, No. 3, 1952, pp. 141-173.
67
Spalding, D. B., Some Fundamentals of Combustion, Butterworths, London, 1955.
68
Spalding, D. B., "Theory of Particle Combustion at High Pressure," ARS Journal, Vol.
29, 1959, pp. 828-835.
69
Faeth, G. M., and Lazar, R. S., "Fuel Droplet Burning Rates in a Combustion Gas
Environment," AIAA Journal, Vol. 9, No. 11, 1971, pp. 2165-2171.
70
Douglas, J., and Gunn, J. E., "A General Formulation of Alternating Direction
Method—Part I—Parabolic and Hyperbolic Problem," Numerische Mathematik, Vol. 82,
1964, pp. 428-453.
71
Briley, W. R., and McDonald, H., "Solution of the Multidimensional Compressible
Navier-Stokes Equations by a Generalized Implicit Method," Journal of Computational
Physics, Vol. 24, 1977, pp. 372-397.
72
Choi, Y. H., and Merkle, C. L., "The Application of Preconditioning in Viscous Flows,"
Journal of Computational Physics, Vol. 105, 1993, pp. 207-223.
73
Buelow, P. E. O., Venkateswaran, S., and Merkle, C. L., "Effect of Grid Aspect Ratio
on Convergence,'MMA Journal, Vol. 32, No. 12, 1994, pp. 2401-2408.
74
Venkateswaran, S., and Merkle, C. L., "Dual Time Stepping and Preconditioning for
Unsteady Computations," AIAA Paper 95-0078, 1995.
75
Rai, M. M., and Chakravarthy, S., "Conservative High-Order Accurate Finite-Differ-
ence Methods for Curvilinear Grids," AIAA Paper 93-3380, 1993.
Chapter 12
Numerical Simulation of Deformed Droplet

Dynamics and Evaporation
San-Mou Jeng*
University of Cincinnati, Cincinnati, Ohio 45221
and
Zhengtao Deng1
Alabama A&M University, Normal, Alabama 35762
I. Introduction
T HE exchange of mass, momentum, and energy between the gas and liquid
phase in a spray combustor are critical features of the combustion process.
Since the spray consists of droplets, it is reasonable to expect that the collective
properties of the droplets would influence the bulk spray vaporization characteris-
tics, which in turn determine the combustor performance. The rate of combustion
in part of the combustor will be controlled by the rate of vaporization. Droplet
trajectories affect the local vaporization rates and droplet drag affects drop trajec-
tories. To understand spray combustion and develop a predictive capability, the
details of the aerodynamic and thermal phenomena in the gas boundary layer,
the wake surrounding the droplet and in the liquid droplet interior must be fully
understood.
Two essential types of droplet dynamics studies are required for successful
spray combustor analysis. First, simplified but accurate vaporization models must
be developed to be used as subgrid models in spray combustor studies. These
models must be efficient to calculate the average characteristics of droplet group
behavior in the combustor with reasonable approximations. A second type of study
is single droplet dynamics that provides detailed information about the fine scales
of the relevant phenomena. Information such as drag coefficient and heat transfer
rate from dynamically deforming drops is required to form the simplified models.
This paper is focused on this second type of study.
Over the past few years, significant progress1"5 has been made in understanding
single droplet behavior and droplet/droplet interactions within a spray combustion
rights reserved.
'Associate Professor, Department of Aerospace Engineering and Engineering Me-
chanics.
t Associate Professor, Department of Civil Engineering.
305
306 S.-M. JENG AND Z. DENG
system. Sophisticated computational models have been used to derive the inter-
face exchange coefficients, e.g., drag coefficient, mass transfer rate, and energy
transfer rate, between droplets and the surrounding hot combustion gas products.
The detailed information about droplet behavior derived from these computational
fluid dynamics (CFD) codes has been compiled and used to establish more ac-
curate physical submodels for spray combustion in practical devices. Because
of the complexity of deformed droplet surface treatment, however, most of the
droplet dynamics studies are limited to the assumption of spherical droplet, with
empirical corrections for nonspherical effects and empirical expressions for drag
and convective effects. Research on deformed droplet modeling has advanced

significantly over the past decade, but the majority of existing studies are limited
to droplet oscillation simulations,6"9 the determination of drag coefficients, and
criteria for droplet breakup predictions. The mass and heat transfer rates for de-
formed and vibrating droplets have only recently been addressed.10"15 To resolve
the fundamental difference between spherical and nonspherical droplet behavior,
comprehensive computationally based analyses are needed.
The major objective of the current study is to understand deforming evaporating
droplet dynamics with convective flows when surface tension/discontinuities and
phase change effects exist at the arbitrarily shaped moving fluid interface. De-
formed droplet dynamics is a problem involving complex multicomponent two-
phase flows with interface discontinuities and phase change. In the present study,
droplet deformation and breakup is studied computationally as a model problem
to gain insight into mechanisms governing complex physical phenomena.
II. Computational Model

The main difficulty of the CFD model on the deformed droplets is that the
droplet shape is unknown and is required as part of the solution of the problem.
As a consequence, the boundary conditions at the droplet surface are nonlinear,
and, in addition, the solution depends on the prior history of the droplet motion
and interface shape. Therefore the crux of any numerical scheme that models
the interaction of two immiscible fluid phases with a sharp interface is its ability
to track an arbitrarily shaped moving interface or discontinuity. In the following
sections, a model of the multicomponent gas flow over an evaporating deforming
incompressible droplet will be discussed.
A. Governing Equations
The model used in this work considers an incompressible liquid droplet in
forced convective gas flow with finite interfacial surface tension. The liquid droplet
consists of only a single species. The compressible gas consists of two species —
one species from liquid droplet evaporation and another species, the freestream
gas. The integrated mass, momentum, and energy equations over a moving control
volume for gas/fluid may be written as follows.
Continuity equation:
p(u-U).dA+ 111,8^1=0 (1)

V(t) JA(t)
NUMERICAL SIMULATION 307
Species equations:
VM
= I! A + iMmi (2)
I AM
J JA \P
Momentum equations:
-^ / 1 1 pudV + 1 1 pu(u - £7) •

at J J Jv(v(tt) J JAM
AM
(3)
AM
Energy equation:
^ I II pEdV + pE(u - U) • dA
of J J JVVM J JAM
Pu dA + / / / r:VudV - J-dA + Es (4)
'VM J JAM
r = /x(Vw + (Vu)T) + AV • ul
where p is local fluid density, u is local fluid velocity vector, U is moving boundary
velocity, E is specific internal energy not including chemical energy, P is static
pressure, r is viscous stress tensor, / is heat flux vector, D is mass diffusivity
coefficient, k is thermal conductivity, and A. and /z are dilatation and shear vis-
cosity coefficients, respectively. Hm is species enthalpy, with subindex m = 1, 2
for different species, ms is rate of change of total mass resulting from droplet
evaporation, Smn is Kroneker delta function, subindex nxl is droplet surface in-
dex number, and species (1) represents the species of which the liquid droplet is
composed. Fs is momentum per unit time transferred from evaporation of droplet
to the fluid. Es is energy transfer rate source term associated with the interaction
between the droplet and the fluid. The ideal gas equation of state is also assumed
for the freestream gas and the vapor of the liquid droplet. For a liquid, similar
governing equations were considered except that incompressible flow assumptions
were invoked with the equation of state p = constant.
B. Boundary Conditions
Gas and liquid interface properties (velocity, temperature, pressure, species par-
tial density, enthalpy) are needed to close the governing equations. By considering
the surface as a singular surface across which the fluid density, pressure, and
enthalpy (or velocity in inviscid fluid) suffers jump discontinuities, the so-called
jump conditions can be derived. The primary jump conditions are directly derived
from the global conservational laws of species mass, total mass, linear momen-
tum, total energy, and entropy. The secondary jump conditions are derived from
the primary ones with respect to kinematics or mechanical energy, internal energy,
enthalpy, and entropy. Generally, the primary conditions can be used to define the
jump conditions with the aid of secondary ones for some special problems. These
jump conditions specify the exchange of mass, momentum, and energy across the
interface and stand as matching conditions between the two phases. They also
determine the interface exchange coefficients (drag coefficient and heat and mass
transfer rates). In the following sections, the interface jump conditions including
surface tension force and phase change thermodynamics are developed.

The boundary conditions at the interface between an arbitrarily shaped and
moving droplet surface that is evaporating and an external viscous gas can be
obtained by applying integral principles of continuity, momentum, and energy
to the control volume containing the interface, as shown in Fig. 1. Consider an
interface with surface tension and phase change across the interface. Take the fluid
of each phase as a viscous and thermally conducting fluid, and assume that two
species exist in the gas phase, one from liquid droplet evaporation and one from
the external freestream flow. No gas species is allowed to dissolve into the liquid,
which simplifies the liquid to be a single species. Letting the interface particles
move with velocity £7, using subscripts g and / to denote gas and liquid phases,
respectively, / to denote fuel species, m to represent species index, m = 1, 2 to
denote fuel and gas species, respectively, a to denote surface tension coefficient,
<M'k*1) ™ (M.k-1)
Fig. 1 Interface control volume to derive boundary jump conditions.
n to represent surface outward normal unit vector, t to represent surface tangential

unit vector, m to represent liquid droplet evaporation rate, and denoting species
mass fraction by F, the jump conditions can be obtained as follows.
Species conservation:
mYgf-mYlf = pgDgv -H (6)

\ Pg /
Continuity of mass flux:
pg(ug - U) n = PI(UI -U)-n = m (7)

Continuity of normal momentum flux:
m(ug -ui)-n= Pg + - Pin-Tgn>n + Tin-n (8)
Continuity of tangential momentum flux:
pgug • t(ug - U) - 1 - pmi • t(m- U)-t- Tgtt + rltt (9)

Continuity of energy flux:
m(Hg - Ht) = -ug '(Pg-Tg)'n + Ui.(Pi-Ti)'n-Jg-n+Ji'n (10)

where Hg and HI are gas and liquid total enthalpy.
In general the preceding interfacial jump conditions do not constitute sufficient
matching conditions necessary to define the problem uniquely. Consequently, they
should be supplemented by an additional boundary condition such as entropy in-
equality that restricts the kinematical, dynamical, and thermal relations between
two phases. From the second law of thermodynamics, we know that entropy can
be carried but not destroyed, i.e., that any real process must result in a net pro-
duction of entropy. The analysis16 based on the detailed entropy condition across
the interface are too complicated for direct application, however, and simplified
entropy conditions can be written as:
Ugt = Hit
that have been defined as thermal and no-slip conditions, recpectively. Tx is the
surface temperature and subscript t represents interface tangential direction.
For the current low-pressure conditions, the governing equations with derived
boundary conditions can be closed by assuming thermodynamic phase equilibrium
at the interface. Applying the Clausius-Clapeyron vapor pressure equations:
Pv Pvcr /T-TQ\Tcr
= (
~P~
ûO ~P~~
FvO I\*cr
T———— T~
*0/I~T~
1 '
where Tcr and TQ are critical and reference temperatures.

C. Numerical Methods
The numerical scheme used for integrating the governing equations follows
closely the arbitrary Lagrangian Eulerian (ALE) method.6'7>17>18 The calculations
necessary to advance a solution one time step are separated into four distinct
phases. The first phase consists of an explicit Lagrangian calculation that calculates
terms other than pressure and convections. In this phase, a contribution of diffusion
fluxes are explicitly added to the appropriate dependent variables. In the second
phase, an implicit Lagrangian that adjusts the pressure field to an advanced time
level and the computational grid vertices are temporarily assumed to move with the
local fluid velocity. There are no mass interchanges among the computational cells
and the new interface positions are calculated. In the third phase of calculations,
the computational meshes both inside and outside of the droplet surface are forced
to move to a new user-defined configuration while interface vertices are kept in
their Lagrangian-determined positions. The donor-acceptor method17 is used to
calculate the dependent variables at the new grid positions defined at the third
phase calculation. The last phase uses the iteration methods to derive the interface
temperature and provides the interface mass and heat transfer rate.
The finite difference equations are written as approximations to the integral
form of the conservational equations in which the integration control volumes
are moving cells of the finite difference mesh. The physical interface boundaries
between two the fluids coincide with a set of cell boundaries. Scalar fluid variables
such as pressure, specific internal energy, density, and cell mass are defined at the
cell center. Vector quantities such as coordinates and velocity components are
defined at the cell vertices.
A detailed description of the numerical method can be found in Ref. 13. The
multiphase procedures used for each time-step integration of the governing equa-
tions can be summarized as follows.
1. First Phase—Explicit Lagrangian Calculation

Contributions to the dependent variables from diffusion terms in the governing
equations are calculated in a fully explicit manner. For example, the density and
mass of cell j, k can be obtained as
^D)«" [v (7
- (Mm)£
where superscript (1) represents step 1, A is the cell surface, and M is the cell
mass. The species partial density in this step can then be obtained as
The velocity and temperature are calculated similar to the species density, and the
pressure field P^ is calculated from equation of state after the first step.
2. Second Phase—Implicit Lagrangian Calculation

The pressure field, including the effect of surface tension force, is solved im-
plicitly. In this step, the grid systems are moving with the same velocity as the
NUMERICAL SIMULATION 31 1
local fluid velocity. From the continuity equation (1), the total mass of each cell
should be conserved between the previous time step and the current phase since
the vertices are temporarily assumed to move with the local fluid velocity. The
calculation mass residue should be zero at each time step for each cell, i.e.,
(Residue)/,* = / / pu - dA -> 0
)/,* = (15)
J JA
JAW
This equation is solved implicitly by using the conjugate-gradient methods in
conjunction with Jacobian matrix preconditioning.19 Once the converged pressure
field is obtained, the fluid Lagrangian velocity and grid system following the
Lagrangian description can be established.
3. Third Phase—New Grid System

At the end of the third phase calculations, the grid system moves with the same
velocity as the local fluid and no convective transport occurs. The adoption of a
purely Lagrangian coordinate to complex flows has been very difficult because
the Lagrangian description of the fluid particle motions is limited to well-behaved
flows.18 To avoid grid crossing and to keep the benefit of Lagrangian surface
tracking, a desired new grid system adapted to the interface shape is constructed
in this phase calculation, and the dependent variables are updated at the new grid
positions. To keep the interface resolution, the interface grids are redistributed
uniformly along the interface at each time step. Once the interface and boundary
grids are determined, the new inferior grid system is generated using a Poisson
solver. Once the new grid system is defined, a donor-acceptor method17 is used to
calculate advective terms and update the dependent variables.
4. Fourth Phase—Interface Matching

An iterative method is used to satisfy interface mass, species, and energy bound-
ary conditions, Eqs. (6-11). Temperature and fuel mass fraction at the interface
are updated. The derived surface temperature and fuel mass fraction are then used
to calculate mass and heat transfer rates.

In this section, two-dimensional numerical simulations of droplet dynamics are
discussed in detail. First, the developed computer code was validated using cylin-
der oscillation calculations, by comparing with theoretical results. Second, the
dynamics of deforming two-dimensional droplets was simulated and compared
with experimental criteria for droplet breakup regimes. Third, deforming, evapo-
rating droplet dynamics were simulated to investigate the effects of deformation
on transport rates such as the droplet evaporation rate.
A. Droplet Shape Oscillation

The nature of the vibrations of a cylindrical liquid jet (two-dimensional droplet)
was first investigated by Rayleigh.20 Using the potential flow assumptions, Rayleigh
was able to derive the frequency of oscillation as
where pd and pe are the droplet and external fluid densities, respectively, R is the
radius of the undisturbed cylindrical jet, and a is the surface tension coefficient.
The values n = 0 and n = 1 correspond only to rigid body motion. The funda-
mental mode is n = 2, and the amplitude of oscillation of a cylindrical jet at any
instant is given in polar coordinates as
r = a + 6cos(nO) (17)
where € is the small disturbance amplitude and a is the radius of the undisturbed
cylindrical jet.
To test the surface tension and numerical algorithm, a series of calculations of

shape oscillations of a cylindrical jet in an immiscible fluid of different density,
in a normal mode, were conducted. Results for three different sets of density ratio
conditions are discussed hereafter.
The initial cylinder surface geometry was chosen as:
a = 0.01 cm
cr = 15 dyne/cm (18)
6 = 0.05
The initial cylinder shape was chosen to be mode 2 following Eq. (17). Both the
droplet and exterior fluid were taken to be inviscid to focus on surface tension
effects and to compare with analytical solutions. To see immiscible fluid density
effects on shape oscillations, three different density ratio test cases were examined.
The density ratios used in these three cases were chosen to simulate shape oscil-
lation conditions of a liquid cylinder oscillating in gas (interior to exterior fluid
density ratio 1000:1), a gas cylinder oscillating in liquid (density ratio 1:1000), and
a comparable density ratio (density ratio 2:1). In the current studies, the exterior
fluid domain was chosen to be three times bigger than the droplet dimension. The
primary output from these oscillation simulations were vertex position oscillation
history, oscillation frequency, droplet surface shapes, and the corresponding veloc-
ity vectors. To evaluate the numerical scheme, numerical results were compared
with theoretical results and experimental photos on shape oscillation.
The three different test conditions and oscillation frequencies are given in
Table 1. Figure 2 shows the position, as a function of time, of the rightmost
vertex of the droplet compared with Rayleigh's theory. For density ratios (interior
to external fluid) of 1:1000 and 1000:1, the calculated oscillation periods agree
excellently with Rayleigh's analysis; however, an approximately 4% difference
exists for a density ratio of 2:1 (447:223.5).
It is also noted that the oscillation amplitude should not and did not decay with
time, and that the oscillation period does not change with time, indicating that the
Table 1 Test results on droplet shape oscillation
Density Ratio,
kg/m3 Theore. Period, s Cal. Period, s
1000:1 6.626E-04 6.626E-04
1:1000 6.626E-04 6.626E-04
447:223.5 5.423E-04 5.560E-04
8.0
o Linear theory
6.0 — Numerical
4.0
CD
"B 2-°
"o.
< 0.0
-2.0
*o
</>
O
-4.0
-6.0
o.o o.s 1.0 1.5 2.0 2.5 3.0
a) Period Number (Time/Period)
o Linear theory -
— Numerical -
0.0 0.5 1.0 1.5 2.0 2.5 3.0
b) Period Number (Time/Period)

Fig. 2 The rightmost vertex position changes with time: a) density ratio 1000:1, b)
density ratio 1:1000, and c) density ratio 2:1. (Continued)
8.0
o Linear theory
6.0 — Numerical
CD
"§ 2.0
j£-»
"5-
< 0.0
c
o
1 -2.0
-4.0
-6.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
c) Period Number (Time/Period)
Fig. 2 (Continued).
numerical diffusion and dissipation are almost zero in the numerical algorithm.
Figure 3 shows the velocity fields at different time steps. The velocity field is
symmetric to the vertical plane containing the center of the cylinder, even at the
transition time from short axis to long axis, such as at the t = 0.502 and 1.509
periods. The velocity distribution inside the droplet is similar to the experimental
photographic results obtained by Trinh et al.21 and Trinh and Wang22 for mode 2
small amplitude oscillations.
B. Droplet Deformation in Viscous Incompressible Flows

The competition between aerodynamic forces and surface tension forces and
their effects on droplet deformation can be related to the Weber number, the ratio
of external aerodynamic force to the surface tension force, and the Bond number,
the ratio of internal hydrodynamic pressure to surface tension force.
To determine the breakup mode, the most common criterion is a critical Weber
number although it is not the only controlling parameter for determining droplet
behavior in a gas flow. Basically, three secondary droplet-breakup regimes have
been summarized23 although overlap may occur in these regimes.
Regime 1 involves parachute-type (bag-type) breakup and chaotic destruction
in which the droplet flattens in the direction perpendicular to the flow and forms
a shroud extending back along the direction of gas motion. This shroud either
breaks off, producing a group of small droplets, or several parachutes are formed
from one droplet, which, in the final stage, produces groups of fine droplets. For
this case,

Time= 0.15 period Time= 0.50 period
Time= 0.45 period Tirne= 0.60 period
Fig. 3 Velocity distribution of an oscillating cylinder, density ratio 1000:1.

(Continued)
8 < We < 40
(19)
0.2 < WeRe~Q'5 < 1.6
Regime 2 involves stripping-type breakup in which the gas flow tears off shrouds
from the flattened droplet, followed by the disk-shaped droplet reaching its critical
deformation and decomposing into small droplets. For this case,
20 < We < 2.0 x 104
(20)
1.0< WeRe-°'5 <20
Regime 3 involves explosion-type breakup where the droplet seems to instantly
shatter into many fine particles. For this case,
2 x 103 < We < 2.0 x 105
(21)
20 < WeRe'Q'5 < 200
To study droplet deformation in convective flows, a series of parametric studies
have been conducted based on the computational scheme developed in this work

Fig. 3 (Continued).
for impulsive incompressible viscous (laminar) flow over an initially stagnant

viscous two-dimensional droplet. Three different test cases were examined. The
test conditions for these simulations are given in Table 2. Although liquid viscosity
affects the breakup time, it does not alter the breakup regime, and the same liquid
viscosity was used for all three cases in the viscous calculations. The derived
parameters are shown in Table 3. According to the established criteria for the
initially spherical droplets, case I should result in no breakup, case II should result
in a parachute-type breakup, and case III should result in stripping-type breakup.
Typical velocity vector plots and droplet surface shapes for case I are presented at
four different time steps in Fig. 4. The initially nom-deformed droplet is deformed
by an impulsive flow. At 0.2 ms, the droplet surface becomes flattened normal to
the inlet gas flow direction. The surface geometry changes, which in turn changes
the surface tension force along the droplet surface. Compared to the aerodynamic
force, this surface tension restoring force is not small (because of the smaller
Weber number) and the droplet starts to oscillate. At 0.4 ms, the transition stage
for oscillation occurs. It was observed that the droplet vibrates primarily at the third
harmonic. At 0.6 ms, a typical third mode surface shape is observed. At 0.8 ms,
the droplet begins deforming back, but because of viscous effects, the amplitude

Table 2 Flow conditions for droplet breakup simulations
Case I Case II Case III

Droplet diameter, 200 200 200
/xm
Liquid density, 1000 1000 447
kg/m3
Liquid viscosity, 1.8x 10~3 1.8x 10-3 1.8 x 10-3
Ns/m2
Surface tension, 15 x 10-3 15 x 10~3 15 x 10~3

N/m
Ambient gas 10 27.386 10
inflow velocity,
m/s
Gas density, 1.0 1.0 223.5
kg/m3
Gas viscosity, 5.556 x 10~5 5.4774 x 10~5 5.556 x 10~5
Ns/m2
of deformation will decay and reach an equilibrium position. The droplet will not
breakup for these conditions.
Parachute-type (bag-type) breakup has been observed in case II as depicted in
Fig. 5. The velocity field and surface shape at six different time steps are presented.
At 0.02 ms, the droplet flattens in the direction perpendicular to the inlet flow.
Instead of oscillating, the droplet undergoes continual deformation. At 0.1 ms, a
recirculation zone is well established in the droplet wake. At 0.12 ms, the droplet
starts to break off from the edge and a more flattened disk shape is observed. This
process agrees well with experimental criteria for parachute-type breakup.
The velocity vector and surface geometry histories for the case III simulation are
shown in Fig. 6. In this case, more rapid deformation occurs similar to parachute-
type or bag-type breakup. With greatly increased flattening of the droplet in
the direction normal to the inlet flow, the droplet edge can be stripped away.
The numerical computation cannot show this stripping, however, because this fine
scale of the interface cannot be resolved.
C. Evaporating and Deforming Droplet

The test case selected was a cold two-dimensional n-octane fuel droplet evap-
orating in a hot oxygen gas inlet flow. The physical parameters used in this study
are shown in Table 4. The derived parameters are shown in Table 5. Five different
Table 3 Derived parameters for droplet breakup simulations
Case I Case II Case III

Reference R€ 36 100 36
Referenced, 1.3333 10 298
WeR7°'5 0.2222 1 49.67
Table 4 Physical parameters of evaporating n-octane droplet
Initial droplet diameter, /xm 200

N-octane fuel density, kg/m3 1000
N-octane fuel viscosity, Ns/m2 1.8 x 10~3
Surface tension coefficient, dyne/cm 15 or 150
Droplet initial temperature, K 300
Droplet conductivity, W/m-K 0.106
Freestream inflow temperature, K 500
Freestream inflow velocity, m/s 31.214

Ambient gas pressure, N/m2 1.0 x 105
Gas phase Prandtl number 0.90
Gas phase Schmidt number 0.90
Molecular weight of oxygen gas 32.0
Molecular weight of n-octane 114.2
Time= 0.2 ms Time= 0.6 ms
Fig. 4 Oscillation/deformation: velocity vectors and droplet surface shape for case
I; Re = 36, We = 1.33.
Table 5 Derived parameters of evaporating droplet

Initial Reynolds' number 100.0
Initial Weber's number 10.0 or 1.0
Ambient gas density, kg/m3 0.7697
Density ratio 1299.2
1.0 or 0.1
Time= 0.06 ms
Fig. 5 Parachute breakup: velocity vectors and droplet surface shape for case II;
Re = 100, We = 10.

Time= 0.03 ms
Fig. 6 Stripping type breakup: velocity vectors and surface shape for case III;
Re = 36, We = 298.
cases were studied, based on droplet vibration modes and freestream gas flow
fluctuations, with the conditions described in Table 6. Notice that in all five test
cases, the surface tension coefficient was held constant, and the droplet surface
latent heat (heat of evaporation) and droplet surface vapor pressure were functions
of surface temperature.
The baseline case 1 calculation was for a cylindrical droplet evaporating in a
convective flow. With a large surface tension coefficient, the predicted surface
deformation was small, and the surface geometry changes were only a result

Table 6 Test cases of droplet evaporation
Deformation Inflow fluctuation

Case Re We switch switch
1 100 10 off off
2 100 10 on off
3 100 1 on off
4 100 1 on on
5 100 10 off on
of evaporation. The transient stage, at time 0.02 ms, and steady state, at time
0.2 ms, velocity vectors, temperature, and fuel species concentration contours in
the gas phase are plotted in Fig. 7. The case 1 vaporization history, ranondeformed
(evaporation rate per unit surface area), is used as the baseline comparison for the
deforming droplet computations.
The case 2 condition simulates droplet evaporation with parachute-type defor-
mation. The predicted dynamic behavior of a droplet for this type of breakup is
shown in Figs. 8 and 9. The deformation time (time for the droplet to deform to its
critical stage) is about the same as for the nonevaporating case. This implies that
the dynamics of droplet deformation are basically unaffected by the vaporization
process for these test conditions. Figures 8 and 9 show the temperature and species
concentration contours during the deformation process. Convective currents in the
freestream surrounding the droplet increase the fuel species concentrations inside
the wake region.
The deforming droplet vaporization history, total droplet surface area, and total
evaporation mass flux compared with case 1 (cylindrical droplet) are given in
Fig. 10. With deformation effects, m (per unit surface area) is about 85% of
ranondeformed- This is the result of the larger recirculation zone at the back of the
deformed droplet where the evaporation rate is significantly decreased. At time
0.08 ms, the nondeformed droplet evaporation rate is increasing, but the deformed
droplet internal temperature field is not well established and its evaporation rate is
slightly decreasing. With more severe deformation, more and more energy must
be fed into the liquid droplet to raise its temperature, and the deformed droplet
evaporation rate decreases slightly with time. Droplet surface area ratio (deformed
droplet total surface area divided by nondeformed droplet total surface area) is
a function of time, and it increases with time, which means that the deformed
droplet has increasingly larger surface area than that of the nondeformed droplet.
Initially, the total evaporation mass flux is controlled by the evaporation rate per
unit area. Once droplet deformation becomes severe, it is affected by the droplet
surface area increment, the reduction of the driving force for mass diffusion, and
the fuel vapor concentration around the droplet. The total evaporation rate of the
severely deformed droplet is larger than that of the undeformed droplet.
The case 3 condition simulates a vibrating/deforming droplet undergoing vapor-
ization. The initial droplet surface geometry is an ellipsoid shape with the long axis
length being 110% of the equilibrium diameter. The surface tension coefficient is
15 dyne/cm, which keeps the droplet from breaking up. The predicted variation
of the axial diameter and evaporation rate with time are compared with results
for the nondeformed droplet, as plotted in Fig. 11. Droplet diameter was found
to decrease while oscillating, and the evaporation rate oscillates with a maximum
2
I
a
C
3 Level SPD
O
A 0.23
9 0.21
0.19
g
I
0.16
0.14
0.12
g 0.09
0.07
0.05
0.02
Fig. 7 Droplet evaporation (case 1): velocity vectors, temperature and species con-
centration contours; at time 0.02 ms, transient stage, at time 0.2 ms, steady state.
Time = 0.02 ms Time = 0.10ms
Level TEM
A 498
9 471
8 453
7 438
6 418
5 400
4 382
3 364
2 347
1 329
Time = 0.06 ms Time = 0.12 ms
Time = 0.08 ms Time = 0.14 ms
Fig. 8 Gas temperature contours for parachute-type deformed droplet (case 2).
Level SPD
A 0.18
9 0.16
8 0.14
7 0.12
6 0.11
5 0.09
4 0.07
3 0.05
2 0.03
1 0.01
Time= 0.08 ms
Fig. 9 Fuel species concentration contours for parachute-type deformed droplet

(case 2).
Evaporation (Parachute-type Breakup)

1.40
HlA/(mA)nondeformed
0.00 0.02 0.05 0.07 0.10 0.13

Time (ms)
Fig. 10 Evaporation rate (per unit area) for parachute-type deformed droplet
(case 2).
0.000 0.100 0.200 0.300 0.400 0.500
Time (ms)
Fig. 11 Vibrating/deforming droplet evaporation rate and axial diameter change
(case 3).
1.300
m m
•l/ HOndeformed
•
m2/mn(m(lef()rmed
0,600
0.000 0.100 0.200 0.300 0.400 0.500 0.600
Time (ms)
Fig. 12 Effects of inflow velocity fluctuation on droplet evaporation rate: oscillating
(riii) (case 4) and nonoscillating (m2) (case 5) droplets.
perturbation amplitude of less than 5%. The oscillation period changed slowly
with time because of reduced droplet diameter. The long axis oscillation ampli-
tude decreases because of viscous damping and evaporation. The first oscillation
period was found to be 0.2438 ms which is comparable with the linear theoretical
prediction 0.2095 ms for a second mode freely oscillating drop.
This calculated oscillation frequency was used as an inflow velocity fluctuation
frequency to study gas/liquid coupling effects on droplet evaporation. In case 4, a
vibrating droplet (same conditions as case 3) was acted on by an impulsive flow
with an inflow velocity fluctuation
Win = (22)
where Wino is the time-averaged inflow velocity, co is the fluctuation frequency, and
T is the elapsed time. In case 5, the same fluctuating inlet velocity is imposed on the
initially undeformed droplet of case 1. Figure 12 depicts the predicted vaporization
histories of cases 4 (mi) and 5 (m2) in comparison to case 1. It is clear that the
evaporation rate is significantly influenced by the inflow velocity fluctuation. The
ra oscillation frequency is virtually controlled by the inflow frequency and the
perturbation amplitude of m is almost 13% of the total rate for a nondeformed
droplet. A 15% increase in the oscillation amplitude was found for the vibrating
droplet in comparison to results obtained without an inflow velocity fluctuation.
IV. Conclusions
A computational model based on the ALE numerical algorithm has been estab-
lished to calculate the dynamics of a deforming droplet and the interface exchange
between the liquid droplet and its surroundings. This model solves the flow in two
fluid phases that are separated by a free surface where surface tension force, phase
change, mass transfer, and heat transfer exist. Interface tracking was carefully de-
veloped in the modeling effort with special treatment of surface tension forces and
interface jump conditions. The model can treat incompressible or compressible
flow, which allows freedom of choice of flow simulations at different conditions.
The model has been used to study the dynamics of two-dimensional droplet
shape free oscillation, nonevaporating and evaporating droplets undergoing defor-
mation/breakup, and oscillations under viscous convective flows. The numerically
predicted breakup modes agree well with experimentally derived criteria. The
computational results on evaporating droplets imply that the dynamics of droplet
deformation are basically unaffected by the vaporization. Deformation effects on
the evaporation rate have been examined indicating that the deformed droplet evap-
oration rates (per unit area) are larger than that of nondeformed droplets. Effects
of fluctuating gas velocity and droplet vibration on propellant droplet evaporation
have been investigated.
References
h, G. M., "Current Status of Droplet and Liquid Combustion," Progress in Engi-
neering and Combustion Science, Vol. 3, Permagon, New York, 1977, pp. 191-224.
2
Faeth, G. M., "Evaporation and Combustion of Sprays," Progress in Engineering and
Science, Vol. 9, Permagon, New York, 1983, pp. 1-76.
3
Law, C. K., "Mechanism of Droplet Combustion," Proceedings of the Second Interna-
tional Colloquium on Drops and Bubbles, JPL 82-7, Monterey, CA, 1981, pp. 39-53.
4
Sirignano, W. A., "The Formulation of Spray Combustion Models: Resolution Com-
pared to Droplet Spacing " Journal of Heat Transfer, Vol. 108, No. 3, 1986, pp. 633-641.
5
Prakash, S., and Sirignano, W. A., "Theory of Convective Droplet Vaporation with
Unsteady Heat Transfer in the Circulating Liquid Phase" International Journal of Heat
Mass and Transfer, Vol. 23, No. 3, pp. 253-268, 1980.
6
Fyfe, D. E., Oran, E. S., and Fritts, M. J., "Numerical Simulation of Droplet Oscillation,
Breakup, and Distortion," AIAA Paper 87-0539,1987.
7
Fyfe, D. E., Oran, E. S., and Fritts, M. J., "Surface Tension and Viscosity with Lagrangian
Hydrodynamics on a Triangular Mesh," Journal of Computational Physics, Vol. 76, No. 2,
1988, pp. 349-384.
8
Leppinen, D. M., Renksizbulut, M., and Hay wood, R. J., "The Effects of Surfactants on
Droplet Behavior at Intermediate Reynolds Number-I. Transient Deformation and Evapo-
ration," Chemical Engineering Science, Vol. 51, No. 3, 1996, pp. 491-501.
9
Liang, P. Y., Eastes, T. W., and Gharakhari, A., "Computer Simulations of Drop Defor-
mation and Drop Breakup," AIAA Paper 88-3142,1988.
1()
Unverdi, S. O., and Tryggvason, G., "A Front-Tracking Method for Viscous, Incom-
pressible, Multi-Fluid Flows," Journal Computational Physics, Vol. 100, No. 1, 1992, pp.
25-37.
H
Haywood, D. M., Renksizbulut, M., and Raithby, G. D., "Numerical Solution of De-
forming Evaporating Droplets at Intermediate Reynolds Number," Numerical Heat Transfer,
Vol. 26, No. 3, 1994, pp. 253-272.
12
Haywood, D. M., Renksizbulut, M., and Raithby, G. D., "Transient Deformation and
Evaporation of Droplets at Intermediate Reynolds Number," International Journal of Heat
and Mass Transfer, Vol. 37, No. 9, 1994, pp. 1401-1409.
13
Deng, Z. T., "Numerical Investigation of Non-spherical Droplet Dynamics in Viscous
Convective Rows," Ph.D. Dissertation, Aerospace and Mechanical Engineering Depart-
ment, University of Tennessee, Knoxville, TN, 1991.
14
Deng, Z. T, and Jeng, S. M., "Numerical Simulation of Droplet Deformation in
Convective Rows," AIAA Journal, Vol. 30, 1982, pp. 1290-1297.
15
Deng, Z. T, Litchford, R. J., and Jeng, S. M., "Two-Dimensional Simulation of Droplet
Evaporation at High Pressure," AIAA Paper 92-3122, 1992.

