Alireza Najafi-Yazdi
Doctor of Philosophy
McGill University
Montreal,Quebec
December 2011
To my mother, Sharhzad
ii
ABSTRACT
The objective of the present study was to investigate the mechanisms of sound
generation by subsonic jets. Large eddy simulations were performed along with
bandpass filtering of the flow and sound in order to gain further insight into the role
of coherent structures in subsonic jet noise generation.
A sixth-order compact scheme was used for spatial discretization of the fully
compressible Navier-Stokes equations.
iii
iv
that there are two mechanisms of sound radiation in unheated subsonic jets that
can occur in all scales of turbulence. The first mechanism is the stretching and the
distortion of coherent vortical structures, especially close to the termination of the
potential core. As eddies are bent or stretched, a portion of their kinetic energy is
radiated. This mechanism is quadrupolar in nature, and is responsible for strong
sound radiation at aft angles. The second sound generation mechanism appears to
be associated with the transverse vibration of the shear-layer interface within the
ambient quiescent flow, and has dipolar characteristics. This mechanism is believed
to be responsible for sound radiation along the sideline directions.
Jet noise suppression through the use of microjets was studied. The microjet
injection induced secondary instabilities in the shear layer which triggered the transition to turbulence, and suppressed laminar vortex pairing. This in turn resulted
in a reduction of OASPL at almost all observer locations. In all cases, the bandpass
filtering of the nearfield flow and the associated sound provides revealing details of
the sound radiation process. The results suggest that circumferential modes are significant and need to be included in future wavepacket models for jet noise prediction.
Numerical simulations of sound radiation from nonpremixed flames were also
performed. The simulations featured the solution of the fully compressible NavierStokes equations. Therefore, sound generation and radiation were directly captured
in the simulations. A thickened flamelet model was proposed for nonpremixed flames.
The model yields artificially thickened flames which can be better resolved on the
computational grid, while retaining the physically currect values of the total heat
v
released into the flow. Combustion noise has monopolar characteristics for low frequencies. For high frequencies, the sound field is no longer omni-directional. Major
sources of sound appear to be located in the jet shear layer within one potential core
length from the jet nozzle.
R
esum
e
Lobjectif de cette etude est dobtenir la meilleure comprehension des mecanismes
de generation de bruit par des jet subsoniques. Cette etude est basee sur simulations
aux grandes echelles de jets reactifs et sans reactifs.
Des calculs numeriques employant des scheme compacts de sixieme ordre. Lintegration
temporelle fut execitee `a laide de scheme Runge-Kutta de de quatri`eme ordre. Des
scheme `a faible dispersion et dissipation numerique. Un formulation integrale basee
sur les analogies acoustiques fut developpees pour la prediction du champ acoustique lointain pour les sources et observateure en mouvement dans un fluide avec
vitesse uniforme. La formulation fut implementee a` laide dalgorithmes facilitant
limplementation pour le traitement de donnees decoulement en haute performance
utilisant des outils de simiulation a grande echelle.
Les champs sonore produit par un jet turbulent non-reactif avec nombre de Mach
de 0.9, et un nombre de Reynolds ReD = 400, 000 fut etudie. Leffect de la taille
du maillage sur la precision de lecoulement en champs proche et e champs sonore
loin de source fut analyse. La sous-resolution de la couche decisaillement a` la sortie
du jet mene a` lapparition de structures coherentes et forte radiation qui no sort pas
physiquement realistes. Deux mecanismes principaux de generation sonore par jets
subsoniques furent identifies.
vi
vii
Le premier mecanisme est letirement et la distorsion de structures tourbillonnaires coherentes, en particulier pres de la fin du coere potentiel. Ce mecanisme
est quadripolaire, et emet principalement vers larriere du jet dans la direction de
lecoulement. Le seconde mecanisme semble etre constitue de vibration transversale
de la couche de cisaillement en reponse a la presemce de structures coherentes dans
la jet. Semblable a` la radiation dune plaque a` bonds finis, la contribution de ce
mechanisme est dipolaire et domine la champs sonore dans la direction transversale,
perpendiculaire au jet.
Lutilisation de plusieurs microjet fut investiguee pour la reduction du bruit.
Linjection a` laide de microjets precipite la transition a` la turbulence, favorisent le
melange et la destrcutction de structures coherentes de grande echelle.
Un filtrage en bandes de etroites fut effectue. Ce traitement des donnees numerique
permet de visualiser les relations complexes entre lecoulement et les onds sonores
emises. Les resultats demontrent limportance de modes circumferenciele, ce qui a
des consequenecs pour les modiles dits de paquets donde pour la preediction du
bruit du jet.
Des simulation numeriques decoulement et champs sonore dune flame sans premelange furent aussi executees. Les simulations incluent encore une fois lecoulement
et le champ sonore associe, obtenus directement des equations de Navier Stokes compressibles. Un mod`ele flammelette epaissie fut propose que donne flammes epaissies
artificiellement qui peuvent etre mieux resolus sur le maillage. Le bruit de combustion a des caracteristiques monopolaires aux basses frequences. Principales sources
de bruit semblent etre situe dans la couche de cisaillement.
Acknowledgements
The undertaking of the work presented in this thesis would have not been possible without the support, guidance, and encouragement of my advisor, Prof. Luc
Mongeau. His patience, diligence, willingness, and enthusiasm to explore new ideas
made my doctoral studies a wonderful experience. His being a mentor passes way
beyond daily research matters.
I am also thankful to Prof. Stephane Moreau and Dr. Marlene Sanjose for not
only our academic discussions, but also for their friendship. Sincere thanks also go
to Prof. Jeffrey Berghtorson. A great teacher and a true mentor, he has always been
generous with his time for me. I am also thankful to Prof. Siva Nadarajah who
kindly served on my thesis advisory committee.
Special thanks go to Prof. Thierry Poinsot, and Dr. Benedicte Cuenot who
made my visit to CERFACS possible. They were very generous with their time for
my various questions on turbulent combustion modeling. Their expertises and kindness to share their knowledge helped me to develop a better understanding in the
field of turbulent combustion.
I am also grateful to Dr. Ali Uzun, Dr. Christophe Bogey, and Prof. Christophe
Bailly for sharing their jet results and for their useful comments.
I would like to thank my friends and colleagues, Dr. Phoi-Tack (Charlie) Lew,
Kaveh Habibi, Hani Bakhshaei, and others who provided a fun and stimulating environment during the course of my studies at McGill. I am also grateful to my family
for their unconditional love and support.
viii
Financial support from McGill University, through the McGill Engineering Doctoral Award (MEDA) and Lorne Trottier Fellowship, is gratefully acknowledged.
My stay at CERFACS was made possible through the financial support of the ECCOMET program and the Marie Curie Fellowship. Financial support from the Exa
Corporation, Green Aviation Research & Development Network (GARDN), Pratt
& Whitney Canada, and the National Science and Engineering Research Council
(NSERC) of Canada is also gratefully acknowledged.
The computational resources were provided by Compute/Calcul Canada through
the CLUMEQ and the RQCHP High Performance Computing Consortia.
ix
TABLE OF CONTENTS
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1
1.2
1.3
1.4
1.5
1.6
I
2
Motivation . . . . . . . . . . . . . .
Background on Unheated Jet Noise
Background on Reacting Jet Noise
Objectives . . . . . . . . . . . . . .
Organization of the Thesis . . . . .
Contributions . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Nonreacting Flows
3
3
4
11
14
15
15
20
21
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
.
.
.
.
.
.
.
.
.
22
25
27
28
30
31
32
32
34
36
3.1
3.2
3.3
36
39
45
45
49
Governing Equations . . . . . . . .
Approximate Deconvolution Model
Spatial Discretization Scheme . . .
Spatial Filtering . . . . . . . . . . .
Temporal Integration Scheme . . .
Nonreflective Boundary Conditions
Sponge Zone . . . . . . . . . . . . .
Inflow Forcing . . . . . . . . . . . .
LES Code Parallelization . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Acoustic
. . . . .
. . . . .
. . . . .
. . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Analogy
. . . . .
. . . . .
. . . . .
. . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
51
52
52
54
55
60
4.1
4.2
.
.
.
.
.
.
.
.
61
62
62
62
63
63
66
68
82
5.1
5.2
5.3
82
83
86
3.4
4.3
4.4
5
Computational Setup . . .
Computational Grid Setup
4.2.1 Grid-30 . . . . . . .
4.2.2 Grid-88 . . . . . . .
4.2.3 Grid-380 . . . . . .
4.2.4 Grid Resolution and
Nearfield Results . . . . .
Farfield Sound Predictions
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
Subgrid Scales
. . . . . . . . .
. . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Computational Setup . . . . . . . . . . .
Nearfield Results . . . . . . . . . . . . .
Farfield Acoustics . . . . . . . . . . . . .
Bandpass Filter Visualization of Acoustic
. . . . . .
. . . . . .
. . . . . .
Nearfield
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
94
. 95
. 96
. 98
. 100
II
Reacting Flows
7.3
114
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
116
118
118
119
119
119
7.4
7.5
7.6
7.7
7.8
7.9
7.10
7.11
7.12
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
120
122
123
123
124
126
128
130
130
133
135
135
137
137
III
Flamelet Modeling . . . . . . . . . . . . . . . .
Flamelet/Progress Variable Modeling . . . . . .
The flamelet code . . . . . . . . . . . . . . . . .
Flamelet Modeling for Compressible Flows . . .
Thickened Flamelet Model . . . . . . . . . . . .
Thickened Flamelet vs. PDF modeling . . . . .
Sound Generation by Nonpremixed Flames . . .
Sound Generation by a Reacting Mixing Layer .
7.11.1 Computational Setup . . . . . . . . . . .
7.11.2 Numerical results . . . . . . . . . . . . .
Sound Generation by a Nonpremixed Jet Flame
7.12.1 Computational setup . . . . . . . . . . .
7.12.2 Nearfield results . . . . . . . . . . . . . .
7.12.3 Farfield Sound . . . . . . . . . . . . . . .
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . .
Future Work . . . . . . . . . . . . . . . . . . . . . . . . .
8.2.1 Source location identification . . . . . . . . . . . .
8.2.2 Heated jets . . . . . . . . . . . . . . . . . . . . . .
8.2.3 Effect of chevrons and lobed mixers . . . . . . . .
8.2.4 Wavepacket models . . . . . . . . . . . . . . . . .
8.2.5 Thickened flamelet model . . . . . . . . . . . . . .
8.2.6 Combustion noise models . . . . . . . . . . . . . .
8.2.7 Extension of Formulation 1C for diffraction effects
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Appendices
154
157
157
158
158
158
158
159
159
160
A.5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . .
Dispersion and dissipation of RK schemes . . . . . . . . .
ILDDRK scheme . . . . . . . . . . . . . . . . . . . . . .
Numerical Example . . . . . . . . . . . . . . . . . . . . .
A.4.1 Linear Wave Equation . . . . . . . . . . . . . . . .
A.4.2 Nonlinear Euler equations: One-Dimensional Case
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . .
xii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
161
162
165
168
169
171
172
B.3
B.4
B.5
Introduction . . . . . . . . . . . . . . . . . . . .
Principal Component Analysis . . . . . . . . . .
B.2.1 Background . . . . . . . . . . . . . . . .
B.2.2 Principal direction . . . . . . . . . . . . .
B.2.3 Singular Value Decomposition . . . . . .
IL-FGM . . . . . . . . . . . . . . . . . . . . . .
Numerical Examples . . . . . . . . . . . . . . .
B.4.1 Flamelets with a single-progress variable
B.4.2 Flamelets with multi-progress variables .
Conclusion . . . . . . . . . . . . . . . . . . . . .
xiii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
177
179
179
180
181
183
186
186
187
189
LIST OF TABLES
41 Reported LES of sound radiation from SP07 jet. . . . . . . . . . . . .
61
42 Normalized cut-off wavenumbers for the grid spacing in the jet shear
layer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
65
43 Simulation time steps and runtime. The term AFTT denotes acoustic
flow-through time and is equal to 32 Dj /Uj . . . . . . . . . . . . . . .
66
84
. . . . . . . . . . . . . . . . . . . . . . 171
xiv
LIST OF TABLES
xv
LIST OF FIGURES
11 Schematic of jet flow field. The jet plume is illustrated with contours of
vorticity (hot color scheme) superimposed on the acoustic pressure
(gray-scale color scheme). Microphone locations are commonly
specified by their distance, R, from the jet nozzle, and polar angle,
, with respect to the jet centerline. . . . . . . . . . . . . . . . . .
17
12 Wavepacket concept for jet flows. This figure shows how the existence
of coherent structures results in a wavepacket pressure distribution
in the shear layer. . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
18
19
19
35
xvi
LIST OF FIGURES
xvii
35
57
57
58
58
59
59
72
72
LIST OF FIGURES
xviii
73
73
74
75
47 Contours of normalized mean axial velocity, < U > /Uj , in the x-y
plane; (a): Grid-30; (b): Grid-88; (c): Grid-380. . . . . . . . . . . .
76
77
49 Variations of (a) centerline mean axial velocity, and (b) axial turbulence intensity. Solid line: LES (Grid-380); dashed line: LES
(Grid-88); dashed-dotted line (Grid-30); : LES of Bogey et al.
(2011); N: Arakeri et al. (2003); : Lau et al. (1979). . . . . . . . .
78
LIST OF FIGURES
xix
79
411 The Ffowcs Williams-Hawkings surface setup for farfield sound prediction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
412 Sound pressure spectra obtained at 52Dj from the jet nozzle: (a) at
= 30 ; (b) at = 90 ; solid line: LES (Grid-380); dashed-line:
LES (Grid-30); N:Bogey et al. (2007); H:Tanna (1977). . . . . . . .
80
81
88
. . . .
88
89
LIST OF FIGURES
xx
90
91
. . . . . . . . . . . . . . . . . . . .
92
93
LIST OF FIGURES
xxi
u0x u0x
;
Uj2
(a) base
. . . . . . . . . . . . . . . . . . . 108
LIST OF FIGURES
xxii
611 The Power Spectral Density (PSD) of the farfield acoustic sound
pressure; (a) radiation angle = 30 ; (b) = 90 . The data are
reported for a common distance of R = 100Dj . . . . . . . . . . . . 109
612 The band-pass filtered acoustic nearfield of the base setup; the
frequency band corresponds to St = 0.225 to St = 0.375; (a)
tUj
Dj
= 438; (b)
tUj
Dj
= 443; (c)
tUj
Dj
= 448; (d)
tUj
Dj
= 438; (b)
tUj
Dj
= 443; (c)
tUj
Dj
= 448; (d)
tUj
Dj
= 453. . . . . . . . . 111
614 The band-pass filtered acoustic nearfield of the microjet setup; the
frequency band corresponds to St = 0.225 to St = 0.375 (a)
tUj
Dj
= 272; (b)
tUj
Dj
= 302; (c)
tUj
Dj
= 332; (d)
tUj
Dj
= 362. . . . . . . . . 112
615 The band-pass filtered acoustic nearfield of the microjet setup; the
frequency band corresponds to St = 0.925 to St = 1.075 (a)
tUj
Dj
= 272; (b)
tUj
Dj
= 302; (c)
tUj
Dj
= 332; (d)
tUj
Dj
= 362. . . . . . . . . 113
LIST OF FIGURES
xxiii
LIST OF FIGURES
xxiv
LIST OF FIGURES
xxv
LIST OF FIGURES
xxvi
NOMENCLATURE
Roman Symbols
B
Dj
Jet diameter
Dk
Diffusivity of species k
DT
Thermal diffusivity
Microjet diamater
Da
Damkohler number
Froud number
Acoustic intensity
Waveknumber
Le
Lewis number
Mach number
xxvii
NOMENCLATURE
Pr
Prandtl number
Re
Reynolds number
Entropy
Sij
St
Strouhal number
Temperature
Tij
Time
Uj
Jet velocity
= [u v
xxviii
= [x y
Mixture fraction
002
g
Z
NOMENCLATURE
Greek Symbols
f
Grid spacing
ij
Kronecker delta
Molecular viscosity
Angular frequency
Chemical source term per unit mass; see eq. (7.2) on page 117
Chemical heat release term per unit mass; see eq. (7.2) on page 117
Density
ij
xxix
NOMENCLATURE
(x , y , z ) Partial derivatives of along physical coordinates
(x , y , z ) Partial derivatives of along physical coordinates
(x , y , z ) Partial derivatives of along physical coordinates
Superscripts, Subscripts, and Accents
()0
Ambient condition
()
Freestream condition
()j
Jet
()r
Reference condition
()rms
()sgs
Subgrid scale
()st
Stoichiometric condition
()
f
()
()
Microjet properties
Abbreviations
CFD Computational Fluid Dynamics
dB
Decibel
xxx
NOMENCLATURE
LES
LHS
The sensation of sound is a thing sui generis, not comparable with any of our
sensations.
Lord Rayleigh in The Theory of Sound
CHAPTER 1
Introduction
1.1
Motivation
More than a century after the first powered flight by the Wright brothers, avia-
tion is a dominant industry and a major job creator, specially in the North American
economy. With increases in air traffic, aircraft noise has increased and become a nuisance for communities living close to active airports. This has led governments to
create increasingly stringent regulations for the maximum sound levels allowable during aircraft landing and takeoff. Despite significant improvements over the past few
decades, aircraft noise remains a significant challenge. Governments and aerospace
industries have laid out plans to reduce the current aircraft noise level by 20 EPNdB1
by the year 2020 (Weasoky, 1998).
One of the most significant contributor to aircraft sound emission is the engine.
The sound radiated from the propulsion system consists of fan noise, combustion/core
noise, and jet exhaust noise. The jet exhaust noise and combustion core noise are
especially significant during takeoff phase. After more than half a century, sound
generation by high speed turbulent jet flows remains one of the most recalcitrant and
challenging problems in aeroacoustics. It is widely accepted that coherent structures
Lighthill (Lighthill, 1952, 1954) who introduced the concept of acoustic analogy.
Lighthills idea was to recast the exact equations of motion, i.e., the conservation
of mass and momentum, as an inhomogeneous wave equation where the nonlinear
terms are moved to the right hand side (RHS) and are treated as sources of sound.
The problem of calculating the turbulence-generated sound is then equivalent to
solving the radiation of a distribution of sources into an ideal fluid at rest. Lighthills
inhomogeneous wave equation is
2 Tij
1 2
2
0
p
=
,
c20 t2
xi xj
(1.1)
(1.2)
In eq. (1.2), ij is the shear stress tensor, ui and uj are the flow velocity components,
and p0 , 0 , and c0 are the ambient pressure, density, and speed of sound, respectively.
The first term in Lighthills stress tensor, eq. (1.2), is commonly referred to as the
Reynolds stress, and is a significant source of sound in turbulent flows. The second
term is due to wave amplitude nonlinearity and entropy variations in the source
region. The third term represents the attenuation of sound waves due to viscous
stresses, and is usually neglected. Due to the appearance of the double divergence
operator
2
,
xi xj
(1.3)
radiation in the aft direction, such that the acoustic intensity is equal to
Ia = K
2j Dj2 Uj8
,
2 5/2
0 c50 (1 Mc cos )2 + rlrc0
R2
(1.4)
ui uj
D 1 dS
1 ij
D2
2
c
=
+
,
D t2
xi
xi
xj xi Dt cp dT
xj xj
(1.5)
p
1
ln .
p0
(1.6)
ui uj
,
xj xi
due to velocity fluctuations, while the second and third terms correspond to the
effects of entropy sources and viscosity fluctuations, respectively.
In comparison to Lighthills analogy, eq. (1.1), a convective term has been moved
to the left-hand side of eq. (1.5). This leads to the appearance of a second-order total
time derivative in the wave operator. The spatial dependence of the speed of sound
has also been included in the wave operator to account for refraction effects.
Lilley (1974) argued that the first term on the RHS of Phillips equation contains
propagation effects for shear flows that should be included in the convective wave
operator on the LHS. He then derived a third-order wave equation2 ,
uj
uj uk ui
D D2
2
2
+ ,
c
+2
c
=
Dt Dt
xi
xi
xi xj
xi
xi xj xk
(1.7)
where represents the effects of entropy generation and viscosity fluctuations which
are generally neglected. Some industrial jet noise prediction tools, such as General
Electrics MGB approach (Balsa et al., 1978), are based on Lilleys analogy.