16
Ishii, M., "Thermo-Fluid Dynamic Theory of Two-Phase Row," Eyrolles, France,
1975.
17
Amsden, A. A., Ramshaw, J. D., O'Rourke, P. J., and Dukowicz, J. K., "KIVA: A
Computer Program for Two- and Three-Dimensional Fluid Rows with Chemical Reactions
and Fuel Sprays," LA-10245-MS, 1985.
18
Fritts, M. J., Fyfe, D. E., and Oran, E. S., "Numerical Simulation of Fuel Droplet Rows
Using a Lagrangian Triangular Mesh," NASA CR-168263, 1983.
19
O'Rourke, P. J., and Amsden, A. A., "Implementation of a Conjugate Residual Iteration
in the KIVA Computer Program," LA-10849-MS, 1986.
2()
Rayleigh, L., "On the Capillary Phenomena of Jets," Proceedings of the Royal Society
of London, Vol. 29, 1879, p. 71.
21
Trinh, E., Zwern, A., and Wang, T. G., "An Experimental Study of Small-Amplitude
Drop Oscillations in Immiscible Fluid Systems," Journal of Fluid Mechanics, Vol. 115,
1982, pp. 453^74.
22
Trinh, E., and Wang, T. G., "Large-Amplitude Free and Driven Drop-Shape Os-
cillations: Experimental Observations," Journal of Fluid Mechanics, Vol. 122, 1982,
pp. 315-338.
23
Borisov, A. A., Gel'fand, B. E., Natanzon, M. S., and Kossov, O. M., "Droplet Breakup
Regimes and Criteria for their Existence," Journal of Engineering Physics, Vol. 40, No. 1,
1981.


Chapter 13
Numerical Simulation of High-Frequency Combustion

Instabilities in Liquid Rocket Engines
Jeffrey M. Grenda* and Charles L. Merklef

Introduction
H IGH-FREQUENCY combustion instabilities have often been encountered

in the development of liquid rocket engines and continue to present sig-
nificant difficulties in reliable motor design. There are substantial challenges in
modeling liquid propellant instability due to the inherent complexity and strongly
coupled nature of numerous physical processes present. Key among these are
atomization, vaporization, and combustion. Several competing physical mecha-
nisms acting individually or in combination may lead to disturbance growth under
certain conditions. To further complicate matters, the gas phase motions are mul-
tidimensional and transient and are often characterized by sustained spinning or
standing tangential acoustic modes.
Until recently, the multidimensional nature of the gas phase disturbances prohib-
ited the application of computational fluid dynamics (CFD) to liquid propellant
instabilities. Current computer capabilities and progress in numerical methods
have now advanced to the point that CFD can complement traditional experimen-
tal and approximate analytical investigations as a practical method of analyzing
instability mechanisms. The current paper addresses the capabilities of applying
CFD in combustion instability research. This includes issues such as numerical
accuracy, computational efficiency, and physical modeling. Techniques for ensur-
ing accurate, efficient algorithms are presented along with means of validating
these procedures and representative stability predictions.
A number of researchers1'3 have applied computational fluid dynamics in the
study of instabilities in liquid propellant engines. Most analyses have been limited
to two-dimensional annular or axisymmetric geometries to minimize computa-
tional effort, in spite of the inherent three-dimensional nature of the flowfield.
Related work using hybrid analytical/numerical techniques has met with some
rights reserved.
'Graduate Research Assistant, Department of Mechanical Engineering.
Professor, Department of Mechanical Engineering.
331
332 J. M. GRENDA AND C. L MERKLE
success,4 although limitations are placed on the treatment of the equations of mo-
tion. Other efforts have included substantial physical modeling of the liquid phase5
in an attempt to pinpoint instability mechanisms.
The current work focuses on three simultaneous approaches which are used
to obtain fully three-dimensional simulations of combustion instability. The most
comprehensive and CPU-intensive analysis solves the full nonlinear fluid dynamic
equations using a mixed finite difference/spectral methodology. Liquid phase sub-
models are directly incorporated into the gas/liquid interphase source terms. One
fundamental limitation on numerical stability analysis is that voluminous numbers
of computations must be performed to determine physical trends or stability bound-

aries, since each computation indicates only the stability at a particular operating
condition. It is also critical to validate the accuracy of CFD solutions. To address
both of these issues, we introduce two simplified analyses to complement the
full numerical methodology. These are a small-amplitude CFD-based linearized
numerical treatment, and an approximate closed-form analytical approach. Both
of these companion linear methods are three dimensional and provide a method of
rapid parametric solution. These approaches provide efficient means for discern-
ing stability trends, also serving as a validation standard, even though physical
modeling is restricted. The full numerical solution permits more complete physics
and nonlinear effects to be modeled.
We begin by introducing the governing equations for two-phase systems typi-
cal of liquid rocket engine combustion chambers. The solution methodology for
the gas phase is outlined, including the details of the finite difference/spectral
procedure. The numerical validation procedures are introduced and are used to
determine the accuracy of the nonlinear three-dimensional approach. More physi-
cally realistic liquid phase effects are then incorporated and used to predict stability
trends for a variety of operating conditions, including the effects of injected droplet
size and liquid temperature.
Theoretical Formulation
A computational method for describing the dynamics of combustion instabil-
ity requires equation sets for the fluid dynamics of the gas phase and the spray
dynamics of the liquid phase. Although the gas phase motions are most properly
described by the full Navier-Stokes and species transport equations, the unsteady
Euler equations give a reasonably accurate description of the wave dynamics of
the combustion chamber. The predominant diffusive mechanisms of mixing on the
droplet scale are incorporated through existing droplet source term correlations
and, consequently, their presence does not detract from the choice of an Euler for-
mulation. Atomization and vaporization are treated in phenomenological fashion
based on classical models from the literature. We begin by describing the equations
for the gas phase followed by methods for treating the liquid phase processes.
Gas Phase Equations of Motion

With interaction terms for the liquid phase included as source terms, the three-
dimensional unsteady Euler equations in cylindrical coordinates become
8Q 8E 1 dFr I dG I
TT
3t + IT
dx + —3-
r dr + —^
r d9 = ~(#gas
r 6 +#iiq)
M (1)
NUMERICAL SIMULATION OF COMBUSTION INSTABILITIES 333
where Q is the vector of primary dependent variables and E, F, and G are the
conservative flux vectors. These are given by the relationships
P pu pv
pu pu2 + p puv
pv puv F = pv2-{- p (2)
pw puw pvw
e (e + p)u_ .(e + p)v_
pw ' 0
puw 0
pvw p + pw2 (3)
pw2 + p —pvw
(e + p)w_ 0
Here, the spatial variables jc, r, and 9 represent the axial, radial, and azimuthal co-
ordinates, whereas u, t>, and w represent the corresponding velocity components.
The density and pressure are given by p and p, and the total energy is defined
as e = pe -f- ^p(u2 + v2 + u;2). The internal energy of the gas is expressed as
6 = RT/(y — 1), where R and y assume their usual definitions as the gas con-
stant and specific heat ratio. For simplicity, we close the system using the perfect
gas relationship for the equation of state.
The source terms appearing on the right-hand side of Eq. (1) have been written
such that Hgas contains the radial and azimuthal terms from the cylindrical form
of the governing equations, and H\\^ contains the general interphase coupling
terms from the liquid analysis. The source term H\\q is determined by integrating
the contributions of mass, momentum, and energy exchange from the droplet
parcels contained within the individual differential volumes of the gas phase.
The instantaneous locations and trajectories of the numerous droplet parcels are
determined from a separate set of liquid phase equations which is outlined in the
following section.
Liquid Phase Equations of Motion

The description of the liquid phase must incorporate substantial physical model-
ing although remaining computationally tractable. For the current work, we choose
a Lagrangian formulation for tracking vaporizing droplets originating from atom-
ization processes near the injector face. As the droplets vaporize, they contribute
mass, momentum, and energy to the gas phase. The dynamic equations that gov-
ern the Lagrangian description of the spray are straightforward applications of the
conservation of mass and force balances for the individual droplets. The equations
for an individual droplet may be written in vector form as
(4)
At
where
Qd = Td] ———-, (5)

md
334 J. M. GRENDA AND C. L. MERKLE
Here, ra</ represents the mass of a specific droplet, and ravap is its vaporization
rate. The quantity HI is the velocity component of the droplet in the ith coordinate
direction, whereas the symbol F represents the vectorial sum of the external forces
acting on the droplet which cause it to accelerate. The change in temperature of
the droplet with time is written in symbolic notation to indicate that an appropriate
energy equation is required to determine the unsteady temperature distribution
within the liquid phase. The current results utilize an infinite thermal conductivity
approximation as an initial investigative assumption.
Link Between Stability Analyses and Engine Configuration

As liquid propellants are injected into a rocket engine, atomization processes
reduce the initially continuous liquid phase (typically liquid jets or sheets) into a
fine spray near the injector face. The spray is entrained in the gas phase and vapor-
izes as the droplets traverse the chamber. Unfortunately, the numerical procedures
needed to model these processes directly remain beyond current or projected com-
puter capabilities. Consequently, computational instability analyses must rely upon
global, phenomenological models which utilize semiempirical correlations based
on experimentally observed behavior. Empirically based atomization models6'7
typically predict a mean droplet size and droplet distribution function near the
injector, often as a function of injector geometry and parameters such as injection
velocity. Substantial research effort has been directed toward developing simpli-
fied models that can be used to describe droplet vaporization.8'9 Other effects such
as injector coupling may also be incorporated.10
An advantage of using CFD procedures for stability analyses is that local,
instantaneous conditions in the gas phase can be used for the evaluation of all
physical submodels. This provides the capability for incorporating true unsteady
effects in the modeling of phenomena such as atomization and vaporization. In
particular, unsteady crossflow velocities could have a dramatic effect on the drop
size distribution produced by the atomization process, thereby resulting in major
changes in the vaporization rate leading to a dramatic impact on combustion
response. Such unsteady effects are not yet available, although certain models
contain inherent assumptions that in effect make them quasisteady. In view of
these issues, the present analysis applies existing models in quasisteady fashion
to provide a logical starting point for CFD-based analyses.
Computational Formulation
Tangential wave motions have typically been the most pervasive and destructive
in liquid propellant instability incidents. The modeling of these disturbances re-
quires the numerical solution of the complete three-dimensional, nonhomogeneous
Euler equations along with the imposition of properly selected physical boundary
conditions. In general, the numerical solution of the three-dimensional unsteady
equations coupled with a detailed treatment of the liquid phase requires signif-
icant computational effort that can become prohibitively expensive. To mitigate
this restriction, we apply a mixed finite difference/spectral method to the gas phase
and a simplified Lagrangian analysis for the liquid phase. The gas phase analysis
utilizes finite differences in the axial and radial directions, and a Fourier expansion
in the circumferential direction. This full nonlinear procedure is outlined in the
following section, followed by a description of the two linearized analyses which
are used in companion with the complete analysis. The final subsection describes
the liquid phase treatment.
Finite Difference/Spectral Nonlinear Analysis

The mixed finite difference/spectral scheme utilizes a Fourier expansion of the
primary variable Q represented by a truncated series of the form
M
Q= [Qm.c(x> r, t)cosmO + Qm,s(x, r, t)smmO] (6)
where the amplitude coefficients

Qm,c = (PC, p5c, pvc* pwc, ec, 0, 0, 0, 0, 0)r (7)
fi w§J = (0, 0, 0, 0, 0, pS9 pus, pi>s, PWS, **f (8)
corresponding to the Fourier mode m are time-dependent functions of both the

axial and radial dimensions. This general series formulation allows both spinning
and standing mode wave solutions to be calculated. The number of terms retained in
the Fourier series representation is selected to resolve the circumferential variation
in the flow properties.
By substituting these expressions into Eq. (1), the general dynamic equations
governing the development of the flowfield in Fourier space can then be written as
dQm,T m m, x , ,
4- - ——-——— = —— G(Qmmj) + -«tot(fim,r
dt dx r dr r r
Here, the vectors E, F, G, and //tot are written as functions of the Fourier coef-
ficients contained in the vector of primary variables Qmj, which is defined as
the sum of the amplitude coefficients Qm,c and Qm,s* The vector //tot includes
both the gas and liquid phase contributions to the source term. The spectral pro-
cedure reduces the three-dimensional equations for the continuum gas phase to
2M — 1 sets of two-dimensional equations. For standing mode calculations, only
M sets of two-dimensional equations need to be evaluated due to redundancy in
the equations. Note that the number of spectral modes is analogous to grid points
in the 9 direction, except that since the spectral decomposition for the 9 direction
converges more rapidly than finite difference procedures, the magnitude of M can
be kept considerably smaller. In both the standing and spinning mode computa-
tions, the calculation of the shape functions in the Fourier domain defines the flow
variables in three-dimensional physical space by the appropriate summation of the
truncated series of the primary vector Qmj-
The solution of the set of dynamic equations given in Eq. (9) is obtained
through an explicit, multistep Runge-Kutta time integration scheme10 utilizing
second-order central differences in space.
Boundary Conditions
The selection and application of physically appropriate boundary conditions is
crucial to any numerical procedure. Because the Euler equations are hyperbolic,
the method of characteristics is used to specify the proper number of physical
conditions and the appropriate compatibility conditions at the boundary. For the
unsteady oscillations characteristic of combustion instability, we specify the in-
coming vorticity and entropy of the gas phase and the mass flux of both the liquid
and gas phases. Because we do not use an unsteady injector model, both the
liquid and gas flow rates are held constant. Semiempirical droplet size distribution
functions obtained from available atomization models are used to determine the
entering droplet diameters and velocities.10
At the downstream boundary, either a choked converging-diverging nozzle or
a "short" nozzle representing a choked flow boundary condition of fixed velocity
is coupled with downstream-running characteristics to compute the solution.Wall

boundary conditions are determined by specifying the normal component of ve-
locity to be zero and using four Riemann invariants. The spectral procedure auto-
matically ensures that periodicity is satisfied in the azimuthal direction.
Linearized Three-Dimensional Analysis

To complement the solutions to the complete nonlinear Euler equations, we
also obtain solutions to a linearized Euler system, using corresponding linearized
expressions for the physical submodels. To obtain these disturbance equations,
we separate the equations into mean and unsteady parts by writing the vectors
Q, E, F, G, and H (the sum of Hgas and H^q) in Eq. (1) as
Q(x, r, 9, t) = g(jt, r, 0} + Q'(x, r, 0, t) (10)
f
F = F+BQ (11)
H = H+DQ' (12)
Here, the overbar indicates that the quantities are to be evaluated at the mean flow
condition, whereas the superscript primes refer to the unsteady portion of the vari-
able. The specific form of the Jacobians is given elsewhere.10 By substituting Eqs.
(11) and (12) into Eq. (1) and retaining only linear terms, we obtain an expres-
sion that admits a separable solution of the form Q' = (pRZ, pw/?Z, pu/?'Z/A,
pwRZ'/Xr, eRZ)T for a one-dimensional mean flow with zero mean radial and
tangential velocities. Note that the caret quantities are functions of x and t only.
The radial dependence R is a Bessel function of integer order /z, R = J M (Xr),
where the wave number A. is determined by enforcing zero normal velocity fluc-
tuations at the wall. The tangential dependence is the trigonometric functions
Z = C\ cos(/z#) -h Ci sin(/z#), where IJL has been defined previously, and C\ and
€2 are constants.
The resulting equation for the unsteady perturbations becomes
_
where Q = (/3, pu, pv, pw, e)T and B is a modified Jacobian obtained from the
separation of variables procedure.10 The boundary conditions for the linearized
problem are analogous to those used in the full nonlinear solution.
Because of the variable coefficients, Eq. (13) cannot be solved in a closed-
form manner and must be solved numerically. The results represent full three-
dimensional (perturbation) solutions, but the computational effort is equivalent
to solving an unsteady, one-dimensional problem. This simplification, therefore,
provides dramatic advantages over the full nonlinear formulation. It provides an

efficient means of performing exploratory surveys of three-dimensional instability
prior to attempting nonlinear studies and can be used effectively as a diagnostic
tool. The linear calculations are also useful for validating the nonlinear numerical
procedure at small amplitudes.
Analytical Three-Dimensional Analysis

By further approximating the mean flow as constant, the three-dimensional
equation given by Eq. (13) can be solved in closed form nonetheless still retaining
distributed atomization-vaporization zones. These analytical solutions are very
useful for validating the numerical results and also provide intuition as to the
extent to which the proposed physical models influence stability. The analytical
approach can also be used to perform initial stability studies before more expensive
detailed investigations are carried out using the nonlinear analysis. The analyt-
ical solution procedure involves finding an eigenvalue relationship between the
complex frequency (period and growth rate) and wave numbers of the problem.10
Liquid Phase Computational Procedure

Before unsteady time-accurate computations can be performed, the steady-state
solution of Eq. (1) coupled with the liquid phase equations through the source term
//iiq must be determined. As each droplet parcel is introduced into the chamber,
the time history of the droplet lifetime can be determined from the integration of
the Lagrangian equations given in Eqs. (4) and (5). The droplet distribution and
initial velocities are specified at (or near) the injector face from an appropriate
atomization model. Steady-state mean flow solutions are obtained using a time-
marching iterative procedure by updating the liquid phase periodically until the
correct gas phase solution is obtained.
For the linearized numerical and closed-form analytical solutions where only
axial variations are permitted in the flow quantities, a purely axial mean Lagrangian
droplet field with no unsteady fluctuations in the droplet velocities is used to
determine the droplet locations. Variations in vaporization rate result from local
fluctuations in the gas phase pressure and velocity. For computational efficiency,
a similar approximation is used for some of the full nonlinear solutions, but
the nonlinear formulation also allows the complete fluctuating trajectories to be
incorporated. In the closed-form solutions, the mean vaporization zone is collapsed
to the injector face although the effects of vaporization fluctuations are considered
throughout the chamber volume.
Results
To demonstrate the capability of the models, we present in this section a series
of numerical results that start from restricted physics and run through increasingly
more realistic combustion chamber conditions. We begin by demonstrating the
level of accuracy that is attained by the computational scheme as a function of the
number of grid points. These demonstrations are based on comparing numerical
calculations with linearized analytical solutions. The full nonlinear procedures
are then validated at small amplitudes using single mode initial conditions from
the closed-form analysis. The effects of a distributed vaporization zone are also
presented to illustrate the effect of vaporization phenomena on the stability char-

acteristics. Finally, the results are extended to include cases which incorporate the
fully integrated Lagrangian liquid phase equations and variations in the mean flow
characteristics.
Grid Resolution Issues and Solution Accuracy

The accuracy of numerical combustion instability models is particularly im-
portant because it must be carefully established that disturbance growth or decay
in a CFD solution is the result of the physics of the model, and not because of
numerical instabilities or numerical dissipation. In general, the accuracy of numer-
ical stability solutions depends on the solution algorithm (which is second-order
central difference here) and the number of grid points per wavelength of the dis-
turbances. Appropriate grid resolution is particularly an issue in multidimensional
computations.
To demonstrate the accuracy that can be obtained with sufficient grid resolution,
Fig. 1 presents the results of the calculation of a single-mode oscillation using the
nonlinear analysis on a 201 x 51 finite difference grid in the (x, r) plane with three
Fourier modes in the 0 direction. This is noted by the nomenclature 201 x 51 x
(3 modes) in further discussions. The following computations are for a cylindrical
combustion chamber of length 0.36 m and radius 0.23 m which was used as
a baseline geometry. The mean vaporization effects have been collapsed to the
injector face to produce a constant gas flowfield comprised of H2/O2 combustion
products at a temperature of 3000 K and a static pressure of 40 atm. The unsteady
vaporization contributions are considered over a fraction of the vaporization length
which is determined by the mean droplet vaporization rate and injected droplet
characteristics.
Figure 1 compares the time trace of the numerical pressure oscillations with
the analytical solutions at the injector face/wall location at an angular location of
0.8
___ Numerical
___ Analytical
0.4
x
ICU
-0.4
-0.8
2 4 6 8 10
Nondimensional Time. tc/L
Fig. 1 Comparison of nonlinear numerical and analytical pressure fluctuation vs time
traces at injector face/wall location, 0 deg; IT,1R,1L single-mode initial conditions;
201 x 51 x (3 modes) grid; constant mean flow.
0 deg. The numerical solution is started from an initial condition that corresponds
to an analytically calculated 17\1/?,1L mode. As time progresses, the disturbance
experiences a steady growth of approximately 5% per cycle. The numerical so-
lution is indicated by the solid line, and the analytical result is shown with the
dashed line. The numerical results are quite accurate as the comparison in the
figure demonstrates. The only discrepancy is a slight underprediction of the ana-
lytical amplitudes by 0.3% per cycle of oscillation. Similar accuracy is obtained
for other locations in the combustion chamber. The results from the linearized
numerical solution (not shown) agree even more accurately because of the lower
dimensional nature of the CFD solution procedure.

Although the results shown in Fig. 1 demonstrate excellent agreement, this
level of accuracy is obtained only when adequate grid resolution is employed.
Figure 2 presents results for this problem on three different grid sizes to indicate
the manner in which the solution accuracy varies with grid refinement. This
figure compares the numerical growth rate of the positive pressure peaks envelope
for a three-dimensional pressure oscillation with the corresponding analytical
solution as a function of the number of cycles of oscillation (time) for grids of
1) 25 x 13x (3 modes), 2) 51 x 26x (3 modes), and 3) 201 x 51 x (3 modes)
points. For the sparse 25 x 13 x (3 modes) grid calculation, the numerical results
grow very slowly and severely underpredict the analytical growth rate. The denser
grids improve the numerical predictions dramatically, where the agreement for
the 201 x 51 x (3 modes) grid is excellent. This case shows an error of only 4%
after 15 cycles of oscillation. In general, some 75 grid points per wavelength are
2.5 -,
1o
<L>
"O
Analytical Solution
// 201 x51 x (3 modes)
3
51 x 25 x (3 modes)
1.5 -I
25 x 13 x (3 modes)
10 15 20
Number of Cycles
Fig. 2 Comparison of nonlinear numerical and analytical pressure fluctuation
envelopes, positive pressure peaks vs oscillation cycle for various grid densities; time
traces indicate injector face/wall location, 0 deg; IT,1R,1L single-mode initial condi-
tions; constant mean flow.
0 Degrees
36 Degrees
108 Degrees
162 Degrees
216 Degrees
& 3 06 Degrees
.S
Number of Cycles
Fig. 3 Phase errors between the numerical and analytical solutions vs oscillation
cycle; 201 x 51 x (3 modes) grid; injector face/wall location, 0 deg; ir,l/?,lL single-
mode initial conditions; constant mean flow.
needed to obtain reasonably precise solutions. Experience with higher order spatial
accuracy indicates this same accuracy can be obtained with about one-fourth as
many grid points when fourth-order differencing is used.
The corresponding dispersion errors for the fine grid solution of 201 x 51 x (3
modes) are presented in Fig. 3 for several tangential locations around the periphery
of the injector face/wall. The phase angle on the dense grid are again very small,
and result in an error of only one-tenth of a degree per cycle of oscillation. The
erratic nature of the phase angle curves in Fig. 3 arise because of the approximate
procedure used to calculate these very small errors. Clearly, the 201 x 51 x (3
modes) grid resolution is adequate for obtaining accurate solutions to the stability
model.
The preceding results are for a single-mode initial condition. A more pertinent
calculation is the response to an arbitrary initial disturbance which is introduced
in the common practice of "bombing" or "pulsing" the combustion chamber. To
demonstrate the capability of modeling such behavior, Fig. 4 presents a long-time
trace of the pressure fluctuation at the injector face/wall (0 deg) location using
an initial arbitrary, broad mode initial condition with disturbances in all three
coordinate directions. For this simulation, we pick combustion conditions such
that there is a range of unstable modes present. In the long-time solution, we
expect the most unstable mode to dominate. The short-time response in Fig. 4
indicates that the initial disturbance excites a wide range of modes in the chamber,
but the long-time solution produces a single predominant mode, as expected. A
power spectrum analysis of the temporal response indicates that the bulk of the
energy in the long-time solution is centered in a narrow frequency band around
6010 Hz, in excellent agreement with the analytical solution for the most unstable
frequency.
Stability Results with Constant Mean Flow

Having established representative numerical guidelines for ensuring accurate
solutions, we now investigate the effects of liquid phase processes on the stability
o.i ,——————,——————,—————,—————,
0.05
-0.05
-0.1
10 20 30 40
Nondimensional Time, tc/L
Fig. 4 Pressure fluctuation vs time for Gaussian pulse, mixed tangential standing
mode solution; 101 x 51 x (3 modes) grid; injector face/wall location, 0 deg; constant
mean flow.
behavior of the combustor. For our initial study, we consider a monodisperse

droplet size and a uniform initial axial velocity near the injector face that ignores
unsteady atomization effects. The Priem-Heidmann infinite conductivity model8
is used for describing the vaporization process. A constant mean flowfield is
specified so that closed-form analytical solutions can be obtained to assess global
stability characteristics and to validate the numerical calculations. The effects of
vaporization are taken to occur in a short length adjacent to the injector face, and
the liquid droplets are assumed to remain at their injection velocities throughout
their lifetime. For this problem, two distinct methods for treating the liquid droplet
temperature are considered. In the first, the liquid temperature is fixed at a constant
value. The second method considers droplet temperature oscillations in response
to changing heat transfer and vaporization conditions at the liquid surface.
A stability map outlining the trends in stability behavior for different drop
sizes and temperatures is shown in Fig. 5. These results were determined by
performing a parametric study using the linearized numerical analysis and closed-
form analytical solutions for liquid droplet temperatures that oscillate in time.
Positive contours in Fig. 5 indicate unstable growth of disturbances in percent
growth per cycle, and negative values indicate stable decay. The neutral stability
curve is indicated by the solid contour and separates the unstable and stable
regions which are labelled U and S, respectively. These results correspond to the
most unstable \T,\R,nL modes located at each operating condition. The triangular
upper bound of the stability is determined by the constraint that the droplets must
totally vaporize within the chamber. The minimum drop size computed was 20 /zm,
and for this droplet size the resulting vaporization length was 14% of the chamber.
The gas phase consists of H2/O2 combustion products at a temperature of 3000 K
and a pressure of 40 atm. The gas and liquid phase velocities are 150 and 100 m/s,
respectively.
The results in Fig. 5 indicate that in the presence of droplet temperature oscilla-
tions, smaller initial droplet sizes and colder initial droplet temperatures give rise
to less stable combustor behavior. As the mean droplet temperature is increased,
400r
"7.-5,
v
S~ -~
UnsieaoyAtomraoon
/ ^
s,)» «>;
Steady AtomuazJon b
~"^'
?i115 120 125 130 135 140 145
Droplet Temperature (K)

Fig. 5 Stability map as a function of droplet diameter and liquid temperature; vari-
able temperature droplets; three-dimensional gas phase lT,lRjiL instability modes;
rgas = 3000 K, Pgas = 40 atm.
stable combustor response is observed for the entire range of droplet sizes con-
sidered. For a given mean droplet temperature, larger drop sizes are more stable,
possibly because of the increased size of the vaporization region. As the drop size
is gradually decreased, the stability characteristics exhibit a minimum value for
moderate drop sizes, whereas small droplets predict increasingly stable behavior.
This may result from phase changes between the vaporizing mass (and energy
addition) and the acoustic pressure and velocity fluctuations. Efforts are currently
directed at investigating this behavior. If liquid droplet temperature variations are
ignored and a constant droplet temperature is used in the parametric solution, all
of the operating conditions shown in Fig. 5 result in stable behavior, regardless of
the droplet size or temperature.
These stability map predictions represent an excellent example of the manner
in which the three complementary approaches are used together. The eigenvalue
search routine of the closed-form analysis returns a combination of real and
spurious solutions, the latter of which are numerical in character and have no
counterpart in the physical solution. This search process is extremely efficient
and may be performed in a matter of several CPU minutes. The linearized nu-
merical solution is then used to distinguish the real eigenvalues and to determine
to which eigenfunctions (modes) they correspond. Although the linear numerical
procedure is sufficiently economical to do the entire stability map alone, this is not
practical because the linear solution may converge to alternative eigenvalues in
the neighborhood of its initial guess, and so might entirely miss the solution under
investigation. A stability map like that shown in Fig. 5 requires approximately 3 h
of CPU time (workstation), corresponding to perhaps 50 eigenvalue solutions. The
analysis is then completed by using the full nonlinear formulation to assess the
effects of various approximations required in the simpler solution characteristics.
By contrast, the solution of a single nonlinear calculation with a reasonably dense
grid may take about 60 CPU h on the same workstation.

a) Nondimensional Time, tc/L

0.001
• Numerical
Analytical
0.0005
P/P 0.0
-0.0005
-0.001
( 20
b) Nondimensional Time, tC/L

0.001
- Numerical
Analytical
0.0005
P'/P o.o
-0.0005
-0.001
c) Nondimensional Time, tcVL
Fig. 6 Comparison of nonlinear numerical and analytical pressure fluctuation vs

time traces at injector face/wall location, 0 deg; single-mode initial conditions: a)
variable temperature droplets f\\q = 125 K, d = 110 /xm, b) constant temperature
droplets f^q = 125 K, d = 110 ^m, and c) variable temperature droplets I^q = 125 K,
d = 163 /mi.
To demonstrate the capability of the models to predict temporal flow oscillations,

several pressure-time traces are shown in Fig. 6 for the identical mean conditions
used in Fig. 5. These results are predicted by the full nonlinear analysis, focussing
initially on small amplitude disturbances. The grid size for these comparisons is
again 201 x 51 x (3 modes).
Three different operating conditions are contrasted in Fig. 6. The first operating
condition shown in Fig. 6a specifies a monodisperse droplet size of 110 /zm with
a mean temperature of 125 K. The vaporization length for this condition is 50%
of the chamber length. (For these constant mean flow conditions, the effects of
vaporization were collapsed to the inlet in the mean flow in an averaged sense, but
the transient effects were distributed throughout the chamber.) The temperature
344 J. M. GRENDA AND C. L. MERKLE
of the droplets in this case fluctuates with oscillations in the gas phase. This
operating condition results in an unstable IT,IR,IL instability mode that grows
at approximately 6% per cycle. As before, the numerical and analytical solutions
are identified by solid and dashed lines, respectively.
The second operating condition considered in Fig. 6b is for an identical droplet
size to that in Fig. 6a except that the droplet temperature is held constant and
does not fluctuate. In contrast with Fig. 6a, the constant droplet temperature
treatment predicts a stable (here a decaying 17,1 R,2L) mode. The third operating
condition is again identical to Fig. 6a except that a larger drop size of 163 /xm
is injected which increases the vaporization length to coincide with the overall
combustion chamber length. As noted in Fig. 6c, the larger droplet size stabilizes
the combustion response which was originally unstable for the smaller initial drop
size. Note that for each of the operating conditions, excellent agreement between
the numerical and analytical solutions is obtained.
Stability Results with Variable Mean Flow

To this point, we have utilized a constant mean flowfield to permit the analytical
solution to be used as a guide in determining regimes of stable and unstable
behavior. The next logical step in the stability investigations is to gradually relax
the simplifying assumptions used in the previous solutions.
As the liquid propellant droplets move through the combustion chamber, they
vaporize and burn so that the amount of mass in the gas phase steadily increases.
The gas consequently accelerates in response to the additional mass and heat
release. It is important to assess what the effect these mean flow variations have on
the stability behavior and to note that the flow acceleration has an indirect effect
on stability through the changing mean flow conditions which impact the local
droplet temperature, velocity, and vaporization rate.
As a representative problem, we consider the case where the inlet gas flow is
composed purely of fuel and all of the oxidizer enters as liquid droplets. The
temperature of the entering fuel (hydrogen gas) is 1000 K whereas the oxidizer
(LOX) enters at 100 K. The Priem-Heidmann model8 is again utilized to model
the droplet vaporization and heat up. An upper-limit distribution function10 with a
mean droplet size of 130 /zm and standard deviation of 2 is used to set the inlet
droplet sizes. The axial component of the droplet velocities entering the combus-
tion chamber is distributed uniformly between the values of 100 and 140 m/s. To
be consistent with the linearized numerical solutions, the radial and tangential ve-
locity components are set to zero. A total of 8 droplet classes is used to model the
distribution function, and the liquid mass is partitioned equally among each class.
The overall oxidizer/fuel (O/F) ratio is chosen as 6, and the inlet static pressure
is 40 atm. An infinite rate chemistry model is used to produce water vapor as the
vaporized oxygen is mixed with the surrounding hydrogen gas.
Steady-state solutions of the gas/liquid flowfield are obtained by the coupled
Eulerian/Lagrangian method described earlier. In this example, the combustion
chamber is taken to be cylindrical with an aspect ratio (L/D) of 1.67 and the
steady-state flowfield is a function of the axial coordinate only because of the
constant area geometry. Droplet acceleration occurs because of relative velocities
between the gas and liquid phases.
The mean flow solutions for this example case are shown in nondimensionalized
form in Fig. 7 as a function of axial location in the combustion chamber. The
1.0 n
Water Mass Fraction — ~"
« 0.75 -
s 0.5 -
0.25 -
0.0
0.0 0.25 0.5 0.75 1.0
Nondimensional Axial Location

Fig. 7 Nondimensionalized steady-state gas phase values for coupled liquid phase
analysis; final mass flow rate m0 = 7.8 kg/s, O/F = 6, rinlet = 1000 K, Piniet = 40 atm,
^Hq, / = 100 K, d = 130 /im, u = 120 m/s.
nondimensionalized mass flow rate in the gas starts at a value of 0.16 and reaches
a maximum value of 1.0 at the exit where the dimensional mass flow rate is 7.8 kg/s.
The profiles of the H2O and H2 mass fractions show the rapid changes in gas phase
composition with chamber location. The H2O mass fraction increases rapidly from
zero and reaches the fully burned condition approximately half-way along the
combustion chamber length. The H2 mass fraction decreases correspondingly as
the fuel is consumed. The changes in gas flow rate and composition, as well the
corresponding changes in temperature, increase the Mach number from an initial
value of 0.05 to a maximum value of 0.12 at the chamber exit.
In contrast to the constant mean flow examples presented earlier, the variation
in the mean flow solutions for this problem does not allow closed-form analytical
solutions to be obtained. Consequently, single mode initial conditions are not
available and the stability response must be determined numerically by pulsing
the mean solution with an arbitrary disturbance. Here, we choose to pulse the
combustor with a disturbance that models a bomb or explosive charge.
The results in Fig. 8 indicate the long-time pressure response for such a pulsed
initial condition near the injector face. The pressure-time trace is given at the
injector face/wall zero-deg location. For this calculation, the droplet temperatures
and velocities remain at their local values at each axial location and are not
allowed to oscillate in time. In the short-time solution, Fig. 8 indicates that a broad
spectrum of frequencies is produced by the pulsed initial condition. In the long-
time solution, the pressure disturbance exhibits stable behavior for all generated
frequencies and gradually decays back to the uniform flow state. Comparisons
with other constant droplet temperature solutions indicates that this decay rate is
0.004 -,
0.002 -
0.0 -
-0.002 -
-0.004
10 20 30
Nondimensional Time, tc7L

Fig. 8 Pressure fluctuation vs time for Gaussian pulse for flowfield including mean
flow variations, 201 x 51 x (3 modes) grid; injector face/wall location, 0 deg; O/F = 6,
rinlet = 1000 K, P^ = 40 atm, rliq) , = 100 K, d = 130 /un, and u = 120 m/s, constant
droplet temperatures.
greater than that observed when no unsteady vaporization effects are present. The
unsteady vaporization, therefore, is an attenuating mechanism within the chamber
for this set of operating conditions. The stable pressure response in Fig. 8 is
qualitatively consistent with the behavior predicted previously by the approximate
analyses when the liquid droplet temperature was held constant. The effects of
mean flow variations in pressure and temperature increase the decay rate of the
pressure oscillations when compared to analogous uniform mean flow solutions.
Summary and Conclusions

Three synergistic levels of analysis have been presented for the investigation
of combustion instability mechanisms. This hierarchy of methods can be used in
an efficient fashion to predict stability regimes and global trends. This combined
numerical/analytical approach offers substantial insight and provides intuition
regarding the physical processes contributing to combustion instability. In addition,
computations can be validated through the companion closed-form solutions,
thus providing increased confidence in the numerical approaches. Stability maps
obtained by applying the hierarchy of solution procedures for constant mean flow
conditions indicate the droplet diameter and liquid temperature are important
variables in determining stability behavior. The addition of mean flow variations
does not change the qualitative response, although mean flow gradients appear to
be slightly stabilizing. The present results show that the numerical approaches of
the type introduced here represent a validated testbed upon which more realistic
modeling of physical processes can be incorporated.
Important additional work with the models remains to be done. Perhaps of most
interest is a careful issue-by-issue study of the effects of various phenomena on
the stability characteristics of a given engine. Following the procedures described
herein, these should include qualitative, global studies of the effects of various
parameters on stability using the two simpler analyses, followed by verification
and sensitivity testing with the full nonlinear analysis. Of particular interest are
improved unsteady droplet vaporization models that provide additional amounts
of time lag in the analysis and unsteady atomization effects that cause the drop
size to vary throughout the pressure fluctuation cycle and/or with the amplitude
of the disturbances. Any such analyses should be closely coupled to experimental

studies which are expected to provide the foundations for the required submodel
improvements.
References
âbiballah, M, Lourme, D., and Pit, F, "PHEDRE-Numerical Model for Combustion
Stability Studies Applied to the Ariane Viking Engine," Journal of Propulsion and Power,
Vol. 7, No. 1, 1991, pp. 322-329.
2
Kim, Y., Chen, C, and Ziebarth, J., "Prediction of High Frequency Combustion Insta-
bility in Liquid Propellant Rocket Engines," AIAA Paper 92-3763, July 1992.
3
Litchford, R., and Jeng, S., "Liquid Rocket Spray Combustion Stability Analyses,"
AIAA Paper 92-3227, July 1992.
4
Priem, R., and Breisacher, K., "Nonlinear Combustion Instability Model in Two- and
Three-Dimensions," NASATM-102381, October 1989.
5
Liang, P., Fisher, S., and Chang, Y., "Comprehensive Modeling of a Liquid Rocket
Combustion Chamber," Journal of Propulsion and Power, Vol. 2, No. 2, 1986, pp. 97-104.
6
Ingebo, R. D., "Dropsize Correlation for Cryogen Liquid-Jet Atomization," NASA
TM-102432, 1990.
7
Mayer,E., "Theory of Atomization in High Velocity Gas Streams" ARS Journal, Vol. 31,
No. 12, 1961, pp. 1783-1785.
8
Priem, R., and Heidmann, M., "Propellant Vaporization as a Design Criteria for Rocket
Engine Combustion Chambers," NASA TR-67, 1960.
9
Abramzon, B., and Sirignano, W., "Droplet Vaporization Model for Spray Combustion
Calculations," Journal of Heat and Mass Transfer, Vol. 32, No. 9, 1989, pp. 1605-1618.
1()
Grenda, J. M., "Computational Analyses of Combustion Instabilities in Liquid Propel-
lant Rocket Engines," Ph.D. Dissertation, Department of Mechanical Engineering, Penn-
sylvania State Univ., University Park, PA, Dec. 1994.

Chapter 14
Performance and Stability Characterization of Liquid

Oxygen/Kerosene Injectors at Aerojet
Jerry Pieper,* Thong Nguyen,f and James Hulka*

PRA-SA-APP-OSD 94-S-0469
I. Introduction
H YDROCARBON propellants, and in particular liquid oxygen/kerosene

(LOX/RP-1), have been used extensively for booster propulsion systems
since the beginning of the Space Age. Most of the rocket engines for these
systems were developed using trial and error methods with little emphasis on
understanding the combustion processes that influenced the injector performance
and combustion stability. During the last two decades, however, many research
and combustion technology efforts were conducted in the U.S. to improve the un-
derstanding of LOX/hydrocarbon combustion phenomena. This paper reviews the
results of those technology programs conducted at Aerojet that used LOX/RP-1
propellants, with particular emphasis on determining the impact on performance
and stability of the geometric and fluid dynamic parameters that affect the spray
combustion processes.
II. Historical Background

Development of rocket engines within the U.S. that used LOX and hydrocarbon
(HC) propellants began with the earliest research programs that emerged after
World War II. Booster and sustainer rocket engines were developed for rocket
launchers such as Jupiter, Thor, Atlas, Titan I, Saturn I, and Saturn V specifically to
use RP-1 fuel, a kerosene as specified in Military Specification MIL-P-25576C. All
of these new rocket engines were developed at full scale using highly empirical cut
and try methods. Chamber diameters ranged from 15 to 40 in., chamber pressures
from 500 to 1100 psia, and engine sea level thrusts from 60 to 1540 klbf.
The combustion stability of these engines that utilized mostly axial, like-
impinging injector elements was achieved without the use of injector face baffles.
Flow rate per element ranged from moderate (~1 Ibm/s) to very large (~8 Ibm/s).
rights reserved.
*Manager, Product Analysis. Member AIAA.
technical Principal, Product Analysis.
^Senior Engineer, Strategic and Space Propulsion. Member AIAA.
349
350 J. PIEPER ET AL
Combustion was statistically stable, i.e., stable without being subjected to artificial
perturbations (intentionally bombing for stability rating was not yet developed).
Injector performance efficiencies were low to moderate (94—97% characteristic
velocity (c*) efficiency) except for the Titan I injectors that achieved high perfor-
mance (98-99%) with lower flow rate per element (~1 Ibm/s).
This early LOX/HC engine development is interesting from a historical per-
spective but is of limited use because the data are sketchy or nonexistent, having
been generated during intensive engine development programs with little emphasis
on development of fundamental liquid propellant rocket combustion technology.
Also, much of the data that was generated is no longer available.

Aerojet participated in this initial period with development of the Titan I weapon
system, a two-staged missile developed for the Air Force.1 Engines for both
stages (LR87-AJ-3 and LR91-AJ-3) were pump fed and powered by fuel-rich gas
generators. Two first-stage thrust chambers delivered 149,400-lbf thrust each at sea
level whereas a single second-stage thrust chamber delivered 80,000-lbf thrust at
vacuum conditions. Thrust chambers for both stages used dish-faced injectors with
like-on-like (LOL) doublet impinging elements. The first- and second-stage thrust
chambers operated at chamber pressures of 585 psia and 682 psia, respectively,
and both produced high-performance efficiency (98-99% c* efficiency). Neither
thrust chamber contained combustion stability damping devices, such as baffles
or acoustic cavities, and each operated with stable combustion in over 100 full
duration tests and many flights conducted during the period from 1960 to 1963.
The Titan system was then converted to storable propellants to improve its launch
readiness as an intercontinental ballistic missile (ICBM).
A second period of evaluation of LOX/RP-1 propellants for liquid rocket ap-
plication at Aerojet occurred during the late 1970s and early 1980s.2"7 During this
period, the development of combustion technology using LOX/HC propellants
was emphasized rather than the development of flight rocket engines themselves.
A variety of hydrocarbon fuels, including RP-1, propane, methane and liquified
natural gas, JP-20, methanol, and ethanol, was investigated. Preburner or gas gen-
erator combustors, as well as main combustors, were tested, along with a variety
of injector element concepts including showerhead, LOL doublet, concentric tube
(coaxial) with and without swirl, impinging triplet [either two fuel on one oxidizer
(FOF) or two oxidizer on one fuel, (OFO)], quadlet, and unique self-atomizing
elements (X doublet, splash plate, and transverse LOL doublet) constructed from
stacks of thin bonded platelets. Smaller combustion chambers (1000-40,000 Ibf)
were used with chamber pressures evaluated as high as 3000 psia.
In most cases, the combustion stability of these combustors was enhanced us-
ing near injector face acoustic resonators (IFAR). Moderate- to high-performance
efficiency (~97-99%) was obtained using lower flow rate per element than was
typical of the earlier engine development period and unlike-propellant impinge-
ment configurations. An overview of this era's technology developed at Aerojet
and elsewhere is presented in Ref. 8.
By the mid-1980s, most of the LOX/HC combustion technology work was di-
rected toward preliminary development of a low-cost, heavy lift launch system for
the U.S. Strategic Defense Initiative (SDI). Hydrocarbon fuels of specific interest
for SDI included RP-1, propane, and methane. By 1987, however, LOX/hydrogen
propellants were selected for the future National Launch System (NLS), and
the combustion technology work with LOX/HC propellants became less critical.
LIQUID OXYGEN/KEROSENE INJECTORS AT AEROJET 351
Fortunately, existing LOX/HC programs were allowed to continue, some to com-

pletion. A substantial and well-documented LOX/HC combustion database was
generated at Aerojet through these efforts.9'12
Two of these programs developed and expanded methodologies to characterize
the combustion stability and performance of large LOX/HC rocket combustion
chambers. The first of these programs9 developed a characterization methodology
that combined analytical performance and stability models with hot-fire testing in
strategically reduced size combustion chamber hardware to characterize full-scale
engine performance and stability. The second program10 expanded the analytical
capability by combining existing industry-wide performance and stability models

into a single computer code, the rocket combustor interactive design (ROCCID).
The code provided the user with an interactive tool to design a combustor and
evaluate desired performance and combustion stability margins. Both of these
programs designed hardware using their respective characterization methodolo-
gies, which were hot fired with LOX/RP-1 propellants. Combustion instabilities
were successfully characterized with the methodologies, and successfully damped
with baffles and acoustic resonators. An OFO triplet element achieved high per-
formance (—99-100% c* efficiency) with a high flow rate per element (~4 Ibm/s)
(Ref. 9).
Another program evaluated the feasibility of designing a 750-klbf, pressure
fed booster engine injector using an array of smaller injector modules.11 A 150-
klbf subscale injector consisting of 12 of the injector modules was designed
using the ROCCID methodology to characterize the performance and combustion
stability. Hot-fire testing was conducted over a range of operating conditions with
a variety of acoustic resonator configurations. A core injector using OFO triplets
with a flow rate per element higher than the F-l obtained high performance (97-
99% c* efficiency) and was (dynamically) stabilized using an appropriately tuned
resonator.
A summary of these programs including pertinent design and operating char-
acteristics is provided in Table 1. Parameters for RP-1 injection only are provided
since it is the rate controlling propellant for these systems.
III. Visualization of LOX/HC Combustion Sprays

One of the first LOX/HC technology programs conducted at Aerojet at the
start of the second era of LOX/HC propellant evaluation was directed at the
visualization of combusting sprays.2 Figures 1-4 show photographs from single-
element combusting sprays of the four primary injection elements used at Aerojet
for LOX/RP-1 combustion. Figure 1 shows the LOX/propane spray and near
combustion field resulting from a pair (one each fuel and oxidizer) of LOL doublet
elements that are commonly used for impinging element injectors. Figure 2 shows
an OFO impinging triplet element LOX/RP-1 combustion field above supercritical
chamber pressure. These photographs clearly show that vigorous combustion is
occurring as soon as the jets or fans impinge. The presence of this early combustion
shows the critical importance of obtaining good mixing within an individual
element spray itself. Orifices for both the LOL doublet and OFO triplet elements
shown in Figs. 1 and 2 were produced using the electron discharge machine (EDM)
process.
£
10
Table 1 LOX/RP-1 thrust chambers designed and tested at Aerojet
RP-1 RP-1
Vacuum Chamber RP-1 orifice orifice Stability
Engine thrust, pressure, flowrate, diameter, Reynolds Element c* damping Reference
system Ibf psia Ibm/s in. no. type efficiency devices no.
Titan I 172,000 585 175 0.055- 04 x 105 LOL doublet 98% None 1
LR87-AJ-3 0.082 c_
Titan I 80,000 682 74 0.039- 05 x 105 LOL doublet 99% None 1 33
LR91-AJ-3 0.057 m
T3
LOX/HC 50,000 2,000 45 0.058 0:7x 105 LOL doublet 96% Acoustic 9 m
ID
methodology cavities or
baffles ^
LOX/HC 50,000 1,800 39 0.125 1 4 x 105 OFO triplet 99-100% Acoustic 9 r~
methodology cavities or
baffles
ROCCID 50,000 1,250 40 0.090 08 x 105 OFO triplet 92-96% Acoustic 10
validation cavities
Pressure-fed 150,000 720 158 0.250 1 5 x 105 OFO triplet 97-99%a Acoustic 11
booster cavities
a
Adjusted to exclude inefficiency resulting from intentional fuel film cooling usage.
LOX
PC = 800 psia O/F = 2.90
Fig. 1 View of the combusting spray from opposing LOX and propane like-on-like
doublet elements (from Ref. 2).
Figures 3 and 4 show combusting sprays from two unique injection elements
created at Aerojet using platelets. The flow passages in these injectors are cre-
ated by etching features into thin metal sheets that are then diffusion bonded
together. The two elements shown replicate the LOL and FOF triplet element
design features, but provide early atomization of the individual jets using internal
impingement on solid surfaces or with similar like-propellant streams, as shown in
Figs. 5 and 6. Thus, for the preatomized triplet (PAT) element, showers of droplets
of unlike-propellants impinge, rather than coherent jets.
LOX
RP-1
LOX
PC = 1505 psia O/F = 2.60

Fig. 2 View of the combusting spray from a single LOX/RP-1 OFO triplet element
(from Ref. 2).
354 J. PIEPER ET AL
LOX
RP-1
PC = 800 psia 0/F = 2.8
Fig. 3 View of the combusting spray from a LOX/RP-1 transverse like-on-like dou-
blet pair element (from Ref. 2).
C3H8
LOX
C3H8
PC = 800 psia 0/F = 2.90
Fig. 4 View of the combusting spray from a LOX/propane FOF preatomized triplet
element (from Ref. 2).
Element Element
Inlet Inlet
Injector—*
Face
Fig. 5 Illustration of the platelet TLOL doublet element.
Fuel Element Oxidizer Element Fuel Element

(Manifold Side)
Fig. 6 Illustration of the platelet FOF PAT element.

356 J. PIEPER ET AL.
Figure 4 shows a partially separated combusting flowfield for this element that
may have been caused by premature combustion due to the enhanced, preimpinge-
ment atomization. This separated flowfield, or striation of the oxidizer and fuel
fans, is known as reactive stream separation (RSS), or blowapart, a combustion
phenomenon originally observed with hypergolic propellants.12
FV. Combustion Efficiency

Combustion performance efficiency in early LOX/RP-1 rocket engines was low
to moderate (^94-97%) because of the use of mostly axial (i.e., noncanted),

like-impinging injection elements, and the inclusion of injector face baffles for
combustion stability. Notable exceptions were the Titan I injectors that achieved
high performance with low flow rate per element, like-impingement elements, and
did not include baffles. In the past two decades, the switch to unlike-propellant
impingement elements, such as OFO triplets, has not uniformly resulted in high
(~98—100%) performance, although the use of injector face acoustic resonators
instead of baffles has reduced the performance loss due to damping devices. For
unlike-propellant impingement elements, mixing occurs in the liquid phase rather
than the vapor phase. With an explosive propellant combination such as LOX/RP-
1, this type of mixing may be sensitive to many parameters, which may explain
the varied combustion performance and stability achieved.
In all of these LOX/RP-1 combustors, there are two main sources of combustion
performance loss. One is the chemical energy loss due to macroscale nonuniform
propellant distribution that may be created intentionally for chamber and baf-
fle wall cooling or unintentionally due to injector manifolding deficiencies. The
second is due to the inefficiencies of the spray combustion process.
For LOX/RP-1 propellants, the combustion inefficiencies of the spray combus-
tion processes can be attributed primarily to insufficient intraelement mixing and
incomplete RP-1 vaporization. At low (fuel-rich) mixture ratios, the combustion
efficiency is additionally limited by low chemical kinetic reaction rates, resulting
in the formation of solid carbon. Oxygen vaporization has not been a significant
factor in LOX/RP-1 combustion inefficiency because its vaporization rate is much
faster than RP-1 with conventional impinging element designs. The evidence for
these general observations is discussed in the following sections using results from
several LOX/RP-1 combustion technology programs conducted at Aerojet.
A. RP-1 Vaporization Limitations

Although the vaporization rate of liquid droplets has not yet been measured
directly in a liquid propellant rocket chamber, the effect of global vaporization
inefficiency can be derived from hot-fire testing. One method is testing with thrust
chambers of different chamber lengths. Since most propellant mixing occurs in the
upstream region near the injector face where the flow is highly turbulent, mixing
is assumed not to be a factor when performance changes with increased chamber
length. This principle was observed in hot-fire testing of gaseous oxygen/hydrogen
rocket injectors.13 Hence, a measured difference in performance when chamber
length is changed can be attributed to a change in vaporization efficiency. An
example of this method, taken from Ref. 3, is shown in Fig. 7, where the differences
in efficiency between tests 056 and 070 at any mixture ratio are attributed to the
difference in RP-1 vaporization level, assuming the mixing efficiency is the same
100
100%FVap
99 —— Mixing Loss o
73% Em (Constant) o-
98 99% F Vap
o
97% F Vap
O 97
A Experimental C* Efficiency
96 ~0— Predicted C* Efficiency

L T
' FJ
95 Test No. (cm)(in.) (°K)(°F) Eg %F Vap
056 27.9(11) 283(49) .73 92
070 37.5(15) 293(67) .73 97
94
2.0 2.5 3.0 3.5 4.0 4.5
Mixture Ratio
Fig. 7 Correlation of propellant vaporization and mixing efficiencies with test mea-
sured performance efficiency (from Ref. 3).
for both tests. The good match between test data and prediction show that the
assumption of constant mixing efficiency is satisfactory.
The global vaporization efficiencies can also be derived from measuring the
axial static pressure profile in a thrust chamber, an example of which (from Ref. 9)
is shown in Fig. 8. A vaporization model is required to calculate the creation of
propellant vapors that are then reacted in a steady-state combustion model to match
the static pressure profile as a function of axial distance. Carrying the prediction
out to the throat plane determines the total vaporization efficiency.
The influence of RP-1 vaporization on overall efficiency can most easily be
demonstrated by the Priem generalized length (Lgen) vaporization parameter.14
This parameter is shown in Fig. 9 for LOL doublet elements.3'9 Despite the presence
of other phenomena affecting the combustion efficiency in the data, such as mixing,
the Lgen parameter is found to correlate with the combustion performance of these
types of elements. Elements that have more influential mixing effects, such as OFO
triplets, did not show such a clear correlation with Lgen.
B. Injector Element Mixing Limitations

The mixing rate of impinging liquid jets or fans cannot currently be measured
directly in a liquid propellant rocket chamber. Consequently, the mixing rate
is either predicted from unielement cold-flow mixing data using an empirical
correlation relating the hot-fire multielement mixing effect, or derived from the
observed mixture ratio sensitivity of injector performance in which the macroscale
mixture ratio maldistribution effects have been factored out.
Unielement mixing efficiencies are currently measured in cold flow because
hot-fire measurements are not practical, although improvements in laser-based
diagnostic techniques may soon make them possible. Typically, a patternator test
with nonreacting liquid/liquid simulants is used to measure an intraelement mixing
efficiency for LOX/RP-1 injection elements.
358 J. PIEPERETAL
1700
Test 5, O/F = 2.77
ROCCID Prediction
1650
" 1600
Q.
1550 Lox/RP-1
0) O-F-O Tripiet Elements
20 in. Chamber Length
1500
Test 7, O/F = 3.57
1450 ERE = 100%
CO ROCCID Prediction
1400
1350
2 3 4 5 6 7 8 10
Axial Distance From Face, in.
Fig. 8 Correlation of propellant vaporization efficiency with measured axial chamber
pressure profiles (from Ref. 9).
An example of deriving injector mixing performance from the observed mixture

ratio sensitivity is provided in Ref. 3 and is shown in Fig. 7. In this example, the
experimental data of two variable mixture ratio tests with the same injector were
modeled with a constant two-zone multielement mixing efficiency13 and varying
vaporization efficiencies. A unique mixing loss was determined by matching the
shape of the measured curve, while not exceeding 100% vaporization efficiency.
In this particular case, the mixing efficiency was not a function of the mixture
ratio, as shown subsequently, but may be for other element types.
1.05 1.05
>1.00 1.00
0)
5 0.95
£ <D
$ 0.90 0.90 o
£
CO
DC 0.85 1 0.85
>
0.80 0.80
LU
0.75 0.75
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Lgen Parameter
Fig. 9 Effect of the generalized length parameter on LOL injector performance.
I.UO " • i.uo

a
X
I .UU
o "
Q) S^m * *£«< ^^ ** * °
/°»« »«>lh' â « *
... . .. .................. ..... .............jg...^... ....... — 1*~ ..............................................
.S> 0.95 0.95 >.
£ / f
$ 0.90 * -0.90|
(Q 1 OFO Triplet Injector
UJ
........../................„........„...._........._._.......„.... Design From Ref. 9 ......
& 0.85 0.85 ^

L orifice
S> / \ . . - •»»" - ---»•
g 0.80 "0.80
LU / \:alculated With Nonreacting Data from Ref. 15
1.0 2.0 3.0 4.0 5.0 6.0

O/F Momentum Ratio
Fig. 10 Effect of O/F momentum ratio on OFO injector performance.
Multielement mixing efficiencies can also be derived from the axial static pres-
sure profile, such as Fig. 8. After the total vaporization efficiency is determined, as
described earlier, the remaining performance loss is then attributed to incomplete
mixing.
For unlike-propellant triplet elements, unielement mixing is often described
as a function of the oxidizer-to-fuel momentum ratio. This relationship can be
determined in nonreacting flow tests.15 As shown in Fig. 10, OFO triplet elements
from a variety of programs7'9'11 were found to have obtained maximum injector
performance over the same range of momentum ratios for which the maximum
mixing efficiency was obtained in nonreacting flow tests.15
However, proper spacing of impinging elements is also important for obtaining
optimum combustion performance. With proper spacing, a large orifice (0.125
in. diameter) OFO triplet element with moderate unielement mixing efficiency
(80%) obtained high overall combustion efficiency (97-99%) (Ref. 9); however,
a smaller orifice (0.090 in. diameter) OFO triplet with much closer spacing had
the overall combustion efficiency significantly reduced (92-96%) (Ref. 10). This
performance drop occurred at higher momentum ratios, as shown in Fig. 11, where
high performance with similar elements, shown in Fig. 10, had previously been
observed. This result may be due to interference with neighboring elements16 from
the excessive spreading of the fan edges at the higher outer-to-inner momentum
ratios.
Unlike-propellant impinging elements with LOX/HC propellants have also been
observed to display RSS, as noted in the LOX/HC combusting spray visualization
studies.2 This phenomenon that results in striation of the oxidizer and fuel fans
increases as the unlike impingement becomes more parallel.
C. Kinetic Limitations at Low O/F

At fuel-rich mixture ratios, RP-1 combustion can be rate limited by reaction
kinetics. The large diluent of unreacted RP-1 vaporizes and undergoes a two-step
gas-phase thermal decomposition, the net effect of which is a reduction in gas
360 J. PIEPER ET AL
l.UO l.UO
I *
>. 1.00 ^.o—— —^- 1.00
.2 o"/' ° *
« 0.95 0.95
* / D Q * a
/ M "* a»
0)
$ 0.90 ' 0.90 o
E
(0
0 a/ ° OFO Triplet Injector
. ..... ............................................................_ Q0g|gp prom Ref. 10 "**••
CC 0.85 r
0.85
o L Dor|fjce = .090 in.
x_ .......L.T\... ... ............................. ......
g 0.80 M
h \ '" '"" •••••••———•---"""--"• 0.80
LU / ^Calculated With Nonreactlng Data from Ref. 15
0.75
1.0 2.0 3.0 4.0 5.0 6.0
O/F Momentum Ratio
Fig. 11 Intraelement mixing may have limited the performance efficiency of an OFO
triplet injector.
temperature. At mixture ratios less than one, LOX/RP-1 combustion gas temper-
atures are substantially lower than the standardized JANNAF one-dimensional
equilibrium (ODE) combustion model17 prediction, as shown in Fig. 12.7 This
phenomenon does not occur at oxidizer-rich mixture ratios.6
The appearance of solid carbon strongly depends on mixture ratio, and was
found experimentally to occur at these fuel-rich mixture ratios. This mechanism
has been modeled at Aerojet using the fuel-rich combustion model (FRCM).7
Figure 13 shows the production of carbon in a liquid rocket plume as a function
of mixture ratio.7 No carbon appears in the plume at nominal main combustion
chamber mixture ratios,7 nor at LOX-rich mixture ratios.6 Carbon deposition in
j. 1800
CO
o>1600 THighT
FRCM J A v g T
Prediction-L Low T
1400
1200 D P c s1500psia
O P c =1000psia
= 1000
A Pc = 750 psia
o A
O 800
0.2 0.3 0.4 0.5 0.6

LO2/RP-1 Mixture Ratio, W ox/Wf
Fig. 12 LOX/RP-1 gas temperatures are kinetic rate limited at low mixture ratios
(from Ref. 7).
3000 MR = 0.55
LOX/RP-1
Pc = 1000psia
MR = 0.35
2500
2000
CL
1500
1000
o
O
Tc = 6420 °F
500 J_ _L _L I
0.1 0.2 0.3 0.4 0.5 0.6 0.7
LOX/RP-1 Mixture Ratio, W ox /W f
Fig. 13 Production of carbon in a LOX/RP-1 plume as a function of mixture ratio
(from Ref. 7).
some historical main combustion chambers, then, was the result of such poor local
mixing efficiencies (either intentional or unintentional) that local mixture ratios
were less than one even though the overall mixture ratio was much higher.7
V. Combustion Stability
Combustion instabilities in a liquid propellant rocket combustion chamber are
characterized by well-organized, large-amplitude oscillations of the chamber pres-
sure. They result from periodic energy addition that is in resonance with the
chamber pressure oscillations. Three criteria must be satisfied for combustion
instabilities to occur: 1) the addition of energy must be periodic, 2) the energy
addition must be in phase with the pressure oscillations, and 3) the energy addi-
tion rate must be greater than the damping rate. The sources for periodic energy
generation can only come from the combustion process. The oscillations in the
combustion rate, however, can result from one or more of the physical and chemi-
cal processes taking place prior to and during combustion. Some of these processes
are inherently steady, some are inherently periodic, and some are normally steady
but can become periodic if a parameter driving the process is periodic.
There are two general types of combustion instabilities. The first is a non-
acoustic, or chug, instability caused by coupling between the propellant injection
process and the chamber gas dynamics. The chamber characteristic length is
much smaller than the oscillation wavelength. Mechanisms causing this type of
instability are well understood, and the instabilities can be eliminated usually by
increasing injection pressure drop or reducing element orifice size.
The second type of instability is acoustic, where oscillation frequencies match
one or more acoustic resonance frequencies of the combustion chamber. These
instabilities can damage hardware. Unfortunately, the detailed mechanisms that
362 J. PIEPER ET AL
cause acoustic instabilities are not well understood, so they are more difficult to
control or avoid during engine development.
The effects on combustion stability of different element types [OFO, LOL,
TLOL, PAT], element sizes, chamber sizes, chamber shapes (cylindrical or rectan-
gular), operating conditions (chamber pressure, mixture ratio, injection pressure
drop or injection velocity), and damping devices (baffle, monotuned and bituned
cavities) were investigated at Aerojet in recent technology programs.3-12 These
effects are discussed in the following sections.
A. Nonacoustic or Chug Instability

Because of the use of high injection pressure drop to ensure a large throttling
range, chug instabilities were seldom encountered on LOX/RP-1 programs, even
with a variety of element types, element sizes, and chamber sizes. Chamber pres-
sure reductions to 50% of nominal,11 40% of nominal,9 and 15% of nominal10
failed to produce chug instabilities, even with mixture ratios from main chamber
injectors as low as 0.54.
A few tests briefly encountered chug instabilities during the early part of
the startup transient where the chamber pressure is extremely low. The insta-
bility, however, disappeared quickly as the mean chamber pressure rose to steady
state.
B. Acoustic Instability
The damaging nature of acoustic instabilities led to the development of rating
techniques and criteria18 that could compare stability margins on some basis other
than statistical test history. Many recent LOX/RP-1 test programs at Aerojet9"11
used the artificial perturbation, or bomb technique, to rate combustion instabil-
ity. The use of the bomb technique for LOX/RP-1 combustors was to induce a
phenomenon (a pop) that LOX/RP-1 combustion seemed to display inherently.
The combustion was judged unstable if the induced oscillations did not damp
within a certain time after the induced disturbance,18 or if hardware was damaged.
However, the bomb completely altered the landscape, even temporarily, on which
the spray combustion processes operated. The process judged unstable may have
nothing to do with normal operation of the combustor. For this reason, correlation
of combustion stability with bombed unstable data is difficult.
Early LOX/RP-1 combustors, such as those on Titan I, that used mostly axial,
like-impingement injection elements, were not bombed, but operated stably in
many hundreds of full-duration tests even without stability aids. Their bombed
stability characteristics are unknown. In a recent program, however, LOL el-
ements, although stable during normal operation, could be bombed unstable,9
whereas other element types were found to be spontaneously unstable.3'9 Unlike-
impingement element injectors in recent programs were also found to be sponta-
neously unstable,9-10 whereas others were stable until bombed unstable.11
In general, differences between programs and tests involved changing more than
one significant parameter. As a result, the effects of each parameter on combustion
stability could not be established with certainty. The following discussion describes
the general trends of the effects of engine design, damping devices, and operating
parameters on acoustic combustion instabilities.
L Effects of Element Type and Size

Triplet element designs for LOX/RP-1 injectors at Aerojet have used both the
OFO and the FOF configurations. The OFO is more optimized for LOX/RP-1
propellant's normal operating momentum ratio. Some of the triplet injectors were
fabricated by EDM methods,7'9"11 and some used platelets.3'4 Orifice size among
all of the programs varied by an order of magnitude.
Stability characteristics were compared from three programs that tested com-
bustors without stability damping devices.9"11 As the orifice diameter increased,
the combustion instability initiation characteristics changed from spontaneously
unstable to only capable of being initiated by a combustion perturbation bomb.