More recently, Goldstein (2003) has proposed a generalized acoustic analogy in
which the Navier-Stokes equations are recast as a set of linearized inhomogeneous
Lilleys equation, as published in the original paper (Lilley, 1974), reads slightly
different from eq. (1.7). In the original paper, it was assumed that the mean flow is
parallel to x1 axis.
Euler equations with source terms representing shear-stress and energy-flux perturbations. This approach requires the proper choice of a base flow around which the
Navier-Stokes equations are linearized.
Acoustic analogies have some intrinsic drawbacks. The linearization of the governing equations and the dissociation of the propagation effects (i.e. refraction of
sound waves) from the source terms are usually somewhat arbitrary without any a
priori knowledge of the sound field. This means that the source terms in acoustic
analogy theories are not necessarily true sources of sound. Moreover, the source
terms are assumed to be known, making acoustic analogies dependent on separate
experimental measurements or theoretical or numerical calculations. With the exception of Lighthills equation, almost all acoustic analogies involve nonlinear differential operators which generally cannot be integrated analytically. This implies
that a quantitative prediction of the farfield sound requires the numerical solution
of the acoustic analogy partial differential equations, which can be computationally
intensive.
As an alternative to acoustic analogy theories, some researchers have investigated phenomenological models based on the observation of coherent structures in
turbulent jets. Mollo-Christensen (1967) suggested a wavepacket concept to model
sound generation by coherent structures. Figure 12 illustrates how the existence of
coherent structures in the shear layer results in a pressure distribution that can be
modeled as a wave packet.
In supersonic jets, such coherent structures generate sound through Mach wave
radiation to the farfield, a process analogous to the sound radiation by a supersonic
wavy wall (Tam, 1995). The introduction of the wavepacket concept led researchers
to explore the flow stability theory as a theoretical framework to analyze the nearfield
dynamics and its relation to farfield sound radiation (Michalke, 1970, 1972; Mankbadi
& Liu, 1984). Models based on linear convecting instability waves were developed
by Tam & Morris (1980) and Tam & Burton (1984a,b). More recently, Wu (2005)
investigated the sound radiation from nonlinearly evolving instabilities.
The relevance of models based on instability wave radiation for subsonic jet noise
is debatable. This is because the energy in wave-numbers with supersonic phase
speeds appears only with the growth and decay of subsonically convected instability
waves. In other words, it can be argued that instability waves do not propagate
supersonically to radiate Mach waves. The appearance of supersonic phase speeds
may be explained as an artifact of the Fourier transform (cf. Fig. 13). Despite this
fact, several authors have investigated the sound emissions from wave packet pressure
fields (Crighton & Huerre, 1990; Avital & Sandham, 1997; Le Diz`es & Millet, 2007;
Obrist, 2009, 2011; Cavalieri et al., 2011).
Following an exhaustive review of supersonic jet noise data, Tam et al. (1996)
showed that far-field sound pressure spectral densities can be fit using two similarity
spectra. The spectrum shape that dominated the aft angles was associated with
Mach wave radiation by large scale structures. Hence, this spectrum is commonly
referred to as the Large Scale Similarity (LSS) spectrum (Morris, 2009). The second
spectrum was used to fit the sound pressure spectra radiated toward sidelines. Tam
et al. (1996) argued that the sound radiated toward sidelines is generated by smallscale structures. Hence, the second spectrum is commonly designated as the Fine
10
11
2004). The interaction of acoustic sound waves with the flame can also lead to
combustion instability in gas turbine engines and industrial furnaces. Depending
on the generation mechanism, combustion noise can be characterized as direct or
12
indirect (Strahle, 1971, 1978). Direct noise is caused by volumetric expansion due to
heat release, and constitutes a monopole source. Indirect sound is generated by the
differential acceleration of entropy non-uniformities and their interaction with solid
boundaries. This mechanism constitutes a dipole source (Howe, 1998).
A review of literature shows that the majority of the combustion noise studies
are either theoretical or experimental. Numerical simulations of sound radiation by
open flames can help improve our understanding of broadband combustion noise. In
the absence of a retro-active feedback leading to instability, the flame can be excited
in a controlled manner to measure the flame transfer function, a quantity useful
for combustion instability studies. This approach is called system identification in
numerical studies of combustion instability (Poinsot & Veynante, 2005).
Zhao & Frankel (2001) preformed the DNS of sound radiation from an axisymmetric premixed reacting jet. They used a 6th-order compact scheme (Lele, 1992) for
spatial discretization, and a 4th-order Runge-Kutta algorithm for temporal integration. A generic one-step global reaction was considered to model the chemistry. Their
computational domain included both the nearfield and the farfield. They observed
that combustion heat release had a significant effect on the vortical structure of the
jet, as well as the radiated sound pressure level and directivity. Lighthills acoustic
analogy was used to identify apparent sound source locations. It was concluded that
heat release stabilized the jet, enhanced sound radiation levels, and altered the frequency of the most unstable modes to lower values, which led to a broader sound
spectrum.
13
Flemming et al. (2007) performed an LES of the H3 flame (Tacke et al., 1998).
The simulation was based on the low-Mach-number assumption in which the pressure work is neglected in the energy equation (Poinsot & Veynante, 2005). This
assumption reduces the computational cost by relaxing the CFL restriction. But,
the results can no longer be used to directly predict the generation and propagation
of sound. A linear wave equation with a monopole source term was used to estimate
the radiated sound pressure level.
Bui et al. (2007) used a hybrid LES/CAA approach, in which a low-Mach number LES was combined with an acoustic perturbation equation modified for reacting
flows. M
uhlbauer et al. (2008) used a RANS/statistical model to simulate broadband combustion noise of the open non-premixed DLR-A jet flame. Ihme et al.
(2009) also used the low-Mach-number assumption to perform an LES of an N2 diluted CH4 H2 /air flame. They developed a modified form of Lighthills acoustic
analogy to predict combustion generated sound. Their acoustic analogy utilized the
flamelet/progress-variable model to formulate the excess density.
So far numerical studies of nonpremixed flames have used the low-Mach-number
assumption to simulate the hydrodynamic field. The radiated sound has been calculated based on simplified models. It is interesting to directly simulate the sound
production and radiation by diffusion flames. Such studies can enhance our current understanding of turbulence/acoustic/flame interactions, and can be used as
benchmarks to further validate combustion noise models or develop new ones.
1.4. OBJECTIVES
1.4
14
Objectives
The literature review presented in Sec. 1.2 suggests that further studies on the
nature of jet noise sources and their dynamics are required. While acoustic analogies
seem to be incapable of providing more insight into the physics of sound sources,
phenomenological models, such as wavepacket theory, seem promising as models
of generation by shear flows. Developing such models, and calibrating them for
practical noise predictions require a better understanding, both qualitatively and
quantitatively, of sound radiation by coherent structures at different scales.
Although two-point space-time correlations provide some information about the
relation between the nearfield flow and the sound perceived in the farfield, they
do not provide a comprehensive picture of sound radiation patterns by turbulent
structures at different scales. The objective of the present study was to investigate
a new approach, bandpass filtering of the flow field, to provide further insight into
the dynamics of sound radiation by coherent structures of subsonic jet flows. This
approach shows how different scales contribute to farfield sound radiation. In the
present study, Large-Eddy Simulations (LES) of high-speed subsonic jet flows were
performed to obtain both the nearfield sources and the farfield radiated sound. Bandpass filtering was then used as a post-processing tool to visualize and characterize
the radiated acoustic field in different frequency bands. The result of this study were
used to identify source mechanisms in subsonic jets. The results suggest that Tam
et al. (2008) description of jet noise generation may be oversimplified.
Similar analysis was also used to perform a preliminary study of the sound
radiation by a nonpremixed reacting mixing layer, and a diffusion jet flame. The
15
nonreacting, high speed, subsonic jets is studied in Part I, while combustion noise is
the subject of Part II. Chapter 2 presents the governing equations, and the numerical
schemes used in the LES of nearfield flow. The farfield sound prediction method is
presented in Chapter 3. The LES results for a Mach 0.9 jet at ReD = 4 105 are
presented in Chapter 4. Chapter 5 describes the bandpass filtering procedure, and the
major findings obtained from this approach. Using the LES and bandpass filtering,
the use of mircojets for jet noise suppression is studied in Chapter 6. Numerical
simulations of sound radiation from a reacting mixing layer, and a jet diffusion flame
are presented in Chapter 7. Some concluding remarks, and suggestions for future
work are presented and in Chapter 8.
1.6
Contributions
The following list summarizes the major contributions of the present work:
High-fidelity, direct computations of sound radiation from reacting and nonreacting jet flows.
Investigation of grid resolution effects on the accuracy of jet noise simulations.
A band-pass filter visualization and analysis of the source region and the acoustic field of subsonic jets.
Development of an Implicit, Low-Dispersion, Low-Dissipation Runge-Kutta
(ILDDRK) scheme of fourth order accuracy.
1.6. CONTRIBUTIONS
16
Development of a new surface-integral acoustic analogy formulation, Formulation 1C, to predict the sound radiated by moving sources in uniformly moving
media.
Large Eddy Simulation of jet noise suppression by impinging microjets.
Development of a thickened flamelet model for simulations of turbulent nonpremixed flames.
Development of a systematic method to define progress variables and the Intrinsically Low-Dimensional, Flamelet Generated Manifold (IL-FGM) modeling for
chemistry tabulation in LES of turbulent flames
Direct noise computation of a reacting mixing layer, and a jet diffusion flame.
FIGURES OF CHAPTER 1
17
R
Figure 11: Schematic of jet flow field. The jet plume is illustrated with contours of
vorticity (hot color scheme) superimposed on the acoustic pressure (gray-scale color
scheme). Microphone locations are commonly specified by their distance, R, from
the jet nozzle, and polar angle, , with respect to the jet centerline.
p(x)
x
Figure 12: Wavepacket concept for jet flows. This figure shows how the existence of
coherent structures results in a wavepacket pressure distribution in the shear layer.
FIGURES OF CHAPTER 1
18
q0
|q(x)|
2l
(a)
downstream
upstream
q(k)
k
ka
ka
k0
(b)
Figure 13: A one-dimensional wavepacket illustration in (a) the physical and (b) the
frequency domain. The gray area corresponds to the range of radiating wavenumbers.
The wavenumber corresponding to sonic propagation is denoted by ka .
FIGURES OF CHAPTER 1
19
10
SPL [dB]
-5
-10
-15
-20
0.1
10
f / fp
Figure 14: Similarity spectra suggested by Tam et al. (1996) for turbulent mixing
noise: solid line: large-scale similarity spectrum ; dashed line: fine-scale similarity
spectrum.
120
115
SPL [dB] / St
110
105
100
95
90
85
80
0.01
0.1
StD = f D / Uj
Figure 15: Sound pressure spectra measured by Bogey et al. (2007) at R = 100D
of a subsonic jet (Mj = 0.9); solid line: measurements at = 30 ; dashed line:
measurements at = 90 .
Part I
Nonreacting Flows
20
CHAPTER 2
Governing Equations and Numerical Methods
In the present study, the Navier-Stokes equations are numerically solved in the
context of Large-Eddy Simulation (LES).The basic idea behind LES is to decompose
the flow properties into a large-scale or resolved component, , and a small-scale or
subgrid component, sg . This decomposition is achieved by applying a spatial-filter
(Pope, 2000),
Z
(t, x) =
(2.1)
where is the filter size, and G is the filter function satisfying the normalization
condition
Z
G(x, y; ) = 1 .
(2.2)
For compressible and reacting flows where the density changes significantly, Favre
filtering is employed. A Favre-filtered quantity, e is defined as
1
=
e =
Z
(t, y)(t, y)G(x, y; )dy .
(2.3)
It is customary to filter the Navier-Stokes equations first, and then solve the
resulting equations for the filtered quantities. This procedure, also known as implicit
filtering, always results in the so-called closure problem, i.e. some terms remain
unclosed and need modeling. These terms correspond to the effects of subgrid-scale
dynamics. Commonly used subgrid scale models include the classic Smagorinsky
21
22
model (Smagorinsky, 1963), and the dynamic Smagorinsky model (Germano et al.,
1991; Moin et al., 1991).
In the implicit filtering, the filter function is not necessarily known as it is defined
by the numerical grid (Kravchenko & Moin, 1997; Meyers & Sagaut, 2007; Lund,
2003). Hence, the filter effect on energy dissipation cannot be quantified. This can
be alleviated by explicitly filtering the flow properties at each step as an integral part
of the numerical simulation (Lund, 2003; Bose et al., 2010). In the present study, the
explicit filtering technique is adopted in order to quantify the energy dissipation by
the LES filter and assess the effects of filtering on the quality of the LES results. The
filtering procedure adopted in this work is based on the approximate deconvolution
model (ADM) (Stolz & Adams, 1999; Mathew et al., 2003; Bogey & Bailly, 2006) of
subgrid scales.
In this chapter, the governing equations, the LES subgrid modeling through
ADM, and some details on the numerical schemes and boundary conditions are presented.
2.1
Governing Equations
The unsteady, non-dimensional, compressible form of the Navier-Stokes equa-
tions were solved on curvilinear grids. The governing equations, in the generalized
coordinates, (, , ), are given by
1 Q
+
J t
F Fv
J
G Gv
J
H Hv
J
=
S
,
J
(2.4)
1
[
J
E]T ,
(2.5)
23
E=
The inviscid flux vectors are
uU + x p
F=
vU + y p
wU + z p
(E + p)U
(2.6)
uV + x p
G=
vV + y p
wV + z p
(E + p)V
(2.7)
and
W
uW + x p
H=
vW + y p
wW + z p
(E + p)W
(2.8)
where the variables U , V , and W are the transformed velocity vectors in the computational space given by
U = ux + vy + wz ,
(2.9)
V = ux + vy + wz ,
(2.10)
W = ux + vy + wz .
(2.11)
and
24
Fv1
x xx + y xy + z xz
Fv2
x yx + y yy + z yz
=
Fv =
F
v3
x zx + y zy + z zz
Fv4
G
v1
Gv2
Gv =
Gv3
Gv4
Gv5
(2.12)
x xx + y xy + z xz
x yx + y yy + z yz
=
x zx + y zy + z zz
(2.13)
and
H
v1
Hv2
Hv =
Hv3
Hv4
Hv5
x xx + y xy + z xz
x yx + y yy + z yz
=
x zx + y zy + z zz
(2.14)
1
Sij Skk ij ,
3
(2.15)
25
ui uj
+
xj
xi
.
(2.16)
,
qi =
2
( 1)Mr Re P r xi
(2.17)
Ur
.
RTr
(2.18)
(2.19)
where f (u) is a nonlinear flux function. Applying a low pass filter, G, to eq. (2.19)
yields
u
f (u)
+G
= 0,
t
x
(2.20)
where G (.) denotes low pass filtering by convolution as outlined in eq. (2.1).
Eq. (2.20) can be recast as
u f (u)
+
= Rsgs
t
x
(2.21)
26
f (u)
f (u)
G
,
x
x
(2.22)
needs to be modeled. The subgrid model should describe the subgrid scale residual, Rsgs as a function of the filtered solution u. The approximation deconvolution
method models Rsgs with the following relation
Rsgs =
f (u)
f (u )
G
,
x
x
(2.23)
(2.24)
(2.25)
u f (u )
u u
+
=G
.
G
t
x
t
t
(2.26)
Since G u G u, the RHS of eq. (2.26) may be set equal to zero, i.e.
u f (u )
G
+
= 0.
t
x
(2.27)
27
which implies to
u f (u )
+
= 0.
t
x
(2.28)
Thus eq. (2.28) is solved instead of eq.s (2.20) or (2.21). The numerical implementation is as follows.
Consider the filtered solution at time step n denoted by u(n) . The unfiltered
solution is obtained from
u (n) = Q u(n) ,
(2.29)
which is used to numerically integrate eq. (2.28) and obtain u (n+1) . The filtered
solution is then obtained from
u(n+1) = G u (n+1) .
(2.30)
When executed sequentially, the filtering step, eq. (2.30), and the deconvolution
step, eq. (2.29), can be combined to form a single filtering step, Q G u (n+1) .
Since Q is not the exact inverse of G, the operator Q G removes high wavenumber
components of the solution. In effect, the filter H = Q G is a low pass filter
similar to G, but with a higher cut-off frequency. In summary, the approximate
deconvolution model is implemented through explicitly applying a low-pass filter
during time integration.
2.3
+
i1
+
i
=a
i1
fi+1 fi1
fi+2 fi2
+b
,
2
4
(2.31)
28
was used for spatial discretization where = 1/3, a = 14/9, and b = 1/9. For
the boundary points i = 1 and i = N , the following third-order one-sided compact
scheme equations were used, respectively:
+2
=
1
(5f1 + 4f2 + f3 ) ,
2
(2.32)
1
(5fN 4fN 1 fN 2 ) .
2
(2.33)
and
+2
N
N 1
+
1
1
+
4
2
=
3
3
(f3 f1 ) ,
4
(2.34)
and
1
4
+
N 2
1
+
4
N 1
=
N
3
(fN fN 2 ) .
4
(2.35)
The above equations form a tri-diagonal system of linear equations which can
be efficiently solved using the Tridiagonal Matrix Algorithm (TDMA) (Conte &
Boor, 1980). The combination of the sixth-order scheme for interior points and the
third-order, one-sided scheme for boundary points results in a fourth-order global
accuracy.
2.4
Spatial Filtering
Since the central difference compact scheme, eq. (2.31), is non-dissipative, spatial
filtering of the solution is required at each time step to remove high wavenumber
components and keep the solution stable. The filtering also serves as the ADM
subgrid modeling for LES as described in Sec. 2.2.
29
The following sixth-order, tri-diagonal filter (Visbal & Gaitonde, 2002) was used:
f f i1 + f i + f f i+1 =
3
X
an
n=0
with coefficients a0 =
11
16
+ 5
, a1 =
8
15
32
17f
,
16
a2 =
(fi+n + fin )
3
16
3f
,
8
and a3 =
(2.36)
1
32
f
16
. The
filter coefficient f (0.5, 0.5) can be used to adjust the filter strength and cut-off
frequency.
As for the spatial discretization scheme, the filter equation needs to be modified
for points near the boundaries. The following sixth-order filter equation with a onesided stencil was used at grid points near the left boundary (Visbal & Gaitonde,
2002),
f f i1 + f i + f f i+1 =
7
X
an,i fn
i = 2, 3 ,
(2.37)
n=1
with
1
31f
15 17f
3
+
, a2,2 = 29
+ 16f , a3,2 =
+
,
32
64
32
64
32
3 3f
5 5f
15
32f , a6,2 =
=
+
, a5,2 = 15
+
,
64
16
8
32
16
a1,2 =
a4,2
a7,2 =
1
64
f
32
(2.38)
and
49 15f
1 f
13
3
+
, a2,3 = 32
+ 16f , a3,3 =
+
,
64
32
64
32
5
3f
3
3f
15
=
+
, a5,3 = 15
+ 32f , a6,3 =
64
16
8
32
16
a1,3 =
a4,3
a7,3 =
1
64
f
32
(2.39)
30
A similar filter was used for points near the right boundary,
f f i1 + f i + f f i+1 =
6
X
aN n,i fN n
i = N 1, N 2 ,
(2.40)
n=0
where
aN n,i = an+1,N i+1 .
(2.41)
(2.42)
the RK4 scheme can be implemented in the following low-storage format (Mitchell,
1995)
Step 1 (Euler Predictor):
Q0 = F(Q(n) ; t)
t 0
Q
2
t 0
Q
U2 = Q(n) +
6
U1 = Q(n) +
(2.43)
31
Q0 = F(U1 ; t +
(2.44)
t
)
2
U1 = Q(n) + tQ0
U2 = U2 +
(2.45)
t 0
Q
3
t 0
Q
6
(2.46)
A fourth-order Implicit, Low-Dispersion, Low-Dissipation, Runge-Kutta scheme (ILDDRK4) (Najafi-Yazdi & Mongeau, 2010a) was also designed for problems which involve sources of stiffness such as combustion, and solid boundaries. The derivation
of the ILDDRK4 scheme and its implementation are presented in Appendix A.
2.6
Characteristic Boundary Conditions (NSCBC) of Poinsot & Lele (1992) were used.
Since the LES code solves the governing equations in the computational space, the
32
NSCBCs are recast and implemented for the computational space over a curvilinear
grid as provided in Jiang et al. (1999).
2.7
Sponge Zone
To further minimize reflections at boundaries, sponge zones were used to atten-
uate strong vortices and waves convected through the boundaries as suggested by
Colonius et al. (1993). In the present work, the sponge zone technique was used to
force the flow field towards a smooth target solution. This was achieved through
grid stretching and adding a damping term to the governing equations in the sponge
zones,
Q
= RHS (x) (Q Qtarget ) .
t
(2.47)
The damping coefficient, , is increased from zero, at the interface between the
sponge zone and the computational domain, to max , at the computational domain
boundary, following a cubic relation. For jet flow simulations, the target solution
at the streamwise outlet boundary is commonly specified based on the self-similar
solution of an isothermal, fully developed, incompressible jet.
2.8
Inflow Forcing
If not already turbulent, the shear flow exiting jet nozzles contains some level of
disturbances that trigger the transition of the shear layer to turbulence. In simulations with high order of accuracy, the numerical error is usually not large enough to
trigger a transition to turbulence. In such cases, disturbances are artificially added
to the jet shear layer in the upstream. In the present work,the inflow forcing technique of Bogey et al. (2003) was adopted. The inflow forcing is achieved by adding
random velocity fluctuations, in the form of a vortex ring, to the jet shear layer at
33
every time step. The velocity perturbations are given by the following equations:
vx0
= Ux
m
X
n cos(n + n ) ,
(2.48)
n cos(n + n ) ,
(2.49)
n=0
vr0
= Ur
m
X
n=0
where vx0 and vr0 are velocity perturbations in streamwise and radial directions, respectively. The azimuthal angle is defined as
y
= arctan .
z
(2.50)
r
(x,
r)
r
0
0
Ux = 2Uj
exp ln(2)
,
r 0
0
"
2 #
x
x
(x,
r)
r
0
0
Ur = 2Uj
exp ln(2)
,
r 0
0
for r =
(2.51)
(2.52)
p
y 2 + z 2 6= 0, where r0 is the vortex ring radius, 0 is the minimum grid
spacing in the jet shear layer, and x0 is the streamwise location of the center of vortex
ring. The parameter (x, r) is given by
2 (x, r) = (x x0 )2 + (r r0 )2 .
(2.53)
34
The vortex ring forcing described above results in approximately solenoidal perturbation velocities to minimize spurious sound radiation.
2.9
35
Figure 21: The seven-point overlap at the interface between two adjacent blocks.
Computation cost
2.5
2
1.5
1
0.5
0
128
256
512
1024
No. of cores
Figure 22: The computation time needed to simulate 30 time steps in a jet flow
simulation vs. increasing number of processors.
CHAPTER 3
Convective Ffowcs Williams-Hawkings Equation: Formulation 1C
Farfield sound pressure calculations (Najafi-Yazdi & Mongeau, 2010b) were performed using a surface integral acoustic analogy, called Formulation 1C (Najafi-Yazdi
et al., 2011). Formulation 1C is based on the convective wave equation, which takes
into account the presence of a mean flow. The formulation is derived to be easy to
implement as a numerical post-processing tool for CFD codes.
In what follows, the derivation and implementation of Formulation 1C are presented.
3.1
surfaces in the field. A more general formulation which is applicable to the cases
where a stationary solid surface is present in the flow field was developed by Curle
(1955), and later generalized by Ffowcs Williams & Hawkings (1969) for moving and
permeable surfaces.
The Ffowcs Williams - Hawkings (FW-H) acoustic analogy (Ffowcs Williams &
Hawkings, 1969) involves enclosing the sound sources with a control surface which is
mathematically represented by a function f (x, t) = 0.
36
Lij nj
dS
xi f =0 4|x y| e
Z
Tij
2
dV ,
+
xi xj f >0 4|x y| e
p (x, t) =
t
0
where [
(3.1)
(3.2)
(3.3)
(3.4)
The vectors u and v are the flow and the surface velocities, respectively. The compression tensor Pij0 is defined as
Pij = (p p0 )ij ij .
(3.5)
The third term integral should be evaluated in the region outside of the surface,
f > 0, to account for the contribution of sound sources outside the FW-H surface. If
the FW-H surface coincides with stationary solid boundaries, the flow velocity, u, is
39
+ U0
,
t
t
x1
(3.6)
where x1 is the direction of the mean flow. This seems to be in contradiction with
Brentner & Farassat (2003) on using the Formulation 1 and 1A for the moving
observer problems. Casalino (2003) provided no rigorous mathematical derivation
in support of this hypothesis. The formulation of the present work circumvents this
difficulty by directly taking into account the presence of a mean flow rather than
solving a moving observer problem.
3.2
& Han (1995). The original derivation of Wells & Han (1995) did not include the
quadrupole noise term, and was for a moving observer problem by considering a
moving frame of reference. This situation is equivalent to a stationary observer in
a moving medium, i.e., wind-tunnel problems. For the sake of completeness and to
clarify the notation, the convective FW-H equation is directly derived for a windtunnel problem in this section. Moreover, the quadrupole noise term is retained.
Consider the motion of an acoustic perturbation in a medium moving at constant
velocity U0 . The flow velocity at each point is U0 + u1 , where u is the local
The bold font is used to denote vector quantities. Such quantities are also
denoted as tensors of rank one and follow Einsteins index notation.
40
(3.7)
0
uj
0
+ U0 j
+
= 0,
t
xj
xj
(3.8)
can be simplified to
where the primed variables denote perturbed quantities. The conservation of momentum,
[ (U0 i + ui )] +
[ (U0 i + ui ) (U0 j + uj )] =
[(p p0 ) ij ij ] , (3.9)
t
xj
xj
is recast as
[ui ] + U0 j
[ui ] +
[(p p0 ) ij + ui uj ij ] = 0 .
t
xj
xj
(3.10)
The surface f (x, t) does not need to coincide with the solid surface as shown in
Fig. 31. Indeed, the surface can be considered permeable or solid depending on the
problem at hand.
41
(3.11)
where H is the Heaviside function. Moving H(f ) inside the differential operators
yields3
H(f ) 0
H(f )0 H(f )uj
H(f )
H(f )
H(f )0
+U0 j
+
= 0
+ U0 j
+ uj
. (3.12)
t
xj
xj
t
xj
xj
Using
H(f )
f
=
(f ) = vj nj (f ) = vn (f ) ,
t
t
(3.13)
and
H(f )
f
=
(f ) = nj (f ) ,
xj
xj
(3.14)
H(f )0
H(f )0 H(f )uj
+ U0 j
+
= Qj nj (f ) ,
t
xj
xj
(3.15)
Qj = [ (uj + U0 j vj ) + 0 (vj U0 j )] .
(3.16)
eq (3.12) is recast as
where
42
H(f )0
(3.17)
where
Tij = ui uj + (p p0 ) c20 ( 0 ) ij ij .
(3.18)
(3.19)
(3.20)
2
2
2
2
2
0
c
+
U
[H(f )ui ] =
[H(f
)
]
U
0j
0j
0
t2
xj xj
t xj
xi xj
2
[Qj nj (f )]
[Lij nj (f )] +
[H(f )Tij ] .
t
xi
xi xj
(3.21)
43
Finally, eq. (3.15) is used to replace the second term on the left hand side of eq. (3.21)
and obtain the convective FW-H equation,
[H(f ) ] =
+ U0 j
[Qk nk (f )]
t
xj
[Lij nj (f )]
xi
2
[H(f )Tij ] ,
+
xi xj
2
(3.22)
c
+
2U
+
U
U
0
j
0
i
0
j
0
t2
xj xj
t xj
xi xj
2
.
(3.23)
44
Without loss of generality, it can be assumed that the mean flow velocity is along
the positive x1 -direction.4
free-space Greens function for the convective wave equation is (Blokhintsev, 1956)
G(x, t; y, ) =
( t + R/c0 )
,
4 R
(3.24)
where
M0 (x1 y1 ) + R
,
2
(3.25)
p
(x1 y1 )2 + 2 [(x2 y2 )2 + (x3 y3 )2 ] ,
(3.26)
R=
R =
and
q
= 1 M02 .
(3.27)
Using the above Greens function yields the solution to the convective FW-H differential equation,
c20
Z t Z
(g) 3
Qj nj (f )
(x, t) =
+ U0
d yd
t
x1 R3
4R
Z t Z
(g) 3
[Lij nj (f )]
d yd
xi R3
4R
Z t Z
(g) 3
2
[H(f )Tij ]
d yd ,
+
xi xj R3
4R
0
(3.28)
If the mean flow is not along the x1 direction, the reference frame can be rotated
to satisfy this condition.
3.3. FORMULATION 1C
45
where
g = t+
R
.
c0
(3.29)
Equation (3.28) shows that the thickness noise temporal derivative in the nonconvective FW-H equation is indeed replaced by a Lagrangian derivative as suggested
by Casalino (2003); however, this substitution is not sufficient to obtain correct results. In addition, Qj and Lij should be modified to include U0 j terms, as shown
in eq. (3.16) and (3.19). Moreover, the convective Greens function should be used,
and not the free-space Greens function.
3.3
Formulation 1C
As for the originalal non-convective FW-H equation, the numerical solution of
the convective FW-H equation, eq. (3.28), is challenging due to the concurrence of
spatial and temporal derivatives. This problem can be circumvented as presented
next. The contribution of quadrupole noise terms is neglected in the following5 , and
the thickness and loading terms are treated separately.
3.3.1
Thickness Noise
p 0T
+ U0
=
t
x1
Z
Z
Qj nj (f )
R3
(g) 3
d yd .
R
(3.30)
Di Francescantonio (1997) and Morgans et al. (2005) discuss in details that the
permeable control surface can be chosen such that the important noise sources are
enclosed without the need to calculate the quadrupole noise.
3.3. FORMULATION 1C
46
The surface f = 0 can can defined in a frame of reference fixed to the surface, .
Any point on the surface is defined by a fixed value of the variable, , irrespectively
of the motion of the surface. The motion of each point on the surface is then fully
described in terms of the translation and the rotation of the frame of reference. It
is also assumed that the FW-H surface does not undergo deformation, expansion, or
contraction. Since the transformations
y = y(, )
(3.31)
= (y, )
(3.32)
and
are isometric, the Jacobian of the transformation is unity. Hence, eq. (3.30) can be
written as
4
p 0T
+ U0
=
t
x1
Z
Z
Qj nj (f )
R3
(g) 3
d d .
R
(3.33)
xi
(g)
R
i 0 (g) R (g)
1 R 0 (g) R (g)
1R
=
=
i 2
c0 xi R
xi R
c0 R
R
(3.34)
and
1
c0 t
i (g)
R
R
!
=
1
c0
i 0 (g)
R
R
!
(3.35)
3.3. FORMULATION 1C
47
yields
xi
(g)
R
1
=
c0 t
i (g)
R
R
Ri (g)
,
R 2
(3.36)
3 = x3 y3 .
R
R
(3.37)
x3 y 3
R 3 = 2
.
R
(3.38)
is
where the radiation vector R
1
R1 = 2 M0 + R1 ,
2 = x2 y 2 ,
R
R
and
x2 y 2
R2 = 2
,
R
x1 y 1
,
R1 =
R
p 0T
=
t
(g) 3
d dd
R
R3
!
Z Z
1 (g)
R
t
Qj nj (f )
M0
d3 dd
t R3
R
Z t Z
R (g)
U0
Qj nj (f ) 1 2 d3 d .
R
R3
Qj nj (f )
(3.39)
c0 yi
(3.40)
where MR is defined as
MR =
1 yi
1
Ri = vi R
i.
c0
c0
(3.41)
1
dg .
1 MR
(3.42)
3.3. FORMULATION 1C
48
p 0T
Qj nj
d
f =0 R (1 MR ) ret
#
"
Z
1 Q j nj
R
d
M0
t f =0 R (1 MR )
"
# ret
Z
R1 Qj nj
U0
d ,
2
f =0 R (1 MR )
=
t
(3.43)
ret
R
e = t
c0
.
(3.44)
The quantity R is no longer the geometric distance between the observer and
the source, but the acoustic distance between the two, as defined in eq. (3.25) (Wells
& Han, 1995; Najafi-Yazdi et al., 2011). In the limit of zero mean flow velocity,
the acoustic distance and the geometric distance are equal. The observer temporal
derivative,
,
t
Some difficulties may arise in the reconstruction of the signal using the advanced
time approach. The input flow properties are usually provided at equally spaced time
steps in the source time domain; however, the uniform discretization of the source
time domain does not necessarily yield a uniform discretization of the observer time
domain due to the Doppler effect. This inconvenience can be avoided by moving the
observer temporal derivative inside the integral, and transforming it into a source
temporal derivative. The thickness noise contribution to the farfield sound pressure
3.3. FORMULATION 1C
49
2
f =0 R (1 MR )
e
Z
R
Qi ni
+
d
R2 (1 MR )2 e
f =0
Z
Qi ni
MR
+
d
3
f =0 R (1 MR )
e
#
"
Z
1 Qi ni + R
1 Q i ni + R
1 Qi n i
R
d
M0
R (1 MR )2
f =0
ret
"
#
Z
R
R1 Qi ni
+M0
d
2
R (1 MR )2
f =0
"
#ret
Z
MR R1 Qi ni
M0
d
R (1 MR )3
f =0
ret
#
"
Z
R1 Qi ni
d .
U0
2
f =0 R (1 MR )
(3.45)
ret
where dots over quantities denote temporal derivatives with respect to the source
time. The term Q i is obtained numerically, while other temporal derivatives are
obtained either numerically or analytically.
3.3.2
Loading Noise
=
xi
Z
[Lij nj (f )]
R3
(g) 3
d yd .
R
(3.46)
As for the thickness noise, all terms under the integral are functions of and only,
with the exception of (g)/R . Therefore, eq. (3.36) is used to transform the spatial
3.3. FORMULATION 1C
50
Z Z
i (g)
t
R
1
=
[Lij nj (f )]
d3 yd
t R3 c0
R
Z t Z
(g)
R
+
[Lij nj (f )] i 2 d3 yd .
R
R3
L
n
R
ij
j
d
4p 0L =
c0 t f =0 R (1 MR )
"
# e
Z
Lij nj Ri
+
d .
2
f =0 R (1 MR )
(3.47)
(3.48)
It can be verified that letting U0 = 0 recovers the loading noise term in Farassats
Formulation 1. The loading noise contribution can also be written in terms of source
temporal derivatives as follows:
4p 0L
i + Lij n j R
i + Lij nj R
i
L ij nj R
R (1 MR )2
f =0
"
#
Z
1
R
Lij nj R
d
c0 f =0 R2 (1 MR )2
"
#e
Z
1
MR
Lij nj R
+
d
c0 f =0 R (1 MR )3
e
"
#
Z
i
Lij nj R
+
d .
2
f =0 R (1 MR )
1
=
c0
"
#
d
e
(3.49)
This equation is equivalent to the loading noise term of Farassats formulation 1A,
and is convenient for the implementation of the advance time algorithm.
3.3. FORMULATION 1C
3.3.3
51
In the particular case where both the source and the observer are stationary
in a wind tunnel, simplifications in Formulation 1C lead to increased computational
efficiency.
In Formulation 1C, the distances R and R , defined by eq.s (3.25) and (3.26),
are constant and do not vary with time. The same observation applies to the radi i , the local normal vector components ni , and MR = 0.
ation vector components R
Therefore, these variables can be evaluated and stored at preprocessing, rather than
computed at every time step, and their source temporal derivative is zero. Likewise,
the source Mach number M and the unit normals n are not functions of time, so
M r = 0 and n = 0. These simplifications lead to the following expressions for the
thickness noise
"
4 p 0T =
f =0
i ) Qi ni U0 R1 Q2i ni
(1 M0 R
R
R
#
d ,
(3.50)
Z
f =0
"
i Lij nj R
1 L ij nj R
i
+
c0 R
R 2
#
d .
(3.51)
The terms containing R vanish quickly in the farfield as R , and thus they
are significant only in the nearfield. It is important to note that the amplitude of the
thickness term contribution to the farfield noise decreases with increasing the mean
flow Mach number. Therefore, the loading noise becomes the dominant source term
for high Mach number flows.
52
The input to the code is a surface mesh, which defines the FW-H surface, and the
time accurate flow solution obtained from CFD. Surface elements (panels) may be
triangular, quadrilateral, or polygonal. The flow properties are specified at the center
of each panel (for first order accuracy) or at the vertices of the panel (for higher order
accuracy) at each time step of the solution. The temporal derivatives are obtained
from a finite difference approximation. Finite-element type shape functions are used
for the spatial interpolation. The integral terms are then obtained from a Gaussian
quadrature with an appropriate number of points to obtain the desired order of
accuracy. The FW-H code was written based on the advanced time algorithm.
The canonical problems of sound radiation by a monopole and a dipole in a
uniform mean flow were solved for the numerical verification of the formulation.
Source terms were computed over a fictitious, closed surface surrounding the source.
In all test cases, the mesh resolution of the FW-H surface and the source and
observer time steps were chosen such that the wavelength and period of the radiated
waves were resolved with at least eight points.
3.4.1
The monopole sound-field is characterized by a simple harmonic velocity potential function (Lockard, 2002)
R
A
exp i t
.
(x, t) =
4 R
c0
(3.52)
53
(3.53)
The induced pressure and density, for the case with a mean flow U0 along the x1 direction, are related to the velocity potential through
p (x, t) = 0
+ U0
,
t
x1
(3.54)
(3.55)
and
p0
(x, t) = 2 ,
c0
0
(3.56)
respectively.
To keep a focus on the formulation and its implementation, and to avoid any bias
related to the accuracy of the flow solver, the flow properties on the FW-H surface
were obtained from the exact solution of the flow field generated by the monopole,
i.e., eq.s (3.52) to (3.56). These properties were then used as the input to the code.
The monopole was located at the origin of the coordinate system. The velocity
potential amplitude was A = 1 m2 /s. The pulsation frequency was 5 Hz. The radiated sound pressure was calculated at a geometrical distance 340 l from the source,
where the characteristic length of the FW-H surface, l, was set to unity. The speed
of sound was 340 m/s. Figure 32 shows the directivity pattern of the sound pressure
measured in presence of a uniform mean flow along the the positive x1 -direction for
M0 = 0.5 and M0 = 0.85. In this polar plot, the radius indicates the RMS pressure at
54
the observer location, while the polar angle corresponds to the radiation angle (measured from the x1 axis) between the source and the observer. In a moving medium,
the directivity pattern is directional due to the convective amplification caused by
the mean flow. Convective amplification causes the sound pressure level upstream of
the source to be greater than that downstream at the same distance from the source.
This convective effect increases with the mean flow Mach number. The ratio of the
sound pressure at = 180[deg] (upstream) and that at = 0[deg] (downstream) is
equal to the theoretical value of (1 + M0 )/(1 M0 ) with less than 0.3 % error.
The FW-H equation applies in principle for observer locations in both the
nearfield or the farfield, and involves no farfield approximation. The accuracy of
the sound prediction in the nearfield was also verified. Figure 33 shows the directivity pattern measured in the nearfield with microphones located at r = 5 l.
The exact solution and the numerical predictions are in very good agreement. The
small deviation from the exact solution is due to the fact that measurements were
made very close to the FW-H surface, where the size of the panels (mesh cells) is
comparable to the distance between the source and the observer. This means that
the panels are no longer acoustically compact. A finer surface mesh yielded more
accurate results (results not shown here).
3.4.2
The second validation test case is sound radiation from a point dipole in a moving
medium. The dipole axis was aligned with the x2 -axis. The velocity potential for
55
(x, t) =
x2
A
R
.
exp i t
4 R
c0
(3.57)
The particle velocities and pressure fluctuations were obtained from eq.s (3.53) and
(3.54), respectively. The same potential amplitude, frequency, and position as test
case 1 were prescribed. Figure 34 shows the directivity pattern of the sound pressure
at r = 30l. The mean flow causes the direction corresponding to the maximum
acoustic pressure to move in the upstream direction. The FW-H code and the exact
solution yield nearly identical results.