Thus, a larger diameter orifice element is found to be relatively more stable even
though the chamber diameter is much larger, which provides for lower acous-
tic frequencies. Larger orifice elements have been previously shown to be more
stable,19 but generally at the expense of reduced performance efficiency.
A moderate-size OFO triplet element was also found to be susceptible to spon-
taneous instability over a wide range of frequencies.9 The same element type and
size is responsible for the spontaneous instability in the first tangential (3500 Hz)
and higher modes (6900 Hz) of a 7.68-in.-diam chamber, the first width mode
(1600 Hz) of a 14.95-in.-wide rectangular chamber, and the first tangential mode
(1200 Hz) of a 17.53-in.-diam chamber with short (1 and 3 in. high) baffles.
A relatively small OFO triplet element was found to be susceptible to sponta-
neous instability at several resonance modes simultaneously.10 The primary res-
onant frequencies in most of the unstable combustion tests corresponded to the
first tangential (IT), acoustic frequency of the combustion chamber, which was
approximately 3300 Hz. On some tests, such as those shown in Fig. 14, multiple
FS1 +1.37-1.40 sec

KFN7-DO1-1J -014
Window = None
Distinct Avg = 5
Bandwidth = 165
Variance = 2.943E+05
O
101
101-
15 30 45 60 75 90 105 120
2
Frequency (10 )
Fig. 14 Power spectral density analysis of chamber pressure showing the existence
of IT harmonic (from Ref. 10).
364 J. PIEPERETAL
75.0
KFN7-D01-1J-021
Window = None
62.5 Distinct Avg - 1
Bandwidth = 88.9
Variance = 2.425E+06
^50.0
co
37.5
.-2
c
(^25.0
12.5
J____L J\
10 20 30 40 50 60 70 80
2
Frequency (10 )
Fig. 15 Power spectral density analysis of chamber pressure showing a large 2T
instability in an undamped combustion chamber (from Ref. 10).
resonant frequencies including the second tangential (2T) and possibly a harmonic
of the IT mode were observed using the power spectral density (PSD) analysis of
the measured chamber pressure. In another test, shown in Fig. 15, a dominant 2T
instability was clearly encountered.
There were no combustion instabilities in the other OFO triplet programs,3'4'7 but
all employed injector face acoustic resonators and were never tested without them.
Thus, the comparative stability characteristics in smaller chambers with smaller
orifice diameter elements and with platelet elements could not be determined.
LOL doublet element designs for LOX/RP-1 injectors at Aerojet have been
fabricated with EDM methods9 or with platelets.3 A similar comparison between
combustion instability initiation characteristics and orifice diameter is observed
for these elements. Spontaneous instabilities occurred in a 4.8-in. chamber with the
platelet transverse LOL element, even with acoustic resonators,3 and in a 14.95-in.-
wide rectangular chamber with EDM doublets without stability aids.9 With larger
EDM orifice diameter doublets, however, instabilities were only initiated in a 9.2-
in.-diam chamber without damping devices by use of a combustion perturbation
bomb. Thus, a larger diameter orifice element is found to be relatively more stable
for LOL doublets also.
A comparison between the combustion stability of LOL doublet elements and
OFO triplet elements based on this data, however, has been inconclusive. Both
types of elements displayed spontaneous instabilities with smaller orifice sizes
and susceptibility only to bombs with larger orifice sizes. With comparable orifice
sizes or flow rate per element, however, OFO triplet elements appear to be less
stable, although they are higher performing.
2. Effects of Damping Devices

The effectiveness of damping devices was clearly demonstrated during the
Oxygen/Hydrocarbon Injector Characterization Program.9 Test results showed that
the combustion was dynamically (bomb) stable in a chamber with either bituned
cavities or baffles, or both, but spontaneously unstable in a chamber with neither
cavities or baffles.
An incorrectly tuned cavity may not provide sufficient damping to prevent
instabilities, as was found during ROCCID verification testing.10 Although the
amplitudes and growth coefficients of the chamber pressure oscillations were
generally reduced in tests with cavities, indicating that some damping was provided
to the system, the instabilities were not completely suppressed.
Even with correctly tuned cavities, instabilities occurred that were severe enough
to cause significant face erosion.3
Destabilizing effects of acoustic cavities were also observed.9 Test results
showed that the combustion was spontaneously unstable in a chamber with cav-
ities, but bombed stable in a chamber without the cavities. The cavities might
have lowered the chamber acoustic frequencies closer to the driving frequencies
of instability.
3. Effects of Operating Conditions

To determine the effects of operating condition, only stable, bomb stable, and
spontaneously unstable tests were examined. Preliminary attempts to correlate
bomb unstable test data where no spontaneous instabilities occurred have been
inconclusive. This may be due in part to the profound effect, discussed earlier,
that the bomb has on the spray combustion processes. In addition, only tests
using configurations without damping devices were used. Data from Ref. 10 were
included because the acoustic cavities, although present, had negligible damping.
800
stable
700
Unstable
600
2 500
"to
0.
Q 300
200
100
0 200 400 600 800 1000 1200

DPF, psld
Fig. 16 Correlation of combustion stability observations of an OFO triplet injector
as a function of injection pressure drop (from Ref. 10).
366 J. PIEPER ET AL.
500
a Stable
v~
• Unstable
450
400
2
.
£ 350 •+**
^•^
0. o
o oV
300 ———i
^te^r~
a
250 LJ
200
3()0 350 400 450 500 55
DPF, psid
Fig. 17 Correlation of combustion stability observations of a TLOL injector as a
function of injection pressure drop (from Ref. 3).
Oxidizer and fuel injector pressure drops for an EDM triplet injector10 are shown
in Fig. 16. A distinct stability boundary is found with the unstable region in the
area of high simultaneous fuel and oxidizer pressure drops.
A similar relationship is observed for a platelet like-on-like doublet injector,3
as shown in Fig. 17. These instabilities were all spontaneous, and all tests, except
the first two, had the same damping devices. This map shows a distinct unstable
region at high simultaneous pressure drops that is remarkably similar to Fig. 16.
These results show that the LOX/RP-1 combustion stability with these OFO
triplets and transverse LOL doublets is affected by not one, but both oxidizer and
fuel injection processes; however, work to date with these data has not revealed
whether the pressure drops themselves, or some other parameters such as injection
velocities, are the key parameters that more strongly influence the combustion
stability.
VI. Conclusions
Over the past 15 years, a significant quantity of LOX/RP-1 liquid rocket injector
performance and combustion stability data has been generated at Aerojet, encom-
passing a large variety of injection schemes, chamber dimensions, and operating
conditions. Because of the experimental nature of these technology programs,
rather than their being carried out for purposes of engine development, the in-
jector efficiencies and combustion stability characteristics have varied widely,
showing the effects of sensitivities of many important parameters.
The intraelement mixing and the RP-1 vaporization were shown to have domi-
nant effects on the injector performance of various injector types. Sensitivity to va-
porization limitations appeared strongest in like-impinging elements such as LOL
doublet pairs, although combustion chamber characteristics such as smaller con-
traction ratio and longer chamber length improved the vaporization efficiency for
all element types. Like-impinging elements also had limited absolute intraelement
mixing potential, especially in edge-on unlike-propellant fan configurations, and

thus required some degree of interelement mixing to achieve high-performance
efficiencies. Unlike-impinging elements such as OFO triplets were sensitive to
intraelement mixing through parameters such as momentum ratio, but absolute
intraelement mixing potential could be made quite high. Proper element spac-
ing to preclude adverse element-to-element fan interaction was shown to still be
important for OFO triplet elements to maximize performance. Since all of these
processes are, in fact, highly coupled, the ultimate understanding of spray pro-
cesses on performance may require advanced computational fluid dynamics (CFD)
codes and nonintrusive combustion diagnostic techniques. At present, theoretical

and empirical atomization, vaporization, and mixing models for combustion per-
formance calculations are moderately coupled, such as in ROCCID.10
The acoustic combustion stability margin was found to improve with larger
orifice diameters and with lower injection pressure drops of both propellants. This
latter result showed that combustion stability of impinging element LOX/RP-1
systems is affected by not just one, but both oxidizer and fuel injection processes.
In addition, the chamber size (diameter) and the use of acoustic damping devices
such as acoustic resonators or injector face baffles had first-order influences on
the combustion stability of LOX/RP-1 combustors.
Correlation of combustion stability with bombed unstable data was inconclu-
sive. This may be due to the bomb completely altering, even temporarily, the
importance or even the existence of certain spray combustion processes that might
otherwise be critical. The process judged unstable may have nothing to do with
normal operation of the combustor.
In general, the combination of such widely varying and influential parameters
precludes most simple characterizations of combustion stability, so that complex
models like ROCCID10 are ultimately required for predictions.
The database generated during technology programs at Aerojet for impinging
element injectors and LOX/RP-1 propellants indicates that both high performance
and stable combustion can be achieved, but only through proper consideration and
tradeoffs of the many design parameters affecting the spray combustion processes
and chamber acoustic mode, frequency levels, and damping.
References
n., "Weapon System 107A-2," Aerojet-General Corp., Final Rept. LRP 260, Tech-
nical Documentary Rept. BSD-TRD-62-51, Contract AF 04(645)-8, Sacramento, CA, April
27,1962.
2
Judd, D. C, "Photographic Combustion Characterization of LOX/Hydrocarbon Type
Propellants," Final Rept. MA-129T, Contract NAS 9-15724, Aug. 15, 1980.
3
La Botz, R. J., Rousar, D. C., and Valler, H. W., "High Density Fuel Combustion and
Cooling Investigation," Final Rept. NASA CR 165177, Contract NAS 3-21030, Sept. 1980.
4
Schoenman, L., and Gross, R. S., "Design, Fabrication, Test and Delivery of a High-
Pressure Oxygen/RP-1 Injector," Final Rept. 3365IF, Contract NAS 8-33651, Sept. 30,
1981.
5
Schoenman, L., "Fuel/Oxidizer-Rich High-Pressure Preburners," NASA CR-165404,
Contract NAS 3-21753, Oct. 1981.
6
Lawver, B. R., "Testing of Fuel/Oxidizer-Rich High-Pressure Preburners," NASA CR-
165609, Contract NAS 3-22647, May 1982.
368 J. PIEPER ET AL
7
Hernandez, R., Ito, J. I., and Niiya, K. Y., "Carbon Deposition Model for Oxygen-
Hydrocarbon Combustion," Interim Final Rept. 2427-IFR, Contract NAS 8-34715, Sept.
1987.
8
Mercer, S. D., and Rousar, D. C, "Aerojet TechSystems Company Contribution to
LOX/HC Combustion and Cooling Technology/' The 1986 Advanced Earth-to-Orbit
Propulsion Technology Conference, NASA CP2437, Vol. II, 1986, pp. 393-438.
9
Pieper, J. L., "Oxygen/Hydrocarbon Injector Characterization," Final Rept. PL-TR-
3029, Contract F04611-85-C-0100, Oct. 1991.
10
Pieper, J. L., and Walker, R. E., "LOX/Hydrocarbon Rocket Engine Analytical De-
sign Methodology Development and Validation," Final Rept. NASA CR-191058, Contract
NAS3-25556, May 1993.
H
Dunn, G. M., "Pressure Fed Thrust Chamber Technology Program," Final Rept. NASA
CR-190666, Contract NAS8-37365, July 1992.
12
Lawver, B. R., "High-Performance N2O4/Amine Elements—Blowapart," Final Sum-
mary Rept. 14186-DRL-5, Contract NAS 9-14186, March 1979.
13
Calhoon, D.F., Ito, J. I., and Kors, D. L., "Investigation of Gaseous Propellant Combus-
tion and Associated Injector/Chamber Design Guidelines," Final Rept. NASA CR-121234,
Contract NAS 3-14379, July 31, 1973.
14
Priem, R. J., and Heidmann, M. F, "Propellant Vaporization as a Design Criterion For
Rocket-engine Combustion Chambers," NASA TR-R-67, 1960.
I5
Ferrenberg, A., Hunt, K., and Duesberg, J., "Atomization and Mixing Study," NASA
CR-178751, RI/RD85-312, Contract NAS 8-34504, Dec. 30, 1985.
16
Giuliani, J. E., and Klem, M. D., "Effect of Model Selection on Combustor Performance
and Stability Predictions Using ROCCID," AIAA Paper 92-3226, July 1992.
17
Nickerson, G. R., Dang, L. D., and Coats, D. E., "Two-Dimensional Kinetic Reference
Computer Program," Final Rept. SN 63, Contract NAS 8-35931, April 1985.
I8
Reardon, R. H., "Guidelines for Combustion Stability Specifications and Verifica-
tions Procedures for Liquid Propellant Rocket Engines," Chemical Propulsion Information
Agency, CPIA Pub. 247, Laurel, MD, Oct. 1973.
19
Reardon, F. H., and Smith, A. J., "The Sensitive Time Lag Theory and Its Application
to Liquid Rocket Combustion Instability Problems," Air Force Rocket Propulsion Lab,
AFRPL-TR-67-314, March 1968.
Chapter 15
Acoustic Sensitivity Measurements of Injectors

in a Small Liquid Propellant Rocket Engine
R. Lecourt* and R. Foucaud*

Office National d*Etudes et de Recherche Aerospatiales, Chdtillon, France
Nomenclature
Ap = nozzle admittance
a = speed of sound
Cp — specific heat at constant pressure
D = chamber diameter
L = chamber length (cylindrical part)
M = Mach number
Re = real part
a — damping
ofBL = damping caused by the viscous boundary layer
y = specific heat ratio
A = thermal conductivity
/x = dynamic viscosity
p = density
I. Introduction
T HE occurrence of instabilities in a liquid propellant rocket engine constitutes

a major risk for the life of the engine and the vehicle it is propelling. The
destructive effects may be indirect, i.e., the vibrations induced on the engine
equipment, but are more often direct. This is true for high-frequency instabilities
(over 1000 Hz) that are associated with a significant increase in heat transfers to
the chamber wall and that generally erode part of the injector face. Many rocket
engines have been subject to this type of problem during their development, such
as the Fl engine built for the first stage of the Saturn-5 and the Viking-5 first
stage engine of the Ariane-1. Much experimental and theoretical research has
been conducted over the past 30 years1-2 to learn how to prevent the occurrence of
rights reserved.
'Research Scientist, Energetics Department, Propulsion Laboratory of the FAUGA-
MAUZAC, 31410 NOE, France.
369
370 R. LECOURT AND R. FOUCAUD
instabilities right from the design stage, to explain the phenomena generating and
supporting these instabilities, and to propose remedies.
The operating instabilities are still not completely understood despite the
progress that has been made. The models are still not refined enough and have to be
readjusted by many tests. This is essentially due to the many and complex physical
and chemical phenomena at play that are very difficult to measure in detail because
of the severe pressure and temperature conditions in the combustion chamber. The
combustion zone near the injector is still rather poorly understood at the present
time, despite its importance in the organization and combustion of the propellant
spray. Even for steady processes, current theory is not yet capable of describing
the formation and interaction of the jets, the establishment and disintegration of
the propellant sheets at the same time as the initial stages of the combustion that
causes recirculation of droplets and combustion products and heat transfers. For
unsteady processes, one possible way to improve the knowledge of this particular
region is to attempt to globally quantify the response of the zone near the injector
to an acoustical excitation.
II. Definition and Basic Principle of the Modulated Exhaust

Rocket Motor Method
Many experiments on high-frequency combustion instability were conducted on
full-scale simulators, because full scale is the best way to obtain the most realistic
operating conditions. As full-scale tests are expensive, however, it was decided to
develop a small-scale simulator. The most desirable small-scale simulator should
reproduce the tangential acoustic modes phenomena of a full-scale rocket engine
with axial injection, be a small-scale device, allow quantitative measurements, and
produce coherent results with full-scale observation.
We chose the intermittent modulated exhaust motor technique because it was
well characterized at ONERA from solid propellant studies3'4 and, from this ex-
perience, it had additional advantages compared with the T-burner.
As in the case of the T-burner, the basic principle of the modulated exhaust
motor is to investigate acoustic modes of full-scale rocket motors by the first
longitudinal mode of a small motor. The small-scale motor is designed so that the
frequency of its first longitudinal mode is the same as that of the acoustic mode of
interest of the full-scale motor. That is to say, to study a first tangential mode, the
length of the small-scale motor is related to the diameter of the full-scale motor:
a/2L =0.586(a/D) (1)

small-scale motor full-scale motor
The intermittent modulated exhaust technique has been shown previously to be

a simple, versatile, and effective means of destabilizing a motor and measuring
its damping.3 The nozzle throat, with its divergent part lopped off, is periodically
obstructed at a frequency close to that of the first longitudinal mode, using a
rotating wheel with just a few teeth concentrated over a small sector as shown in
Fig. 1.
Each excitation phase, occurring once during each revolution of the wheel, is
followed by a period of natural operation during which the motor damping can
be measured. It is a low-cost tool because of the small scale of the device. As it
includes only a few actual injectors of the full-scale engine injection face, a good
representation of the injection phenomena is obtained. It is also possible to keep
ACOUSTIC SENSITIVITY MEASUREMENTS OF INJECTORS 371
axial
injection
first longitudinal mode
Fig. 1 Sketch of the experimental device and of the acoustic field.
the same Mach number of the flow in the full-scale engine. It is easy to modify
the frequencies of the longitudinal mode by varying the length of the combustion
chamber. The rotating wheel is a very convenient means of exciting the first
longitudinal mode: it allows an easy control of the excitation parameters (frequency
and amplitude). This technique allows accurate measurements (frequency, peak-to-
peak amplitude, and damping of the pressure oscillations). As shown in Fig. 1, it is
possible to imagine the study of the pressure coupling with an axial injection when
the spray is mainly located in the pressure antinode and the velocity coupling with a
radial injection when the spray is mainly located in the velocity antinode. This way
it is assumed that both couplings, occurring simultaneously in a tangential acoustic
mode, can be studied separately with a longitudinal acoustic mode. Finally, it
allows an accurate estimation of the various acoustic losses.
The method has a few drawbacks. For example, only the longitudinal mode
can be excited and, as only a few injector elements are used, all the interactions
between the injector elements cannot be reproduced.
III. Description of the Experimental Setup

Schematically, the test setup consists of a small engine, a modulation system
that is used to destabilize it, and a test bench to supply the propellants.
A. Description of the Test Bench

For each propellant, there is one cylinder in which the liquid is delivered and
stored. The tank capacity is large enough to carry out several tests without refilling.
Each supply line is equipped with two Bell and Howell gauge bridge pressure
transducers, a thermocouple, a flowmeter operating through the range from 5 to
50 cm3/s, a 20-ju,m filter, a controlled injection valve, and circuits for draining and
scavenging the leftover propellants remaining at the end of the test by inert gases.
The transducers are for the initial adjustements and for calculating the mass flow
rates of each propellant, and thereby their mixture ratio, after the test.
For the sake of simplicity the pressurization is provided by large-capacity buffer

tanks (20-liter cylinders) limiting the pressure drop during the test. The pressur-
ization gas is nitrogen and the two propellants are pressurized by two totally
independent circuits.
B. Rocket Engine
The required length of the combustion chamber was determined from the fre-
quency of the acoustic mode of interest of the full-scale engine. The main pur-
pose of this study was to investigate the first tangential mode at 6500 Hz of a
storable liquid propellant rocket engine under development. For a sound veloc-
ity of 1150 m/s, the combustion chamber of the small-scale engine was required
to be 89 mm long. At the beginning of the study, the contraction ratio of the
nozzle of the small-scale engine was chosen close to that of the full-scale en-
gine to reproduce the Mach number of the actual flow. Several injectors designs
were tested, each composed of one or a few injection monoelements, typically
from one to six. The number of monoelements resulted from a compromise be-
tween the size of the small-scale rocket engine and the representativity of the
spray phenomena, for example, including the effects of fans interactions. For
the study of the pressure coupling, the injectors were positioned for axial in-
jection at the head end; for the study of the velocity coupling, they were po-
sitioned for radial injection laterally, at midlength of the combustion chamber.
This way, it is implicitly supposed that the processes interacting with the acoustic
field (pulverization, evaporation, and combustion) take place very close to the
injector.
As the durations of the tests were to be short, about 1 s, the combustion chamber
was built as a heat sink. One configuration of such a small-scale rocket engine is
presented in Fig. 2.
C. Modulation System
This system consists of an electric motor driving a wheel carrying three teeth,
all close together. The width of each tooth and the spacing between them are equal
to the throat diameter.
Initially, for the sake of safety, the teeth obstructed half the throat at most.
The electric motor was able to run as fast as 13,000 rpm. As the circumference
of the modulation wheel is long enough to carry 60 equally spaced teeth, these
characteristics allow the rocket engine to be excited at frequencies up to 13,000 Hz.
The teeth are made of molybdenum and the wheel and teeth were balanced before
use.
J Combustion
Injection • chamber Modulator
pressure wheel
transducers
— Tooth
Injector
Chamber
pressure
transducers
Fig. 2 Test motor longitudinal section.
D. Measurements
1. Instrumentation
The purpose of the instrumentation is to control the mean operation of the engine
and acquire the desired unsteady data. The unsteady pressure measurements were
made with very wide passband Kistler 6001 and 7001 transducers. Two of them
were located at each end of the chamber near the pressure antinodes of the first
longitudinal mode, as shown in Fig. 2, to measure the damping of the pressure
oscillations in the combustion chamber. Two other Kistler transducers were located
in the injector, as close as possible to the injection holes, to record the possible
propagation of waves from the combustion chamber.
2. Data Acquisition Systems

The data acquisition systems were both digital and analog because of the very
different characteristics of the steady and unsteady signals, especially the wide
passband required for the unsteady measurements. The only signal acquired di-
rectly in digital mode was a global chamber pressure signal and its unsteady
filtered component between 2 and 20 kHz. All of the other channels were recorded
on a wide passband Honey well 101 analog tape unit. The unsteady components
of the Kistler transducer signals were filtered above 200 Hz.
The signals recorded in digital mode were rapidly analyzed after the test to
judge the validity of the experimental conditions and of the findings. If the test
was deemed satisfactory, the unsteady signals were digitized off-line, reading back
the analog tape at a rate eight times slower than the recording speed before analysis.
On-line and off-line acquisitions were verified to produce the same results. This
way, spectral analyses up to 50 kHz were possible on all of the channels of interest,
and the oscillation damping could be measured precisely up to 12 kHz.
E. Optimization of the Device

Initial tests showed that the pressure oscillations of the first longitudinal mode
were damped very quickly, in a few periods. With these conditions it was not
possible to measure the damping with the desired accuracy. Consequently, the
combustion chamber was enlarged to increase the contraction ratio of the nozzle.
This results in the decrease of the acoustic losses of the rocket engine that are
related to the Mach number of the flow in the combustion chamber and the
geometry of the nozzle. The contraction ratio increased from about four to about
eight. Thus, the small rocket engine did not reproduce the aerodynamic field of
the full-scale rocket engine as desired to reproduce the flow effects on the spray,
but lower damping values were obtained that allowed accurate measurements.
The geometric modification of the rocket engine also increased its characteristic
length, resulting in a higher combustion efficiency than that obtained in the full-
scale rocket engine and at a higher frequency of the first longitudinal mode, closer
to the first tangential frequency of the full-scale engine.
IV. Description and Analysis of Characteristic Tests Results

A. Unmodulated Operation of the Small-Scale Engine
Each new configuration of the rocket engine is first tested without modulation
to verify the natural stability of the rocket engine, to check the operating point, to
measure the combustion efficiency, and to analyze the natural combustion noise
produced by each configuration. A pressure trace of an unmodulated operation of
the rocket engine is given Fig. 3.
Test 24/85
PCAR
Index16001
10
0 Oia 0.6 0.9 1.2

Fig. 3 Combustion chamber pressure (test without modulation).
The pressure curve indicates a very quick ignition, a good stabilization of the
operation, and a very healthy burnout. This figure also reveals the natural noise
coming from the injection and combustion, which affects the pressure measure-
ments. Spectral analysis of the combustion noise did not generally reveal the
existence of any particular frequency. Only some configurations of the small-scale
rocket engine always happened to be spontaneously unstable at their own first
tangential mode.5
B. Modulated Operation
1. Low Amplitude Modulation
The mean combustion performance of the rocket engines was not affected by the
modulation. The variation of the steady value of the pressure remained quite similar
to the variation without modulation except for the pressure peaks corresponding to
the excitation phases at each turn of the wheel, as shown in Fig. 4. Figure 5, with
its expanded time scale, suggests that the engine quickly becomes stable again
between two excitations.
Test 05/84
PCAR
20_[Pressure (bar).
Index 16000
10
Time (s)
0 0.3 0.6 0,9 1.2
Fig. 4 Combustion chamber pressure (test with modulation).

0 0.0015 0.003 0.0045

Fig. 5 Magnification of a modulation of the chamber pressure (low-level excitation).
Figure 5 also shows that steady pressure is increasing during the modulation
phase. This increase is due to the reduction of the sonic throat area by the passage
of the three teeth in front of it. The slope of this drift can be decreased by
diminishing the height of the teeth. Consequently, to obtain the same amplitude
as the harmonic excitation, the number of teeth has to be increased. These two
parameters are also sensitive to the distance between the modulation teeth and the
rocket engine throat. We adjusted this distance to obtain a harmonic excitation
of roughly 10% (peak-to-peak) of the steady pressure to provide linear pressure
excitation. Depending on the mass flow rate exhausting from the rocket engine,
this distance was set at values between 0.5 and 1 mm.
Figure 6 shows the unsteady signal from the chamber as a function of the
digitization index (this index measures the number of time steps of the digital
acquisition). We observe that the wideband filtered signal stands out clearly from
the noise, which in this case represents no more than 10% at most of the initial
peak-to-peak amplitudes of the oscillations. The signal can be broken down into
four parts: 1) up to index 200 the signal corresponds to the natural stable operation,
2) 3 oscillations of increasing amplitude occur due to the passage of the modulator
wheel teeth, 3) 12 oscillations occur that correspond to the engine response to the
excitation and, finally, 4) beyond index 650, the engine returns to its natural stable
operation. The signals of the two extremities of the chamber are quite comparable,
but are opposite in phase. The phase relationship and the fact that the frequency
of oscillations was close to the expected value (6500 Hz) demonstrated that the
first longitudinal acoustic mode was triggered as desired.
Comparison of the unsteady components of the chamber pressure signals with
the fuel pressure signal indicated an acoustical coupling of the chamber with
the supply system for some injectors. The unsteady pressure in the fuel line was
similar to the chamber pressure. In other cases, the amplitudes of the oscillations
test 10/86 '

PCAR
Index 367177"
-0.8-
0 256 513 769

Fig. 6 Unsteady component of the chamber pressure signal during a modulation.
were reduced by a factor of three with respect to the chamber and drowned out by
the noise.
Finally, it should be noted that one experiment gives several tens of excitations
of the first longitudinal mode of the rocket engine, so that it is possible to check
the reproducibility of the unsteady measurements and, in that case, to submit them
to a statistical treatment.
2. High Amplitude Modulation

The objective of experiments with a high amplitude modulation was to obtain
a nonlinear behavior of the wave in the rocket engine and, consequently, measure
the sensitivity of the injector to nonlinear conditions.
This modulation was accomplished by obstructing the full area of the throat
with the teeth. Figure 7 shows that the evolution of the steady pressure caused
by the modulation is very large, about 0.65 MPa peak-to-peak, or about 40% of
the mean chamber pressure. Thus, the sensitivity measurement is not performed
at a specific operating point. Moreover, the full obstruction roughly doubled the
peak-to-peak amplitudes of oscillations, as expected, but their damping kept an
exponential form, suggesting the response to the perturbation was still linear. The
damping value was also lower in that case than that in the same test with half
obstruction. This result can be explained by the influence of the large evolution
of the steady chamber pressure during the excitation phase, as shown in Fig. 7.
The same phenomenon was observed when an excitation phase occurred during
the ignition or burnout of the engine.
Thus, it was concluded that good nonlinear operating conditions should be ob-
tained by increasing the number of the teeth rather than increasing the obstruction
of the throat to keep the evolution of the steady pressure small.

0 0.0015 0.003 0.0045

Fig. 7 Magnification of a modulation of the chamber pressure (high-level excitation).
C. Frequency and Damping Measurements

1. Frequency Measurements
To measure the acoustic response frequency each damping period is processed
by a numerical Fourier transform program. Each window of analysis yields an
amplitude spectrum by which the predominant frequency can be known accu-
rately along with any other frequencies or harmonic components that may be
present.
As several tens of excitations were performed for each experiment, the evolution
of the frequency can be studied for the duration of each experiment. Figure 8, which
is representative of all of the tests, shows nonsystematic variation of the frequency
during the test. Furthermore, in most of the cases the deviations from the mean
value, 6460 Hz here for head end measurements, were low. Finally, this same
figure shows that the aft and head end pressure measurements match each other
well.
2. Damping Measurements
Determining the damping is a more difficult problem considering the observed
signal-to-noise ratio. To isolate the effects of the dominant frequency component
of the signal, the unsteady signal is first filtered numerically in a narrow band to
either side of the frequency calculated previously. Figure 9 illustrates the results
of this operation for the signal given in Fig. 6. The maximum amplitudes in the
damping phase are then computed automatically and an exponential regression of
these amplitudes yields the damping value directly.
To obtain damping values with a maximum of reliability, the code keeps the
value for the number of extrema that gives the best correlation factor of the
exponential regression. For example, a variation of the correlation factor from
0.967 to 0.997 was obtained. Damping results are presented Fig. 10 for two
6600
g 9 n
6500 D "a ' * **
m ffl*8 H *b
6400 * 9 **» m 9 *bam 5m
i
f
BB 9
Cl S] G2
6300
6200 D head end ^ aft end

(mean value=6460 Hz) (mean value= 6454 Hz)
6100
6000 - . - . . . .,___._
i ———— i,————————,,———— .———
, ,—————
0 10 20 30 40
wheel rotations
Fig. 8 Evolution of the frequency measurements on the chamber pressure signal
during an experiment.
Test 10/86
PCAR
Index 36717T
0.7 •
256 513 769

Fig. 9 Unsteady component of the chamber pressure signal after numerical filtering
for the damping measurements.
low damping
experiment
high damping
experiment
10 20 30 40 50
wheel rotations
Fig. 10 Evolution of the damping measurements of two typical experiments.
configurations of the rocket engine, one that produced a high damping, the other
a low damping. As well as for the frequency, the damping value is nearly constant
for the test duration with low variation. The measurements made on both pressure
lines match each other well, as shown in Fig. 11. Each excitation is a realization
of the same event and a statistical treatment can be applied to all of the data of one
experiment.
3. Analysis of the Acoustic Response

The experimental measurements provide overall damping values. They take
into account the injection and burning spray processes and the acoustic losses of
the combustion chamber. In these experiments, almost all of the chambers used
were similar from an acoustic standpoint, such that the damping measurements of
different configurations could be readily compared. But a test series was conducted
with combustion chambers of different lengths with, therefore, different acoustic
losses. To compare the results in that case it was necessary to calculate the acoustic
losses of the combustion chambers. This procedure was generalized to all results
and the acoustic losses of each combustion chamber was calculated and substracted
from the experimental measurements to reveal the acoustic response of the injector.
acoustic response = experimental damping — chamber acoustic losses (2)

Two methods were used to evaluate the acoustic losses of the rocket en-
gines. The first was the acoustic balance developed by Culick.6 For a first lon-
gitudinal acoustic mode in a cylindrical combustion chamber and a converg-
ing nozzle, the damping due to the acoustic losses is the summation of three
contributions:
= (a/L)M + (XBL (3)

3500
• D
JD m •
3000 -- D
« D
D
D
D
2500 -
2000 -- head end * aft end

(mean value=3034 s-1) (mean value=3108 s-1)
1500 -4- -H
10 20 30 40 50
wheel rotations
Fig. 11 Comparison of the damping measurements on both chamber pressure
signals.
where M = Mach number of the combustion chamber, AT = nozzle admittance

(evaluated with the quasimonodimensional theory of Croeco-Sirignano7), and aBL
is evaluated from the boundary-layer admittance8
(4)
The first contribution of damping is due to the convection of the waves out of
the engine. The second contribution is due to the reflection of the waves on the
converging nozzle. The third is the damping caused by the aerodynamic boundary
layer.
The second method used to evaluate the acoustic losses of the rocket engine
was computational fluid mechanics. This was done during the verification phase of
the acoustic properties of a code solving the two-dimensional axisymmetric Euler
equations.9 After that it was verified that standing waves in a closed cavity were
not damped or amplified by the code, computations were performed to simulate
the gaseous flow in the rocket engine during the steady state and during an acoustic
excitation similar to an experimental one. For medium size combustion chambers,
from 70 to 120 mm, the agreement was very good between the acoustic balance and
quasimonodimensional computations. For shorter combustion chambers, compu-
tations needed a fine two-dimensional grid because the converging nozzle became
large in view of the cylindrical chamber and the wavelength. It is thought, in the
case of the short combustion chambers, that the two-dimensional computations
are more reliable than the acoustic balance because of two-dimensional acoustic
effects10 that cannot be taken into account by the acoustic balance method.
The acoustic losses of the modulated exhaust rocket engine can thus be accu-
rately calculated. If the acoustic response of the tested configurations is defined
as in Eq. (2), these configurations can be compared even though they have been
fired in different combustion chambers.
V. Systematic Studies
In this chapter, only the results obtained in the rocket engines with a nozzle
contraction ratio of eight are presented. The damping measured in the engine
with a nozzle contraction ratio of four were too high and thus not believed to be
accurate.
The experiments were conducted with several types of unlike impinging jet
injectors, including doublet and quintuplet (pentad or four-on-one) injectors. Ni-
trogen tetroxide was the oxidizer and either UDMH, Aerozine 50, UH25 (75%
UDMH, 25% hydrazine hydrate), or MMH was used as fuel.
The modulation system was designed for one value of the throat diameter.
The width of the teeth was equal to this throat diameter. Consequently, to obtain
reproducible conditions of modulation, the throat diameters could not vary over
a large range. Therefore, the number of monoelements for each configuration
of injector was chosen to have a roughly constant mass flow rate and operating
pressure, whatever the kind of injector used. All of the injectors were tested in the
axial position. Two injectors, the doublet and one quintuplet, were tested in the
radial position. The operating chamber pressure was varied from 0.9 to 2.5 MPa.
In almost all of the experiments, the length of the combustion chamber was
chosen to trigger the first longitudinal acoustic mode at about 6500 Hz. One
injector was fired with combustion chambers of different lengths to measure the
damping of the first longitudinal acoustic mode with frequency ranging from 5 to
12kHz.
Table 1 describes the experimental program and Table 2 the injector designs.
In the following description, damping values are always written positive. The
response values are written positive if they have an amplifying effect, negative if
they have a damping effect.
Table 1 Overview of experimental program
Chamber, Mixture
Injector Number of Fuel pressure, 1L frequency, Injection ratio
type elements type MPa kHz type O/F
UDMH 1.5-2.5 axial
Doublet D 6 A50 1.5 6.5 axial 1.8
UH25 1.5 axial
MMH 1.5-2.5 axial and radial
UDMH 1.5-2.5 axial
Quintuplet SQ0 3 MMH 1.5-2.5 6.5 axial and radial 1.8
MMH axial
MMH 1.5-2.5 5 to 12 axial
Quintuplet SQi 3 MMH 0.9-1.2-2.2 6.5 axial 1.8
Quintuplet SQib 3 MMH 0.9-1.2-2.2 6.5 axial 1.8
Quintuplet LQi 1 MMH 0.9-1.2-2.2 6.5 axial 2.1
Quintuplet LQ2 1 MMH 0.9-1.2-2.2 6.5 axial 2.1
Table 2 Description and schematics of the injectors
Injection diameter Injection velocity, m/s