3.4.3
56
The effect of the mean flow is illustrated in Fig. 36(b) where the time histories of
the sound pressure measured upstream ( = 0[deg]) and downstream ( = 180[deg])
of the monopole at a distance of r = 2 l are compared. As expected, the amplitude of
the peak pressure for the upstream observer is greater than that for the downstream
observer due to convective amplification. The pressure ratio is no longer (1+M0 )/(1
M0 ), as in case 2, due to the motion of the monopole.
57
Figure 31: Flow over a rigid body whose motion is defined by f (x, t).
90
90
120
120
60
150
180
150
30
0.002
0.004
0.006
0.008
0.01
0.012
60
180
30
(a)
0.1
0.2
0.3
0.4
0.5
0
0.6
P rms
prms
(b)
58
90
90
120
120
150
150
30
180
60
60
30
180
1.1
0.5
1.5
2.5
3.5
P rms
prms
(a)
(b)
90
90
120
60
120
150
180
30
0.001
0.002
0.003
0.004
prms
(a)
0.005
0.006
60
150
180
30
0.002
0.004
0.006
0.008
0.01
0.012
prms
(b)
Figure 34: The directivity pattern of a point dipole, measured at r = 30l, radiating
in (a) a medium at rest, and (b) a flow at M0 = 0.5 moving in the positive x1 -direction
( = 0 [deg]); solid line: exact solution; symbols: FW-H code.
59
10
10
= 180 [deg]
= 0 [deg]
-2
-2
-4
-4
-6
-6
-8
-8
-10
-10
(a)
(b)
Figure 36: Time history of the sound pressure generated by a rotating monopole
in a moving medium (test case); (a) at (2l, 0, 0) where solid line corresponds to the
exact solution and symbols represnt the results obtained from the FW-H code; (b)
comparison between the results at (2l, 0, 0), = 0[deg], and (2l, 0, 0), = 180[deg].
CHAPTER 4
Large Eddy Simulations of an Isothermal High Speed Subsonic Jet
The sound radiation from an isothermal turbulent jet at a Mach number of 0.9
was investigated. The flow condition corresponds to the SP07 case in the experimental test matrix of Tanna (1977). This test case was chosen because of the availability
of high quality experimental data, and numerous numerical studies. Table 41 summarizes the computational setup of selected LES studies of SP07 jet, and provides a
comparison with those of the present work.
The jet was simulated on three grids. The grid resolution needs to allow the
growth of high-frequency instability waves and the generation of fine-scale turbulence
in the shear layer to preclude unphysical vortex pairing as recently confirmed by
Bogey & Bailly (2010) and Bogey et al. (2011). Three grids with 30 million, 88.5
million, and 380.1 million nodes were considered. These grids are designated as
Grid-30, Grid-88 and Grid-380, respectively. The grid resolution was chosen based
on available computational resources, and the previous work in the literature as listed
in Table 41. The simulations on the three grid were used to study the effect of grid
resolution, and LES filter on the shear layer development.
60
61
ReD
Grid size
400,000 390 200 200 15.6M
88,000
128 32 240 0.98M
400,000 255 221 221 12.45M
200,000
292 128 128 4.8M
100,000 252 1024 962 252.1M
400,000 462 256 256 30.2M
400,000 1350 256 256 88.5M
400,000 1450 512 512 380.1M
Farfield
FWH-H
Kirchoff
Linearized Euler
FW-H
Linearized Euler
FW-H
FW-H
FW-H
4.1
Computational Setup
The main jet stream was initialized through the following tangent-hyperbolic
r rj
20
,
(4.1)
where Uj is the inflow jet centerline velocity, and 0 is the initial momentum thickness
of the shear layer. A value of 0 = 0.06rj was chosen for the initial momentum
thickness, where rj is the initial jet diameter. The inlet density profile was imposed
based on the Crocco-Buseman relation,
(r) = j
1 2 u(r)
Mr
1+
2
Uj
1
u(r)
,
1
Uj
(4.2)
62
circumferential modes were omitted to minimize the sound radiation by the forcing
procedure as suggested by Bogey & Bailly (2005).
4.2
4.2.1
Grid-30
For Grid-30 test case, the computational domain extends to 40Dj in the streamwise, x, direction, and from 15Dj to 15Dj in the transverse, y and z, directions. The mesh consists of 482 256 256 grid points. The first 436 nodes in the
x -direction correspond to the physical domain. The last 46 grid points fall in the
sponge zone where a 4% grid stretching rate is applied from x = 25D to x = 40D.
Another sponge zone is also considered in 0 x 2Dj to damp the fluctuations
before interacting with the inlet boundary. The damping coefficient, , was increased
from zero, at the interface between the sponge zone and the computational domain,
to max = 0.8, at the computational domain boundary, following a cubic relation.
The minimum grid spacing in the x -direction is approximately 0.065Dj . This
grid spacing is kept constant in the physical computational domain, and then is
gradually stretched close to the sponge zone. The grid is exponentially stretched from
the centerline to transverse boundaries. The minimum and maximum grid spacing
in the transverse directions are 0.0365D, and 0.3632D, respectively. Figure 41
shows the distribution of the grid spacing in the axial and transverse directions. The
computational grid is also shown in Fig. 42.
4.2.2
Grid-88
The computational domain of Grid-88 extends to 35Dj in the streamwise direction, and from 10Dj to 10Dj in the transverse directions. The mesh consists of
63
1350 256 256 grid points with the first 1200 nodes, in the x -direction, falling in
the physical domain. In the sponge zone, the grid is stretched at a rate of 1.8% from
x = 20D to x = 35D. Another sponge zone is also considered in 0 x 0.5Dj .
The streamwise grid spacing in the physical domain is approximately 0.065Dj . The
minimum and maximum grid spacing in the transverse directions are 0.0365D, and
0.3632D, respectively.
4.2.3
Grid-380
The initial simulations in the present work, and findings of other researchers
(Bogey & Bailly, 2010; Bogey et al., 2011), show that the grid resolution in the jet
shear layer has significant effects on the growth of instability waves and transition to
turbulence. Insufficient grid spacing in the shear layer results in unphysical damping
of the instability waves and in the generation of laminar vortex pairing.
64
The grid spacing has a direct effect on the LES filter width, and hence on
the maximum resolved wavenumber. A comparison between the attenuation due to
molecular viscosity and that due to the LES filter provides useful information about
how well the turbulence is resolved for different wavenumbers.
2
u
Consider the velocity attenuation due to the viscous term xi x
, where is
i
the kinematic viscosity. The nondimensional transfer function of this term is then
equal to (Bogey et al., 2011)
T (k) = k 2 ,
(4.3)
Dj
2
(4.4)
and
1
T (k)Dj
=
(k2 )
T (k) =
Uj
ReD
2
Dj
0
2
,
(4.5)
where corresponds to the grid spacing at a given location in the domain. For jet
flow simulations, it is recommended to measure in the shear layer which corresponds to r = rj .
The attenuation due to the LES filter, eq. (2.36), is given by
TLES (k) = 1
(4.6)
where f is the attenuation factor, and ai coefficients are given by eq. (??). A value
of f = 0.47 was used for the simulations based on others studies (Uzun, 2003; Lew,
2009).
Dj
at rj
0.0651
0.0217
0.0097
65
k
kDj
Table 42: Normalized cut-off wavenumbers for the grid spacing in the jet shear
layer.
tUj
Dj
Grid-30 0.0106
Grid-88 0.0063
Grid-380 0.0035
66
Runtime in AFTT
25.6
14.4
13.32
72
90
240
200
400
1200
Table 43: Simulation time steps and runtime. The term AFTT denotes acoustic
flow-through time and is equal to 32 Dj /Uj .
4.3
Nearfield Results
Each simulation was evolved over a sufficient time to obtain converged statistics.
Table 43 shows the time step, the simulations runtime in acoustic flow-through
times, and the computational cost of each simulation.
Given the high Reynolds number of the jet flow, the shear layer was expected
to become turbulent rather quickly with fine-scale structures. Figure 46 shows the
snapshots of the instantaneous vorticity field superimposed on the acoustic pressure
for the three test cases. The test case with Grid-30 shows a relatively laminar shear
layer with a transitional vortex pairing which is considered to be unphysical for such
a high Reynolds number. The increase in mesh density substantially improved the
simulation quality as can be observed in test case with Grid-88. The laminar vortex
pairing is no longer observed, and the shear layer development seems to be more
physical. Higher frequency sound waves appear to be captured better with the fine
grid. Further increase in the mesh density to 380 million grid points, resulted in
capturing even finer turbulent structures in jet the shear layer, and higher frequency
sound waves.
Figures 47 and 48 show the contours of the normalized mean axial velocity,
< U > /Uj , and Reynolds stress, < xx >=< uu > /Uj2 , for the three test cases,
67
respectively. The jet does not spread immediately in the simulation with Grid-30
because of the unphysical delayed transition to turbulence, a feature of many similar
numerical simulations. In contrast, the jet starts to spread almost immediately after
forcing for the other two cases. However, the spread rates of the jets for test cases
Grid-80 and Grid-380 are smoother than that in Grid-30. This in turn results in
longer potential core lengths for the two former simulations. The potential core
length in Grid-30, Grid-88 and Grid-380 are 7.1Dj , 7.2Dj and 7.7Dj , respectively.
These values are in good agreement with the experimental value of 7Dj reported by
Fleury et al. (2008).
The increase in the potential core lengths may be explained by considering the
entrainment process of the ambient irrotational flow into the jet shear layer. In the
case of Grid-30, the large scale vortex pairing due to Kelvin-Helmholtz instability is
an efficient mechanism to entrain the ambient irrotational flow inside the jet plume.
This mechanism is much weaker in the two other simulations where such vortex
pairing is absent. The formation of laminar vortex pairing has also resulted in higher
levels of Reynolds stress in the Grid-30 test case as (cf. Fig. 48).
Figure 49 shows the mean axial velocity, < Uc > /Uj , and turbulence intensity,
urms /Uj , as a function of axial distance along the jet centerline. The plots compare
the results obtained from the present work with experimental data as well as LES
results from Bogey et al. (2011). The centerline turbulence intensity obtained from
Grid-88 undergoes a rapid increase starting from x/Dj 2 that is not observed in
experiments and the cases with higher mesh density. This unphysical increase in
turbulence intensity is believed to be associated with vortex pairing, albeit weaker
68
than that in Grid-30 test case. The results obtained with Grid-88 and Grid-380 are in
significantly better agreement with experimental data for both the mean centerline
axial velocity, and centerline turbulence intensity. However, the peak turbulence
intensity in Grid-380 is observed at x 14.5Dj which is further downstream from
the jet nozzle than the location obtained in the experiment (x 10Dj ). The reason
for this discrepancy is not clear and needs further investigation. The LES results
of Bogey et al. (2011) seem to follow a similar trend. However, since no data point
was reported by Bogey et al. (2011) for x > 12Dj , a solid conclusion cannot be
drawn about whether this discrepancy is a general artifact of high order LES or it is
particular to the present computational setup.
The jet decay rate can be quantified according to (Hussein et al., 1994)
BDj
Uc
=
,
Uj
x x0
(4.7)
where B is the jet decay rate parameter, and Uc is the mean centerline axial velocity.
The decay rate parameter, B, is equal to 6.5 for Grid-30. This value is approximately
equal to 9.0 for Grid-88 and Grid-380. Fleury et al. (2008) reported a value of 9.1
for parameter B. Figure 410 shows the distribution of the inversed mean centerline
velocity obtained from LES (Grid-380), the linear fit obtained based on eq. (4.7),
and the experimental results of Arakeri et al. (2003).
4.4
pressure levels for Grid-30 and Grid-380. A funnel-shaped, fictitious control surface
was considered around the the turbulent jet plume (c.f. Fig. 411). In both cases,
69
the inlet and outlet surfaces which cut through the plume were excluded for the
FW-H calculations.
For Grid-30, the surface starts at an axial distance of around 2Dj from the
inlet boundary, and extends to 24Dj along the streamwise direction. The cross-wise
extend of the surface increases from 6Dj at its inlet to 12Dj , at its outlet. The flow
properties on the FW-H surface were recorded every 10 time steps over a period
of 16000 time steps. This time period corresponds to four acoustic-flow-through
time1 . Based on the grid resolution around the control surface, and assuming that
at least 6 points per wave-length are required to resolve acoustic waves, the grid cutoff frequency is estimated to correspond to a Strouhal number of St = f Dj /Uj 3.
The farfield computation was performed using 16 cores and took approximately 30
minutes.
For Grid-380, the FW-H surface starts at an axial distance of around 1.2Dj
from the inlet boundary, and extends to 22Dj along the streamwise direction. The
transverse width of the surface increased from 4Dj at its inlet to 7Dj , at its outlet.
The flow properties on the FW-H surface were recorded every 20 time steps over a
period of 32000 time steps. This period corresponds to approximately three acousticflow-through time. The grid cut-off frequency corresponds to a Strouhal number of
St = f Dj /Uj 7. The farfield computation was performed using 24 cores and took
approximately 20 minutes.
70
The sound pressure levels were calculated for virtual microphones located along
an arc with a radius of 52Dj from the jet exit. The power spectral densities (PSD) of
the sound pressure at emission angles = 30 and 90 are shown in Fig.s 412(a),
and 412(b), respectively. The spectra obtained from Grid-30 show unphysical tones
at low frequencies because of the vortex pairing. The spectra also fall rather quickly
for high values of Strouhal number because of the coarse grid resolution.
The spectra obtained from Grid-380 are in much better agreements with experimental results over a wide range of Strouhal number. The discrepancy observed
in the low frequency region of the spectra is due to the relatively short duration of
the farfield sound pressure recorded. The agreement could improved by evolving the
simulation for longer period.
Figure 413 shows a comparison between the overall sound pressure levels (OASPL)
obtained from the simulations and experiments. The results of the simulation with
Grid-30 show an overall overestimation of about 4 dB for almost all radiation angles
(cf. Fig. 413(a)). This overestimation is believed to be due to the laminar vortex
pairing.
Figure 413(b) shows the results obtained from the LES with Grid-380. The initial results of the simulation with Grid-380 showed an unexpected overestimation of
the OASPL. A careful examination of the time history of farfield pressure showed an
unphysical, low-frequency tone with a period of about 60Dj /Uj . Careful examination
of the nearfield pressure field showed that this unphysical tone was generated because
of the excessive grid stretching at x/Dj 27. A high-pass filter was employed to
remove the effect of this tone. The filtered results showed much better agreement
71
with the experimental data. In fact, the difference between the experimental data
and the predicted pressure levels are less than 1 [dB] for almost all emission angles.
72
0.5
0.5
0.4
0.4
y / Dj
x / Dj
0.3
0.2
0.3
0.2
0.1
0.1
0
0
10
15
20
25
30
35
0 2 4 6 8 10 12 14
y / Dj
40
x / Dj
Figure 41: Grid stretching for Grid-30; left: axial direction; right: transverse direction.
20
15
10
-5
-10
-15
10
15
20
25
30
35
Figure 42: Grid-30 shown in x-y plane; every 4th node is shown.
73
0.2
y / Dj
x / Dj
0.15
0.1
0.05
0
0
10
15
x/
20
Dj
25
30
35
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0
10
y / Dj
Figure 43: Grid stretching for Grid-380; left: axial direction; right: transverse
direction.
20
15
10
-5
-10
10
20
30
Figure 44: Grid-380 shown in x-y plane; every 6th node is shown.
74
0.1
0.01
0.001
0.0001
1e-05
1e-06
/36
/16
/8
/4
/2
k
(a)
kDj
10
100
0.1
0.01
0.001
0.0001
1e-05
1e-06
1e-07
0.1
10
k
(b)
Figure 45: Variation of the attenuation due to molecular viscosity and LES filter
with wavenumber, k. In (a), solid line: numerical filter; dashed line: molecular
viscosity (Grid-30); dashed-dotted line: molecular viscosity (Grid-88); dotted line:
molecular viscosity (Grid-380); the symbol corresponds to the lateral grid spacing
at r = Dj /2. In (b), solid line: molecular viscosity; dashed line: LES filter (Grid-30);
dashed-dotted line: LES filter (Grid-88); dotted line: LES filter (Grid-380).
75
(a)
(b)
(c)
Figure 46: Snapshots of vorticity field superimposed on acoustic pressure; (a): Grid30; (b): Grid-88; (c): Grid-380.
76
(a)
(b)
(c)
Figure 47: Contours of normalized mean axial velocity, < U > /Uj , in the x-y plane;
(a): Grid-30; (b): Grid-88; (c): Grid-380.
77
(a)
(b)
(c)
Figure 48: Contours of normalized mean axial Reynolds stress, < xx >=< uu >
/Uj2 , in the x-y plane; (a): Grid-30; (b): Grid-88; (c): Grid-380.
78
1.4
1.3
1.2
1.1
Uc / Uj
1
0.9
0.8
0.7
0.6
0.5
0.4
0
10
15
10
15
x / Dj
(a)
0.2
urms / Uj
0.15
0.1
0.05
0
0
x / Dj
(b)
Figure 49: Variations of (a) centerline mean axial velocity, and (b) axial turbulence
intensity. Solid line: LES (Grid-380); dashed line: LES (Grid-88); dashed-dotted
line (Grid-30); : LES of Bogey et al. (2011); N: Arakeri et al. (2003); : Lau et al.
(1979).
79
Uj / Uc
0
0
10
15
20
x / Dj
Figure 410: Inverse of mean streamwise velocity along the centerline normalized by
the inflow jet velocity. Legend: solid line, eq. (4.7); , Arakeri et al. (2003); N, LES
(Grid-380).
Figure 411: The Ffowcs Williams-Hawkings surface setup for farfield sound prediction.
80
120
110
100
90
80
70
60
50
0.1
10
StD = f Dj / Uj
(a)
120
110
100
90
80
70
60
50
0.1
StD = f Dj / Uj
(b)
Figure 412: Sound pressure spectra obtained at 52Dj from the jet nozzle: (a) at
= 30 ; (b) at = 90 ; solid line: LES (Grid-380); dashed-line: LES (Grid-30);
N:Bogey et al. (2007); H:Tanna (1977).
81
90
60
30
90
95
100
105
0
115
110
OASPL [dB]
(a)
90
60
30
90
95
100
105
110
0
115
OASPL [dB]
(b)
Figure 413: Directivity of overall sound pressure level (OASPL) in [dB] at 52Dj
from the jet nozzle; (a): LES (Grid-30); (b): LES (Grid-380). Legend: , LES
with Grid-30; , LES (unfiltered) with Grid-380; , LES (filtered) with Grid-380;
N,Bogey et al. (2007); J, Mollo-Christensen et al. (1964); I, Lush (1971).
CHAPTER 5
Sound Generation by Subsonic Jets: A Band-Pass Filtering Study
In this chapter, Tam et al. (2008)s hypothesis is revisited. The dynamics of
sound radiation pattern by coherent structures at different scales is studied throught
the band-pass filtering of the pressure field in the nearfield. When visualized and
studied with the jet nearfield, the band-pass filtered acoustic field can provide useful
information about the location and dynamics of different sound sources, and their
dominant radiation directions for different frequencies.
For the present study, the nearfield data from the LES of an isothermal round
jet at Mj = 0.9, and ReDj = 400, 000 is considered. The nearfield data from the
Grid-380 case, as described in Chapter 4, is used for bandpass fitlering.
5.1
within a specific frequency band, f [f1 , f2 ]. For the present study, Finite-ImpulseResponse (FIR) filters were considered. Consider a discrete signal in time domain,
x(t). The FIR filtered signal at time step n, x(n), is obtained from the following
equation:
x(n) =
M
1
X
(5.1)
k=0
where h(k)s are the filtering coefficients, also known as impulse response coefficients
of the filter. The operator denotes that the FIR filtering is essentially a convolution
82
83
procedure. Equation (5.1) represents an FIR filter of order M which requires the
use of M data points, x(n k) where k = 0 M 1. Increase in the order of an
FIR filter results in improved accuracy of the filtering procedure, but at a higher
computational cost.
Several methods exist to design efficient and accurate FIR filters1 . In the present
work, the classic window method of FIR filter design (IEEE, 1979) was adopted. The
R
fir1 function of MATLAB
2
was used to obtain the impulse response coefficients,
R
MATLAB
is a registered trademark of the Mathworks, Inc.
84
Filter order
128
128
Table 51: Parameters used in the design of the bandpass filters. The frequencies
are cast in nondimensional form as the Strouhal numbers, f Dj /Uj .
Given the relative small value of the center frequency, the sampling frequency was
reduced to a value equivalent to Sts = fs Dj /Uj = 7.14 to further increase the
frequency resolution and accuracy of the band pass filter. Figure 51 shows the
response magnitude of the filter in the frequency domain. The response magnitude
is close to unity, 0 [dB], within the frequency band of interest, and falls rapidely to
a value around 103 = 60[dB] for low frequencies, and to 1.78 104 75[dB]
for high frequencies.
Another one-third-octave band filter, with center frequency equivalent to Stc =
4.0, was considered to study the sound radiation at high-frequencies. This frequency
band corresponds to St = f Dj /Uj [3.56, 4.48]. The filter order was the same as
that of the former one, but the sampling frequency was increased to a value equivalent
to Sts = fs Dj /Uj = 14.28. Figure 52 shows the response magnitude of the latter
filter in the frequency domain. Table 51 summarizes the parameters used in the
bandpass filtering study.
Large scale sound radiation is associated with relatively low frequencies (low
values of Stouhal number). Therefore, the resulting pressure field from the lowfrequency bandpass filtering (cf. Table 51) is hypothesized to represent radiation
by large scale coherent structures. Figure 53 shows snapshots of the vorticity field
85
superimposed on the bandpass filtered pressure field in the x-y plane. The unfiltered
field for the same time step is shown in Fig. 53(a) for comparision.
Figure 54 shows six snapshots of the bandpass-fitlered pressure field in sequence. These snapshots illustrate the evolution of the pressure field during a period
of T = 5Dj /Uj which is twice the period corresponding to the band-center frequency,
f = 0.4Uj /Dj . Coherent wavepackets are observed in the bandpass filtered pressure
fields from the inlet to the end of the potential core (cf. Fig 55(a)). It is also
interesting to note strong radiation along directions close to the jet axis, = 40 , as
the coherent wavepackets undergo distortion in the region close to the end of the potential core. Although weaker, significant radiation is also observed along directions
normal to the jet axis.
The high frequency sound field also exhibits strong sound radiation toward =
50 . Fine-scale sources appear to be more confined to the early stages of the shear
layer development (cf. Fig. 53(c)). Complex interference patterns are also observed.
A comparision between the bandpass filtered pressure and vorticity fields is
shown in Fig. 56. It is interesting to note that both fields show wavepacket structures for both low frequency and high frequencies. Figure 56(a) clearly shows that
the vorticity and pressure wavepackets coincide and represent the same coherent
structures in the flow.
Figure 57 shows the passband overall sound pressure levels measured on an
arc with radius 52Dj from the jet exit. The low-frequency radiated OASPL remains
almost constant at 88 [dB] as the emission angle is decreased from = 90 to 60 .
The OASPL then increases to its peak value of 97 [dB] at = 40 and then decreases
86
for smaller emission angles. The high-frequency OASPL is also quite directive. It
first increases from = 90 to = 50 where it reaches its maximum value. The
OASPL then decreases as the emission angle is further decreased. This type of
directivity is quite similar to that of a point quadrupole. It is also interesting to note
that the low-frequency OASPL is larger than the high-frequency one for almost all
emission angles including angles close to = 90 . This suggests that the large-scale
structures are more efficient in radiating sound regardless of the emission direction.
The observed directivity trends are in contrast with Tam et al. (2008)s hypothesis in
which fine scale sources are associated with radiated to the angles close to = 90 .
5.3
are two mechanisms of sound radiation in subsonic jets. The first mechanism is the
stretching and distortion of coherent structures, especially close to the termination
of the potential core. As eddies are bent or stretched, some portion of their kinetic energy is radiated as sound to the farfield. This vortex sound radiation can
be formulated using Lighthills analogy (Howe, 1998, 2003), and is quadrapolar in
nature. The strong directive sound radiation toward 45 may be explained by
this mechanism of sound generation.
The second mechanism of sound generation seems to be associated with the
vibration of shear-layer interface with the ambient quiescent flow, and constitutes
dipole sources. As turbulent coherent structures role up and are convected downstream inside the shear layer, they exert force on the ambient flow by inducing
periodic ripples and pressure fluctuations on the shear layer interface. The sound
87
88
-20
-40
-60
-80
-100
-120
-140
0
0.5
1.5
2.5
3.5
StD = f Dj / Uj
Figure 51: The magnitude response, in dB, of the one-third-octave bandpass filter
with center frequency equivalent to Stc = fc Dj /Uj = 0.4. This frequency band
corresponds to St = f Dj /Uj [0.36, 0.45].
0
-20
-40
-60
-80
-100
-120
-140
0
StD = f Dj / Uj
Figure 52: The magnitude response, in dB, of the one-third-octave bandpass filter with center frequency equivalent to Stc = fc Dj /Uj = 4. This frequency band
corresponds to St = f Dj /Uj [3.56, 4.48].
89
(a)
(b)
(c)
Figure 53: Snapshots of the vorticity field superimposed on the acoustic pressure;
(a): Unfitered field; (b): bandpass filtered around Stc = 0.4; (c): bandpass filtered
around Stc = 4.
(a)
(b)
(c)
(d)
(e)
(f)
90
Figure 54: Snapshots of the vorticity field superimposed on the bandpass filtered
acoustic pressure; The band-center frequency was Stc = f Dj /Uj = 0.4; snapshots
are shown in sequence with a time difference of t = 0.84Dj /Uj .
L
-
H
HH
H
HH
91
*
H
j
H
(a)
(b)
Figure 55: Wavepackets and radiated sound obtained from the low-frequency bandpass filtering; (a): pressure and vorticity field shown together; (b): pressure field
only. The potential core length, L, and dominant radiation directions are also shown
with arrows for reference.
92
(a)
(b)
Figure 56: Wavepackets obtained from the bandpass filtering of vorticity and pressure fields; the color scheme corrosponds to bandpass filtered vorticity field while the
gray scale color scheme corrosponds to the bandpass-filtered pressure contours; (a):
low-frequency passband; (b): high-frequency passband.
90
93
80
70
60
50
40
30
20
10
70
75
80
85
90
95
0
100
OASPL [dB]
Figure 57: Directivity of passband overall sound pressure level (OASPL) in [dB]
at R = 52Dj ; Legend: , low-frequency radiated sound; N, high-frequency radiated
sound.
CHAPTER 6
Large Eddy Simulation of Jet Noise Suppression by Impinging Microjets
The use of microjets for jet noise suppression has earned attention over the past
few years. Experimental studies (Greska et al., 2003, 2004; Arakeri et al., 2003;
Castelain et al., 2008) have shown that the inclusion of a circumferential array of
microjets may lead to a decrease in overall radiated sound pressure levels between 1-3
dB. Arakeri et al. (2003) studied the effect of microjets on turbulence characteristics
of a Mach 0.9 jet. The microjets resulted in a significant reduction of the nearfield
turbulent intensities. It was also observed that that the peak normalized vorticity in
the shear layer was significantly reduced in the presence of the microjets.
Castelain et al. (2008) studied the impact of the number of microjet nozzles and
the microjet mass flow rates on the radiated noise. They found that high values of
injected mass-flux promote high-frequency noise generation due to the interaction
between the microjets and the jet mixing layer. It was also noted that the modification of coherent structures in the flow was influenced by the spacing, diameter,
number, and velocity of the microjets.
Alkislar et al. (2007) studied the role of streamwise vortices on the aeroacoustics
of a Mach 0.9 jet. They compared the effect of microjets and chevrons on the coherent
structures. Their measurements showed a reduction in noise levels at low frequencies
for both cases. The reduction in sound levels was associated with a diminished
growth of the large-scale coherent structures at the jet exit. They also reported an
94
95
increase in the high-frequency sound pressure content due to small scale structures
generated by mixing.
Only a few numerical studies (Huet et al., 2008, 2009; Lew et al., 2010; Shur
et al., 2010) of noise suppression by microjets have been reported so far. In this
chapter, the effect of microjets on the flowfield and acoustic field of a high-Machnumber subsonic jet is studied.
6.1
Computational Setup
A circular jet with Mj = 0.9 and ReD = 400, 000 was considered for the present
study. The computational setup was similar to test case Grid-30 of Chapter 4 (cf.
Table 41 on page 61).
Eight microjets, with D /Dj = 0.05, were positioned at regular circumferential
intervals around the core jet with an impingement angle of 60 . The microjets
had a flow Mach number of around M = 1.42, yielding an injected-to-main mass
flux ratio of m
/m
j = 0.004. The momentum flux ratio, u2 /j u2j , was 2.5. These
parameter are comparable to those in the experimental study of Alkislar et al. (2007).
The diameter of microjet nozzles was too small to accurately resolve the microjet
structure with a reasonable grid size. As for the studies of Huet et al. (2008) and
Shur et al. (2010), the effect of microjet injection was taken into account by the
adding a source term to the Navier-Stokes equations.
96
The source term S (cf. eq. (2.4)) takes into account the effect of microjet
injection and has the following form
S = A(x)
m
U
m
V
m
W
(6.1)
m
(cp T0, )
The function A is a Gaussian distribution,
A(x) =
1
( 2 )
e
3/2
|xx |2
2 2
(6.2)
which ensures that the source term has a finite value close to the microjet position,
x , and decays exponentially away from where the microjet exists. The parameter
can be used to control the distribution of the source term. A value of = D
was chosen for the present study. It should be noted that integrating eq. (6.1) over
the whole volume around the microjets yields the desired amount of total mass,
momentum, and energy injection to the flow.
6.2
Nearfield Results
The simulation was conducted for 72,000 time steps which was equivalent to 24
acoustic-travel-time periods between the boundaries. The flow statistics were gathered after 11,000 time steps to ensure that the initial transients left the computational
domain. The nearfield computation took 72 hours using 200 cores.
Figure 61 shows instantaneous snapshots of the sound pressure field generated
by the base round jet, Fig. 61(a), and the effects of microjets setup, Fig. 61(b).
97
Instantaneous contours of vorticity are also shown to visualize the coherent structures. The nozzle location is shown schematically for reference, although it was not
included in the computational domain.
In the case of the base round jet, the pressure field shows strong sound generation
from vortex pairing, and the formation of large scale coherent structures in the jet
shear layer. This mechanism leads to the very efficient radiation of low frequency
sound waves to the farfield. In contrast, the microjets seems to enhance the transition
to turbulence, and the creation of smaller scale structures which hampers the low
frequency sound generation by vortex pairing.
Figure 62 shows the centerline distribution of mean axial velocity and axial
turbulence intensity for the base round jet. The profiles obtained from the present
LES are in agreements with experimental data.
Figure 63 shows the distribution of the inversed mean centerline velocity. A
value of 6.5 was obtained for jet decay parameter (i.e. parameter B of eq. (4.7)),
which is in agreement with the measured value of 6.25 reported by Zaman (1998).
The potential core length for the base round jet is around 7.1 D which is in
agreement with both experimental and numerical data in literature. The potential
core length for the case with microjets is around 6.55 D (c.f. Fig. 64) . Contours
of mean axial velocity, with and without microjet injection, are also compared in
Fig. 65.
Figure 66 and 67 show the effect of microjets on the distribution of the mean
jet velocity. The penetration of microjets in the core jet plume seems to induce
98
secondary instabilities in the shear layer which triggers the transition to turbulence
and result in a shorter potential core length.
Figure 68 shows contours of axial Reynolds stress, xx =
u0x u0x
,
Uj2
in the vicinity of
the nozzle exit. In the case of a base round jet, high levels of Reynolds stress appear
near the vortex pairing region and occupy a relatively wide portion of the shear layer.
The use of microjets, on the other hand, results in reduced levels of Reynolds stress
which appear closer to the nozzle. The streamwise distributions of peak turbulence
intensities are shown in Fig. 69. It is observed that the microjets injection resulted
in a rapid increase of turbulent intensities up to a distance of 1.3Dj . This confirms
that the use of microjets yields an earlier transition to turbulence. However, the
maximum level of turbulence intensity seems to be reduced compared to the case
without any microjet injection.
6.3
Farfield Acoustics
Farfield sound pressure levels were calculated for virtual microphones located
along an arc with a radius of 100Dj from the jet exit. Figure 610 shows the farfield
overall sound pressure level (OASPL) for the two cases. The present LES results (line
1 and 2) show an average reduction of about 4 dB in OASPL for almost all emission
angles. In comparison, Alkislar et al. (2007) reported a reduction of about 0.5 to 1 dB
in OASPL. A 6 dB difference between the present LES results and the experimental
data reported by Alkislar et al. (2007) was observed. It is interesting to note that
the base jet OASPL reported by Alkislar et al. (2007) is itself 3 to 4 dB less than
the reported values by Bogey et al. (2007) and Lush (1971). This discrepancy could
be associated with the fact that the nozzle used in the experiment had a small dent
99
which broke the symmetry of the jet. The effect of the dent, observable in Fig. 10
Alkislar et al. (2007), might have been quite significant in triggering the transition to
turbulence, and on farfield sound pressure levels. Indeed, Bogey & Bailly (2010) and
Bogey et al. (2011) have recently shown the significance of nozzle-exit boundary-layer
condition on the flow and acoustics of initially laminar jets. It may be concluded
that the shear layer of the base round jet in Alkislar el al.s experiment had higher
levels of turbulence intensity than the simulated jets. Hence, it was less effected by
the microjet injection.
The power spectral densities (PSD) of the sound pressure at emission angles
= 30 and 90 are shown in Fig.s 611(a), and 611(b), respectively. The spectra
at = 30 reach their maxima at around St = 0.3, and then drop quickly with
frequency while the spectra at = 90 , on the other hand, show a shallower drop
rate. This is in agreement with experimental observations (c.f. Tam et al. (2008)
and the references therein). The spectral levels for the case with microjets were
reduced by approximately 4 to 6 dB at very low frequencies compared to those of
the base round jet. The difference between the spectra of the base round jet and
those of the microjet setup decreased with increasing frequency. It is interesting to
note cross-overs at St 0.8, for = 30 , and St 3, for = 30 , above which
the microjet spectra show higher energy content than those of the base round jet.
These values for cross-over Strouhal numbers are in very good agreement with those
reported by Alkislar et al. (2007).
100
FIGURES OF CHAPTER 6
101
(a)
(b)
FIGURES OF CHAPTER 6
102
1.2
0.8
0.6
0.4
0.2
10
15
20
15
20
(a)
0.2
0.15
0.1
0.05
10
(b)
Figure 62: Centerline distribution of (a) mean axial velocity and (b) axial turbulence
intensity for the base round jet; solid line: present LES; : Zaman (1986), N: Lau
et al. (1979). : Arakeri et al. (2003).
FIGURES OF CHAPTER 6
103
6
5.5
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
10
15
20
25
30
Figure 63: Inverse of mean streamwise velocity along the centerline normalized by
the inflow jet velocity; solid line: LES of the base round jet; dashed dotted line:
linear regression.
1.2
0.8
0.6
0.4
0
10
15
20
Figure 64: Centerline distribution of mean axial velocity; solid line: base round jet;
dashed dotted line: with microjets.
FIGURES OF CHAPTER 6
104
(a)
(b)
Figure 65: Contours of normalized mean axial velocity; (a) Base round jet; (b) with
microjets.
FIGURES OF CHAPTER 6
105
Figure 66: The three-dimensional spatial evolution of the mean flow velocity with
streamwise distance in the presence of microjets.
Figure 67: The effect of microjets on the mean streamwise velocity at 1D from the
nozzle exit.
FIGURES OF CHAPTER 6
106
(a)
(b)
u0x u0x
;
Uj2
FIGURES OF CHAPTER 6
107
0.3
rms(ux)peak
0.2
0.1
10
15
20
15
20
(a)
0.3
rms(ur)peak
0.2
0.1
10
(b)
Figure 69: The streamwise distribution of peak (a) axial turbulence intensity and
(b) radial turbulence intensity; solid line: base round jet ; dashed-dotted line: with
microjets
FIGURES OF CHAPTER 6
108
120
1
2
3
4
5
115
OASPL [dB]
110
105
100
95
90
20
40
60
80
100
[deg]
Figure 610: Measured and predicted far-field noise directivity; 1: base round jet
(present LES); 2: with microjets (present LES); 3: base round jet (Bogey et al.,
2007); 4: base round jet (Alkislar et al., 2007); 5: with microjets (Alkislar et al.,
2007). The data are scaled for a common distance of R = 100Dj .
FIGURES OF CHAPTER 6
109
120
100
80
60
40
20
10
-1
10
10
(a)
120
100
80
60
40
20
10
-1
(b)
Figure 611: The Power Spectral Density (PSD) of the farfield acoustic sound pressure; (a) radiation angle = 30 ; (b) = 90 . The data are reported for a common
distance of R = 100Dj .
FIGURES OF CHAPTER 6
110
(a)
(b)
(c)
(d)
Figure 612: The band-pass filtered acoustic nearfield of the base setup; the fretU
tU
quency band corresponds to St = 0.225 to St = 0.375; (a) Djj = 438; (b) Djj = 443;
tU
tUj
Dj
FIGURES OF CHAPTER 6
111
(a)
(b)
(c)
(d)
Figure 613: The band-pass filtered acoustic nearfield of the base setup; the fretU
tU
quency band corresponds to St = 0.925 to St = 1.075; (a) Djj = 438; (b) Djj = 443;
(c)
tUj
Dj
= 448; (d)
tUj
Dj
= 453.
FIGURES OF CHAPTER 6
112
(a)
(b)
(c)
(d)
Figure 614: The band-pass filtered acoustic nearfield of the microjet setup; the
tU
tU
frequency band corresponds to St = 0.225 to St = 0.375 (a) Djj = 272; (b) Djj = 302;
(c)
tUj
Dj
= 332; (d)
tUj
Dj
= 362.
FIGURES OF CHAPTER 6
113
(a)
(b)
(c)
(d)
Figure 615: The band-pass filtered acoustic nearfield of the microjet setup; the
tU
tU
frequency band corresponds to St = 0.925 to St = 1.075 (a) Djj = 272; (b) Djj = 302;
(c)
tUj
Dj
= 332; (d)
tUj
Dj
= 362.
Part II
Reacting Flows
114
CHAPTER 7
Reacting Flow Simulations
The objective of the present study is to use a high-order compressible NavierStokes solver to study the sound radiation by a diffusion flame in a planar mixing
layer, and by a nonpremixed jet flame. Such simulations with realistic chemistry pose
two major challenges. The first challenge is associated with incorporating complex
chemistry mechanisms for realistic flames. Combustion of conventional hydrocarbon
fuels involves complex chemistry mechanisms with many species and hundreds of
chemical reactions. Solving an additional transport equation for each species is
not affordable at present. Several methods have been introduced to address this
problem. These approaches include reducing chemical mechanisms in intrinsic low
dimensional manifolds (ILDM) (Maas & Pope, 1992a,b), or using flamelet-based
approaches such as the flamelet/progress variable (FPV) (Pierce & Moin, 2004; Ihme
et al., 2005; Ihme, 2007), flamelet prolongation of ILDM (FPI) (Gicquel et al., 2000),
and flamelet generated manifolds (FGM) (Van Oijen & De Goey, 2000). In these
approaches, the thermo-chemical states of interest are calculated and tabulated as
functions of a set of independent variables prior to the simulation. One additional
transport equation is solved for each independent variable during the simulation. All
other thermo-chemical variables of interest are retrieved from the tabulated thermochemistry during the course of the simulation. However, using the conventional
tabulated method implicitly requires the use of low-Mach-number assumption. The
115
116
energy, along with the equation of state, constitute the governing equations describing the physics of a compressible reacting flow. The conservation of total mass retains
the same form as the continuity equation for nonreacting flows (Poinsot & Veynante,
2005)
uj
+
= 0,
t
xj
(7.1)
117
+
=
(Vk,j Yk ) + k ,
t
xj
xj
(7.2)
where Vk,j is the the j- component of the diffusion velocity of species k, and k
denotes the production rate of species k due to the chemical reactions.
The conservation of momentum reads
N
X
ij
ui uj ui
+
=
+
Yk gi ,
t
xj
xi
k=1
(7.3)
where the last term on the right hand side takes into account the body forces, namely
buoyancy due to gravity.