Injector Impingement
type Fuel Oxidizer Fuel Oxidizer angle
Doublet D 4> .25* 26 18 Of
Quintuplet SQ0 0.934> .6* 25 20 l.5a
Quintuplet SQi 0.933> .6* 25 20 OL
Quintuplet SQib 1.1* .9* 17 14 a
Quintuplet LQi 3.1* .4* 13 20 l.5a

Quintuplet LQ2 2.2* .4* 25 20 l.5a
doublet quintuplet
A. Influence of the Injector Design and of the Chamber Pressure

Two series of experiments were carried out. The first tested the doublet injector
D and the SQ0 quintuplet injector with UDMH and MMH as fuel. The second one
used the quintuplet injectors SQ0, SQls SQlb, LQ1? and LQ2, with MMH as fuel.
All configurations were realized at different operating combustion pressures.
1. Comparison of the Injectors D and SQ0 ( UDMH Fuel)
The damping measurements and calculated responses (experimental measure-
ment minus acoustic loss) are presented Table 3. The doublet injector is found
to be more stable than the SQ0 quintuplet, regardless of the combustion chamber
pressure. Also the SQ0 injector was spontaneously unstable on the first tangential
acoustic mode of the small-scale rocket engine. Nevertheless, the results showed
that it was possible to classify the different injectors according to the acoustic
sensitivity to the first longitudinal mode, because of the difference in damping at
a chamber pressure of 1.5 MPa.
Table 3 Damping and response results for SQ0 and D injectors

UDMH fuel
Damping results (s-1) Response results (s-1)

Pen, MPa 1.5 2.5 1.5 2.5
Quintuplet SQ0 740 unstable 805
Doublet D 1990 1350 -485 130
Table 4 Damping and response results for all

the injectors—MMH fuel
Damping (s- 1)
Pch, MPa 0.9 1.2 1.5 1.8 2.2 2.5
Injector:
Quintuplet SQ0 1650 1110
Quintuplet SQi 1830 1400
Quintuplet SQ!b 2540 1810

Quintuplet LQi 1980 1850 940
Quintuplet LQ2 2210 1870 1170
Doublet D 3060 2090
Response results (s- 1)
Quintuplet SQ0 -95 320
Quintuplet SQi -480 -80
Quintuplet SQib -1190 -490
Quintuplet LQi -680 -570 305
Quintuplet LQ2 -910 -590 75
Doublet D -1555 -610
2. Comparison Between All of the Injectors (MMH Fuel)

The results are recapitulated on Table 4 and Fig. 12. The doublet injector is
again more stable than the SQ0, even with a different fuel, and more stable than
any of the quintuplet injectors regardless the pressure level.
Figure 12 shows the influence of the chamber pressure on the acoustic response
of the injectors. Increasing the chamber pressure was destabilizing for all injec-
tor configurations. In these experiments, the chamber pressure was increased by
increasing the mass flow rate of the propellants. Thus, not only was the chamber
pressure modified, but also the injection and spray parameters (liquid jet velocities,
drop sizes, velocities, and distributions). The combustion pressure is generally in-
creased in the same manner on a full-scale rocket engine when increasing thrust
is needed. For a different comparison, one experiment increased the chamber
pressure by diminishing the throat area, keeping the injection parameters and the
chamber Mach number constant. The exact configuration and its damping and re-
sponse results are shaded on Tables 1 and 4. These damping and response results
fall between the neighboring results of 1.5 and 2.2 MPa chamber pressures. This
suggests that the chamber pressure has more influence on damping than the in-
jection characteristics. Thus, whatever method is used to increase the combustion
pressure level, the damping is lowered. The increase of the combustion pressure
is clearly destabilizing.
Comparing the results of the SQ0, SQ], and SQlb injectors reveals trends among
the effects of the injector design. As shown in Table 2, these injectors differ by
only one design parameter. The results in Table 4 show that a stabilizing effect
is brought about by diminishing the impingement angle (SQ0 vs SQt results at
2.2 MPa) and by enlarging the injection holes (SQl vs SQlb results at 1.2 and 2.2
MPa).
• SQO
D SQ1
* SQlb
O LQ1
A LQ2
A D
chamber pressure (MPa)

Fig. 12 Responses of the injectors according to the chamber pressure level.
B. Influence of the Fuel Type

The doublet injector was tested with four hydrazine derivatives fuels: UDMH,
MMH, Aerozine 50 (50% UDMH, 50% hydrazine), and UH25 (75% UDMH, 25%
hydrazine hydrate). The damping measured on these experiments showed a large
stabilizing effect of the MMH fuel with regard to UDMH, as shown in Table 5.
This stabilizing effect of MMH was also seen on the triggering of spontaneous
high-frequency instabilities. In the first small-scale engines with radial injectors,
spontaneous instabilites occurred with UDMH as fuel and none with MMH.5
Unfortunately, recording difficulties arose during the experiments with Aerozine
50 and UH25 fuels, and the damping measurements were not accurate, especially
for those with Aerozine 50. The results of Table 5 thus indicate only trends for
these two fuels. The damping obtained with the UH25 is higher than that obtained
Table 5 Evolution of damping

according to the fuel type
Damping (s - 1)
Doublet D
Pch, MPa 1.5
Fuel
UDMH 1990
UH25 2300
A50 3000
MMH 3060
with UDMH. With Aerozine 50 the damping is still higher than with UH25 so that
its value is close to the one obtained with MMH. But because of the recording
difficulties, the discrepancy between the measurements of each modulation is large
and thus it is not possible to say if the stabilizing effect of Aerozine 50 is higher
or lower than that of MMH.
C. Influence of the Frequency of the First Longitudinal Mode

This study was done with the SQ0 quintuplet injector, MMH as fuel, and at
two chamber pressure levels, 1.5 MPa and 2.4 MPa. The frequency of the first
longitudinal acoustic mode was varied from 5 to 12 kHz simply by using different
combustion chamber lengths. The decrease and increase of the length occurred
on the cylindrical part of the combustion chamber. Thus, all of the chamber
configurations kept the same nozzle and consequently the same Mach number.
Also, in spite of the decrease of the chamber length the operating point and the
combustion efficiency were kept nearly constant for each pressure level.
Very accurate frequency and damping measurements were obtained. The evo-
lution of the damping vs the frequency is displayed in Fig. 13. The variation of
the damping is generally monotonic except during the experiment at 11 kHz. A
tentative explanation will be provided later.
Because the combustion chambers were geometrically different, it was not pos-
sible to direcly interpret the damping measurements. The acoustic losses of each
combustion chamber was evaluated and then subtracted from the damping mea-
surements to obtain the acoustic response of the injector. The acoustic response was
considered, in a first approach, independant of the combustion chamber geometry.
It was explained that the acoustic losses were evaluated by two methods, the
acoustic balance and nonviscous flow computations. As such flow computations
4500
3500
2500
D a
1500 -- a
a
a
500 D
6 8 10 12
frequency (kHz)
Fig. 13 Evolution of the average dampings for the SQo injector vs the frequency of
the acoustic mode (1L).
Table 6 Results of the acoustic losses evaluation by the acoustic

balance and the CFD methods for the combustion chambers
of different lengths
Pch = 1.5 MPa Pch = 2.4 MPa

Chamber length, Acoustic Computed Acoustic Computed
mm balance results balance results
120 1015 1125 1000 1110
102 1245 1345 1225 1325
87 1540 1605 1525 1590

77 1845 1870 1820 1845
69 2200 2165 2175 2140
63 2590 2435 2565 2410
57 3125 2840 3100 2815
52 3775 3235 3750 3210
were computer intensive at that time, they were done in a quasimonodimensional

grid to save time and money. The results of both methods are presented in Table 6.
The agreement is good when the combustion chamber is long and worse when
it becomes shorter, because neither of the two methods (acoustic balance and
quasimonodimensional flow computations) can take into account two-dimensional
effects. A two-dimensional flow computation that was considered more reliable
than the previous calculations, done for the shorter combustion chamber, gave
an acoustic loss close to the average of the values given by the acoustic balance
and quasimonodimensional flow computation methods. Thus, it was decided to
average the two methods to evaluate the acoustic losses. Subtracting them from
the damping measurements gave the response as a function of the frequencies of
the longitudinal mode, shown in Table 7 and Fig. 14. Figure 14 shows that up to
10 kHz for the lower chamber pressure and 8 kHz for the higher chamber pressure,
Table 7 Evolution of the acoustic sensitivity of the

SQ0 injector with the frequency of the first
longitudinal mode
Damping (s-1) Response (s-1)

Pch, MPa 1.5 2.4 1.5 2.4
Frequency, kHz
4.94 1110 610 60 445
5.87 1190 750 105 525
6.88 1530 1150 45 410
7.81 2060 1400 -200 435
8.91 2220 1550 -40 600
9.72 2565 1730 -50 760
10.87 3770 2265 -790 690
12.08 2965 2325 540 1155
1500
P
1000
D
D
D
500
D
-500
-1000
8 10 12
frequency (kHz)
Fig. 14 Evolution of the average responses for the SQo injector vs the frequency of
the acoustic mode (1L).
the acoustic response is roughly constant. The response is nearly zero for the low
chamber pressure. That is, there would be no coupling between the burning spray
and the acoustic field in the aggregate. The response is positive (about 500 s"1)
for the high chamber pressure, which means that the burning spray promotes the
amplification of the acoustic field. Above 8 and 10 kHz, respectively, with an
exception for the point at 11 kHz, there is approximately parallel behavior of the
responses toward an amplification of the acoustic field. As the damping values
obtained for 10, 11, and 12 kHz, were measured with the best correlation factors,
above 0.99, it is inferred that these results are quite accurate.
Concerning the 11-kHz results, the following explanation is proposed. At 10
kHz the burning spray is still well in the upstream pressure antinode of the acoustic
field and there is pressure coupling. At 11 kHz, because the combustion chamber
is shorter, the burning spray is close to the pressure node and a lower pressure
coupling occurs, so the overall damping is higher. At 12 kHz, because the com-
bustion chamber is shorter again, the burning spray is located in the downstream
pressure antinode and pressure coupling can occur again.
D. Injector Location
To investigate different couplings between the burning spray and the acoustic
field (pressure and/or velocity coupling), several rocket engines and injectors were
built to position the burning spray in different locations of the acoustic field. The
work was performed both with the doublet and the SQ0 quintuplet injectors. The
combustion chambers were designed to place the injectors at midlength and at
the first quarter of the combustion chamber, requiring the monoelement injectors
to be placed radially. The doublet injectors were oriented so that the spray fans
in the small-scale engine were positioned in the same way relative to the first
longitudinal acoustic field as in the full-scale engine, relative to the first tangential
acoustic field. For the quintuplet injector that produces a rather axisymetric spray,
such considerations are not relevant. This way, it was hoped that the burning spray
would experience two different acoustic couplings: mainly velocity coupling when
the injector was located at the middle of the combustion chamber L/2, and a mixed
pressure and velocity coupling when the injector was located at the first quarter of
the combustion chamber L/4.
The systematic study was done with MMH-N2O4 propellants. The rocket en-
gines operated well with the doublet injector. The combustion efficiency was
high for both chamber configurations (L/2 and L/4) and the frequency of the first
longitudinal mode reached 6500 Hz in both cases. With the quintuplet injector,
however, combustion efficiency was lower and decreased when the injector was
moved from the L/4 position to the L/2 position. The first longitudinal mode
frequency also decreased from 6100 to 5500 Hz when this occurred. Because
of these varying steady-state operating conditions, the unsteady measurements
obtained with the quintuplet injector were thought to be suspect compared to the
other results. Consequently, only the results of the doublet injector are presented. In
Table 8 the results obtained with the radial injector are compared to the results
of the axial injector. For the injection location L/2, the damping measurement
remained constant for both combustion pressure levels. For the L/4 injection lo-
cation, there was a slight decrease in damping when the pressure was increased.
With axial injection there is a larger decrease in damping with pressure increase.
For the injector located at L/2, the absence of an influence of the combustion
pressure level on the damping is surprising. Even if there is no unsteady acoustic
coupling related to the pressure, the chamber pressure modifies the steady char-
acteristics of the spray. The higher pressure was obtained by increasing the flow
rate of the propellants, and hence the injection velocities. This produces smaller
drops and it is surprising that it has no effect on the damping, unless opposite
phenomena counterbalance the effect of the smaller drops.
The analyses of the acoustic responses were calculated as before. Since the
geometry of the combustion chambers with the radial injectors was identical to that
of the axial injector, the acoustic balance method necessarily gives the same result
for the acoustic losses (approximately 1500 s"1). Computed flow simulations were
performed as they were earlier, but were two dimensional because of the radial
injection. Surprisingly, these computations gave acoustic losses evaluations of
around 2200 s"1, higher than those calculated with the acoustic balance and close
Table 8 Comparison of the damping and response

of the axial and radial doublet injectors
Injector Pch, Damping C.A.L.a Response

location MPa (s-1) (s-1) (s-1)
0 1.5 3060 1520 -1555
2.5 2090 1495 -610
L/4 1.5 2210 1850 -360
2.5 1900 1800 -100
^2 1.5 2250 2275 25
2.5 2250 2210 -40
a
C.A.L.: computed acoustic losses + viscous boundary layer losses.
to the experimental damping results when the injection was at the middle of the
combustion chamber, as shown in Table 8. Analysis of the nozzle admittance in
the computed flow simulations showed that its value was close to that calculated
with the Crocco-Sirignano theory,7 as was the case in Ref. 10. Therefore, a two-
dimensional effect on the nozzle admittance cannot explain the large difference
between the acoustic balance and the computations. It is thought that this difference
was due to flow turning6 because of the radial injection. If this effect is well
simulated according to the characteristics of the computations (nonviscous flow,
mesh refinement), the obtained acoustic losses can be used to determine the
responses, as shown in Table 8. According to these results, axial injection at the

head end of the combustion chamber is more stabilizing than radial injection
at the L/4 and L/2 locations. But, because of the small number of experiments
performed, as well as the doubts about the evaluation of the acoustic losses in
radial injection configurations, this comparison is suspicious.
VI. Application to a Full-Scale Rocket Engine

All of the experiments described in this paper were performed in parallel with
the development of a full-scale engine. These experiments on small-scale engines
were intended to aid the rocket engine manufacturer in changing or redesigning the
injector if high-frequency combustion instabilities occurred. Therefore, damping
measurements obtained from the modulated exhaust engines were compared with
the behavior of the full-scale engine.11 For the comparison, one must keep in
mind that, on a small-scale, the damping measurements were performed on a
first longitudinal acoustic mode. Generally, the full-scale engine was unstable on
tangential modes, usually the first one. The frequencies of the acoustic modes
were identical, approximately 6500 Hz, but the mode shapes were different.
A. Comparison According to the Injector Type

The full-scale engine was more stable with the doublet injector than with the
SQ0 quintuplet injector, which correlates well with the small-scale damping mea-
surements. In full-scale tests the SQ0, SQ l5 LQ2 injectors were all stable in static
conditions and unstable, generally on first and second tangential modes, in bomb
tests without the presence of acoustic cavities.11 Because of the low number of
full-scale tests performed and the limited data gleaned from them, however, it was
not possible to discriminate between these injectors according to their sensitivity
to high-frequency instabilities in full-scale conditions. Therefore, it is not possi-
ble to compare the full-scale behavior of the quintuplet injectors to the damping
measurements made in small-scale experiments.
B. Comparison According the Chamber Pressure

Small-scale experiments always showed a decreasing damping when the pres-
sure was increased, except for radial injection at the middle of the chamber. In the
full-scale tests the doublet injector was stable at 1.2 MPa, spontaneously unsta-
ble at 2 MPa, but on the first longitudinal mode (2760 Hz). Full-scale tests with
SQ0 and LQ2 injectors showed no influence of the pressure in the studied range.
Nevertheless, the destabilizing effect of the combustion pressure is in agreement
with an accepted rule for a bipropellant rocket engine: the probability of the oc-
currence of instabilities increases with the chamber pressure. The other injectors
were not tested at full-scale for the pressure effect.
C. Comparison According to the Fuel Type

On the full-scale rocket engine, the doublet injector was tested with MMH,
UDMH, Aerozine 50, and hydrazine as fuels; the SQ0 injector was tested with
MMH and UDMH. With both injectors, the engine was clearly more unstable with
UDMH than with MMH. The behavior of the engine with the doublet injector was
as stable using Aerozine fuel as it was with MMH fuel. UH25 was not used, but
a stabilizing effect was noticed in small-scale tests with UH25 as it was for the
Viking IV engine with the same fuel.12 The small-scale experiments are thus in
good agreement with the full-scale tests.
VII. Conclusions
The modulated exhaust engine is a convenient, accurate, reliable, and low-cost
tool for evaluating the sensitivity of bipropellant injectors to an acoustic field. For
axial injectors and small amplitude excitation, accurate and reproducible damping
measurements, accurate evaluations of acoustic losses, and, consequently, intrinsic
acoustic responses of the injectors were obtained.With the small-scale engine, the
stabilizing effect of the use of MMH fuel instead of UDMH fuel, the destabilizing
effect of the mean chamber pressure, the influence of the injector design, and
the frequency of the longitudinal acoustic mode were demonstrated. Rather good
agreement was found with the behavior of a full-scale engine.
Improved results could surely be obtained with radial injectors and large am-
plitude excitation. In both cases, the work should be carried on to increase our
knowledge of velocity and nonlinear couplings.
A more detailed interpretation of the data would require a major effort of analysis
and research into the basic phenomena. These studies should include an investi-
gation of characteristics of the sprays in cold flow simulation and/or in burning
conditions such as liquid spatial distribution, local mixture ratio distribution, evap-
oration, and combustion processes (single-drop or group combustion model13).
Some of these processes should be studied in unsteady conditions as well
as in steady-state conditions to determine, if they exist at all, the key amplifying
processes in combustion instabilities. For example, transition from one combustion
model to another (group model to single-drop model) may occur under the action
of the acoustic waves.
Likewise, a computational tool should be developed to take into account the
basic knowledge obtained by the experimental program. This tool would serve to
accurately evaluate gains and losses in order to reduce the unknown part in the
damping measurements obtained with the modulated exhaust engine. That is to
say, to separate, in which is now globally attributed to the response of the injector,
which gains and losses in the measured damping values are ruled by the different
processes of the spray combustion (atomization, evaporation, combustion).
Acknowledgments
This work was funded by the Direction des Recherches, Etudes et Techniques
(DRET) and by the Direction des Missiles et de 1'Espace (DME). We would
also like to acknowledge helpful discussions with Paul Kuentzmann, Fran9ois

Vuillot, and Gerard Guillement; the participation of Francois d'Herbigny and Serge
Midorge in the experimental work; and the Societe Europeenne de Propulsion
(SEP), manufacturer of the full-scale engine, for its cooperation.
References
l
Harrje, D. T., and Reardon, F. N., Liquid Propellant Rocket Combustion Instability,
NASASP-194, 1972.
2
Barrere, M., Jaumotte, A., Fraeijs de Veubeke, B., and Vandekerkhove, J., Rocket
Propulsion, Elsevier, New York, 1960.

3
Kuentzmann, P., and Laverdant, A., "Experimental Determination of the Response of
a Solid Propellant to High-Frequency Pressure Oscillations," La Recherche Aerospatiale,
Vol. 1984-1, No. 1, English edition, pp. 39-55.
4
Vuillot, F, and Kuentzmann, P., "Flow Turning and Admittance Correction: An Exper-
imental Comparison," Journal of Propulsion and Power, Vol. 2, No. 4, 1986, pp. 345-353.
5
Lecourt, R., and Foucaud, R., "Experiments on Stability of Liquid Propellant Rocket
Motors " 23rd Joint Propulsion Conference, AIAA Paper 87-1772, July 1987.
6
Culick, F. E. C, "Stability of Three-Dimensional Motions in a Combustion Chamber,"
Combustion Science and Technology, Vol. 10, 1975, pp. 109-124.
7
Crocco, L., and Sirignano, W. A., "Behaviour of Supercritical Nozzles Under Three
Dimensional Oscillatory Conditions," AGARDograph, No. 117, 1967.
8
Morse, P. M., and Ingard, K. L, Theoretical Acoustics, McGraw-Hill, New York, 1968.
9
Kuentzmann, P., "Combustion Instability," AGARD Lecture Series 180, July 1991, pp.
7-26, 7-27.
10
Lupoglazoff, N., and Vuillot, F, "Numerical Simulation of Unsteady Two-Dimensional
Rows in Solid-Propellant Rocket Motors," La Recherche Aerospatiale, Vol. 1992-2, No. 2,
English ed., pp. 21-41.
H
Melchior, A., and Lecourt, R., "Combustion Stability Tests of a 6 kN Engine and Com-
parison with Subscale Tests," 41 st IAF Congress, IAF-90-251, International Astronautical
Congress, October 1990.
12
Souchier, A., Lemoine J. C., and Dorville, G., "Resolution du probleme des instabilites
sur le moteur VIKING," 33eme Congres IAF, IAF-82-363, International Astronautical
Congress, 1982.
13
Chiu, H. H., Kim, H. Y., and Croke, E. J., "Internal Group Combustion of Liquid
Droplets," 19th Symposium (International) on Combustion, Combustion Inst., Pittsburgh,
PA, 1982, pp. 971-980.


Chapter 16
Spray Combustion Processes in Internal

Combustion Engines
Song-Charng Kong* and Rolf D. Reitz*

University of Wisconsin—Madison, Madison, Wisconsin 53707
Nomenclature
A = pre-exponential constant of reaction rate, combustion
model constant, total flame area per unit volume
Af = frontal area of drops
a = parent drop radius
B — mass transfer number, branching agents in ignition
model, combustion model constant
BO = drop breakup size constant
B\ = drop breakup time constant
b = drop collision impact parameter
CD = drag coefficient
Cf = concentration of fuel
Cm = concentration of species m
Co2 = concentration of oxygen
Cp = concentration of product
Q 1,2,3, C^, C^, Cs — turbulence model constants
C2 = combustion model constant, 0.142
c = specific heat
D = diffusivity
E = activation energy
F = force, time rate of change of drop velocity
/ = spray model function, combustion model delay
coefficient
/ii /2, /3» /4 = kinetic parameters in ignition model
/bu = drop breakup source term
/coii = drop collision source term
rights reserved.
'Graduate Student, Engine Research Center, Department of Mechanical Engineering.
Professor, Engine Research Center, Department of Mechanical Engineering.
395
396 S.-C. KONG AND R. D. REITZ
g = specific body force

An° = enthalpy of formation at 0 K
/ = specific internal energy
In = nth-order modified Bessel function of the first kind
i = ^l
J = heat flux
Kc = concentration equilibrium constant
Kn = nth-order modified Bessel function of the second kind
Kp, Kt, Kq, Kb = kinetic parameters in ignition model
k — turbulent kinetic energy, wave number, 2n/A.

/ = wave number
M = third party molecules in ignition model, species mass
m = mass
min = minimum of two quantities
N = number of drops
Nu = Nusselt number
n = number of drop collisions
n^ = surface normal vector
P = fluid pressure, cylinder pressure, ignition model products
Pr = Prandtl number
p = probability function
Q = energy, intermediate species in ignition model
q = random number
R = gas constant, rate of change of drop radius, real part of
complex number
R* = radical in chemical reaction
Re = Reynolds number, pVa/IJL
RH = hydrocarbon fuel
r = drop radius, stoichiometric mass ratio of 0)2 to fuel,
combustion products fraction
Sh = Sherwood number for mass transfer
T = temperature, Taylor parameter
t = time
U — relative velocity between drops and gas
u = gas velocity
u' = turbulent velocity
V = drop volume, local consumption/production rate per
unit flame area
Vol = volume
v = drop velocity
XQ = oxygen mole fraction
x = drop position
Ym = species mass fraction
y = distance from walls, drop distortion parameter
W = work, molecular weight, reaction rate
We = Weber number, pV2a/a
w — relative velocity of drops and gas
Z = Ohnesorge number
SPRAY COMBUSTION PROCESSES 397
a — thermal diffusivity
y = drop radius ratio,
A = incremental amount
8 = Kronecker delta function
s = turbulent kinetic energy dissipation rate
rj = surface wave amplitude
A = wavelength of the most unstable surface wave
A = wavelength
\JL, — gas dynamic viscosity
v\ = gas kinematic viscosity

v 12 = collision frequency
p = species density
a = surface tension, viscous stress tensor
r = liquid breakup time
rc, n, rt = characteristic time scales
£2 = growth rate of the most unstable surface wave
co — wave growth rate
V = gradient operator
Subscripts
bu = drop breakup
coll = drop collision
crit = critical quantity
/ = soot formation
fv = fuel vapor
/ = incoming drop
k — turbulent kinetic energy
m = species index
o = outgoing drop, soot oxidation, oxygen atom
sphere = spherical drop
s = dissipation of turbulent kinetic energy
1. d, I = liquid drop, drop number 1
2. g = gas, drop number 2
Superscripts
c = chemical reaction
s = spray
* = equilibrium state; radical species
= first-order derivative
= second-order derivative
' = first-order derivative
I. Introduction
H IGH efficiency and low emissions have long been the major concerns of
internal combustion engine designers. Nowadays, in particular, engine man-
ufacturers are facing more and more stringent emission standards due to rising
environmental awareness. This has resulted in increasing interest in engine re-
search activities, especially for diesel engines. In fact, understanding the physics
and chemistry involved in diesel combustion processes is challenging since the

diesel engine represents one of the most sophisticated combustion systems. More
significant diesel emission reductions can be achieved only if the fundamental the-
ories of combustion are explored in detail. A good combustion model would make
prediction of emissions possible and engine designers could use numerical models
as a tool to help predict emissions and other engine performance parameters.
Specifically, the major complexity of diesel engines comes from transient effects
and the inhomogeneity of spray combustion. For a typical heavy-duty truck diesel
engine at a rated speed of 2000 rev/min, the injection pressure is as high as
100 MPa. At these high pressures, the fuel spray atomizes, breaks up into small
droplets, vaporizes, and mixes with the air. At some point in the process of
fuel-air mixing, ignition takes place, followed by a fast flame spreading process.
The burning that takes place shortly after ignition, it is believed, is of premixed
components, whereas the subsequent burning is thought to be mixing-controlled
combustion which can be characterized as diffusion burning. Therefore, an overall
description of these combustion processes requires an understanding of three major
processes: spray dynamics, ignition kinetics, and postignition combustion.
In this study, a theoretical investigation of spray combustion processes in diesel
engines is performed. An integrated numerical model will be presented that in-
cludes the state-of-the-art spray, ignition, combustion, and emission models. These
models are implemented into the computer code KTVA-II,1 which was developed
for simulating three-dimensional, transient, chemically reactive flows with sprays.
The validation of models is to compare the numerical results with experimental
measurements which include both spray ignition delay in combustion bombs and
cylinder pressure, heat release rate, and emission data of diesel engines. It is hoped
that this work can provide both a fundamental understanding of diesel combustion
and yield some strategies to reduce engine emissions.
II. Governing Equations

Since the computer code KIVA-II1 is used as a framework in this study, the
governing equations of the fluid phase in KIVA-II will be described in this section.
The mass, momentum, and energy equations coupled with the turbulence model
are solved. The interaction terms between the spray droplets and the fluid phase
will be presented in the next section.
The mass conservation equation for species m has source terms arising from
the vaporizing spray and chemical reactions, i.e.,
+ V . (pmu) = V . pDv ^) j +pcm + pXi (D

dt
where pm is the mass density of species m, p is the total mass density, u is the fluid
velocity, D is the (turbulent) diffusion coefficient of Pick's law, and pcm and ps8m\
are source terms that refer to particular chemical reactions and spray conditions.
The species 1 corresponds to the fuel.
The momentum equation for the fluid mixture is
^ + V - (puu) = - VP - v(^pK)\ + V • a + Fs + pg (2)

dt \3 /
where P is the fluid pressure, k is the turbulent kinetic energy, a is the (turbulent)
viscous stress tensor in Newtonian form, Fs is the rate of momentum gain per unit
volume due to the spray, and g is the specific body force which is assumed to be
constant.
The internal energy equation is
+ V.(pw/) = -PV - u - V - J + pe + Qc + Q* (3)

at
where / is the specific internal energy exclusive of chemical energy, / is the heat
flux vector including turbulent heat conduction and enthalpy diffusion effects,
and Qc and Qs are the source terms due to the chemical heat release and spray
interactions.
Turbulence is modeled by the k-e model with the transport equations for tur-
bulent kinetic energy k and its dissipation rate £ as follows:
s
—— + V • (puk) = ~pkV • u + o'.u + V • I ( JL ))v*
v* 1-pe
- + W (4)
dt 3 L\rrk; J
+ |[C£lcr:V« - Ce2pe + C,W'] (5)
These are standard k—s equations with some added terms. The source term
— (f Ce\ — C£i)peVu accounts for length scale changes when there is velocity
dilation. Source terms involving the quantity Ws arise due to interactions with
the spray. The model constants use the values from the standard k—s model,
i.e., Cei = 1.44, Ce2 = 1.92, Ce3 = -1.0, Prk = 1.0, Pre = 1.3, and C5 = 1.5
as suggested by Amsden et al.1 The turbulent transport of mass, momentum,
and energy is controlled by diffusion coefficients of the form D = C^k2 /s and
Cp = 0.09.
In addition, the state equations are assumed to be those of an ideal gas mixture
and corresponding thermodynamic properties are taken from the JANAF tables.
The chemical reaction source terms will be defined together with combustion
models.
In the present study, rigid wall boundaries are used together with the turbulent
law-of-the-wall conditions. Temperature boundary conditions use fixed temper-
ature walls. In calculations of turbulent flow, boundary conditions for turbulent
kinetic energy k and its dissipation rate s are taken to be
V* - it = 0 and e = C^/y (6)

where k and s are evaluated at a distance y from the wall and C^ = 0.38. The
effect of compressibility on the turbulence is accounted for by using a transformed
distance y' = f p / p 0 d y in both Eq. (6) and in the turbulent law of the wall.2
III. Spray Modeling

A. Spray Equations
A spray equation, which is formulated from a droplet distribution function, is
solved to describe the spray dynamics. The droplet distribution function / has
1 1 independent variables including three droplet position components jt, three

velocity components v, radius r, temperature Td (assumed to be uniform within
the drop), distortion from sphericity y, the time rate of change dy/dt = y, and
time t. Therefore,
f(x, v, r, Td, y, y, t) dv dr dTd dy dy (7)

is the probable number of droplets per unit volume at position x and time t with
velocities in the interval (v, v -f dv), radii in the interval (r, r -f dr), tempera-
tures in the interval (Tdl Td -f dTd) and displacement parameters in the intervals
(y,y + dy) and ( y , y H- dy). The time evolution of / is obtained by solving a
form of the spray equation,
f\ f f\ O CJ
- + V, • (/v) + Vv • (/F) + —(//?) + r(/r rf ) + — (/y)
/coll + /bu (8)

y
In Eq. (8) the quantities F, R, td, and >> are the time rates of changes fol-
lowing an individual drop, of its velocity, radius, temperature, and oscillation
velocity y. The terms /coii and /bu are the sources due to droplet collisions and
breakup.
The spray equation is not solved using finite differences. Instead, the equation is
solved using the method of characteristics in a Lagrangian formulation.1 The tra-
jectories of spray drops are traced in phase space once they are injected. The spray
model3 considers the drop interactions with turbulence and walls and calculates
the drop momentum change, breakup, collision and evaporation, which results
in the change of drop locations, sizes, velocities, temperatures, and distortions.
The status of drops, i.e., function /, is then updated, and the contribution of fuel
spray to the gas phase is also be obtained since mass, momentum, and energy is
transferred between the phases.
From the spray model solutions, the exchange functions ps 9 F v , Qs', and Ws
can be obtained for the use in Eqs. (1-5). These are obtained by summing the rate
of change of mass, momentum, and energy of all droplets at position x and time t .
ps = - [ fpd4nr2RdvdrdTddydy (9)
F9 = - f fpd(±nr3F' -f- 4n r2 Rv) dv dr dTd dy dy (10)
Q* = - j fpd{4nr2R[ll + {(v - u)2}
+ \nr\ClTd -f F' - (v - u - u')]}dvdrdTddydy (11)
Vs = - / fpd\7ir^F'
W ±nr3F' - u'AvdrdT ddydy
a'dvdrdTidjydy (12)
where F' = F — g, (v — u) is the relative velocity, u' is the turbulence velocity

normally distributed with a variance of |&, and // and c\ are the internal energy
and specific heat of liquid drops, respectively. Physically, W,s is the negative of
the rate at which the turbulent eddies are doing work in dispersing the droplets.
Through these exchange functions, interactions between spray droplets and gas
phase are then accounted for.
B. Breakup, Collision, and Coalescence of Droplets

Solving the dynamics of the fuel spray and its interactions with the gas is
essential for diesel engine modeling. To calculate the mass, momentum, and energy
exchange between the spray and the gas, one must account for a distribution of
drop sizes, velocities, and temperatures. Additional consideration is also needed
for the distortions, breakup, collisions, and coalescences of drops.
To describe the atomization processes, two models are currently available: the
Taylor analogy breakup (TAB) model4 and the wave breakup model.3 The first of
these, the TAB model, compares an oscillating-distorting drop to a spring-mass
system.5 In this model, the gas aerodynamic force on a drop, the liquid surface
tension force, and the liquid viscosity force of a drop are seen as analogous to the
external force acting on a mass, the restoring force of a spring, and the damping
force, respectively. The model constants have been determined from theoretical
and experimental results. The distortion parameter y is calculated by solving the
following spring-mass analogy equation:
2 p w2 8a 5/z
r r
3 Pd Pd Pdf
where pd,a,[L are the droplet density, surface tension, and viscosity, respectively;
and w is the relative velocity of the droplet and surrounding gas. If the value of
y exceeds unity, the droplet breaks up into smaller droplets with radius chosen
randomly from a /-squared distribution and with a specified Sauter mean radius
that is obtained from considerations of surface energy conservation.4
The spray model of Reitz3 considers the unstable growth of Kelvin-Helmholtz
waves on a liquid surface. In this wave breakup model, Reitz3 used results from a
linear stability analysis of liquid jets to describe the breakup details of the injected
liquid blobs. This stability analysis leads to a dispersion equation relating the
growth of an initial perturbation on a liquid surface of infinitesimal amplitude to
its wavelength and to other physical and dynamic parameters of both the injected
liquid and the ambient gas.
In this study, to model diesel spray dynamics, the wave breakup model of Reitz3
is adopted to replace the TAB model in KIVA-II. The stability analysis starts by
imposing on the liquid surface an infinitesimal axisymmetric displacement of the
form (see Fig. 1)
rj = R(êz+a)t] (14)
where ry0 is the initial amplitude of the disturbance. Associated with the disturbance
is a small axisymmetric fluctuating pressure, axial velocity, and radial velocity for
both the liquid and gas phases. These fluctuations are described by the continuity