The conservation of total energy is expressed as
E uj E
+
= T +
t
xj
xi
+
T
xi
xi
N
X
!
hs,k Yk Vk,i
k=1
N
X
Yk gi (ui + Vk,i ) ,
(ij ui ) +
xj
k=1
(7.4)
where T denotes the heat release due to the combustion. The third term on the
P
h
Y
V
RHS, x i N
k=1 s,k k k,i , is the differential diffusion of sensible enthalpy. The
P
last term on the RHS, N
k=1 Yk gi (ui + Vk,i ), is the power generated by the body
forces.
Considering an ideal gas mixture, the equation of state is
p = RT ,
(7.5)
118
where the gas constant is related to the universal gas constant, R, and mean molecular weight, W , through
R=
7.2
N
X
Yk
R
=R
.
W
W
k
k=1
(7.6)
Simplifying assumptions
A set of assumptions are made to neglect some physical effects which are unim-
portant for the problems being considered in the present study. These assumptions,
which simplify the governing equations and make the computational cost affordable,
are presented below.
7.2.1
Diffusion Fluxes
Using the Ficks law (Williams, 1985), the diffusion velocity can be approximated
by
Vk,j = Dk
ln Yk
.
xj
(7.7)
Yk
Dk
+ k .
xj
(7.8)
Yk uj Yk
=
+
t
xj
xj
DT
=
,
cp Dk
Dk
(7.9)
+
=
t
xj
xj
1
Yk
DT
Lek
xj
+ k .
(7.10)
119
It is customary to assume unit lewis numbers for numerical simulation of turbulent flames (Poinsot & Veynante, 2005). The numerical simulations of a turbulent
CH4 /H2 /N2 air diffusion flame (Pitsch, 2000) show that despite using unit lewis
numbers, good agreement with experimental data can be obtained especially at high
Reynolds numbers.
7.2.3
Buoyancy effects
Given the large density variation in a turbulent reacting flow, the effects of
buoyancy can be significant especially for low-Mach number flames. The effect of
buoyancy force can be characterized with Froud number,
Uref
.
Fr =
gDref
(7.11)
2/3
x
Dref
ref
1/3
(7.12)
to quantify the local buoyancy effect. The effects of Buoyancy force can be considered
negligible for < 1. Considering a turbulent jet flame with equal values of density
for the fuel and oxidizer, eq. (7.12) can be used to show that buoyancy effects are
negligible for x < 25Dref (Ihme, 2007).
7.3
Mixture Fraction
The analysis of nonpremixed flames can be simplified by defining a conserved
scalar called the mixture fraction. The mixture fraction is usually defined as a linear
120
combination of species mass fraction such that it takes values from zero, in the
oxidizer stream, to unity, in the fuel stream.
The mixture fraction can be considered as a nondimensional coordinate to describe the structure of a nonpremixed or partially-premixed flame. For further details
on how to define the mixture fraction for general fuels, the reader is referred to Chapter 3 of Poinsot & Veynante (2005).
7.4
Flamelet Modeling
In studies of turbulent flames, it is customary to characterize chemical reaction
rates with the Damkohler number. The Damkohler number, Da, is defined as the
ratio between the flow time scale and the chemistry time scale,
Da =
Lref Tref
tF
,
2
tc
Uref ref Uref
(7.13)
where Lref is the characteristic length of the flow, and Tref is the heat-release term
at a reference temperature.
For sufficiently large Damkohler numbers, turbulent flames may be treated as
ensembles of locally laminar flamelets (Peters, 2000; Poinsot & Veynante, 2005).
Furthermore, assuming that chemistry takes place fast enough such that the flame
structure is essentially in steady-state, the thermo-chemical states defining the flame
structure can be obtained from the steady flamelet equation (Peters, 2000; Poinsot
& Veynante, 2005),
1 2
2 = ,
2 Z
(7.14)
121
2Z
.
xi xi
(7.15)
The vector represents the temperature and the species mass fraction,
= [Y1
...
Yn
T ]T ,
(7.16)
and represents the corresponding source terms due to chemical reactions. The
scalar dissipation rate, , appears as an independent parameter in eq. (7.14). Indeed,
it is not totally independent of the mixture fraction. The usual practice is to express
the scalar dissipation rate in terms of the mixture fraction, using the solution of a onedimensional problem. For example, Peters (1983) showed that the scalar dissipation
rate in a one-dimensional mixing layer is given by
(z) = st exp 2 erfc1 (2Zst ) [erfc1 (2Z)2 ,
(7.17)
(7.18)
known as the flamelet library. Using the flamelet approach, one needs to solve
the Navier-Stokes equations, and only one additional transport equation, for the
122
conserved scalar Z. The flamelet library is then used to retrieve the thermo-chemical
variable of interest.
7.5
manifold as shown in Fig. 71. The upper branch of the S-shaped curve corresponds
to a stable burning conditions, while the lower branch represents non-reacting steady
states. The middle branch corresponds to unstable solutions of the steady flamelet
equations. Starting from an initial condition above the middle branch, the flame
tends to reach stable burning condition. With initial conditions below the middle
branch, the flame will eventually quench to reach a stable non-reacting condition on
the lower curve. The turning point between the upper and middle branches, q , is
the onset of flame quenching. For conditions with st > q , the heat dissipation from
the reaction zone is too great to maintain a stable burning condition. The turning
point between the lower and middle branches, i , represents the onset of re-ignition.
For conditions with st < i , the heat dissipation cannot offset the heat release by
chain branching reactions and the mixture will undergo a chemical run-off to reach
a stable burning condition.
Capturing the middle branch and flame quenching/re-ignition are of importance
for the accurate prediction of sound radiation from a turbulent flame. The middle
branch is inaccessible if the flamelet is tabulated in terms of Z and st , as represented
by eq. (7.18). This is because a single value of st corresponds to multiple solutions
in the flamelet library (cf. Fig. 71(a)). Therefore, a unique parametrization of the
flamelet library is not possible.
123
ber assumption where the density is assumed to change with temperature but remains independent of pressure (Poinsot & Veynante, 2005). This assumption is
realized by neglecting the pressure work terms in the conservation of energy. The
use of low-Mach number assumption eliminates the acoustic waves, and hence the
CFL restriction is relaxed. In fact, the energy equation might not even be solved
124
directly during the simulation (Poinsot & Veynante, 2005). Instead, the temperature is retrieved from the tabulated flamelet library. This method is interesting for
many applications, but cannot be used to directly simulate the combustion noise or
flame/acoustics interaction.
In this study, the fully compressible form of energy equation was solved along
with conservation of mass, momentum, and progress variable. However, the heat release because of the chemical reactions was retrieved from the flamelet library. Using
this approach, the tabulated chemistry was still used for computational efficiency,
while sound generation and propagation were captured.
This approach yields a small difference between the temperature obtained from
the simulation and the tabulated temperature in the flamelet library. This difference
is ignored in this study because it is only restricted to the flame region, and is also
very small for low-Mach number flows. In cases for which strong compressibility
effects are present, the flamelet library can have an additional dimension to take this
difference into account.
7.8
125
flame speed is retained to correctly predict the value of the fuel consumption and the
total heat release into the flow. Unfortunately, this model does not yield accurate
results for diffusion flames, because nonpremixed flames are non-propagating and the
concept of laminar flame speed does not apply to them.
In the present study, a new model is proposed in which the reaction zones of
diffusion flames are thickened in the mixture fraction space rather than the physical
space. In other words, a thicker flame is obtained in the physical space by increasing
the width of the reaction zone in the mixture fraction space as schematically shown
in Fig 72. Therefore, the model is called thickened flamelet model as compared to
the thickened flame model of Colin et al. (2000) for premixed flames.
The reaction zone in the flamelet is thickened by a change of variables from the
physical mixture fraction, Z, to a sudo-mixture fraction variable, Z 0 . This transformation can be considered as grid clustering near the stoichiometric mixture fraction,
Zst , in the mixture fraction space. The thickened flamelet model should also yield
the same value of total heat release as the original flame. Thus, the modified reaction
rate should satisfy the following equation,
T =
Z
T (Z)dZ =
dZ
T (Z) 0 dZ 0 =
dZ
0 T (Z 0 )dZ 0
(7.19)
dZ
.
dZ 0
(7.20)
In the new approach, the conventional flamelet equations, eq. (7.14), are initially
solved. Then, a proper transformation from Z to Z 0 is chosen to thicken the flamelet
126
through the use of eq. (7.20). The transformation is chosen in such a way that the
flame is properly resolved on the computational grid, while it can still be considered
acoustically compact compared for the shortest wavelength of interest. During the
course of simulation, the modified reaction rate is retrieved from the flamelet library
to yield a thickened flame in the physical domain.
7.9
Thickened Flamelet Model vs. Presumed PDF Modeling for Nonpremixed Flames
To better understand the effects of flame thickening, the thickened flamelet
model is compared with the widely used presumed Probability Density Function
(PDF) modeling for LES of nonpremixed turbulent combustion.
In conventional LES of turbulent reacting flows, assumptions of conventional
subgrid models are generally no longer valid. This is because the structure of the
flame is dominated by the effects of reaction and diffusion, and cannot generally
be determined by assuming a cascade process. Pope (1991) introduced a filtered
density function (FDF) to take into account the subgrid-scale (SGS) fluctuations in
a statistical manner.
In the context of the flamelet-progress-variable (FPV) approach, any filtered
such as
f
thermo-chemical quantity, ,
T or Yei , is modeled as,
=
Z Z
(7.21)
(7.22)
127
A transport equation can be solved to obtain the FDF function, P (Z, ). However, it
is customary to presume a shape for the FDF to avoid solving an additional transport
equation. Since the progress variable, , is supposed to be statistically independent
from Z, the joint FDF function can be simplified to
P (Z, ) =
P (Z)P () .
(Z, )
(7.23)
Assuming that the subgrid state of at a grid point corresponds to only one particular
flamelet (Pierce & Moin, 2004) yields
.
P () = ( )
(7.24)
Equations (7.21) and (7.24) can be combined to calculate the filtered heat-release
rate,
T =
Z Z
T (Z, )
dZd ,
P (Z)( )
(Z, )
(7.25)
T (Z, )
e
(Z, )
P (Z)dZ.
(7.26)
A comparison of eq. (7.26) and eq. (7.19) shows that the proposed thickened flamelet
model and the conventional presumed-PDF subgrid modeling become identical if the
flame is thickened in such a way that
dZ
=
P (Z) .
0
dZ
(Z, )
(7.27)
This analogy provides a rationale to chose a proper function for the transformation
between Z and Z 0 in the thickened flamelet model.
128
2
2
p
1 2
0
2
(ui uj ij ) 2 2 ,
p =
c20 t2
xi xj
t
c0
(7.28)
where c0 is the speed of sound in the ambient fluid. Introducing the excess density
(Morfey, 1973),
e = ( 0 )
1
(p p0 ) ,
c20
(7.29)
the following generalized form of Lighthills equation can be derived for a reacting
flow (Bailly et al., 2010)
1 2
2
2
0
p '
(ui uj ij )
c20 t2
xi xj
0 1
T
+
t c2
+
(e uj ) .
xj t
(7.30)
The first term on the right hand side of eq. (7.30) constitutes a quadrupole aerodynamic source of sound as in nonreactive flows. The second term is a monopole
source associated with direct combustion noise due to the volumetric expansion of
fluid. The third term represents indirect combustion noise sources and is a dipole
129
xi xj
Re ij
0 c20
( 1) 2 T
+Da
c
( u ) .
+
xj e j
(7.31)
M0 2 p ' Da( 1) 2
+ Da T ( 1)
.
c
c2
(7.32)
The first term on the RHS of eq. (7.32) denotes the contribution of the temporal
change of energy production term. The second term denotes a source term due
to the non-uniformity of the flow field. For isobaric combustion, the second term
contribution is less significant, and hence is neglected in what follows.
Using the chain rule to expand the temporal derivative of heat-release term in
eq. (7.32) yields
M02
Z T
0 c20 st T
2
2
0
+
p ' Da( 1) 2
.
2
c
st
Z
(7.33)
The first term in the brackets denotes the variation of heat release due to scalar dissipation rate fluctuations which is mostly due to motion of small scale structures. The
second term is associated with the heat release change due to temporal fluctuation
of mixture fraction which is mostly due to wrinkling of the flame surface by large
structures.
130
The sound radiation from a reacting mixing layer was invistigated. A planar
mixing layer with the following hyperbolic tangent profile for streamwise velocity
was considered:
U1 + U2 U2 U1
+
tanh
u(y) =
2
2
2y
,0
,
(7.34)
where U1 and U2 are the velocities of the low-speed and high-speed streams, respectively. The parameter ,0 represents the shear layer vorticity thickness at the inlet
boundary.
A one-step, global reaction of the following form was considered:
F + r O
(1 + r)P .
(7.35)
All species were assumed to have equal molecular weights, and diffusivity.
The reaction progress rate is given by (Lee & Pope, 1995)
(1 YP )
1 r+1
= A exp
exp
YF YO YP
,
1 (1 YP )
K
(7.36)
Tb Tu
,
Tb
(7.37)
131
and
=
Ta
,
Tb
(7.38)
respectively.
For this nonpremixed system, the mixture fraction is defined as (Poinsot &
Veynante, 2005)
rYF YO + YO0
,
Z=
rYF0 + YO0
(7.39)
where YF0 , and YO0 are the mass fractions of the fuel and the oxidizer in pure fuel and
oxidizer streams, respectively. The value of the mixture fraction at stoichiometry
would then be equal to
Zst =
1
.
1+r
(7.40)
In the present study, the values r = 1, A = 8 104 , = 0.87, = 4.0, and K = 100,
were considered which are in agreement with those considered by Sripakagorn et al.
(2004). Since the energy equation was already among the governing equations which
would be solved, temperature was used as the progress variable.
Figure 73 shows the heat-release source term before thickening as is obtained
from solving the flamelet equations. The flamelet library was modified for flame
thickening through the following transformation:
0
Z = A sinh [(Z 0 Zst
)] + B ,
(7.41)
where
1
0
0
A = {sinh [(1 Zst
)] + sinh( Zst
)}
(7.42)
132
and
0
B = A sinh ( Zst
).
(7.43)
0
= Zst were chosen for this study. Figure 74
The stretch parameter = 5.5, and Zst
shows the effect of the transformation in spreading the reaction zone in the mixture
fraction space. Note that the maximum value of the heat-release term is decreased to
maintain the same area under the curves which corresponds to the total heat release
into the flow.
The fuel and oxidizer were treated as ideal gases with = 1.4, and the cold
initial temperature T . The convective velocity was
Uc =
U1 + U2
= 0.375c ,
2
(7.44)
U2 U1
= 0.125c .
2
(7.45)
(U2 U1 ),0
= 1200 .
(7.46)
133
(7.47)
where = 0.0045, and is a random number between -1 to 1. The fundamental frequency, f0 corresponds to the most amplified instabilities, and is given by (Michalke,
1964)
f0 = 0.132
Uc
.
,0
(7.48)
Numerical results
The flow was initially assumed to be nonreactive, and was evolved to form a
fully developed, naturally growing, mixing layer as shown in Fig. 75. Once the
nonreactive, mixing layer was established, the simulation was switched to the reactive mode. During initial simulations, it was found out that the mixing layer could
restablize due to the decrease in the local Reynolds number caused by the combustion. Therefore, the heat-release source term was taken into account for points with
134
x > 40,0 to avoid the restabilization of the flow. The simulation was continued to
160,000 timesteps, which is equivalent to 6 acoustic-flow-through time.
Figure 76 shows the instantaneous vorticity contours of the reacting mixing
layer. The flow field shows severe vortex break down and shear layer instability in
comparison to the nonreactive case shown as in Fig. 75. The increased level of vortex
break down enhances the mixing of the fuel and the oxidizer,which can be observed
by studying the mixture fraction field as show in Fig. 77. Temperature contours
are shown in Fig. 78. The temperature increases from ambient cold temperature to
almost the adiabatic flame temperature within the mixing layer due to heat release.
Figure 79 shows the dilatation field along with the temperature contours highlighting the mixing layer. As expected, the combustion generated noise exhibits a
monopole type behavior. An interesting observation is the existence of strong sources
at three distinct locations. The first strong source seems to be located at the very
early stages of vortex break down at about x = 65,0 which is also shown by the
leftmost arrow. The second source, which is the strongest, seems to be located at
x = 92,0 where the vortex pairing is seen. The third strong source seems to be
located at x = 110,0 where a pair of vortices coalesce into a less coherent structure.
A zone of silence is noticeable downstream of x = 110,0 where shear layer width
increases, and it exhibits sever mixing and formation of less coherent structures. Ihme
et al. (2006) also report the existence of such a zone of silence due to the refraction
of sound waves caused by the sound speed gradient.
To better understand the dynamics of the noise generation, pressure fluctuations
were measured at 36 points, with 10 [deg] intervals, on a circle with radius R = 70,0 .
135
The circle was centered at x = 92,0 , and y = 0 which is approximately the location
of vortex pairing (c.f. Fig. 710). Figures 711 to 714 show the frequency content
of the measured pressure fluctuation signals at different angels from the positive xdirection. The location of these points are illustrated with the red circles in Fig. 710.
Strong low frequency tones are observed in the zone of silence ( = 10 [deg]).
These tones appear at the fundamental frequency of vortex breakdown, f0 , and its
first subharmonic, f0 /2, which corresponds to the vortex pairing frequency. These
tones, which are also observed for other radiation angles, suggest that flame wrinkling
by vortex breakdown and vortex pairing is a very effective source of sound. This is
also in agreement with the acoustic wave pattern seen in Fig. 79.
The fact that the acoustic energy level is relatively small in higher frequencies,
at = 10 [deg], is believed to be due to the refraction of high frequency waves.
Measurements show an increasing dominance of higher frequency tones with increasing radiation angle. The high frequency tones appear at subsequent harmonics of f0 ,
and f0 /2. The most dominant frequency at = 90 [deg] was found to be 4f0 (c.f
Fig 713). Fig. 715(a) and 715(b) show the directivity pattern of radiated sound
pressure level (SPL) at frequencies f0 , and 4f0 , respectively. Both figures show almost omni-directional patterns. A decrease of almost twenty decibels in radiated
SPL at 4f0 is observed in the zone of silence.
7.12
7.12.1
136
with realistic chemistry for methane-air combustion (e.g. DLR-A and Sandia D
flames). Unfortunately, these attempts were unsuccessful because of the prohibitive
computational cost of a high-order simulation.
Methane-air mixtures have typically low values of stoichiometric mixture fraction, and high values of Damkohler number. This in turn leads to excessively large
values of flame length and small values of reaction zone thickness. Simulating such
flames requires excessively large number of grid points, especially if a high-order
scheme is used. It was observed that insufficient grid resolution would result in the
generation of unphysical saw-tooth waves. The existence of these waves would later
make the simulation unstable. Therefore, it was decided to simulate a diffusion jet
flame with the simplified chemistry described in Sec. 7.11. A value of = 0.87 was
chosen for the heat-release parameter. All other parameters retained their value of
Sec. 7.11.
Grid-88 of Chapter 4 was chosen for the computation. The fuel jet stream was
initialized through the tangent-hyperbolic profile of eq. (4.1). The jet velocity was
chosen to yield Mj = Uj /c0 = 0.3, and ReD = 20, 000. The mixture fraction field
was initialized through the following relation,
Z(r) = 0.5 1 + tanh
r rj
2Z
,
(7.49)
where Z was chosen to be equal to the initial momentum thickness of the shear
layer.
137
Nearfield results
The simulation was evolved for 125,000 time steps which is equivalent to eighteen
acoustic-flow-through time. The simulation took approximately 8 days on 432 cores.
Figure 716 shows instantaneous snapshots of the jet flame flow field. Contours
of temperature field are superimposed on those of acoustic pressure to visualize the
sound radiation from the flame. Strong omni-directional sound radiation from the
shear layer is observed. Some high frequency radiation is also observed near the inlet
region which is due to the forcing procedure.
Since it was not intended to simulate a realistic flame, no comparison with
experimental data is provided. Centerline mean-flow properties, and their respective
rms values are shown in Fig. 717. The potential core length was found to be equal
to 6.2Dj . The maximum value for the centerline turbulence intensity was found to
be equal to 0.14, which was observed at around 12Dj from the jet nozzle. The mean
mixture fraction profile is similar to that of the axial velocity. The mean temperature
increases from the inlet, where < T > /T = 1, to reach its maximum value,
< T > /T 1.7, which is close to the adiabatic flame temperature. The maximum
temperature is observed at x/Dj 15 where the mean mixture fraction reaches
the stoichiometric value, Zst = 0.5. This trend is in agreement with experimental
observations of realistic flames.
7.12.3
Farfield Sound
Flow properties were recorded on an FW-H surface which extended from 2.2Dj
from the inlet boundary to 22Dj along the streamwise direction. The transverse
width of the surface increased from 4Dj at its inlet to 7Dj , at its outlet. The flow
138
properties on the FW-H surface were recorded every 20 time steps over a period
of 32000 time steps. This period corresponds to approximately three acoustic-flowthrough time. The grid cut-off frequency corresponds to StD = f Dj /Uj 11. The
farfield sound pressure history was calculated for virtual microphones located at a
distance of 100Dj from the jet flame nozzle. The farfield computation was performed
using 12 cores and took approximately 20 minutes.
Figure 718 shows the farfield sound pressure spectra at = 30 and = 90 .
The spectra show similar sound pressure levels for low values of Strouhal number.
This confirms the monopole-type characteristics of combustion noise at low frequencies (Strahle, 1971, 1978). It is also interesting to note that the sound pressure levels
measured at = 30 fall below those measured at = 90 for high frequencies.
This is believed to be due to high-frequency refraction effects, and is also reported
in experimental studies (Singh et al., 2003, 2004).
Figure 719 shows the directivity plot of the overall sound pressure level measured at a distance of 100Dj from the jet nozzle. The directivity plot is approximately omni-directional which is typical of monopole sources. The OASPL increases
for about 2 [dB] in the region close to 50 , which is probably due to sound
refraction effects.
The results from one-third octave bandpass filtering of the flame sound field, in
three different frequency bands, are shown in Fig. 720. The low frequency radiation looks quite omni-directional which is in agreement with the findings of spectral
139
analysis. As the radiation frequency increases, the pressure field becomes more directional. It is also interesting to note that sources seem to be located within one
potential core length from the jet nozzle.
FIGURES OF CHAPTER 7
140
2400
Upper branch
2200
Tst [K]
2000
1800
1600
Middle branch
1400
1200
Lower branch
1000 1
10
100
1000
st
(a)
2000
T [K]
1500
1000
500
0.2
0.4
(b)
0.6
0.8
Figure 71: Solution of the steady flamelet equations for partially premixed CH4 /air
combustion; the flame condition corresponds to the experiment of Cabra et al. (2005).
(a): S-shaped curve showing the variation of the temperature at stoichiometry as a
function of stoichiometric scalar dissipation rate, st ; (b): the flamelet solution in Z
space at st = 100s1 .
FIGURES OF CHAPTER 7
141
Figure 72: A schematic of the effect of flamelet thickening; left: the original flamelet;
right: the thickened flamelet.
FIGURES OF CHAPTER 7
142
1.2
st
2.5
0.8
0.6
1.5
0.4
0.2
0.5
0.2
0.4
0.6
0.8
Figure 73: The heat-release source term, T obtained from solving the flamelet
equations.
0.3
0.25
0.2
0.15
0.1
0.05
0
0
0.2
0.4
0.6
0.8
Figure 74: The heat-release source term, T at st = 0.6; solid line: original flamelet
solution as shown in Fig. 73; dashed line: thickened flamelet.
FIGURES OF CHAPTER 7
143
FIGURES OF CHAPTER 7
144
Figure 77: Mixture fraction contours for the reacting mixing layer.Pure fuel corresponds to Z = 1, while pure oxidizer corresponds to Z = 0.
Figure 78: Mixture fraction contours for the reacting mixing layer.Pure fuel corresponds to Z = 1, while pure oxidizer corresponds to Z = 0.
FIGURES OF CHAPTER 7
145
Figure 79: Temperature contours and dilatation field for the reacting mixing layer.
FIGURES OF CHAPTER 7
146
Figure 710: Locations of the virtual probes for acoustic pressure measurements.
FIGURES OF CHAPTER 7
147
0.00018
0.00016
0.00014
FFT( pac )
0.00012
0.0001
8e-05
6e-05
4e-05
2e-05
0
0.1
10
f/f0
Figure 711: The Fourier transform of the pressure fluctuation, pac = p/p 1,
measured at R = 70,0 , from x = 92,0 , and y = 0; solid line: = 10 [deg]; dashed
line: = 10 [deg].
0.0003
0.00025
FFT( pac )
0.0002
0.00015
0.0001
5e-05
0
0.1
10
f/f0
Figure 712: The Fourier transform of the pressure fluctuation, pac = p/p 1,
measured at R = 70,0 , from x = 92,0 , and y = 0; solid line: = 50 [deg]; dashed
line: = 50 [deg].
FIGURES OF CHAPTER 7
148
0.00012
0.0001
FFT( pac )
8e-05
6e-05
4e-05
2e-05
0
1
10
f/f0
Figure 713: The Fourier transform of the pressure fluctuation, pac = p/p 1,
measured at R = 70,0 , from x = 92,0 , and y = 0; solid line: = 90 [deg]; dashed
line: = 90 [deg].
0.00014
0.00012
FFT( pac )
0.0001
8e-05
6e-05
4e-05
2e-05
0
1
10
f/f0
Figure 714: The Fourier transform of the pressure fluctuation, pac = p/p 1,
measured at R = 70,0 , from x = 92,0 , and y = 0; solid line: = 140 [deg];
dashed line: = 140 [deg].
FIGURES OF CHAPTER 7
149
90
90
120
60
150
120
30
60
150
30
180
20
40
60
80 100 120
180
SPL [dB]
210
300
40
60
80 100 120
SPL [dB]
330
240
20
210
330
240
300
270
270
(a)
(b)
Figure 715: The directivity of radiated sound pressure level measured at R = 70,0 ,
from x = 92,0 , and y = 0; (a): at frequency f0 ; (b): at frequency 4f0 . The ambient
pressure was assumed to be atmospheric.
FIGURES OF CHAPTER 7
(a)
150
(b)
Figure 716: Instantaneous snapshots of jet flame flow field; (a): contours of temperature (hot color scheme) superimposed on the acoustic pressure (gray-scale color
scheme); (b): mixture fraction.
FIGURES OF CHAPTER 7
151
1.2
0.2
0.15
0.8
urms / Uj
< Uc > / Uj
0.6
0.4
0.1
0.05
0.2
0
0
0
10
15
20
1.2
0.3
0.25
0.8
0.2
0.6
15
20
15
20
15
20
0.15
0.4
0.1
0.2
0.05
0
0
10
15
20
x / Dj
10
x / Dj
0.5
1.8
0.4
Trms / Tj
< T > / Tj
10
x / Dj
zrms
<Z>
x / Dj
1.6
1.4
1.2
0.3
0.2
0.1
0
0
10
x / Dj
15
20
10
x / Dj
Figure 717: Mean and rms statistics of (a) axial velocity, (b) mixture fraction, and
(c) temperature along the jet flame centerline.
FIGURES OF CHAPTER 7
152
100
90
80
70
60
50
40
0.1
10
StD = f Dj / Uj
Figure 718: Sound pressure level spectra at a distance of R = 100Dj from the jet
flame nozzle; : radiation angle = 30 ; N: radiation angle = 90 .
90
60
30
60
70
80
90
0
100
OASPL [dB]
Figure 719: Directivity of overall sound pressure level (OASPL) in [dB] at 100Dj
from the jet nozzle
153
(a)
(b)
(c)
CHAPTER 8
Conclusions & Future Work
8.1
Conclusions
Although having been studied for about six decades, an unanimously accepted
description of jet noise generation mechanism has not been provided yet. The reported experimental and numerical studies of jet noise problem suggest that coherent
structures play a significant role in sound generation. The objective of the present
work was to use large eddy simulation and bandpass filtering of the flow field to
provide further insight into the role of coherent structures in subsonic jet noise generation. Bandpass filtering provides useful information about the location of sources
and the directivity of the radiated sound for a particular frequency band. This
interesting feature of bandpass filtering facilitates the study of sound radiation by
coherent structures at different scales.
Sound radiation from an unheated, Mach 0.9 jet at ReD = 400, 000 was studied
using state of art LES techniques. Computational grids with 33 million, 88 million,
and 380 million grid points were considered to examine the effect of grid resolution
on the shear layer development and flow properties. It was observed that a poor discretization of the shear layer may result in unphysical vortex pairing and high levels
of overall sound pressure. The grid with 380 million grid points provided the best
resolution, and yielded nearfield results with very good agreement with experimental
154
8.1. CONCLUSIONS
155
data. The predicted farfield sound pressure spectra were also in very good agreement with the reported experimental results for a wide range of frequencies. Some
discrepancies were observed for low values of Strouhal number which are believed
to be due to the relatively short time length for which the farfield calculations were
performed.
Finite Impulse Response (FIR) bandpass filters were used to study the sound
radiation from the jet at low and high frequencies. Low-frequency results showed
strong radiation along directions near the jet axis, 40 , from sources located
close to the termination of the potential core where coherent wavepackets undergo
distortion. This seems to be in agreement with reported results from two-point
correlation analysis (Panda & Seasholtz, 2002; Panda et al., 2005; Bogey & Bailly,
2007). Although weaker, significant low frequency radiation was also observed along
directions normal to the jet axis. This observation suggests that large-scale structures
significantly contribute to farfield sound radiation along sideline directions. The high
frequency sound field was strong towards 50 . Fine-scale sources were more
confined to the early stages of the shear layer development. Complex interference
patterns were also observed.
Careful examination of bandpass filtered pressure field suggested that there are
two mechanisms of sound radiation in subsonic jets that can occur in all scales of
turbulence. The first mechanism is the stretching and distortion of coherent structures, especially close to the termination of the potential core. As eddies are bent or
stretched, a portion of their kinetic energy is radiated as sound waves to the farfield.
Such vortex sound radiation may be formulated using Lighthills analogy (Howe,
8.1. CONCLUSIONS
156
1998, 2003). It is quadrapolar in nature and could explain the strong sound radiation towards 45 .
The second mechanism of sound generation seems to be associated with the
vibration of shear-layer interface with the ambient quiescent flow and has dipolar
characteristics. As turbulent coherent structures role up and are convected downstream inside the shear layer, they exert forces on the ambient flow by inducing
periodic ripples and pressure fluctuations on the shear layer interface. This mechanism is believed to be responsible for sound radiation along sideline directions.
Jet noise suppression through the use of microjets was also studied using a grid
with 30 million nodes. An unheated jet at Mj = 0.9 with a total of 8 equally spaced
microjets was considered. The test case conditions were comparable to those of
the experiment conducted by Alkislar et al. (2007). The microjet injection induced
secondary instabilities in the shear layer which triggered the transition to turbulence.
This, in turn, resulted in a slight reduction of the potential core length. In the case
of the base round jet, high levels of Reynolds stress appeared near the vortex pairing
region and occupied a relatively wide area in the shear layer. The use of microjets,
on the other hand, resulted in a narrow region of high-level Reynolds stress which
appeared closer to the nozzle. Microjet injection resulted in a reduction of about 4
[dB] in OASPL for almost all observer locations. This reduction is expected to be
less significant for better resolved turbulent shear layers on finer grids for which the
laminar vortex pairing is absent. The band-pass filter analysis showed that microjet
injection resulted in a low-frequency radiation pattern that was quite anti-symmetric
157
at some instances. This could be attributed to the presence of some helical waves
excited by the presence of the microjets.
Numerical simulations of sound radiation from nonpremixed flames were also
performed in the present work. The simulations featured the solution of the fully
compressible Navier-Stokes equations. Therefore, sound generation and radiation
were part of the solution. Given the small thickness of the flame reaction zone, resolving the flame structure on the grid was found to be the most significant challenge for
such studies. A thickened flamelet model was proposed that yields artificially thickened flames which can be resolved on the grid, while the physical value of total heat
release into the flow is retained. Combustion noise has monopolar characteristics for
low frequencies. For high frequencies, the sound field is no longer omni-directional.
Major sources of sound seems to be located in the jet shear layer within one potential
core length from the jet nozzle.
8.2
Future Work
The present work can be continued in many aspects for both reacting and non-
reacting flows. Some recommendations for future directions are discussed below.
8.2.1
= c0 .
k
(8.1)
158
Fluctuations satisfying the above equation are most likely to contribute to the farfield
sound radiation (Goldstein, 2003). An interesting alternative approach is to employ
the wavelet transform for such spatial filtering. Wavelet transform provides an interesting tool to study sources at different scales.
8.2.2
Heated jets
The bandpass filtering study presented in this thesis considered isothermal jets
only. It is interesting to conduct similar analysis for heated jets to study the effect of
entropy fluctuations. Such a study may clarify whether there is a separate entropy
source mechanism for sound generation, or heated jets also radiate sound through
the two suggested mechanisms for isothermal jets.
8.2.3
The bandpass filtering approach can also be used to study the effect of noise
suppressing devices on sound radiation at different frequencies. In the present work,
the effect of microjets on jet noise suppression was studied. It is interesting to pursue
similar studies for other noise suppressing devices such as chevrons and lobed mixers.
8.2.4
Wavepacket models
The bandpass filtered databases generated in the present work can be used to
develop wavepacket models for jet noise. A superposition of sound field generated
by wavepackets at different scales may provide a promising jet noise model.
8.2.5
It is desirable to have further verification of thickened flamelet model with complex chemistry, and realistic flame conditions. If proven accurate, the thickened
flamelet model can be a promising alternative to conventional subgrid models.
159
Direct simulations of flames and their radiated sound field, such as the ones
presented in this thesis, can be used to assess the accuracy of different acoustic
analogies and models modified for reacting flows. Some of such analogies are recently
presented by Bailly et al. (2010).
8.2.7
Formulation 1C can be extended to include the diffraction effects. One can use
the resulting formulation to study the diffraction of sound waves by fuselage, wing
pylons, etc. without the need to include them directly in the CFD simulation.
Part III
Appendices
160
APPENDIX A
A Low-Dispersion and Low-Dissipation Implicit Runge-Kutta Scheme
A.1
Introduction
Low-dispersion and low-dissipation schemes are needed for simulation of wave-
propagation phenomena such as those in computational aeroacoustics and electromagnetics. Efficient schemes with low dissipation and dispersion are available for
spatial discretization (Lele, 1992; Tam & Webb, 1993). But, optimization of temporal integration techniques to minimize dispersion and dissipation errors have received
comparatively less attention. The first attempt to develop a low-dispersion and lowdissipation Runge-Kutta (LDDRK) scheme was reported by Hu et al. (1996). Singlestep and two-step alternating methods were discussed. For single-step algorithms,
optimized four-, five-, and six-stage explicit Runge-Kutta schemes were introduced.
Fourth-order accuracy was obtained only for the six-stage scheme. Optimal 4-6 -and
5-6-stage alternating schemes were also developed which could retain fourth-order
accuracy. Bogey & Bailly (2004) developed explicit five- and six-stage Runge-Kutta
schemes. Both algorithms were of second order accuracy. More recently, Berland
et al. (2006) introduced a low-storage, fourth-order accurate optimized scheme.
The use of explicit schemes is not always desirable because of numerical stability
limitations. For example, including the nozzle geometry for jet noise simulations, and
simulating flame sound radiation and acoustic interaction are among the applications
for which a small CFL requirement may lead to prohibitive computational costs.
161
162
y(0) = y0 ,
(A.1)
P
X
b F ,
(A.2)
=1
where
F = f (yn +
P
X
a F , t = tn + c t)
= 1...P .
(A.3)
=1
163
(A.4)
bT
where the vector c corresponds to the positions of stage values within the time step.
Matrix A is the matrix of a s in eq.(A.3). A lower triangular matrix A with null
elements on the diagonal corresponds to an explicit scheme. Vector b corresponds
to the weight coefficients, b , in eq.(A.2).
Consider the following linear differential equation
dy
= i y(t) .
dt
(A.5)
For a P -stage Runge-Kutta method, the vector Yn+1 , made up from the stage solutions for the time step n + 1
T
Yn+1
= [y1
y2
...
yP ] ,
(A.6)
(A.7)
= t .
(A.8)
where
(A.9)
164
(A.10)
(A.11)
(A.12)
G()
= |R()| ei ,
Ge
(A.13)
such that the amplification (dissipation) and phase (dispersion) errors are represented
by |R(z)|, and the phase angle, , respectively. A low-dissipation and low-dispersion
scheme should ensure that |R()| = 1 and = 0 up to reasonable values of which
is related to the CFL number. In fact, it can be shown that
= (k x) CFL
(A.14)
for a linear wave equation, where k is the modified wavenumber owing to the spatial
discretization scheme.
165
ponents of vectors c, b, and matrix A should be determined such that the dispersion
and dissipation errors are minimized and the desired order of accuracy is maintained.
In the present work, a three-stage scheme of the following form is considered:
c1 a11
c2 a21 a22
0
0
(A.15)
b2
b3
Elements a12 , a13 , and a23 are set to zero for that the solution at one stage to be uncoupled from the solution at future stages. This enhances the ease of implementation,
and the convergence of the iterative method. Such schemes with a lower-triangular
A are called diagonally implicit Runge-Kutta (DIRK) schemes (Hairer & Wanner,
2010). If, in addition, all aii values are equal, the scheme is called singly diagonally
implicit Runge-Kutta (SDIRK) (Hairer & Wanner, 2010). The present schemes is
not considered to be SDIRK. This is because the additional assumption of SDIRK
schemes, which requires the diagonal elements to be equal, leaves us with fewer
coefficients for optimization.
To retain at least fourth-order accuracy, the coefficients should satisfy the following conditions:
First order:
3
X
=1
b = 1
(A.16)
166
Second order:
3
X
b c =
=1
1
2
(A.17)
Third order:
3
X
b c2 =
1
3
(A.18)
b a c =
1
6
(A.19)
=1
3 X
3
X
=1 =1
(A.20)
Fourth order:
3
X
b c3 =
1
4
(A.21)
b c a c =
1
8
(A.22)
b a c2 =
1
12
(A.23)
b a a c =
1
24
(A.24)
=1
3 X
3
X
=1 =1
3 X
3
X
=1 =1
3 X
3 X
3
X
=1 =1 =1
G() ei 2 d = min ,
(A.25)
167
where is specified according to the range of for which the method is optimized.
Moreover, to ensure the stability of the scheme, the amplification error should not
exceed the unity for any value of , i.e.
|R()| 1
Re() [0, ) .
(A.26)
1/2
/2
(1 + )/2
(1 )/2
1+
(1 + 2)
(1 + )/2
1/(62 )
1 1/(32 )
1/(62 )
(A.27)
where
2
= cos
.
18
3
(A.28)
168
Parameter
Value
a11
0.377847764031163
a21
0.385232756462588
a22
0.461548399939329
a31
0.675724855841358
a32
-0.061710969841169
a33
0.241480233100410
b1
0.750869573741408
b2
-0.362218781852651
b3
0.611349208111243
c1
0.257820901066211
c2
0.434296446908075
c3
0.758519768667167
Table A1: Optimal coefficients for the fourth-order, low-dispersion, low-dissipation,
implicit Rung-Kutta scheme.
The proposed scheme, ILDDRK4, yields significantly reduced dispersion and dissipation errors for all values of compared to the SDIRK scheme. The accuracy of
ILDDRK4 scheme for < 1 is comparable to that of standard explicit RK4 while it
retains better accuracy for > 1. It should also be noted that since the amplification
factor is less than or equal to unity for all values of Re() [0, ) the scheme is expected to be unconditionally stable. A stability analysis confirmed that the scheme
is stable for all negative values of Re(), and it is thus A-stable.
A.4
Numerical Example
In this section the accuracy of the scheme is compared with that of some con-
ventional Runge-Kutta schemes for three problems. A test case involving a linear
wave equation is considered Sec. A.4.1. A test case involving the propagation of an
Gaussian acoustic pulse is presented in Sec. A.4.2.
169
(A.29)
was solved numerically with the classic fourth-order Runge-Kutta , the three-stage
fourth-order SDIRK, and the proposed scheme. In all cases, a 6th-order implicit
compact scheme with a seven-point stencil (Lele, 1992) was used for spatial discretization. An eighth-order implicit filter (Gaitonde & Visbal, 2000) was employed
to ensure the numerical stability of the explicit scheme. Filtering was performed
at each time step for the explicit scheme. The filter was not needed and thus not
employed for the implicit schemes.