Fig. 1 Schematic diagram showing surface waves and breakup on a liquid blob.
equation and equations of motion. These equations are then solved and lead to the
dispersion relation
2kl
o2 + 2vik2a)\ -^
/ 2 / 0 (A:a)/o(/a)
ak
+
Pi
where / 2 = A:2 -f &Vvi an<^ ô» A and KQ, K\ are modified Bessel's functions of
the first and second kinds, respectively. Reitz generated curve fits of the numer-
ical solutions to Eq. (15) for the maximum growth rate (a> = £1} and for the
corresponding wavelength (A = A)
(16)
z-
We2 =
A and £2 characterize the fastest-growing (or most probable) waves on the liquid
surface. Essentially, Eqs. (15-18) constitute an atomization theory which can also
be used as a framework to organize jet breakup regimes.3
1. Liquid Breakup
By applying the preceding atomization theory to diesel spray modeling, fuel
drop parcels are injected with a characteristic size equal to the nozzle hole diameter.
The number of drop parcels injected is determined from the fuel flow rate.
The liquid breakup was modeled by postulating that new drops are formed (with
drop radius r) from a parent drop or blob (with radius a) with
r = £0A (B0A < a) (19a)
f(
r = min / 2 7x0.33 (B<>A > a> (°ne time only) (19b)
where B0 = 0.61. In Eq. (19a), it is assumed that small drops are formed with
drop sizes proportional to the wavelength of the fastest growing or most probable
unstable surface wave; Eq. (19b) applies to drops larger than the jet (low-speed
breakup) and assumes that the jet disturbance has frequency £2/2n (a drop is
formed each period) or that drop size is determined from the volume of liquid
contained under one surface wave.
The mass of new droplets due to breakup is subtracted from the parent drops.
The change of the radius of a parent drop follows the rate equation
da/dt = -(a - r)/r (r < a) (20)
where r is the breakup time, and
r = 3.726fiifl/Af2 (21)
where B\ is the breakup time constant with the suggested value B\ = 20 (Ref. 3).
However, this value depends on the injector characteristics, and other values can
be found in different literature, e.g., B\ = 1.73 by O'Rourke and Amsden.4
2. Drop Collision and Coalescence

A collision frequency v\2 between drops in parcels 1 (with larger drops) and 2
is calculated in each computational cell
v12 = N2n(r} + r 2 ) 2 |v, - v2\/Vol (22)
N2 is the number of drops in parcel 2, v is the drop velocity vector, and Vol is the
volume of the cell. The probable number of collisions n within the computational
timestep Ar is then equal to vÂf . It is further assumed the probability of no
collisions is
p(n) = 6TU'2A' (23)
so that 0 < p(n) < 1. A collision event is assumed if p(n) is less than a random
number generated in the interval (0,1).
A collision impact parameter b is calculated and is compared to the critical value
bcr\t. If b is less than bcrit, the droplets coalesce; if b exceeds bcrit, the droplets
maintain their sizes and temperatures but undergo velocity changes.
r2) (24a)
r 2 ) 2 min[l.0,2.4(x 3 -2Ay2 + 2.1y)/We}] (24b)
where q is the random number in the interval of (0, 1) and / = r\/r2. If coales-
cence is predicted, n drops are removed from parcel 2 and the size, velocity, and
temperature of drops in parcel 1 are modified appropriately.3
C. Drop Drag
The acceleration F in Eq. (8) is obtained from the equation of motion of a drop
moving at a relative velocity U in the gas
'2 (25)
where p\, V, x, Af are the drop density, volume, position, and frontal area, re-
spectively; and where pi is the gas density. To calculate the drop drag, the drop is
usually taken to be a sphere with drag coefficient
CD=\24/Red(l + ±Re!) Red < 1000 (2fi)

[0.424 Red > 1000
where Red is the Reynolds number of liquid drops. However, the fuel drops
undergo high distortion due to the high-injection velocity and the drag coefficient
changes as a drop departs from a spherical shape. Highly deformed drops were
observed when the drops entered the high-velocity gas stream, with velocities up
to 250 m/s (Ref. 6). In fact, a distorted drop has a higher drag coefficient than a
spherical drop. To account for this effect, the distortion of a drop is calculated by
solving the spring-mass analogy equation from the TAB model, i.e., Eq. (13). The
distortion parameter y is obtained for a single drop which should lie between the
limits of a sphere (y = 0) and a disk (y = 1) with CD = 1.54. Therefore, a simple
expression for drag coefficient is formulated as follows.
CD = CD,sphere(l + 2.632?) (27)

This dynamically varying drag coefficient is important especially in diesel engines
which have strong spray/wall impingement interactions.
D. Spray/Wall Impingement
The impact of a drop on a heated surface may lead to instantaneous breakup,
sudden vaporization, development of a thin liquid film on the surface, and slid-
ing or bouncing back with a highly distorted drop. Results from a single-drop
wall impingement experiment7 show that the tangential velocity component of
the rebounding drop does not change but that the normal velocity component
can be evaluated by the initial Weber number of the drop. A correlation of the
arrival (Wet) and departure (We0) Weber numbers was formulated by Gonzalez
etal.8
We0 = 0.618Weie-'0ei (28)

For Wet < 80 the drop rebounds from the wall whereas for Wei > 80 the drop
disintegrates into small drops that move away from the impingement sites parallel
to the surface.
The present spray model uses the results to characterize the drop's behavior
after it hits the wall. For Wet < 80, the drop rebounds from the surface with a
normal velocity calculated from mass and momentum conservation as
vel0 = veli(We0/Wei?-5 (29)

For Wei > 80, the drop is modeled by analogy to the oblique impact of a liquid
jet on the wall.9 The subsequent disintegration of the drop depends on the relative
velocity between the drop and the gas.
However, the wall-impingement effects also represent a sudden disturbance
acting on the drop which is similar to the sudden exposure of a drop to a high-
velocity gas jet.10 Since the spray model3 considers the growth of the liquid surface
wave, which results in further breakup into smaller droplets, the drops are thought
to be more readily broken up when subject to a sudden disturbance. Therefore,
the drop breakup time constant is assigned a different value, B\ = 1 .73, after wall
impingement.
E. Vaporization
The rate of drop radius change R in Eq. (8) due to vaporization uses the Frossling
correlation
(30)
where D is the (laminar) mass diffusivity of fuel vapor in air, B is the mass
transfer number, and Sh is the Sherwood number. The fuel mass fraction at the
drop surface (which appears in B) is obtained by assuming the partial pressure of
fuel vapor equal to the equilibrium vapor pressure at drop temperature. The rate
of change in drop temperature needed in Eq. (11) is calculated from an energy
balance involving the latent heat of vaporization and the heat conduction from the
gas. The rate of heat conduction to the drop is
Q=a(T2-Tl)Nu/(2r) (31)
where a is the (laminar) thermal diffusivity and T2 and T\ are the gas and drop
temperatures, respectively. Other details about drop vaporization are described by
Amsden et al.1
IV. Combustion Modeling

A. Ignition Kinetics
Since diesel engines differ from spark-ignited engines which have prescribed
ignition timing and energies, the specific ignition characteristics need to be de-
scribed. The low-temperature chemistry of hydrocarbon fuels is characterized by
specific features such as degenerate branching reactions, cool flames, and two-
stage ignition phenomena. Several ignition kinetics models have been proposed
to simulate the autoignition processes. Preferred among these are the reduced ki-
netics mechanism models,11'12 which can characterize the important processes by
means of generic reactions and species.
The reaction equations in the ignition model of Zellat and Zeller11 follow
the same approach as the elementary reaction steps, i.e., initiation, propagation,
branching, and termination reactions. This model assumes that the chemical char-
acteristic time preceding autoignition is comparably larger than the turbulent mix-
ing time. The model contains a generalized radical /?* that represents the radical
pool effectively involved in the autoignition processes. The four pseudoelementary
reactions are as follows.
Initiation: fuel -^ 2/?*
Propagation: fuel + C>2 4- /?* -> R* 4- products 4- heat
Branching: fuel 4- /?* -+ 3#*
Termination: 2R* + M -+ out
where M represents any other species. The reaction rate is basically in Arrhenius
form. Some of the model constants are determined on the basis of phenomenolog-
ical analysis, whereas the others are determined by calibration with combustion
bomb experiments. Applications can be found in the literature.13*14
The ignition model by Halstead et al.12 is the one that has become known as
the Shell model. It was developed to predict knock in gasoline engines and was
formulated by simulating the autoignition processes of hydrocarbon fuels in a rapid
compression machine. This model uses a simplified reaction mechanism which
contains five generic species and eight generic reactions based on the degenerate
branching characteristics. The rate constants of every reaction are basically in
Arrhenius form. Extensive experimental data were obtained in developing this
model, and the model has been successfully applied both to engine knock15- 16 and
to diesel ignition study.17
In this study, the Shell model12 is used for ignition kinetics modeling. The model
uses a reduced kinetic mechanism to describe the low-temperature chemistry of
hydrocarbon fuels. The reaction equations are as follows:
O 2 -> 2/T Kq (32)

R* -+ /?* 4- P 4- heat Kp (33)
R*^R* + B frKp (34)
R*-*R* + Q f4Kr (35)
R* + Q _» /?* + B f2Kp (36)
B -> 2/T Kh (37)
/?* -^ termination f$Kp (38)
2/?* -> termination Kt (39)
where RH is the hydrocarbon fuel (CnH2m), R* is the radical formed from fuel, B
is the branching agent, Q is a labile intermediate species, and P is oxidized prod-
ucts, consisting of CO, CO2, and HaO in specified proportions. The expressions
for Kq, Kp, Kh, KtJ f\, /2, /s, /4 etc. are given by Halstead et al.12 In addition,
the local concentrations of 02 and N2 are needed to compute the reaction rates.
The approach of the Shell model and corresponding reactions and species in terms
of their roles in this model are described by Kong and Reitz.18
The Shell model was implemented into the KIVA-II code to predict the ignition
delay of diesel sprays in various combustion bombs19-20 using the kinetic constants
of 90 research octane number (RON) fuel.18 Since the kinetic parameters of the
Shell model were determined under the specific conditions including the fuel char-
acteristics and the corresponding submodels such as the heat transfer model, some
model constants are adjustable when the Shell model is implemented into KIVA-
II for diesel ignition modeling. After the sensitivity study,18 the pre-exponential
constant of the formation rate of intermediate species <2, Eq. (35), was modified,
from 1.88 x 104 to 1.3 x 106, to reproduce the combustion bomb ignition delay
data. Once this value was chosen, it was kept the same for different combustion
bombs and also for diesel engine computations.
B. Postignition Combustion
In diesel engines, the majority of combustion is thought to be mixing controlled.
Hence, the interactions between turbulence and chemical reactions have to be
considered. Several turbulent combustion models which include the effects of
turbulence on mean reaction rates are available for engine applications.21'22 Based
on the fact that the k-s model is used for turbulence modeling, these combustion
models formulate the combustion time scale by making use of the turbulent kinetic
energy and its dissipation rate.
The combustion model of Magnussen and Hjertager21 is an eddy-dissipation
model. It is generally accepted that diffusion combustion is controlled by the mix-
ing of the fuel and oxidizer. If it is further assumed that the fuel or reactive species
and oxidizer are contained in two different eddies, then the rate of combustion is
comparable to the rate of the dissipation of eddies, i.e., e/k, which describes the
intermixing of fuel and oxidizer to a molecular level. However, the concentrations
of the fuel and oxidizer are still controlling parameters in the combustion rate.
Therefore, the reaction rate can be expressed as follows.
rate = A(fi/*)min[C/, C O2 /r, BCp/(l + r)] (40)
where A and B are model constants which depend on both the structure of the
flame and the reaction between the fuel and oxygen, Cf, Co2, and Cp are the con-
centrations of fuel, oxygen, and products, respectively, and r is the stoichiometric
mass ratio of oxygen to fuel. This model was applied to model a prechamber
diesel engine combustion by Zellat et al.13 One difficulty with the application
of this model is that r is generally not known a priori in cases of incomplete
combustion.
The coherent flame model of Marble and Broadwell22 describes more detailed
information of combustion by considering the local laminar flamelet elements.
This model was initially developed for turbulent diffusion combustion in a shear
layer. The turbulent reaction zone is described as an ensemble of flamelet elements
which are stretched by turbulent flows but still maintain their laminar flamelet
structures locally. The reaction rate W, i.e., the source term, is then formulated
from the total flamelet area per unit volume A (cm2/cm3) as
W = pVA (41)
3
where p is the local average mass density (g/cm ) and V is the local consump-
tion/production rate per unit flame area, i.e., cm3/(s-cm2) which has the same
physical significance as the flame speed, cm/s. Separate submodels have to be
developed for V and A.
The flame speed V is assumed to be equal to the laminar flame speed at the
local thermodynamic conditions (temperature, pressure, and mixture fraction). A
submodel of the strained laminar flame is used to calculate the laminar flame speed
using detailed chemistry and transport properties. Specifically, this submodel is
used to construct a flamelet library or algebraic fuel consumption laws.
The important part of the coherent flame model is the calculation of total flamelet
area per unit volume A. The model formulation starts from an evolution equation
of flame surface which corresponds to the flowfield and its coherent structure. This
equation is derived by considering an infinitesimal surface propagating through a
turbulent flow. Finally, a transport equation for flame surface density is formulated,
incorporating the transport, diffusion, production, and destruction of flame area.
This equation accounts for local flamelet and global flow characteristics. The flame
area is created by stretching and annihilated by reactant consumption.
To extend this model to diesel combustion, further modifications have to be
made to account for diesel combustion characteristics: ignition delay and both
premixed and diffusion burn.14 For the ignition delay, a transport equation for the
total area of contact surface (between fuel and oxidizer) is formulated to describe
the mixing when combustion has not yet occurred. The contact surface is not yet
a reactive surface. After the mixing of the reactants, ignition takes place (modeled
by coupling other ignition models), and premixed flame pockets are formed. The
subsequent premixed burn is modeled by a transport equation for the premixed
flame surface evolution. In fact, the premixed flame surface equation is activated
by ignition chemistry. After the premixed burn, diffusion combustion comes into
play, modeled by a diffusion flame surface transport equation. In this equation, the
flame extinction due to the high strain rate is considered. It is noted that flamelets
need a finite time to respond chemically to the high local flowfield stretch in some
areas. High values of the turbulent stretch can be found near the injector where
the turbulent time scale is very short compared to the chemical time scale. The
flamelet is considered extinguished when it is subjected to a critical strain rate,
i.e., 2000s-1.
This modified flamelet model was used to model a Cummins engine by Dillies
et al.14 The agreement between predicted and measured engine data is plausible.
However, the premixed burn portion is not well described, and the cases studied
only considered low-compression ratio engine conditions with low-peak pressures
(from 50 to 75 atm), which are not representative of the conditions of production
heavy duty engines.
A laminar and turbulent characteristic-time combustion model for spark-ignited
engines23 was chosen for modeling the diesel combustion processes in the present
study. This model also uses the eddy-breakup concept and has been demonstrated
to perform well in various applications.2'24'25 With this model, the time rate of
change of the partial density of species m, due to conversion from one chemical
species to another, is given by
c\Y
Y — F*
Q/m
_£^__£m. = (42)
dt rc
where Ym is the mass fraction of species m, Y£ is the local and instantaneous
thermodynamic equilibrium value of the mass fraction, and TC is the characteristic
time for the achievement of such equilibrium. The characteristic time TC is assumed
to be the same for the seven species considered necessary to predict thermodynamic
equilibrium temperatures accurately: fuel, C>2, N2, CC>2, CO, H2, and H2O. Among
these seven species, six reactive species (i.e., all except N2) are accounted for in
order to solve the local and instantaneous thermodynamic equilibrium values Y^.
The model was combined with the Shell ignition model to simulate the overall
combustion processes in a diesel engine. The ignition model was used wherever
the temperature was lower than 1000 K to simulate the ignition chemistry. If the
temperature was higher than 1000 K, the combustion model was activated for
describing high-temperature chemistry.
The most important part of this model is to formulate appropriately the char-
acteristic time rc which is the sum of a laminar time scale and a turbulent time
scale,
TC = TI + frt (43)
The laminar time scale r/ is derived from the correlated one-step reaction rate
from a single droplet autoignition experiment26
rate = A[C14H3o]a25[O2]1-5 exp(-E/RT) (44)

where A = 1.54 x 1010and£ = 77. 3 kJ/mol, considering that tetradecane is used.
By equating this reaction rate to that of Eq. (42), and assuming an equilibrium state
concentration of fuel equal to zero, the following laminar time scale is obtained:
n = A-1[C14H3o]a75[02]-1-5exp(E//?r) (45)
The turbulent time scale rt is proportional to the eddy turnover time and is
equal to Cikje, where C2 = 0.142 is the same for the other engine studies.2'24'25
The turbulence parameters, k and £, are calculated from the k-s turbulence model
in KIVA-II. The delay coefficient / simulates the influence of turbulence on
combustion after ignition has occurred,
/ = (! -O/0.632 (46a)
where r is the ratio of the amount of products to that of total reactive species
(except N2)
2 2 *co 4- Yn2 ,A^\

r = ————— -—— -————— (46b)
1
— * N2
The parameter r indicates the completeness of combustion at a specific region.
Its value varies from 0 (no combustion yet) to 1 (complete consumption of fuel
and oxygen). Combustion occurs at molecular scale as a result of collisions be-
tween reacting molecules. Subsequently, combustion is strongly influenced by
turbulence since the turbulence has significant effects on the transport properties
and the preparation of reactants. The delay coefficient / changes from 0 to 1
accordingly depending on local conditions. In other words, the initiation of com-
bustion relies on laminar chemistry; turbulence starts to have an influence only
after combustion events have already been observed, similar to the model of Mag-
nussen and Hjertager.21 Eventually, the combustion will be dominated by turbulent
mixing effects in the regions of r\ <^rt. However, the laminar time scale is not
negligible near the injector regions where the high-injection velocity makes the
turbulent time scale very small. In view of the overall diesel combustion process,
the general picture—premixed first and then mixing control —coincides with the
given model formulation.
The idea of naming the function / as delay coefficient comes from modeling
spark-ignited engines and the growth of flame kernel. When the flame kernel
grows to be comparable to the turbulence eddy size, it becomes influenced by

the turbulence.23 However, the combustion process is more complicated in diesel
engines than that in spark ignition (SI) engines. The function / must be formu-
lated differently. The wide range of equivalence ratio and nonhomogeneous spray
droplet combustion makes the monitoring of flame kernel growth impossible. But
the idea that laminar chemistry initiates combustion and then turbulence influences
combustion gradually is still adopted. Our way to account for the separate effects
of laminar chemistry and turbulence is to use the appearance of products as an
indicator of mixing following the initiation of combustion events.
By using this combustion model, the chemical source term pcm in the species
continuity equation and the chemical heat release Qc in the energy equation
are replaced. A first-order accurate numerical scheme is employed to calcu-
late both the change in species density and the heat release due to chemical
reactions,
= -p(Ym - Y*)(l - e-*"r<) (47)
where p is the total density, and A Ms the numerical timestep. Thus, the equivalent
source terms are pcm = A/) w /Af and Qc = A jg/Af. Since the total chemical time
scale rc also includes the turbulent time scale, the effects of turbulence on mean
reaction rates are then accounted for.
C. Soot and NO Formation

To model NO emission, the extended Zel'dovich mechanism was used. A sim-
plified mechanism describing the formation and oxidation of soot27 was used here
to simulate the soot evolution in the engine. The net rate of soot formation is taken
to be the formation rate minus the oxidation rate,
dm,. _ dmsf _ dmso
( }
dt ~~ dt dt
which treats both the formation and oxidation processes as different single step
Arrhenius reactions
dm sf
dt
(50)
where Aff v , Msoot, ô» and P are the fuel vapor mass, soot mass, oxygen mole frac-
tion, and pressure, respectively. The values for kinetic parameters are Af = 100,
A0 = 30, Ef = 12500 cal/mol, and Eo = 14000 cal/mol.
V. Comparison with Experiments

A. Combustion Bombs
Here ignition modeling is achieved by implementing the Shell model12 into
the KTVA-II code. For purposes of numerical definition, ignition occurs when the
llatm
10
0)
Q
o>
1.1 1.2 1.3 1.4 1.5 1.6

1/T (1000/K)
Fig. 2 Comparison between predicted (dashed line) and measured19 (solid line)
ignition delay.
temperature exceeds 1000 K due to chemical reactions. This ignition model was
validated using experimental data measuring ignition delay of the diesel spray in
combustion bombs.19'20
Igura et al.19 heated a combustion bomb electrically at different initial tempera-
tures and pressures (ranging from 670 to ~ 900 K and from 1 to ~ 31 atm absolute,
respectively). Ignition was detected by a photodetector such that the illumination
delay time was recorded. The comparisons between measured and predicted igni-
tion delay are shown in Fig. 2. Good agreement was obtained. The decrease of the
ignition delay corresponding to the increase of initial bomb temperature and/or
pressure is fairly well predicted.
Similarly, ignition delay measurements by Edwards et al.20 are also used to
validate the model. The initial bomb temperatures vary from 688 to 1310 K with a
uniform initial pressure of 30 atm. As shown in Fig. 3, a good level of agreement
between measurements and predictions can also be observed.
In a well-characterized environment, such as that of a combustion bomb, the
model constant could be modified to best fit the measured ignition delay. However,
when the model is applied to diesel ignition modeling, it is desirable not to change
the model constants case by case. Therefore, no further modification is made when
the Shell model is used to model diesel engine ignition delays in the following
studies.
B. Diesel Engines
1. Caterpillar Engine
The Shell ignition model was combined with the characteristic-time combus-
tion model to simulate the overall combustion processes of a Caterpillar diesel
engine28 (model E 300, arrangement no. 1Y0540). The specifications and oper-
ating conditions of this engine are listed in Table 1. Notice that the spray angle
is 26.5 deg from the head which makes the spray droplets impinge on the piston
1000/T
0.6 0.8 1 1.2 1.4 16
1 00 P————————I————————I————————I—————————\—————
10 Predicted
I
I
Q)
0.1 - I
1300 1100 900 800 700
Core Temperature at injection (K)
Fig. 3 Comparison between predicted and measured20 ignition delay, initial bomb
pressure 30 atm.
surface shortly after injection. The effects of spray-wall interactions are found to
be important in this engine.
The computational grid together with the droplets distribution in the combustion
chamber is shown in Fig. 4 for the baseline case. The computed pressure is
compared with the measured pressure in Fig. 5. The computed heat release rate data
are also compared with those obtained by applying a zero-dimensional model to
the measured pressure data as seen in Fig. 6. More detailed in-cylinder temperature
contours and droplet distribution in a sequence of crank angles during ignition are
presented in Fig. 7. Ignition can be seen to have occurred by 3 deg before top dead
center (TDC).
Table 1 Caterpillar engine
Specifications:
Cylinder bore, mm 137.6 Stroke, mm 165.1
Compression ratio 15.1 Displacement volume, L 2.44
Number of nozzle orifice 6 Nozzle hole diameter, mm 0.259
Spray angle (from head), deg 26.5 Combustion chamber Quiescent
Piston crown Mexican hat
Operating conditions:
Engine speed, rpm 1610 Overall equivalence ratio 0.46
Fuel Tetradecane Injection duration, deg 21.5
Start of injection Baseline case: — 11 atdc
Other cases: -15; -13; —8; -5 atdc

Fig. 4 Perspective view of computational grid and fuel droplet distribution at 5-deg
ATDC, computation domain in one-sixth of the whole combustion chamber since the
injector has six nozzle holes.
Sensitivity studies using different ignition, combustion, and spray models have
been performed. The results using the standard KIVA-II, which uses a single-
step reaction rate for both ignition and combustion modeling and an unmodified
spray model, are also shown in Fig. 5. The standard KTVA-II model computations
suffer from too fast combustion at the early stage of combustion due both to
the inadequate simulation of the low-temperature chemistry and to the failure
to account for turbulent effects on the mean reaction rate. The high-combustion
temperature resulting from the rapid reactions tends to vaporize fuel spray quickly
so that the liquid fuel is unable to penetrate into the combustion chamber. Hence,
the combustion is restricted to a small region and the flame does not spread around
the chamber. This results in low cylinder pressure and a large amount of unburned
fuel vapor around the injector regions. The accurate prediction of spray penetration
is, thus, crucial to the combustion modeling.
In the unmodified spray model, the drop is taken to be a sphere which has a
lower drag coefficient than a distorted drop. The low-drag coefficient results in
-100 -50 0 50 100

Crank angle (degrees)
Fig. 5 Computed and measured cylinder pressure for baseline case of a Caterpillar
engine.

-20 -10 0 10 20 30 40
Fig. 6 Computed and measured (from zero-dimensional simulation) heat release rate
for baseline case of a Caterpillar engine.
- 3 ATDC
h = 1710 K
1 = 796 K
- 2 ATDC
h = 2000 K
1 = 930 K
TDC
h =2220 K
1 = 955 K
Fig. 7 In-cylinder temperature and droplet distribution in a sequence of crank an-

gles. Since all the droplets are projected on a plane, some droplets appear outside the
piston outline.

-100 -50 0 50 100

Fig. 8 Effects of drop drag and wall impingement on the combustion processes.
high-relative velocity between the drop and the gas so that the breakup droplets
have small sizes [see Eqs. (16) and (19)]. These small droplets vaporize quickly,
and the combustion is enhanced, producing a high cylinder pressure shortly after
ignition. Figure 8 shows the results using the standard drag coefficients. For
spray penetration, it is worth noting that drag coefficients have different effects
for vaporizing and nonvaporizing sprays. For nonvaporizing spray, the higher drag
coefficient of distorted drops tends to reduce their velocity and penetration. But the
low-relative velocity results in larger breakup drops which carry more momentum
and can penetrate farther. These two effects compete with each other, and the
results showed that the modification of drop drag coefficient does not change the
penetration of a nonvaporizing spray.10 On the other hand, for a vaporizing spray,
this study shows that the modified drag coefficient helps the spray penetration
mainly because the large breakup drops do not vaporize easily and can keep
penetrating.
£100 Injected
80
Present
60 Model
O
no wall
4
§
C
° breakup
0)
g 20
Q_
-10 o 10 20 30
Fig. 9 Effects of wall impingement on combustion by enhancing the atomization
(present model).
0.6
0 20 40 60 80 100
Fig. 10 In-cylinder soot and NO evolution compared to engine exhaust measurements
for the baseline case.
However, using the dynamically varying drop drag coefficients alone is not suf-
ficient to match the experimental data in the Caterpillar engine. It is also necessary
to consider the influence of wall-impingement effects as a further breakup mecha-
nism. The situation of a drop impacting a wall is thought to be similar to that of a
drop which is exposed suddenly to a high-speed air jet. Therefore, the breakup time
constant (B\ = 1.73) determined by Liu et al.10 was used after the drops hit the
wall. When wall-breakup effects were not considered, good agreement of engine
pressure data was not achieved. The corresponding fuel consumption rates are
shown in Fig. 9, which shows that reatomization effects due to wall impingement
can enhance the combustion.
By applying the soot and NO models, in-cylinder soot and NO evolution can
also be obtained as in Fig. 10. The emission is plotted in gram per brake horse
power-hour (g/bhp-hr) to allow comparison with the regulations. Measured ex-
haust values by Nehmer28 are also pointed out in the figure, and the agreement
Table 2 Tacom engine
Specifications:
Cylinder bore, mm 114.3 Stroke, mm 114.3
Spray angle (from head), deg 7.5 Combustion chamber Quiescent
Engine speed, rpm 1500 Fuel Tetradecane
Injection timing____Equivalence ratio Intake air temperature
Baseline case: —10.5 ~ 14.5atdc 0.5 334
Case 1: -10.5 ~9.5atdc 0.3 334
Case 2: -10.5- 14.5atdc 0.5 311

0 10 20 30 40 50 60
-50 0 50
Crank angle (degrees) Crank angle (degrees)
co 10
Q_
0 10 20 30 40 50
-50 0 50
0 10 20 30 40 50
-50 0 50
Fig. 11 Comparison of predicted (dash lines) and measured (solid lines) engine data
under different injection timing: (from top) 15,13,8, and 5 degrees BTDC.
is satisfying. The soot formed during the early stage of combustion oxidizes at
the later stage due to high cylinder temperatures. The majority of NO is formed
during the diffusion burning phase, and the reactions freeze during the later part of
combustion. The same NO formation phenomena have been observed in dumping
experiments by Donahue.29
The proposed ignition, combustion, and modified spray models were further
applied to simulate combustion processes for the Caterpillar engine under various
operating conditions, differing mainly by their injection timings. In four different
runs injection started at 15,13,8, and 5 crank angle degrees before top dead center,
whereas the injection rate shape and duration were the same. As can be seen in
Fig. 11, computed cylinder pressures and heat release rates are in good agreement
with the measured data without a case-by-case change in any model constant. In
addition, the corresponding soot and NO computations have been made, and the
measured tradeoff of soot and NO emission is well represented.30"32
2. Tacom Engine
The present model was also applied to model a Tacom diesel engine29 with spec-
ifications and operating conditions shown in Table 2. This engine has a shallower
bowl and smaller displacement volume than the previous Caterpillar engine. Also,
the spray angle is only 7.5 deg from the head, which confines the fuel spray inside
the combustion chamber. In this case, the fuel spray tends to vaporize faster and
no strong wall impingement is observed, except when some droplets hit the far
side of the piston surface, as seen in Fig. 12.
The computed and measured pressure data and heat release rate are shown in
Figs. 13 and 14, respectively. The computed NO evolutions are also compared to
the measured data by the dumping experiment29 as shown in Fig. 15. The model
was also used to simulate two other cases with different operating conditions, case
1 with the equivalence ratio changing from 0.5 to 0.3 and case 2 with the intake
air temperature changing from 334 to 311 K. Computed and measured pressure
data are shown in Fig. 16.
>TDC h = 2400K 1 = 940 K
+5ATDC h = 2420K 1 = 942 K
Fig. 12 In-cylinder fuel droplets distribution and temperature contours of the Tacom
engine. The spray has minimal wall interactions.
Table 3 Cummins engine
Specifications:
Cylinder bore, mm 139.7Stroke, mm 152.4
Spray angle (from head), deg 18 Combustion chamber Quiescent
Engine speed, rpm 1689 Fuel Tetradecane

Injection timing -18-' 1 1 atdc Equivalence ratio 0.4
3. Cummins Engine
The experimental results from a Cummins NH engine by Yan and Borman33
were also used to validate the combined model. The engine specifications and op-
erating conditions are listed in Table 3. Mainly the cylinder pressure data are used
for comparison. Good agreement between the measured and computed cylinder
pressure data is obtained as shown in Fig. 17.
VI. Summary and Conclusion

An integrated model for spray combustion processes in diesel engines is pre-
sented in this study. Good levels of agreement are obtained between computed
and measured data, which include spray ignition delay and engine measurements.
Further investigations of emission models and also the prediction of engine per-
formance are now possible.
The accurate prediction of spray dynamics is crucial to combustion modeling.
The effects of drop distortion on the drag forces must be considered. A distorted
drop has a higher drag coefficient than a spherical drop. The momentum exchange
12
Computed
10
co 8
a.
g> 6
3
•100 -50 0 50 100

Fig. 13 Computed (dashed line) and measured (solid line) cylinder pressure data for
baseline case of the Tacom engine.
1600
°20 -10 0 10 20 30 40 50
Fig. 14 Computed and measured (from zero-dimensional simulation) heat release

rate data for baseline case for the Tacom engine.
between drops and surrounding gas is then enhanced for distorted drops. This
effect slows down the liquid breakup and, consequently, improves the penetration
of combusting diesel spray. A better spray droplet distribution is then obtained
which sets the stage for accurate combustion computations.
Wall impingement effects have a strong influence on the atomization and
breakup processes. Liquid atomization is enhanced due to the strong disturbance
imposed on liquid drops by wall interactions.
To predict the ignition delay accurately, the low-temperature chemistry of hy-
drocarbon oxidation must be treated using multistep ignition kinetics. The reduced
kinetics mechanism, Shell model, is sufficient to predict the diesel spray ignition
in both combustion bombs and diesel engines over a wide range of operating
conditions.
12
10 S S
Q.
.C
.a Computed
S 6
4 ~
2 -
i
0 20 40 60 80
Fig. 15 Computed in-cylinder NO evolution compared to dumping experimental

data for baseline case of the Tacom engine.

12 |——————————I——————————T
-100 -50 0 50 100

Fig. 16 Computed (dashed line) and measured (solid line) pressure for Case 1 (lower
equivalence ratio) and Case 2 (lower intake air temperature) in Table 2 (Tacom engine).
Postignition combustion must include the effects of turbulence on the mean

reaction rate. The eddy breakup time scale (k/s) is incorporated in the combustion
model to account for the turbulence mixing effects. However, the laminar chemistry
still cannot be neglected during the premixed burn phase. An adequate arrangement
is to use the laminar chemistry to initiate combustion and then incorporate the
increasing influence of turbulence on the reactions.
In this study, the overall combustion process and in-cylinder thermodynamic
conditions are well predicted. Strategies for engine emission control can then
be investigated by using a predictive numerical model. Further development of
emission models is needed to study the pollutant formation mechanisms.
3500
-40 -20 0 20 40
Fig. 17 Comparison between predicted and measured cylinder pressure data to-
gether with computed heat release data of the Cummins engine.
Acknowledgments
This work was supported under NASA Lewis Grant NAG 3-1087 (W. T. Win-
tucky and M. Valco, Grant Monitors). Also appreciated is support for computations
from the Army Research Office, Cray Research, Inc., and Caterpillar Engine, Inc.
The authors thank Greg J. Hampson for help with the emission models.
References
^msden, A. A., O'Rourke, P. J., and Butler, T. D., "KIVA-II - A Computer Program
for Chemically Reactive Flows with Sprays," Los Alamos National Labs., LA-11560-MS,
Los Alamos, NM, May 1989.
2
Reitz, R. D., "Assessment of Wall Heat Transfer Models for Premixed-Charge Engine
Combustion Computations," Society of Automotive Engineers, SAE Tech. Paper 910267,
1991.
3
Reitz, R. D., "Modeling Atomization Processes in High-Pressure Vaporizing Sprays,"
Atomisation and Spray Technology, Vol. 3, 1987, pp. 309-337.
4
O'Rourke, P. J., and Amsden, A. A., "The TAB Method for Numerical Calculation of
Spray Droplet Breakup," Society of Automotive Engineers, SAE Tech. Paper 872089,1987.
5
Taylor, G. I., "The Shape and Acceleration of a Drop in a High Speed Air Stream," The
Scientific Papers ofG. I. Taylor, edited by G. K. Batchelor, Vol. Ill, Cambridge Univ. Press,
Cambridge, England, UK, 1963.
6
Liu, A. B., and Reitz, R. D., "Mechanism of Air-Assisted Liquid Atomization," Atom-
ization and Sprays, Vol. 3, No. 1, 1993, pp. 55-75.
7
Wachters, L. H., and Westerling, N. A., "The Heat Transfer From a Hot Wall to Imping-
ing Water Drops in the Spheroidal State," Chemical Engineering Science, Vol. 21, 1966,
pp. 1047-1056.
8
Gonzalez, M. A., Borman, G. L., and Reitz, R. D., "A Study of Diesel Cold Starting
Using Both Cycle Analysis and Multidimensional Calculations," Society of Automotive
Engineers, SAE Tech. Paper 910180, 1991.
9
Naber, J. D., and Reitz, R. D., "Modeling Engine Spray/Wall Impingement," Society of
Automotive Engineers, SAE Tech. Paper 880107, 1988.
IO
Liu, A. B., Mather, D., and Reitz, R. D., "Modeling the Effects of Drop Drag and
Breakup on Fuel Sprays," Society of Automotive Engineers, SAE Tech. Paper 930072,
1993.
H
Zellat, M., and Zeller, H., "Modelisation multidimensionnelle de 1'auto-inflammation
dans un moteur Diesel—Premieres validations experimentales," Rapport Institut Francais
Petrole 35551, 1987.
12
Halstead, M., Kirsh, L., and Quinn, C, "The Autoignition of Hydrocarbon Fuels at
High Temperatures and Pressures—Fitting of a Mathematical Model," Combustion and
Flame, Vol. 30, 1977, pp. 45-60.
13
Zellat, M., Rolland, T, and Poplow, F, "Three Dimensional Modeling of Combus-
tion and Soot Formation in an Indirect Injection Diesel Engine," Society of Automotive
Engineers, SAE Tech. Paper 900254, 1990.
14
Dillies, B., Marx, K., Dec, J., and Espey, C., "Diesel Engine Combustion Modeling
Using the Coherent Flame Model in KIVA-II," Society of Automotive Engineers, SAE
Tech. Paper 930074, 1993.
15
Natarajan, B., and Bracco, F. V, "On Multidimensional Modeling of Auto-Ignition in
Spark-Ignition Engines," Combustion and Flame, Vol. 57, 1984, pp. 179-197.
16
Schaperton, H., and Lee, W., "Multidimensional Modeling of Knocking Combustion
in SI Engines," Society of Automotive Engineers, SAE Tech. Paper 850502, 1985.
17
Theobald, M. A., and Cheng, W. K., "A Numerical Study of Diesel Ignition," American
Society of Mechanical Engineers, ASME Paper 87-FE-2, 1987.
18
Kong, S. C, and Reitz, R. D., "Multidimensional Modeling of Diesel Ignition and
Combustion Using a Multistep Kinetics Model," Journal of Engineering for Gas Turbines
and Power, Vol. 115, 1993, pp. 781-789 (ASME Paper 93-ICE-22).
19
Igura, S., Kadota, T., and Hiroyasu, H., "Spontaneous Ignition Delay of Fuel Sprays in
High Pressure Environments," Transactions of the Japan Society of Mechanical Engineers,
Vol.41, 1975, pp. 1559-1566.
20
Edwards, C. F., Siebers, D. L., and Hoskin, D. H., "A Study of the Autoignition Process
of a Diesel Spray via High Speed Visualization," Society of Automotive Engineers, S AE

Tech. Paper 920108, 1992.
21
Magnussen, B. F., and Hjertager, B. H., "On Mathematical Modelling of Turbulent
Combustion with Special Emphasis on Soot Formation and Combustion," 16th Sym-
posium (International) on Combustion, Combustion Institute, Pittsburgh, PA, 1976, pp.
719-729.
22
Marble, F. E., and Broadwell, J. E., "The Coherent Flame Model for Turbulent Chemical
Reactions," Project Squid Rept. TRW-9-PU, 1977.
23
Abraham, J., Bracco, F. V, and Reitz, R. D., "Comparisons of Computed and Measured
Premixed Charge Engine Combustion," Combustion and Flame, Vol. 60, 1985, pp. 309-
322.
24
Reitz, R. D., and Kuo, T. W., "Modeling of HC Emissions due to Crevice Flows in
Premixed-Charge Engines," Society of Automotive Engineers, SAE Tech. Paper 892085,
1989.
25
Kong, S. C., Ayoub, N., and Reitz, R. D., "Modeling Combustion in Compression
Ignition Homogeneous Charge Engines," Society of Automotive Engineers, SAE Tech.
Paper 920512, 1992.
26
Bergeron, C. A., and Hallett, W. L. H., "Ignition Characteristics of Liquid Hydrocarbon
Fuels as Single Droplets," Canadian Journal of Chemical Engineering, Vol. 67, 1989, pp.
142-149.
27
Yoshizaki, T., Nishida, K., and Hiroyasu, H., "Approach to Low NOx and Smoke Emis-
sion Engines by Using Phenomenological Simulation," Society of Automotive Engineers,
SAE Tech. Paper 930612, 1993.
28
Nehmer, D. A., "Measurement of the Effect of Injection Rate and Split Injections on
Diesel Engine Soot and NOx Emissions," MS Thesis, Mechanical Engineering Dept., Univ.
of Wisconsin, Madison, WI 1993.
29
Donahue, R., "An Experimentally Determined Temporally Based Study of Tur-
bocharged Diesel Nitric Oxide Emissions," MS Thesis, Mechanical Engineering Dept.,
Univ. of Wisconsin, Madison, WI, 1993.
30
Kong, S. C., Hampson, G. J., and Reitz, R. D., "Modeling Diesel Sprays, Combustion,
Soot and NO Emissions," Sixth International Conference on Liquid Atomization and Spray
Systems, Rouen, France, 1994.
31
Hampson, G. J., Kong, S. C., and Reitz, R. D., "Multidimensional Modeling of Diesel
Combustion and Emissions and Comparison with Experiments," Combustion Science and
Technology, 1994, to be published.
32
Patterson, M. A., Kong, S.-C, Hampson, G. J., and Reitz, R. D., "Modeling the Effects
of Fuel Injection Characteristics on Diesel Engine Soot and NOx Emissions," Society of
Automotive Engineers, SAE Tech. Paper 940523, 1994.
33
Yan, J., and Borman, G. L., "Analysis and In-Cylinder Measurement of Particulate
Radiant Emissions and Temperature in a Direct-Injection Diesel Engine," Society of Auto-
motive Engineers, SAE Tech. Paper 881315, 1988.

Chapter 17
Spray Combustion Processes in Regenerative Liquid

Propellant Guns
Gloria P. Wren,* John D. Knapton,f and Terence P. Coffee*

Weapons Technology Directorate, U.S. Army Research Laboratory, Aberdeen
Proving Ground, Maryland 21005-5066
I. Introduction
T HE regenerative liquid propellant gun (RLPG) is a propulsion concept that has

been actively pursued by the U.S. Army for the past 10 years for integration
into a next-generation artillery weapon. Artillery guns are indirect fire weapons
that must be able to hit targets at ranges of 2-50 km or beyond as opposed to tank
guns, which are line-of-sight, direct fire weapons. Research in liquid propellants
(LP) for military applications has been ongoing for the past 50 years. During this
time, a number of designs and propellants have been experimentally tested and
theoretically analyzed both in the U.S. and abroad, leading to significant advances
as the technology has matured. The U.S. Army is now routinely firing large-
caliber RLPGs and evaluating their potential for the battlefield. LP guns have a
unique requirement, compared to most applications that use liquid propellants,
for high pressure that is generated from the combustion of LP. This requirement
necessitates a high combustion rate with concomitant effect on required rates
of liquid injection, breakup, and combustion. In current Army applications, for
a 155-mm gun (i.e., the diameter of the gun tube is 155 mm), as many as 14
liters of LP must be injected and burned over an extreme range of pressure from
atmospheric to 500 MPa in a ballistic time scale on the order of 25 ms. Less
stringent civilian applications for military liquid propellants, such as mining,
tunneling, and automobile air bag inflation, are also being researched. This paper
reviews spray combustion processes in RLPGS (i.e., guns using the combustion
chamber pressure to initiate and sustain the flow of LP) with emphasis on sprays
in current U.S. Army designs. To provide a perspective for the evolution of the
sprays encountered in RLPGs, however, we first review major milestones in the
development.
rights reserved.
Team Leader, Propulsion Modeling and Analysis.
^Team Leader, Liquid Propulsion.
^Senior Research Mathematician.
425
426 G. R WREN ET AL.
II. Historical Perspective

A. Background
Liquid propellants have long been considered an attractive energy source for
propulsion systems.1 As related to guns, the advantages that were recognized in
the early studies continue to be relevant today. Advantages that served to motivate
much of the early work were related to bulk storage, cost, logistics, charge and
projectile loading, safety, high-energy density, and ability to tailor the combus-
tion chamber pressure. In addition to high-energy density, found especially with
bipropellants, some of the early propellants also offered a low flame temperature
and low molecular weight. The early propellants were often toxic and highly
corrosive, however, presenting serious problems that would not be acceptable by
today's standards.
Controlling the combustion of an unstable liquid surface while using a con-
figuration suitable for a practical systems application is not a trivial problem.
The simplest approach is to fill the combustion chamber with the LP and burn
the propellant in a bulk mode (referred to as bulk-loaded); the same chamber
configuration is used in solid propellant guns. Despite the uncertainties in the
generation of the burning surface area and the fundamental combustion mech-
anisms, the bulk approach has worked surprisingly well. Gas generation rates,
however, are susceptible to small and probably unavoidable variations during
ignition and, hence, result in ballistic variations that are not acceptable for gun ap-
plications. The projectile velocity coefficients of variation for groups of repeated,
controlled laboratory firings are generally between 1 and 2%, as much as five
times worse than that achieved with either conventional solid propellant guns or
with the advanced injector system used in today's regenerative guns. Control of
the variation in muzzle velocity is important in applications such as artillery that
utilize muzzle velocity as a major factor affecting precision, and, thus, probability
of hit.
One approach that has demonstrated promise for controlling the combustion
of liquid propellants is the regenerative injection concept in which the LP is
injected into the combustion chamber by moving piston(s). The feasibility of
this concept for controlling combustion was first demonstrated at both the Jet
Propulsion Laboratory2 and Experiment Inc.3 Testing was performed at these
organizations during the late 1940s and early 1950s. Since combustion control
and overall system feasibility are the primary objectives of a practical system, we
include in our review comments about the injector designs that are relevant to the
types of sprays encountered in current RLPGs being studied by the military.
Other important system design issues for the injection orifice, however, are not
discussed. These issues, discussed in detail as early as 1951,4'5 include methods for
injector orifice closure, piston design, type of piston, dynamic seals, method for
piston deceleration, and system implementation. Piston design, for example, has
to be sufficiently rugged for the type of injector to avoid excessive stress concen-
trations. Orifice closure, as another example, was a serious safety concern during
the early studies, especially for those applications using hypergolic propellants.
These issues, together with the combustion characteristics of the propellant and
the chamber boundary conditions, are coupled, and a thorough analysis requires
both a finite element analysis of mechanical components including the injector
and an appropriate multidimensional interior ballistic model.
An additional system-related issue important for applications such as artillery,

requiring a variable injection area to control the pressurization rate in the com-
bustion chamber, was also considered during the early studies. One design used a
movable orifice plug6 for varying the injection area. A second design, described
subsequently, generated an annular injection spray based on a variable diameter
chamber.7 A third design,8 also discussed later, used a movable center piston,
called a valve closure, that was used to initially seal the orifice.
In the remainder of this section we examine some of the early injection concepts
and identify problem areas that eventually led, during the 1980s, to the type of
annular injector that generates annular LP sheets and is the basis for the present
regenerative gun. To illustrate the various injector designs, we select some of the
more interesting concepts, together with test results, from studies conducted at
the Jet Propulsion Laboratory (JPL), Experiment Inc., and the General Electric
Company (GE).
B. External Injection
One of the early concepts consisted of an externally powered injector similar
to what has been used in rocket motors. This approach was abandoned rather
early due to the requirement for a source of high-pressure gas external to the gun
chamber to force the propellant into the chamber. The power requirements for
such a system were considered excessive for a practical system.9 The development
of the regenerative injection concept, however, offered an approach whereby the
high pressure developed during the combustion of the propellant could be used to
supply the necessary pressure for injecting the liquid propellant.
C. Shower-Head Injectors
During the 1940s and early 1950s liquid propellant gun work used both mono-
propellant and bipropellant formulations. Most of the monopropellant (fuel and
oxidizer are combined in a single LP) tests used a hydrazine-based propellant, and
most of the bipropellant (fuel and oxidizer are separate components) tests used
hydrazine and either hydrogen peroxide or red fuming nitric acid for the oxidizer.
Various types of injectors were tested and generated LP sprays with widely varying
characteristics.
The first type of regenerative injector consisted of holes drilled in the head
of a differential area piston. This type of injector has been called a shower-head
injector and is illustrated in Fig. I.10 As the name implies, the sprays generated in
a shower-head injector consist of multiple streams of propellant that may interact
with each other as the LP is injected. The early injection systems often incorpo-
rated more than one injector. Thus, the same fixture could be used to test either
monopropellants or bipropellants with the resultant implications for the combus-
tion process. Monopropellants do not require mixing to combust, whereas the
fuel and oxidizer components of bipropellants must be mixed for combustion to
occur. The shower head injectors tested at JPL2 and later at Experiment Inc. in a
127-mm gun3 were mounted radially on the outside of the gun chamber, shown
conceptually in Fig. 2. Injection results from an increase in chamber pressure,
which causes a displacement of the differential area piston (described later).
Interest in achieving high velocity prompted consideration of approaches that
used shower-head injectors located down the barrel of the gun. A conceptual
428 G. P. WREN ET AL
Fig. 1 Illustration of shower-head type of regenerative injector.10
illustration is shown in Fig. 2. Much later, this type of down-barrel injector was
tested and showed that the chamber pressure-vs-time curve could be altered11;
however, whether there was any significant increase in gun performance in terms
of muzzle velocity was never determined.
D. Jet Propulsion Laboratory Studies of Shower-Head Injectors

Radially mounted injectors, although offering an acceptable approach for con-
trolling combustion, impose severe systems implications due to their size and lo-
cation on the outside of the chamber and, as we shall see, difficulty in developing
a practical orifice closure system with a high discharge coefficient. Nevertheless,
the results are of interest and provided the first indication of the feasibility of the
regenerative concept. The JPL data2 supporting one of the early claims for concept
feasibility are summarized in Table 1. The injector used in these tests is illustrated

Fig. 2 Radially mounted shower-head injectors showing both chamber and down
bore radially mounted injectors.6
in Fig. 3. For expediency, orifice closure in the JPL shower-head injector tests was
achieved by using various methods such as beeswax plugs, rubber plugs, grease,
and electrical tape. The projectile mass for all tests was 110 g with the exception
of group 3, which used a 225-g projectile.
Tests using a solid propellant charge, Table 1, yielded a velocity coefficient
of variation of 0.35%, which compares favorably with 0.37% for one of the
liquid propellant groups (LP 4). Importantly, the authors noted that the pressure-
vs-time characteristics were similar between the solid propellant tests and the
liquid propellant tests. A brief examination of the data in Table 1 is of interest.
Groups LP 1 and 2 used only one piston (eight no. 24 drill-size holes). The same
propellant mixture of hydrazine and water (96% N2H4 -f- 4% H2O) was used in
both groups and was fired under similar conditions except group 2 had 12% more
liquid propellant. As expected, group 2 resulted in a higher pressure. Not expected
was the lower velocity for group 2. It is postulated that the lower velocity for
group 2 may have been associated with the change in the ignition system, from
solid (M52A3 electric primer and 1.2 g of a solid propellant booster) to a liquid
Table 1 Summary of 20-mm reproducibility tests at the

Jet Propulsion Laboratory2
Total injection areab Pressure Velocity

Booster
Propellant charge, No. of Piston 1, Piston 2, Max, /Mean, /Mean,
(group) C/M* 8 tests cm2 cm2 MPa % m/s %
Solid 0.35 — 8 — 379 4.7 968 0.35
LPc 1 0.38 1.2 5 0.937 not used 249 4.3 827 1.14
LP2 0.43 (LP) 5 0.937 not used 308 2.2 811 0.71
LP3 0.28 4.0 5 0.460 0.460 309 6.5 619 2.41

LP4 0.60 4.0 4 0.937 0.669 377 3.7 904 0.37
LP5 0.64 4.0 5 0.937 0.669 418 7.2 915 0.97
LP6 0.71 (LP) 5 0.669 0.460 311 3.1 857 1.06
a
Charge-to-mass ratio (ratio of propellant mass to projectile mass).
b
Differential area of the piston was 1.3.
c
Liquid propellant (for the groups listed, various mixtures of hydrazine, hydrazine nitrate, and water
were tested).
igniter. The liquid igniter was a hypergolic igniter consisting of 1.5 cm3 of red
fuming nitric acid (RFNA) injected into 3 cm3 of (56% N2H4 + 40% N2H4NO3
+ 4% H2O). Test results over the years have often demonstrated a dependence of
ballistic performance on the early gas generation rate. The results from groups 1
and 2 suggest the importance of properly controlling the interaction between the
early injected propellant and the igniter. The velocity coefficient of variations for
groups 1 and 2 were 1.14% and 0.71%, respectively.
Group 3, fired with (58% N2H4 -f 22% N2H4NO3 + 20% H2O) and the heavier
projectile had approximately the same total injection area (0.920 cm2) as groups 1
and 2, but two opposing pistons (each with eight no. 36 drill-size holes) were used.
Assuming unburned propellant traversed the chamber, the propellant jets would
impinge on each other, resulting in a breakup of the jet into finer drops. A result
would be a faster mass burning rate, assuming the linear burning rates for the
different propellants are similar. For these conditions, Wren et al.12 showed, much
later, that such conditions would be desirable for suppressing pressure oscillations.
Unfortunately, the JPL report did not include any chamber pressure-vs-time data
for groups 1-3, so we cannot determine whether or not our interpretation is correct
for a change in the mass burning rate. The velocity coefficient of variation for group
3 was 2.41%, the worst among the six groups.
Group 4 yielded the best uniformity in the velocity coefficient of variation.
This test was performed with two opposing pistons, one piston with eight no. 30
drill-size holes and one piston with eight no. 36 drill-size holes. Without additional
tests, we can only speculate on whether this combination approached an optimized
condition that would more accurately control combustion. A comparison with
group 5 shows that the same type of pistons was used; however, the velocity
coefficient of variation degraded to 0.97%. Test conditions for groups 4 and 5
were similar with the exception of the propellant composition. Group 4 used (85%
N2H4 + 6% N2H4NO3 + 9% H2O), and group 5 used (54% N2H4 + 27% N2H4NO3
4- 19% H2O) for the propellant.
DIFFERENTIAL PISTON
MONOPROPELLANT
CHARGE
LIQUID - IGNITER
FITTING
CO
O
O
N2 PRESSURE
03
CO
no
31
O
O
m
CO
LIQUID IGNITER CO
m
CO
INITIAL COMBUSTION VOLUME
Fig. 3 Illustration of radially mounted injector used at the Jet Propulsion Laboratory.2
Group 6, tested with (58% N2H4 + 21% N2H4NO3 + 21% H2O), was similar
to group 3 but was tested with a lighter projectile and had two opposing pistons—
one with eight no. 30 drill-size holes and one with eight no. 36 drill-size holes.
The velocity coefficient of variation was 1.06%, considerably better than the data
recorded for group 3 using a propellant with a similar composition. The difference
in repeatability between the two groups suggests that ballistic repeatability may
be sensitive to both the injection area and the projectile mass.
Although the shower-head type of injector successfully demonstrated feasibility
of the regenerative type injection concept, there were some important injector-
related problems identified during the JPL tests that involved flashback into the
propellant reservoir. As its name indicates, flashback is a condition in which
combustion gases in the combustion chamber flow backwards into the liquid
propellant reservoir, initiating undesirable and sometimes damaging combustion.
The studies showed that flashback could occur during the startup of the injection
process, potentially the most serious time in the ballistic event due to the large
volume of confined liquid propellant. Flashback can also occur during injection. It
was found that the tendency for flashback could be reduced by using a sufficiently
low startup pressure (less than 14 MPa) that can be controlled by the igniter output,
by the use of small injection orifices, and by the composition of the hydrazine
propellant. Flashback into the reservoir continued to plague the early development
work, and additional examples are shown subsequently.
E. Experiment Inc. Studies of Shower-Head Injectors

For practical implementation, consideration was given in the early injector de-
signs for sealing the propellant using some type of mechanical orifice closure. For
small-caliber guns, however, an orifice closure was not considered necessary. The
first small-caliber (0.60 cal., 15.24 mm) gun using a bipropellant in an automatic
weapon at Experiment Inc.5 did not rely on an orifice closure valve. An analysis of
the dynamic fill-and-fire sequence indicated that the filling of the propellant reser-
voirs and loss of fluid through the small open injection orifices could be neglected.
An illustrative sketch of the injector is shown in Fig. 4. A design requirement
for this particular injector was a firing rate of 2000 rounds/min (Ref. 4) and a
propellant flow rate of 2 1/s (Ref. 5).
Experiment Inc. further developed the radial injector3 by incorporating an ori-
fice closure (Fig. 5) with an adjustable opening pressure, generally set at 1.7 MPa.
This type of injector was tested in a 127-mm regenerative gun using anhydrous
hydrazine and RFNA. The maximum charge weight, fired with two sets of injec-
tors, was approximately 8.16 kg. Average mass injection rate for the final four
tests was estimated at 1020 kg/s with a differential injection pressure of 55 MPa.
The use of radial injectors, although cumbersome and posing serious systems
implications, nevertheless offers an approach for increasing gun performance by
using down-barrel injection that places the high-pressure combustion zone closer
to the base of the projectile. The tests summarized in Table 2 used two stages
of injection; the first stage was located in the chamber and the second stage was
located 1.27 m from the centerline of the pistons in first stage. Each stage consisted
of two hydrazine injectors and one RFNA injector.
The igniter for the 127-mm tests consisted of 55—90 g of Bullseye pistol powder
contained in a cardboard tube and was ignited by an electric match. Igniter pressure
rise rates were rather low, between 0.5 and 0.7 MPa/ms. Maximum igniter pressure
was given at 11 MPa.
LU
1
LU UJ LU
i a-
CD
LU
I
QC
.a
433 SPRAY COMBUSTION PROCESSES

1. Injection Piston 4. orifice closure Springs

2. Retaining Plug 5. orifice Valve
3. seals *• Orifice Plate
Fig. 5 Illustration of a radial shower-head type of injector used in 40-mm and 127-
mm tests.14
The 127-mm gun was fired with a charge-to-mass ratio (C/M, ratio of propellant
mass to projectile mass) up to 1.54 and yielded a maximum velocity of 1522 m/s.
The rather low velocity for such a high C/M is attributed to poor mixing of the fuel
and oxidizer, a problem that was evident in several bipropellant programs. Besides
poor mixing, it was also pointed out by the authors13 that lack of synchronized
displacements of the fuel and oxidizer pistons could also contribute to poor mixing.
The early tests had a low projectile shot-start pressure (the pressure at which
the projectile first begins to move) that was controlled by a break bolt and that also
affected performance, a problem that was corrected by increasing the shot-start
pressure from 18 MPa to about 23 MPa. The fuel-oxidizer mixing problem was
significantly improved with the higher shot-start pressures. For the high shot-start
pressures, it was claimed that chamber pressures were greater than 90% of the
theoretically predicted pressures.
It was noticed that tests up to test 43 produced a yellow or yellowish-white
flash and relatively large amounts of bluish smoke and NO2. All subsequent tests,
although described as being fuel rich, produced an intense white flash and little
smoke.
Perhaps one of the more interesting results of the tests was the generation of
a relatively flat pressure that is associated with ideal ballistic performance. Since
Table 2 Summary of tests with a 40-mm regenerative liquid propellant gun

using two or three injectors8
Chamber Pressure Velocity

Total
injection area,, No. of Mean, /Mean, Mean, /Mean,
Test series C/M cm2 tests MPa % m/s %
I two injectors 0.153 2.568 7 228 6.3 632 0.72
II two injectors 0.153 1.265 3 129 3.1 579 0.79
III two injectors 0.153 1.265 3 110 8.3 580 0.63
IV two injectors 0.153 1.265 5 130 7.9 584 0.53
V three injectors 0.23 1.897 5 175 5.3 670 0.96
the structural elements of the gun such as the tube have a maximum pressure limit,
and the projectile has a maximum load limit, a flat pressure-time curve at the
defined limiting pressure produces the maximum projectile velocity for a given
gun/projectile/propellant system. Test 51 was particularly noteworthy; a flat mean
pressure of about 130 MPa was recorded for almost 2 ms. Test 50, fired under
similar conditions as test 51, gave approximately the same maximum chamber
pressure, but had a higher velocity. The discrepancy is attributed to a different
injector geometry, resulting in a change in the flow and mixing characteristics of
the oxidizer and fuel.
Two flashback problems occurred (tests 23 and 30) during the early tests.
Possible causes were listed as: 1) burning through the orifice plate during steady
flow, 2) sensitized propellant as a result of a contaminant, 3) compression of an
air bubble in the top injector, and 4) burning through the orifice plate as a result of
unsteady flow. The first two possible causes were not considered likely. Ignition
as a result of an air bubble compression also was not considered likely due to past
experience, although the presence of an air bubble could contribute to pressure
oscillations in the reservoir due to a water-hammer effect during startup, thereby
enhancing the possibility of flashback. The problem of pressure oscillations in
the reservoir during the early pressure rise was addressed much later by Knapton
and Coffee.14 To reduce the effect of the oscillations associated with a compliant
system, they recommended that the propellant be prepressurized to at least 1.5
MPa prior to ignition.
It was also reported that some galling or scoring of surfaces occurred during the
tests; however, the problem generally was not severe and could be corrected by
hand polishing. Also, occasional extrusion of high-pressure o-ring seals occurred.
As a test precaution, the seals were inspected after each firing and replaced after two
or three firings. There were few attempts to optimize performance by increasing
the orifice discharge coefficient until the end of the 127-mm program. It was noted
at the conclusion of the 127-mm program that the lack of good agreement between
theory and tests may have been due partly to a low discharge coefficient.13
The last seven tests were fired under similar conditions with the same total
injection area. Small changes were made in the orifice flow geometry to improve
mixing of the bipropellants. The mean velocity for six of the last seven tests,
disregarding test 52, was 1453 m/s with a coefficient of variation of 1.1 %. Including
test 52, the coefficient of variation was degraded to 2.1%. The reason for the
higher velocity for test 52 was not given, but presumably was associated with
the small changes that were made in the orifice geometry. The repeatability of
the tests was rather poor, especially when compared with repeatability tests with
today's regenerative injectors (summarized subsequently). Reasons that probably
contributed to the poor repeatability were likely due to the slow chamber pressure
rise and possible variability on the displacements of the multiple injectors.
Although some significant accomplishments were made at Experiment Inc., it
was apparent that progress on the program was slowed due to the complexity of
multipiston systems, as well as the difficulties in handling hypergolic bipropellants.
An additional concept examined during the 1950s is of interest since it suggests
an approach suitable for automatic projectile loading. Propellant injection was
again through holes; however, with this concept the displacement of the orifice
valve controlled the rate at which the holes opened. This concept, shown in Fig. 6,
has a hollow piston that could be extracted for projectile loading.8 A motivation
for this design was a conclusion from prior tests suggesting that the best operating
injectors have the smallest outside diameter. For this reason, the 40-mm gun
Fig. 6 Schematic of a hollow piston with a movable orifice closure valve used in a
40-mm regenerative liquid propellant gun at Experiment Inc.8
designed by Swanson8 had four small injectors of the type shown in Fig. 6 and
were located in pairs around the chamber. The injector was the same type used
in the 127-mm tests discussed earlier and also in some 37-mm tests conducted
at Experiment Inc. The injection could either be radially into the center of the
chamber, or, as shown in the figure, diagonally with a flow component along the
axis of the chamber. In either case, flow would be displaced through the orifice
valve in the combustion chamber.
This concept, without the orifice plate, was tested much later at General Electric
Company. Both radial and diagonal flow paths were tested at GE, as well as flow
around the outside of the injection piston. After numerous tests, the concept
(referred to at GE as concept V) was abandoned due to a low safety design margin
caused by high stress concentrations near the injection holes. Uncontrolled ignition
around the outside of the injection piston, using a hydroxylammonium-nitrate-
based monopropellant, also occurred on some of the GE tests.
The results of the first tests at Experiment Inc. are summarized in Table 2.
The propellant was anhydrous hydrazine (95% N2H4). The orifice type of injector
shown in Fig. 6 was used, and tests were performed using two or three injectors.
The injection piston had a differential area of 1.43, and outside diameter was 64.5
mm. The initial chamber volume was 190 cm3.

Test series I, Table 2, was based on a relatively large total injection area. The
tests yielded rather long ignition delays followed by a rapid pressure rise. The
chamber pressure-vs-time curve was similar to pressure curves obtained from
bulk-loaded tests. It was postulated that the large injection area resulted in an
excessive flow rate, causing a quenching of the hot igniter gases by the injected
propellant. The test series illustrates one problem that can be encountered with
an injector that has an excessive injection rate. The tests also serve to emphasize
the importance of properly matching the injector with the igniter system and the
ignition characteristics of the propellant.
The tests after series I, listed in Table 2, were performed with smaller injection
orifices. The ignition delay was eliminated, and rates of pressure rise and maxi-
mum chamber pressures were lower. Interestingly, although the combustion during
subsequent test series was more effectively controlled by the injection, there was
not a significant improvement in the coefficient of variation, as summarized in
Table 2, for either the chamber pressure or velocity.
F. General Electric Company Studies of Shower-Head Injectors

Because of an Army interest in high-velocity guns, work on regenerative injec-
tors came to an end during the late 1950s, and interest in the regenerative injector
concept did not resume until the GE company at Burlington, Vermont, began
testing a simple shower-head type of injector, similar to that shown in Fig. I.15
The tests were first performed in small caliber using a 25-mm fixture. The GE
tests quickly demonstrated that combustion control could be achieved and ballistic
performance could be altered depending on the injection area and volume of the
propellant. Figure 7 illustrates the effect of injection area on chamber pressure.
This result, predicted by modeling, provides a clear indication that the injection
area could be used to control the shape of the pressure-vs-time curve.
The success of the early work at GE prompted interest in scaling the shower-head
injector to a size that could be used in a large-caliber gun and also in developing a
practical orifice closure. An interesting orifice closure design by Mayer16 is shown
on the top in Fig. 8. Although the design functioned as intended during firing tests,
it resulted in an increase in the number of flashbacks to the propellant reservoir
and irregular piston displacements.
To more thoroughly characterize the flow through the orifice, Bulman and
Mandzy17 conducted flow-visualization tests using a transparent orifice (Fig. 8)
for simulating the geometry of the spline orifice. The flow-visualization tests
clearly showed cavitation and flow separation in the orifice. Based on the gun
tests and the flow-visualization tests, it was concluded that the complex flow
path with the spline inserts aided in inducing cavitation and flow separation in
438 G. P. WREN ET AL.
INJECTION
AREA 3- 1 7 cm '
2.73 cm 3
^ ' 2.45 cm 3
2.03 cm 3
2.0 3.0 4.0

TIME ( m s )
Fig. 7 Effect of injection area on chamber pressure.15
the injector, conditions that can lead to flashback. Similar problems, including
uncontrolled ignition events in the injector, were encountered several years later
by Klingenberg,11 who tested various types of complex orifice closure mechanisms.
G. Annular Injectors
The low discharge coefficient associated with complex orifice flow channels
prompted a search for a more smoothly contoured injector. An orifice closure
mechanism, however, was still considered necessary, especially for the large ori-
fices required for medium- and large-caliber applications. Before examining the
more recent annular concepts, we first examine an annular injector concept that
was tested during the 1950s.7
H. Experiment Inc. Studies of Annular Injectors

One of the first type of annular sheet injectors (Fig. 9) was proposed by
Fleischhauer.7 This type of injector is similar to the first annular sheet injector
(concept V) used at GE. A total of 51 tests were performed at Experiment Inc. in a
40-mm gun using anhydrous hydrazine. The differential piston area of the piston
was about 1.2.
The preliminary tests with this type of injector resulted in a problem of quench-
ing the injected liquid propellant after ignition, similar to the problems encountered
with the hollow injectors described earlier. The delayed combustion problem again
serves to illustrate the importance of optimizing the orifice together with the igni-
tion system and the type of propellant. One test, at a C/M of 0.153, was performed
with hydrazine nitrate added to the hydrazine (72% N2H4 + 23% N2H4NO3 4- 5%
H2O) with no delayed combustion. The chamber pressure and projectile velocity
were 207 MPa and 745 m/s, respectively.
I. General Electric Company Studies of Annular Injectors

Later tests with the center hollow injector mentioned earlier showed that the
injector functioned properly with the propellant Otto-II; however, the higher pres-
sures associated with hydroxyl ammonium nitrate (HAN) propellants resulted
in poor performance. The injector was eventually abandoned due to low design
stresses that would be exceeded in the event of a flashback into the reservoir.

Fig. 8 Illustrations of spline ball check valve injector used at General Electric in
a shower-head type injector (top) and flow cavitation (bottom) in the same type of
injector.17 The flow rate for the test shown on the bottom was 0.290 1/s and the inlet
pressure was 0.414 MPa; diameter of the ball was 8.7 mm.
Fig. 9 Annular sheet injector.7
Because of the design limitations of the center hollow injector, GE investigated

a second approach consisting of an outer injection piston and an inner piston as
illustrated in Fig. 10. Several variations of this type of injector were considered,
including a concept in which the center piston was fixed and two concepts in
which both the center piston and the outer injection piston were free to move. The
remainder of this chapter focuses on sprays from the current gun configuration in
which both pistons move rearward during the ballistic cycle, creating both variable
flow area and greatly varying injection velocities. The propellant used in most of
the tests is XM46, originally called LGP1846, a liquid monopropellant described
in detail in Ref. 18.
8
I
I
Fig. 10 Illustration of the in-line annular injector (concept VI) with a tapered control
rod tested at General Electric Co. tests in 30 mm.
III. Sprays from Annular Injectors in RLPGs

A. Description of Flow from Reservoir Through Orifice
Present firings use a design referred to as a concept 6C (sometimes written as
concept VIC) regenerative liquid propellant gun that is shown schematically in
Fig. 11. The liquid propellant is initially loaded in the liquid reservoir in front of
the portion of the gun forming the rear of the reservoir, called the transducer or
valve block. This location is also between the two cylindrical, moving pistons. An
external igniter mounted on the side of the combustion chamber is used to initially
pressurize the combustion chamber, initiating the motion of the inner piston. The
inner piston is also called the control piston since its motion determines the mass
NOT DRAWN TO SCALE
Fig. 11 Concept 6C regenerative liquid propellant gun.
flow rate from the reservoir. The motion of the control piston is modulated by a
damper assembly in back of the piston which provides precise metering of the
LP into the combustion chamber. The motion of the outer piston, also called
the injection piston since it primarily forces the propellant into the combustion
chamber, is controlled by the area ratio between the liquid reservoir side and the
combustion chamber side (about 1.3 in current RLPGs). As liquid flows out of
the reservoir, the liquid pressure drops, the injection piston accelerates, and the
injection orifice closes. As the injection orifice becomes smaller, the injection
piston motion pressurizes the reservoir, raising the liquid pressure, decelerating
the injection piston, and opening the orifice wider. The communication between
the pistons via the LP in the liquid reservoir provides a self-correcting feature
in the design which provides a safety margin. In a 155-mm gun, the maximum
charge is approximately 14 liters, but the initial charge size varies depending
on desired projectile velocity. Thus, the initial chamber volume also varies, with
larger chamber volume associated with small initial charge size. At the maximum
separation normally encountered during firing conditions, the injection orifice is
about 2 cm wide.
A representative mean, filtered, pressure-time history from the combustion
chamber and the corresponding curve recorded in the LP reservoir for a 155-mm
concept 6C RLPG is shown in Fig. 12.19 The LP reservoir pressure is higher than
the combustion chamber pressure by a factor approximately equal to the hydraulic
ratio (i.e., ratio of projected injection piston area on the combustion chamber side
to LP reservoir side). The process is initiated by the ignition train that serves
to pressurize the combustion chamber and force both the control and injection
pistons to the rear, compressing the LP in the liquid reservoir. During the second
phase, the early thin spray from the LP reservoir is injected, and the cool liquid both
accumulates in the combustion chamber and becomes heated. When the LP is fully
ignited, the accumulated LP and the injected LP burn rapidly as shown by the steep
rise in the combustion chamber pressure. The area of the injection piston on the
chamber side is larger than the area on the reservoir side, providing the appropriate
250.0
0.0
0.0 5.0 10.0 20.0
time (mo)
Liquid Pressure and Chamber Pressure - Round 51 (line). Model with

Droplet Burning (dot).
Fig. 12 Representative combustion chamber (lower curves) and liquid reservoir
(higher curves) pressure vs time plots from a midcharge, 155-mm RLPG.
force balance required for continued injection of LP. The chamber pressure reaches
maximum pressure, and the pistons reach maximum acceleration. In some firings,
particularly at maximum charge, a pressure plateau can be observed at maximum
pressure, indicating a quasistable equilibrium between the gas being produced in
the combustion chamber and the gas flow into the barrel. The motion of the control
piston is then slowed by the damper, the injection of propellant ends, and all LP
is consumed. Finally, the combustion gases expand, and the projectile completes
its travel down the barrel.
The gun can be simulated accurately, as shown by agreement with a lumped
parameter code.19"23 The flow from the reservoir into the chamber can be approx-
imated as steady-state Bernoulli flow, although early modeling of flow through
the orifice in concept 6 guns theoretically investigated a possible variable dis-
charge coefficient24'26 reported by some studies of experimental data.27'28 Using
the steady-state Bernoulli equation, the mass flow rate is given by
mass flow rate = CDAv^/2gopL(pL — PC) (1)

in which CD is the empirical discharge coefficient to represent flow losses, Av
is the vent area, go is a conversion constant to convert pressure to metric units,
pL is the liquid density, pL is the liquid pressure, and pc is the pressure in the
combustion chamber. All loss terms (entrance losses, frictional losses, inertial
effects, flow separation, and cavitation) are lumped into the discharge coefficient.
In a gun firing, chamber and reservoir pressures as well as piston travels are
measured. Using a modified Tait equation of state (i.e., pressure is a power function
of density) for the propellant (liquid is compressible at gun pressures), the mass
flow rate and the discharge coefficient can be computed. In earlier guns, low
discharge coefficients for liquid flow from the LP reservoir were occasionally
seen. However, the concept 6C guns have not shown noticeable losses in the flow.
When used in simulations, a discharge coefficient of 0.95 has been accurate as
compared to experimental data for all concept 6C guns.19'29
One-dimensional models for the complete gun have also been developed.30 In
addition, lumped parameter24'25 and one-dimensional26 models for the flow from
the reservoir were developed to study possible transient features of the flow. The
models indicated that transient effects were not important for the designs of interest
and that the flow out of the reservoir quickly matched steady-state Bernoulli flow.
Earlier problems with obtaining free flow were possibly due to flow separation.
The concept 6C guns, with a more carefully designed smooth orifice, do not show
evidence of significant flow losses; however, the flow of the propellant at cold
conditions (Army specifications require operation down to — 45 °C) has only been
initially studied.31 The flow is expected to show little loss in current injector designs
from hot conditions to 60°C and down to — 40°C as a result of the almost constant
viscosity-vs-temperature characteristic in this range.32 At very cold conditions to
—50°C, however, the XM46 viscosity increases substantially, and flow through
the injector would likely be affected.
In some firings, however, hot gas from the chamber works its way back into the
reservoir (referred to as flashback); in some cases this leads to a reversal in the
motion of the injection piston. Flow separation has been suspected as a cause of
this anomaly. To investigate flow through the orifice for specific designs, a two-
dimensional finite element model of the reservoir was developed and used to study
flow from the reservoir into the chamber.33'34 The model indicates no tendency for
the flow to separate in the design studied. Another possible cause of flashback is
that vibration of the injection piston could open a boundary layer, allowing hot
gas to work its way gradually back into the reservoir. Work is in progress35-36 to
integrate the reservoir flow model with a model for piston vibration to study these
effects.
B. Description of Spray Combustion in Combustion Chamber

The liquid is injected at a very high rate. For a 7.2-liter charge (of a possible
maximum of 14.1 liters), the entire firing cycle is about 25 ms. The maximum rate
of liquid injection is about 1.5 x 106 g/s. The result is a very dense spray, which
is difficult to study both experimentally and theoretically.
Experimental research has provided some diagnostics of annular sprays at am-
bient pressures as great as 37 MPa (Ref. 37). In addition, diagnostics taken at
the Ernst Mach Institut in Germany have provided a clear indication of the jet
structure.38 The intact core converges toward the centerline of the chamber. Cap-
illary instabilities break up the core. Liquid is stripped from the jet beginning at
the point of injection (atomization regime). With the onset of combustion, large
(however, still submillimeter size) burning drops become visible, although no em-
pirical correlation of droplet sizes has been successful. Apparently, these drops
burn in vortices about the spray periphery. It is believed that these large drops are
due to coalescence of smaller droplets in vortices.
Most of the combustion in guns is at even higher pressures, where the burning
rate is expected to be substantially higher. Some information can be obtained
from gun data. If the rate of injection and the pressure in the combustion chamber
and the gun tube are known, an energy balance analysis reveals approximately
the amount of the injected liquid which has been converted to final combustion
products. For the 7.2-liter shot, at any time step the amount of unburnt liquid in
the combustion chamber has a maximum of around 350 cm3, implying that the LP
burns very rapidly after injection.
In the lumped parameter model RLPGUN developed at the Army Research
Laboratory (ARL), the injected liquid is assumed to instantaneously break up into

droplets, and the existence of an intact core is ignored.39 The assumption appears
reasonable since the flow velocities and pressure are high and the combustion gas
is dense. Droplets are assumed to form by aerodynamic forces. A correlation due
to Wolfe and Andersen40 that has shown good comparison with experimental data
is used. The Wolfe and Anderson correlation gives the droplet diameter d so that
d = 5.1426V-* D*alvlpl*p (2)

in which V is the relative velocity between the gas and liquid, D is the diameter
of the annular jet, OL is the surface tension of the liquid, /x/, is the gas dynamic
viscosity, pi is the liquid density, and pc is the gas density. The correlation predicts
droplets that are too small (liquid burns too rapidly after injection) compared to
experimental data. To maintain the functional form, the droplet diameter from the
above correlation is multiplied by 30 (Ref. 39). This scaling factor is probably
necessary to account for coalescence of droplets that cannot be accurately modeled.
With the preceding adjustment, the simulations agree with experimental data for
all calibers of concept 6C guns studied, including 30-mm, 105-mm, and 155-mm
guns, at all charge sizes.19'41-42
The development of a correlation that was applicable to a range of gun conditions
including the accumulation phase provided a predictive capability that had been
lacking, and lumped parameter models at ARL and Lockheed-Martin Defense
Systems have been used to predict gun performance both for experimental firing
programs and design studies.43'44 Agreement between model and experiment is
easier to obtain after the initial spray has developed into full spray since the
combustion becomes primarily injection driven. That is, the liquid is combusted
very rapidly after it enters the chamber at elevated pressure. Thus, a large change
in the assumed droplet diameter has only a small effect on performance predicted
by the model.
Intact core models, however, have also been investigated39 since diagnostic work
appears to support the formation of a liquid core.37'38 The assumption was made
that most of the liquid accumulation was in the core of the jet. Various intact core
models obtained from a study of the literature were implemented in the code and
assessed compared to experimental data. The predicted liquid accumulation did
not even qualitatively follow the experimental data, and a detailed account is given
in Ref. 39.
On injection into the combustion chamber and decomposition into droplets, the
Sauter mean diameter of the droplets is tracked in the model RLPGUN. The gas
generation rate is determined using a linear, pressure-dependent burning rate of
the form
r = bPn (3)
300.0
ao
.
0.0 10 10.0
time (ms)
Round ITS, 7.2 liter charge. C60 gauge.
Fig. 13 Experimental combustion chamber pressure (using XM46) data from
155-mm RLPG.
in which r is the rate, P is the pressure, b is the coefficient, and n is the expo-
nent. Both b and n are obtained from a fit to the measured pressure history data
obtained from a diagnostic device. The analysis of the data requires a surface
area. Burning rates for XM46 have been experimentally measured using both
closed chamber45 and strand burner techniques.46'47 Early attempts to measure the
pressure-dependent burning rate of HAN-based liquid propellants were only able
to study the regime to about 100 MPa (Ref. 46) and for strand burner measurements
it was necessary to gell the propellant to obtain any estimate of the linear regres-
sion rate of the propellant surface. Gelling the propellant introduced uncertainty
about the effect of the gelling agent on the measured burn rate, and there appeared
to be an indication of a change in the slope of the burn rate curve around 70 MPa
in early data.46 Thus, closed-chamber experiments with ungelled propellant were
conducted since these experiments were able to both eliminate the gelling agent
and dynamically achieve high pressures, although the surface area was not well
known. The measured burning rate had an exponent close to 2.0, indicating high-
pressure sensitivity of the propellant.45 With the availability of new strand burner
techniques that allowed pressures of up to 300 MPa XM46 was retested.47 The
pressure exponent was measured as 1.2, again indicating high-pressure sensitivity
of the propellant. The propellant, however, had to be gelled to maintain a relatively
flat surface area for analysis, possibly biasing the results. Thus, the precise values
for a pressure-dependent burn rate are difficult to measure, but the value for the
pressure exponent is estimated to be between 1.2 and 2.0.
C. Pressure Instability During Combustion

The lumped parameter code provides an accurate simulation of mean, smoothed,
experimental data over a wide range of gun sizes and conditions; however, almost
all liquid propellant gun firings also show large-amplitude, high-frequency pres-
sure oscillations. In the RLPG, combustion instability is exhibited with a rich
spectral history which shows both acoustic velocity wave and noise components.
Pressure-time data taken in the combustion chamber from one 155-mm firing are
shown in Fig. 13, and the associated fast Fourier transform from 15 to 16 ms
is shown in Fig. 14.12 As can be seen, the pressure fluctuates around the mean and
shows high-amplitude (to 30% of mean), high-frequency (to 60 kHz) oscillations.

Acoustic modes for the chamber are expected up to approximately 12 kHz, with
higher frequencies associated with combustion noise and gauge response.
Since the pressure instability exhibits radial and tangential modes, multidimen-
sional models (two dimensional/axisymmetric/three dimensional) have been used
to research sources and control to the oscillations.48'49 At ARL, a two-dimensional
control volume model of the combustion chamber and gun tube has been de-
veloped specifically for the RLPG.48 The code includes generalized geometry to
treat the shape of the internal boundaries as shown in Fig. 15. The model gives
good overall agreement with experimental data as shown in Fig. 16, although the
pressure instability starts somewhat later and dies more rapidly than in the test.
In addition, the structure of the oscillations predicted by the model is simpler
than that recorded in the experiment (see Fig. 17), although the major frequencies
15.0-
10.0-
5.0-
0.0
0.0 10.0 20.0 30.0 50.0
40.0 600
Frequency (kHz)
Round 175, 7.2 liter charge. C60 gauge.
FFT, 15 to 16 m$.
Fig. 14 Fast Fourier transform of the experimental data in Fig. 1 from 15 ms to
16 ms.
•••••••••••••••••••IIIIIIIIIIIIIIIIIIIIIIB
•••••••••••••fir
Fig. 15 Model representation of the internal geometry of a 155-mm RLPG.
generated by the model agree well with the experimental data. It is also noted that
better measurement techniques for pressure oscillations in RLPGs are undergoing
study, and the recorded measurements shown in Fig. 17 are biased by the gauge
mounting techniques; however, measurement of the lower frequency oscillations
are believed to be accurate.
In the model, oscillations are attributed to the pressure sensitivity of the com-
bustion. As liquid is injected, a small-pressure rise is created due to inertial
confinement. This small-pressure wave spreads out, reflects from the combustion
chamber walls, and returns to the injection site. Since the combustion is pressure-
dependent, the combustion increases, increasing the magnitude of the pressure
wave. After a few cycles, pressure waves become very large.
Oscillations are primarily radial. Because the area change from the chamber to
the tube is large, longitudinal modes are not excited. The oscillations are much
larger along the centerline and near the injector than at the chamber wall, where
the experimental data are taken. The very large oscillations near the injector are
believed to contribute to the occasional penetration of hot gas into the reservoir.
The model is extremely simplified, ignoring the intact jet core, the distribution
of drop sizes, coalescence of droplets, heat transfer to the liquid, and kinetics.

300.0
0.0 10.0
time (ms)
Fig. 16 Comparison of simulation and experiment of pressure instability in a 155-mm
RLPG.
The pressure waves may mechanically contribute to the breakup of the jet. The
outer piston may vibrate, modulating the flow into the chamber. The physical
situation is very complicated, and it is fortunate that the simplified models used
represent experimental data as accurately as they do. Work is continuing on a
three-dimensional model to allow study of possible tangential effects.
IV. Summary
Current Army injectors in RLPGs have developed empirically through various
testing programs over the last 40 years. Designs have been motivated by the unique
requirement in guns of high pressure and temperature in an event occurring in a
few milliseconds. In addition, injector designs must satisfy practical considerations
related to safe operation such as sealing. Early sprays were generated from streams
of propellant passing through cylindrical passages.
Sprays in current U.S. Army RLPGs are primarily thick, annular sheets that
encounter a hot, dense environment of reacted and partially reacted products. The
pressure regime is high (atmospheric to 500 MPa), and the sheet varies from thin to
thick (from 0 to 2 cm) as the injection orifice opens and then closes. The propellants
currently being studied are HAN-based monopropellants that exhibit high burn-
rate sensitivity to pressure. Models have shown that the pressure sensitivity of
the propellant combination can lead to large-magnitude (to 50% of mean), high-
frequency pressure oscillations that are observed in experimental data. Although
significant progress has been made over the past several years in understanding
the spray combustion characteristics in RLPGs, research continues in fundamental
models of jet breakup and combustion based on diagnostics, as well as sources
15.0
10.0-
0.0
Oft S.O 10.0 15.0 20.0 25.0 3aO 35.0 40.0 45.0 5 0
a)
Frequency (kHz)
15.0
lao-
a) 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0 45.0 5 0

b) Frequency (kHz)
Fig. 17 Comparison of frequency spectrum of pressure instability in simulation and

experiment for a 155-mm RLPG at two different one-millisecond-wide time windows,
11-12 ms (top) and 15-16 ms (bottom).
and control of the pressure instability. Current spray models are simplified and
compare reasonably well to experimental data. However, diagnostics and models
are particularly needed to predict particle sizes, coalescence of droplets, and
kinetics associated with the propellant combustion process.
Acknowledgments
This work has been supported by the U.S. Army through its research labora-
tories (formerly the Ballistic Research Laboratory and now the Army Research
Laboratory) and program managers. The current Army sponsor of liquid propel-
lant technology development is Program Manager-Crusader, Picatinny Arsenal,
New Jersey, and the authors express their appreciation for the support of this office.
References
ôrrison, W. F., Knapton J. D., and Bulman, M. J., "Liquid Propellant Guns," Gun
Propulsion Technology, edited by L. Stiefel, Vol. 109, Chap. 13, Progress in Astronautics
and Aeronautics, AIAA, Washington, DC, 1988, pp. 413-471.
2
Foster, C. R., Gordon, E., and van De Verg, N., "Experimental Firings of a Regen-
eratively Pumped Liquid Monopropellant Gun," Jet Propulsion Lab., Pasadena, CA, June
1954.
3
Dorsey, E. G., Jr., and Rice, J. E., "Liquid Fuel Catapults With Regenerative Injection.
VII. Construction and Preliminary Test of an Experimental 127 mm Launcher," Experiment
Inc., TP-81, Richmond, VA, July 1954.
4
Niemeier, B. A., Fleischhauer, E. T., Dorsey, E. G., Jr., Forney, H. B., and Wilson,
E. J., Jr., "Liquid Fuel Catapult. VI. Fluid Seal and Impact Absorption Studies," Experiment
Inc., TP-48, Richmond, VA, June 1951.
5
Niemeier, B. A., Fleischhauer, E. T, Lichtenstein, V, Jr., and Wilson, E. J., Jr., "Liquid
Propellant Small Caliber Gun with Repetitive Fire (A Design Study)," Experiment Inc.,
TP-49, Richmond, VA, Feb. 1952.
6
Fleischhauer, E. T., Rathbun, K. C., and Foster, A. T., Jr., "Liquid Fuel Catapult. IX.
Research Directed Toward the Attainment of Ultra High Projectile Velocities," Experiment
Inc., TP-85, Richmond, VA, July 1954.
7
Fleischhauer, E. T., "Application of Liquid Propellants to Guns and Launchers. HI. New
Injection Systems," Experiment Inc., TP-95, Richmond, VA, Oct. 1955.
8
Swanson, D. L., "The Design and Construction of a 40-mm Liquid Propellant Regen-
erative Injection Gun," Experiment Inc., TP-92, Richmond, VA, Oct. 1955.
9
Niemeier, B. A., and Barr, W. J., "Liquid Fuel Catapults With Regenerative Injection, I;
Design and Test of Preliminary Caliber 0.50 Models," Experiment Inc., TP-38, Dec. 1949.
10
Hasenbein, R. G., "Regenerative Liquid Propellant Gun Technology," Benet Weapons
Lab., Watervliet, NY, June 1981.
H
Klingenberg, G., private communication, Ernst Mach Inst., Germany, 1989.
l2
Wren, G., Knapton, J. D., DeSpirito, J., Coffee T., and Klingenberg, G., "Pressure Os-
cillations in Regenerative Liquid Propellant Guns," Propellants, Explosives, Pyrotechnics,
Vol.20, 1995, pp. 225-231.
13
Dorsey, J., and Fuller, E. R., II, "Liquid Fuel Catapults With Regenerative Injection.
VII. Engineering Evaluation of a Test of a 127 mm Regenerative Launcher," Experiment
Inc., TP-90, Richmond, VA, May 1955.
14
Knapton, J. D., and Coffee, T., "Uncontrolled Ignition Mechanisms in Liquid Propellant
Guns," Propellants, Explosives, Pyrotechnics, Vol. 16, 1991, pp. 65-67.
452 G. P. WREN ET AL.
15
Graham, A., and Bulman, M. J., "General Electric 25 mm Regenerative Liquid Pro-
pellant Gun Program," Proceedings of the 12th JANNAF Combustion Meeting, Chemical
Propulsion Information Agency, Vol. I, Laurel MD, 1975, pp. 391-404 (CPIA Publication
273).
16
Mayer, R. E., private communication, General Electric Co., Pittsfield, MA, 1982.
17
Bulman, M., and Mandzy, J., "Technical Notes on Orifice Tests," Ballistic Research
Lab., BRL CR-548A, Aberdeen, MD.
18
Klein, N., "Liquid Propellant for Use in Guns," Gun Propulsion Technology, edited
by L. Stiefel, Vol. 109, Progress in Astronautics and Aeronautics, AIAA, Washington, DC,
1988, Chap. 14, pp. 473^97.

I9
Wren, G., Coffee, T, and Morrison, W., "A Comparison Between Experiment and
Simulation for Concept VIC Regenerative Liquid Propellant Guns. III. 155 mm," U.S.
Army Ballistic Research Lab., BRL-TR-3151, Aberdeen Proving Ground, MD, Sept. 1990.
2()
Morrison, W., Baer, P., Bulman, M., and Mandzy, J., "The Interior Ballistics of Regen-
erative Liquid Propellant Guns," Proceedings of the Eighth International Symposium on
Ballistics, Avco Systems Div., Wilmington, MA, Oct. 1984.
21
Cushman, P., General Electric Ordnance Systems Div., Internal Rept., Pittsfield, MA,
1980.
22
Coffee, T. P., "A Lumped Parameter Code for Regenerative Liquid Propellant Guns,"
U.S. Army Ballistic Research Lab., BRL-TR-2703, Aberdeen Proving Ground, MD, Dec.
1985.
23
Coffee, T. P., "An Updated Lumped Parameter Code for Regenerative Liquid Propellant
In-Line Guns," U.S. Army Ballistic Research Lab., BRL-TR-2974, Aberdeen Proving
Ground, MD, Dec. 1988.
24
Morrison, W., and Wren, G., "A Model of Liquid Injection in a Regenerative Liquid
Propellant Gun," U.S. Army Ballistic Research Lab., BRL-TR-2851, Aberdeen Proving
Ground, MD, July 1987.
25
Wren, G., and Morrison, W, "Extension of a Model of Liquid Injection in a Regen-
erative Liquid Propellant Gun Based Upon Comparison With Experimental Results," U.S.
Army Ballistic Research Lab., BRL-TR-3065, Aberdeen Proving Ground, MD, Dec. 1989.
26
Coffee, T. P., "One-Dimensional Modeling of Liquid Injection in a Regenerative-
Propellant Gun," U.S. Army Ballistic Research Lab., BRL-TR-2897, Aberdeen Proving
Ground, MD, March 1988.
27
Coffee, T. P., "The Analysis of Experimental Measurements on Liquid- Regenerative
Guns," U.S. Army Ballistic Research Lab., BRL-TR-2731, Aberdeen Proving Ground,
MD, May 1986.
28
Edelman, R., Private communication, Science Applications, Inc., Canoga Park, CA,
1985.
29
Coffee, T, and Wren, G., unpublished data, U.S. Army Research Lab., Aberdeen
Proving Ground, MD, 1987-1995.
3()
Gough, P. S., "A Model of the Interior Ballistics of Hybrid Liquid Propellant Guns,"
U.S. Army Ballistic Research Lab., BRL-CR-565, Aberdeen Proving Ground, MD, March
1987.
31
Pate, R., and Schlerman, C., "Flow Properties of LGP 1846 at Reduced Temperatures,"
Proceedings of the 22nd JANNAF Combustion Meeting, Chemical Propulsion Information
Agency Vol. II, Laurel, MD, Oct. 1985, pp. 319-332 (CPIA Publication 432).
32
Leveritt, C., private communication, U.S. Army Research Lab., Aberdeen Proving
Ground, MD, 1993.
33
Wren, G., Ray, S., Aliabadi, S., and Tezduyar, T, "Space-Time Finite Element Com-
putation of Compressible Flow Between Moving Components," Army High Performance
Computing Research Center Rept. 94-043, Minneapolis, MN, August 1994; International
Journal of Numerical Methods in Fluids, Vol. 21, No. 10, 30 Nov. 1995, pp. 981-991.
34
Ray, S. E., "Large-Scale Computational Strategies for Solving Compressible Flow
Problems," Ph.D. Thesis, Dept. of Aerospace Engineering and Mechanics, Univ. of
Minnesota, Minneapolis, MN, 1995.
35
Ray, S. E., private communication, U.S. Army Research Lab., Aberdeen Proving
Ground, MD, 1995.
36
Cook, D., private communication, Martin Marietta Corp., Pittsfield, MA, 1995.
37
Birk, A., and McQuaid, M., "Deliberations on the Dynamics and Core Structure of
Reacting Sprays at Elevated Pressures as Revealed from X-Ray and Laser Scattering
Observations," Recent Advances in Spray Combustion, Vol. I, edited by K. Kuo, Vol.
166, Chaps. 3-7, Progress in Astronautics and Aeronautics, AIAA Reston, VA, 1996,
pp. 57-184.
38
Warker, D., and Krehl, P., "Application of Flash Radiography to the Analysis of
Simulated Liquid Gum Propellant Jets," Ernst Mach Inst., Germany, Rept. No. 5/90, Nov.
1990.
39
Coffee, T. P., Baer, P. G., Morrison, W F, and Wren, G. P., "Jet Breakup and Combus-
tion Modeling for the Regenerative Liquid Propellant Gun," U.S. Army Ballistic Research
Lab., BRL-TR-3223, Aberdeen Proving Ground, MD, April 1991.
40
Wolfe, H. E., and Andersen, W. H., "Kinetics, Mechanism, and Resultant Droplet
Sizes of the Aerodynamic Breakup of Liquid Drops," Aerojet TechSystems, Aerojet Rept.
0395-04(18)SP, April 1984.
4
Coffee, T., Wren, G., and Morrison, W., "A Comparison Between Experiment and
Simulation for Concept VIC Regenerative Liquid Propellant Guns. I. 30 mm," U.S. Army
Ballistic Research Lab., BRL-TR-3072, Aberdeen Proving Ground, MD, Dec. 1989.
42
Coffee, T, Wren, G., and Morrison, W, "A Comparison Between Experiment and
Simulation for Concept VIC Regenerative Liquid Propellant Guns. II. 105 mm," U.S.
Army Ballistic Research Lab., BRL-TR-3093, Aberdeen Proving Ground, MD, March
1990.
43
Anon., "Milestone I Decision," Defense Acquisition Board, Pentagon, Washington,
DC, 1994.
44
Anon., Martin Marietta Defense Systems, "Interior Ballistic Modeling," Contractor
Rept., Pittsfield, MA, 1995.
45
Oberle, W., and Wren, G., "Burn Rates of LGP 1846 Conditioned Ambient, Cold and
Hot," U.S. Army Ballistic Research Lab., BRL-TR-3287, Aberdeen Proving Ground, MD,
Oct. 1991.
46
McBratney, W, "Windowed Chamber Investigation of the Burning Rate of Liq-
uid Monopropellants for Guns," U.S. Army Ballistic Research Lab., BRL-MR-03018,
Aberdeen Proving Ground, MD, 1980.
47
McBratney, W, and Vanderhoff, J., "High Pressure Windowed Chamber Burn Rate
Determination of Liquid Propellant XM46," U.S. Army Research Lab., ARL-TR-442,
Aberdeen Proving Ground, MD, June 1994.
48
Coffee, T. P., "A Two-Dimensional Model for Pressure Oscillations: Extension to
Generalized Geometry," U.S. Army Research Lab., ARL-TR-349, Aberdeen Proving
Ground, MD, Jan. 1994.
49
Madabhushi, R., Hosangadi, A., Sinha, N., and Dash, S., "Large Eddy Simulation
Studies of Vortex Shedding with Application of LPG Instabilities Using the CRAFT Navier-
Stokes Code," U.S. Army Research Lab., ARL-CR-241, Aberdeen Proving Ground, MD,
August 1995.

Author Index
Acker, W. P. ..................................63 Melton, L. A. ..............................143

Avedisian, C. T. ............................31 Merkle, C. L. ..............................331

Berthoumieu, P. ............................91
Chang, R. K...................................63 Nguyen, T. ..................................349
Chen,G. ................................63, 187 Oefelein, J. C...............................263
Coffee, T. P. ................................425 Pal, S. ..........................................233
Deng, Z. T. ..................................305 Pieper, J. ......................................349
Faeth, G. M.....................................3 Presser, C. ......................................31
Foucaud, R. ..........................91, 369 Reitz, R. D. ................................395
Gomez, A.....................................187 Ruff, G. A. ......................................3
Grenda, J. M. ..............................331 Ryan,H. ......................................233
Gupta, A. K. ..................................31 Samuelsen, G. S........ ......... ..113, 159
Hardalupas, Y...............................201 Santoro, R. J. ..............................233
Hodges, J. T...................................31 Serpengiizel, A..... ................... .......63
Hoover, D. ..................................233 Swindal, J. C. ................................63
Hulka,J. ......................................349 Talley, D. G. ................................113
Jeng, S.-M. ..................................305 Thamban,A. T. S. ........................ 113
Knapton, J. D...............................425 Tseng, L.-K. ....................................3
Kong, S.-C...................................395 Whitelaw, J. H. ............................201
Lavergne, G...................................91 Wren, G. P. ..................................425
Lecourt, R. ............................91, 369 Wu,P.-K. ........................................3
McDonell, V. G. ................113, 159 Yang, V. ......................................263


Kuo1996 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

Kuo1996 PDF

Uploaded by

Copyright:

Available Formats

Downloaded by RMIT BUNDOORA LIBRARY on August 16, 2015 | http://arc.aiaa.org | DOI: 10.2514/4.

Recent Advances in Spray

Measurements and Model

Purchased from American Institute of Aeronautics and Astronautics

This page intentionally left blank

Recent Advances in Spray

Paul Zarchan, Editor-in-Chief

Progress in Astronautics and Aeronautics

John J. Berlin Leroy S. Fletcher

Richard G. Bradley Alien E. Fuhs

William Brandon Ira D. Jacobsen

Clarence B. Cohen John L. Junkins

Luigi De Luca Pradip M. Sagdeo

Purchased from American Institute of Aeronautics and Astronautics

This page intentionally left blank

In view of convenience in transport and flexibility in storage of liquid fuels, spray

The objectives of these volumes are: 1) to present various areas of progress

chemical, aerospace, aeronautical, or computer engineering. Engineers and sci-

Subject Area 1. Spray Combustion Measurements

Chapter 1. Continuous- and Dispersed-Phase Structure of Pressure-

L. K. Tseng, Purdue University, West Lafayette, Indiana, G. A. Ruff, Drexel

Chapter 2. Behavior of Droplets in Pressure-Atomized Fuel Sprays with

Chapter 3. Spray Diagnostics with Lasing and Stimulated Raman

Chapter 4. Feasibility of Droblet Size and Concentration Measurements in

Chapter 5. Laser Sheet Visualization of Spray Structure.............113

Chapter 6. Use of Fluorescence Methods for Measurement of Droplet and

Chapter 7. Assessing the Physics of Spray Behavior in Complex

Chapter 8. Structure of a Coflow Laminar Spray Diffusion Flame.... .187

Chapter 9. Coaxial Airblast Atomizers with Swirling Air Stream......201

Chapter 10. Pressure Oscillation Effects on Jet B r e a k u p . . . . . . . . . . . . . . 2 3 3

Subject Area 2. Spray Combustion Modeling and

Chapter 11. Simulation of High-Pressure Spray Field Dynamics . . . . . . . 2 6 3

Chapter 12. Numerical Simulation of Deformed Droplet Dynamics and

Subject Area 3. Instability of Liquid Fueled Combustion Systems

Chapter 13. Numerical Simulation of High-Frequency Combustion

Chapter 14. Performance and Stability Characterization of Liquid

Chapter 15. Acoustic Sensitivity Measurements of Injectors in a Small

Subject Area 4. Spray Combustion in Certain Practical Systems

Chapter 16. Spray Combustion Processes in Internal Combustion

Chapter 17. Spray Combustion Processes in Regenerative Liquid

Purchased from American Institute of Aeronautics and Astronautics

This page intentionally left blank

Purchased from American Institute of Aeronautics and Astronautics

This page intentionally left blank

Continuous- and Dispersed-Phase Structure

of Pressure- Atomized Sprays

4 L.-K. TSENG ET AL.

CONTINUOUS- AND DISPERSED-PHASE STRUCTURE 5

II. Spray Breakup Regimes

CONTINUOUS- AND DISPERSED-PHASE STRUCTURE

DRIP RAYLEIGH WIND- ATOMIZATION

8 L.-K. TSENG ET AL.

is approached, which implies large Ohnesorge number. Thus, in view of the

C. Effects of Jet Exit Conditions

CONTINUOUS- AND DISPERSED-PHASE STRUCTURE 9

breakup regimes, have been dominated by effects of vorticity and turbulence at

10 L.-K. TSENG ET AL.

environment, with the results indicating negligible effects of varying liquid/gas

CONTINUOUS- AND DISPERSED-PHASE STRUCTURE 11

12 L-K. TSENG ET AL.

Fig. 4 Sketch of process of turbulent primary breakup at a liquid surface, from Wu

CONTINUOUS- AND DISPERSED-PHASE STRUCTURE 13

has sufficient mechanical energy to form a drop—whichever is larger. For test