A wave speed C = 1 and spatial step size x = 1 were chosen. The initial
condition was
(xxm )2
u(x, t = 0) = e b2
2k2
2k1
[x xm ] + cos
[x xm ]
,
cos
x
x
(A.30)
where xm = 90, b = 20, k1 = 1/8, and k2 = 1/16. Figure A3 show the results
obtained for t = 300 with CFL numbers equal to 0.5 to 2. Table A2 shows the L2
norm of the error defined as
Z
xmax
(uex unu ) dx
L2 (e) =
1/2
,
(A.31)
where uex , and unu represent the exact and numerical solutions, respectively. The
proposed three-stage implicit scheme yields more accurate results than both the
170
Present scheme SDIRK-4
0.581021
1.89694
1.60723
3.74675
1.98162
4.4246
3.67395
4.84705
Table A2: The L2 norm of the error between the numerical results and the exact
solution at t = 300.
171
L2 (e)
1.17384E-5
2.00805E-4
3.21381E-3
4.79683E-2
Table A3: Error between the numerical results and the exact solution at t = 300
for different CFL numbers.
the error by almost a factor of 16, which is indicative of the fourth order accuracy of
the proposed scheme.
A.4.2
x2
b ,
=
1
+
e
u = 1 ,
p = 1 + e xb2
,
(A.32)
where = 0.01, and b = 3.0. The domain extended from -100 to 100 with x = 0.4.
Nonlinear Euler equations in one-dimension were considered as governing equations.
Nonreflective characteristic boundary conditions (Thompson, 1987; Poinsot & Lele,
1992) were considered at the boundaries. The boundary conditions were implemented
by modifying the flux terms at the boundaries to set the incoming waves to zero at
the boundaries.
The SDIRK and the new ILDDRK schemes were used for time integration. A
6th-order implicit compact scheme (Lele, 1992) was used for spatial discretization.
The implementation of implicit Runge-Kutta time integration with the compact
A.5. CONCLUSION
172
scheme was achieved following the work of Ekaterinaris (1999). Figure A.5 shows the
density profiles obtained from both schemes for t = 26 at different CFL numbers. The
reference solution corresponds to the numerical results obtained from classic explicit
Runge-Kutta scheme at CFL=0.25. Both schemes show very good agreement with
the reference solution at CFL=2 as shown in Fig. A4(a). With increasing CFL,
the accuracy of the solution obtained from SDIRK deteriorates significantly while
ILDDRK retains reasonable accuracy. It should be noted that the pulse is not
symmetric due to the existence of a mean flow. Due to the negative mean flow
velocity, the peak on the left side is propagating at c + |u| toward negative direction
of x axis, while the peak on the right side is moving at c |u| toward the positive
direction of x axis. Because the propagation speed of the peak on the left side is
greater, its corresponding dispersion and dissipation errors are more significant.
Although not shown here, simulations with CFL equal to 100 and more were also
performed to verify the stability of the temporal integration scheme. The simulations
remained stable even for such high values of CFL, although they resulted in very poor
accuracy as expected.
A.5
Conclusion
A three-stage, fourth-order, implicit, low-dispersion, and low-dissipation Runge-
A.5. CONCLUSION
173
A.5. CONCLUSION
174
/40
2.2
2
1.8
/80
1.4
|R()|
1.6
1.2
1
-/80
0.8
0.6
-/40
0.4 0
10
12
14
(a)
(b)
0.1
0.1
0.01
0.01
1-|R()|
Figure A1: Amplification factor (a) and difference in phase of the Runge-Kutta
schemes: : the new ILDDRK scheme, , standard explicit RK4; N, SDIRK.
0.001
0.001
0.0001
0.0001
1e-05
/8
/4
/2
(a)
1e-05
/8
/4
/2
(b)
Figure A2: (a) Dissipation and (b) dispersion error of the Runge-Kutta schemes
in logarithmic scales: : the new ILDDRK scheme, , standard explicit RK4; N,
SDIRK
A.5. CONCLUSION
175
Exact solution
New Implicit LDD-RK4
SDIRK
classic explicit RK4
f(x)
f(x)
Exact solution
New Implicit LDD-RK4
SDIRK
classic explicit RK4
-1
-1
-2
350
360
370
380
390
400
410
420
-2
350
430
360
370
380
390
(a)
3
400
410
420
430
420
430
(b)
3
Exact solution
New Implicit LDD-RK4
SDIRK
classic explicit RK4
f(x)
f(x)
Exact solution
New Implicit LDD-RK4
SDIRK
classic explicit RK4
-1
-1
-2350
360
370
380
390
400
410
420
430
-2350
360
370
380
390
(c)
400
410
(d)
Figure A3: Numerical solution of eq. (A.29) at t = 300; (a) CFL=0.5; (b) CFL=1.0;
(c) CFL=1.5; (d) CFL=2.0.
A.5. CONCLUSION
176
Reference
ILDRK
SDIRK
1.003
1.003
1.002
1.002
1.001
1.001
1
-100
1
-50
50
100
-100
-50
50
(a) CFL = 2
(b) CFL = 4
1.003
1.002
1.002
1.003
100
1.001
1.001
-100
-50
50
100
-100
-50
50
(c) CFL = 6
(d) CFL = 8
100
APPENDIX B
Systematic definition of progress variables and Intrinsically
Low-Dimensional, Flamelet Generated Manifolds for chemistry
tabulation
B.1
Introduction
Numerical simulations of turbulent reacting flows with detailed chemistry remain
177
B.1. INTRODUCTION
178
To overcome this limitation, flamelet-based approaches use the numerical solution of one-dimensional premixed flames or nonpremixed flamelet equations to construct look-up tables which describe the chemistry. In these methods, the thermochemical states of interest are calculated and tabulated, prior to the simulation, as
functions of a set of independent variables, referred to as progress variables.
Consider a flamelet table represented mathematically as
= F (c1 , c2 , , cn ) = F (c) ,
(B.1)
179
In this work, a systematic method to define the progress variables for tabulated
flamelets is proposed. The method is based on a Principal Component Analysis
(PCA) of the composition space, which is itself constructed from one-dimensional
flame simulations. Also known as Proper Orthogonal Decomposition (POD), the
Karhunen-Loe
ve Decomposition, and the Hotelling transform, PCA is widely used
for reduced-order modeling, low-dimensional representations of phenomena, and data
reduction (Berkooz et al., 1993). The method is analogous to the eigenvalue analysis of ILDM, providing the flamelet tabulations methods with a similarly rigorous
mathematical formulation.
Using PCA, the proposed approach provides the minimum number of linearly independent progress variables needed to uniquely identify the thermo-chemical states
of interest. Hence, the flamelet-based tabulations created with the new approach are
designated Intrinsically Low-Dimensional, Flamelet Generated Manifolds (IL-FGM).
A brief introduction to the Principal Component Analysis and the Singular Value
Decomposition (SVD) is provided in Sec. B.2. The new approach to define progress
variables and construct IL-FGM is presented in Sec. B.3. Numerical examples are
provided in Sec. B.4. Conclusions are drawn in Sec. B.5
B.2
B.2.1
Background
Consider the set of n measurements of a physical phenomenon. Each measurement yields m variables which describe the physical state of the system, represented
by an m1 vector, x. In the present study, the vector x consists of the mass fraction
180
Y2
Y m ]T ,
(B.2)
where Yi represents the mass fraction of the species i, and m is the number of species.
The data obtained from the n measurements form an m n matrix,
X = [x1
x2
xn ] .
(B.3)
Let X and C be m n matrices representing the original dataset, and its projection along the principal directions, respectively1 . These two matrices are related
through the following equation:
C = WX,
(B.4)
The mean value of each row of X is assumed to be zero for convenience in the
following derivations. If the dataset does not satisfy this condition, one can always
subtract the mean values from the elements of the matrix X before further analysis.
181
where the matrix W constitutes an orthonormal set on which the original dataset,
X, is projected to obtain C.
For optimal representation of the original dataset, the matrix C should have
minimum redundancy, and the new bases should be linearly independent. This is
achieved by defining W such that the covariance matrix,
CC =
1
CCT ,
n
(B.5)
is diagonal. The covariance matrix of the transformed data set, CC , is related to the
covariance matrix of the original dataset, CX , as follows:
1
CCT
n
1
(WX) (WX)T
=
n
1
T
= W
XX WT
n
CC =
= WCX WT .
(B.6)
This form is used in Sec. B.2.3 to find matrix W using a Singular Value Decomposition procedure.
B.2.3
(B.7)
182
CX =
(B.8)
1 T
SS ,
n
(B.9)
1 T
SS .
n
(B.10)
Since CC = n1 SST is diagonal, it is concluded that the columns of U are the principal
components of the data set X.
Another important property of the SVD is that
T
ek
X
mn = Umk Skk Vnk
k < min(m, n) ,
(B.11)
is the closest approximation of X in the least square sense for a k-dimensional sube k is commonly referred to as the approximation of rank k to X.
space. The matrix X
Matrices Umk , and Vnk are formed from the first k columns of Umm and Vnn ,
B.3. IL-FGM
183
respectively; similarly, the matrix Skk is the diagonal matrix made of the first k
singular values of X.
The SVD procedure yields a set of base vectors for a linear representation of
a given dataset with the fewest possible dimensions. In cases where the manifold
exhibits significant nonlinearities, such as turning points, the PCA might no longer
yield a representation with the fewest possible dimensions. This is because a basic
PCA procedure does not include any information about how the manifold is generated, and the physics of the problem is obscured. Such information can be useful to
better parametrize a dataset. Assume, for example, that all data points of a given set
fall on a circle with a constant radius. A simple PCA procedure yields a dataset with
at least two dimensions, corresponding to the Cartesian coordinates of each point.
But if the fact that all points are located on a circle is taken into account and polar
coordinates is used, the dataset can be represented with only one dimension which
corresponds to the polar angle of each point. This example of nonlinear component
analysis or Kernel PCA (Scholkopf et al., 1996) illustrates how some physical insight
can be used to map a dataset with significant nonlinearities into a new space, where
the dataset can be represented linearly, before the PCA procedure.
B.3
Fig. B1. Such data are used to generate tabulated flamelets such as FGM and FPI.
For a chemical mechanism with m species, the one-dimensional flame is sampled at
B.3. IL-FGM
184
m
X
wij Yj ,
(B.12)
j=1
such that they form the principal directions of the flamelet manifold in the composition space. As mentioned in Sec. B.2, the principal directions yield an accurate
representation of a dataset with the minimum number of dimensions. The magnitude
of each singular value of the matrix X indicates how important the corresponding
progress variable is in the tabulation, and how many progress variables are needed.
This constitutes a systematic way of defining the progress variables for Intrinsically
Low-dimensional, Flamelet-Generated Manifolds (IL-FGM). Equation (B.12) can be
The dataset X can also include samples from simulations with variable equivalence ratio, inlet temperature, or strain rate depending on the intended application.
In any case, the mean value of each row of the matrix X should be subtracted from
the corresponding row before the PCA procedure.
B.3. IL-FGM
185
recast in matrix form to yield eq. (B.4). Thus, the weight coefficients wij s are
obtained from the SVD of the matrix X as discussed in Sec. B.2.3.
For some cases, such as non-adiabatic flamelets (Fiorina et al., 2003), the flamelet
library might exhibit some nonlinearities, such as turning points at quenching and
re-ignition, and thus Kernel PCA is preferable. In such cases, it is desirable to include other scalars such as enthalpy in the definition of progress variables. The scalar
should then be scaled between zero and unity for a balanced weight with respect to
species mass fractions in the PCA procedure. Similarly, other scalar such as strain
rate or scalar dissipation rate can also be used in the definition of progress variables,
although the transport equation for the resulting progress variable might be very
complicated. In general, it is recommended to use the species mass fraction to define
the progress variable as much as possible to avoid such complications.
It should also be noted that the present method is an a posteriori analysis of an
already generated manifold. Thus, it cannot guide the choice of flamelets for which
the PCA procedure is to be performed. The correct choice of famelets requires identifying relevant physical phenomena (e.g. heat transfer at solid boundaries, radiation,
droplet evaporation, and etc.) for a given problem and generating flamelets for the
full range of applicable parameters. Once proper flamelets are generated, the (kernel) PCA can be used to define suitable progress variables and downsize the flamelet
library. An example is provided in Sec. B.4.2 in which a flamelet library is generated for a problem which involves various values of equivalence ratios in fresh gases.
The PCA procedure yielded two progress variables which were successfully used to
interpolate thermo-chemical states at different equivalence ratios along a flame.
186
Numerical Examples
In this section, the application of IL-FGM is demonstrated for both mono-
Two flamelets, each generated from the numerical solution of a freely propagating flame in a CH4 -air mixture, were considered as the first test cases. The GRI-30
mechanism (Smith et al., 1999) was used to describe the complex chemistry. The
fresh gas conditions were set to P = 1 [atm] and T = 473 [K]. The flamelets were
generated for equivalence ratios = 0.85 and = 1.9, respectively. Each flamelet
consists of data from a single equivalence ratio calculation, and thus can be fully
described with one progress variable.
The weight coefficients to define the progress variable for = 0.85 and = 1.9
are presented in Tables B1 and B2, respectively. The entries for species with significant weight are in bold. The method automatically attributes significant weights
to major reactants (O2, CH4), and major products (CO2, H2O). It is interesting
to note that the weight coefficient for CO was smaller for the fuel-lean flamelet than
that for the fuel-rich one as expected. This intrinsic ranking of important species is
a significant feature of the proposed method.
To further verify that the obtained progress variable uniquely describes the
flamelet, a new flame with = 0.85 and similar inlet conditions was solved, but
the computational domain length and the flame position were changed. Figure B2
187
shows the spatial distribution of the temperature and selected species mass fractions.
The numerical solution of the flame with complex chemistry is in excellent agreement
with data retrieved from the previously synthesized flamelet. Figure B3 shows the
same results in the progress variable space. It was also found that minor species with
weights lower than 0.0001 can be neglected from the definition of progress variables
without significant loss of accuracy.
Figure B4 shows the results of a similar test for a flame with = 1.90. Results
obtained from a flamelet with a conventionally defined progress variable are also
shown for comparison. The systematically defined progress variable yields more
accurate results than the arbitrarily defined one. The arbitrarily defined progress
variable does not necessarily increase or decrease monotonically along the flame.
Therefore, for some points in the flame, the value of the progress variable is not
unique, which causes inaccuracies in the interpolation. On the other hand, the
present method tries to define progress variables based on the principal components of
the reference solution which ensures the most accurate representation of the flamelet.
B.4.2
In many practical problems, the flamelet library may include several dimensions.
For instance, the flamelet library may include solutions for various equivalence ratios
and strain rates (c.f. Nguyen et al. (2010)), or two-phase flames. In such problems,
the thermo-chemistry can still be tabulated in terms of several progress variables,
defined from linear combinations of species mass fractions. This is demonstrated by
considering a flamelet library with various equivalence ratios. It is important to note
that no a priori knowledge of chemistry or flame structure is required.
188
A flamelet library was obtained from the solutions of a freely propagating flame
in CH4 -air mixture for {0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5}. The inlet
conditions for fresh gases were P = 1 [atm] and T = 473 [K].
Figure B5 shows the temperature distribution from the flamelet library as a
function of the first two progress variables, c1 and c2 . The weight coefficients for the
definition of c1 and c2 are provided in Tables B3 and B4, respectively. A Delaunay
triangulation of the data set was constructed using the Computational Geometry
Algorithms Library (The CGAL Project, 2010) for the linear interpolations.
Figure B6 shows a comparison between the results from a numerical simulation
of the flame at = 1.15, and the interpolated data from the flamelet library. The
temperature and the mass fractions of major species such as CO and CO2 are in
excellent agreement. The results for minor species such as OH are not as accurate,
but still in overall good agreement. The small discrepancy in the results for the OH
radical is due to the small weight attributed to this species. This highlights the fact
that the PCA procedure does not the practical importance of intermediate species
into consideration. The results can be further improved through the use of higher
order interpolations, and an increased number of flamelet solutions to expand the
flamelet library. Alternatively, one can always explicitly increase the weight of the
intermediate species of particular interest. The latter approach is akin to the kernel
PCA as discussed in Sec. B.2.3.
B.5. CONCLUSION
B.5
189
Conclusion
A systematic method to define progress variables for chemistry tabulation is
presented. The method is mathematically based on the Principal Component Analysis (PCA) of the composition space. The present method constituates a rigorous
mathematical formulation for flamelet-based tabulation methods analogous to the
eigenvalue analysis of the ILDM method.
The new method intrinsically yields the minimum number of linearly independent progress variables to accurately represent a general flamelet. Hence, it is suggested that the flamelet-based tabulations created with the new approach be designated with Intrinsically Low-Dimensional, Flamelet Generated Manifolds (IL-FGM).
Since PCA only tries to find the best linear representation of the dataset, in cases
where the manifold exhibits a significant nonlinearity, such as turning points, a nonlinear component analysis or Kernel PCA (Scholkopf et al., 1996) is preferable.
Numerical examples were presented to verify the accuracy of the present method.
Both mono-dimensional and multi-dimensional flamelet libraries were considered.
It was demonstrated that the new method automatically identifies the species of
significance importance, and takes them into account accordingly in the definition of
the progress variables. Very good agreement was obtained between flamelet-retrieved
results and those obtained from the numerical simulation of freely propagating flames.
This confirmed the effectiveness of the present approach.
Although only premixed flames were considered in the present work, the method
can also be applied in principle for partially premixed and nonpremixed flamelets.
B.5. CONCLUSION
190
The development of the present method involved a unity-Lewis-number assumption. It is desirable to extend the present method to relax this assumption in future
studies. The present approach is applicable for the simulation of turbulent combustion in real combustors, with nonpremixed, and partially premixed and non-adiabatic
regimes.
FIGURES OF APPENDIX B
191
Y
i
1 2 ... j
...
FIGURES OF APPENDIX B
192
H2
H
O
O2
OH
0.0001
0.0004
0.0060
-0.7637 0.0166
H2O
HO2
H2O2
C
CH
0.4140
-0.0002
0.0000
0.0000
0.0000
CH2
CH(S)
CH3
CH4
CO
0.0000
0.0000
0.0002
-0.1989 0.0686
CO2
HCO
CH2O
CH2OH
CH3O
0.4480
0.0000
-0.0008
0.0000
-0.0000
CH3OH
C2H
C2H2
C2H3
C2H4
-0.0002
0.0000
0.0000
0.0000
-0.0001
C2H5
C2H6
HCCO
CH2CO HCCOH
0.0000
-0.0007
0.0000
0.0000
0.0000
N
NH
NH2
NH3
NNH
0.0000
0.0000
0.0000
0.0000
0.0000
NO
NO2
N2O
HNO
CN
0.0002
-0.0000
0.0000
0.0000
0.0000
HCN
H2CN
HCNN
HCNO
HOCN
0.0000
0.0000
0.0000
0.0000
0.0000
HNCO
NCO
N2
AR
C3H7
0.0000
0.0000
0.0106
0.0000
-0.0000
C3H8
CH2CHO CH3CHO
0.0000
0.0000
0.0000
Table B1: The weight coefficients of each species, as obtained from PCA, to define
the progress variable for a premixed CH4 -air flame at = 0.85. The results are
shown to the fourth decimal. The entries for species with significant weight are in
bold.
FIGURES OF APPENDIX B
193
H2
H
O
O2
OH
0.0272
0.0000
0.0000
-0.7328 0.0001
H2O
HO2
H2O2
C
CH
0.4320
-0.0000
-0.0000
0.0000
0.0000
CH2
CH(S)
CH3
CH4
CO
0.0000
0.0000
0.0015
-0.3239 0.3706
CO2
HCO
CH2O
CH2OH
CH3O
0.3558
0.0000
-0.0006
0.0000
-0.0000
CH3OH
C2H
C2H2
C2H3
C2H4
-0.0002
0.0000
0.0395
0.0000
0.0017
C2H5
C2H6
HCCO
CH2CO HCCOH
0.0000
-0.0004
0.0000
0.0008
0.0000
N
NH
NH2
NH3
NNH
0.0000
0.0000
0.0000
0.0000
0.0000
NO
NO2
N2O
HNO
CN
0.0000
0.0000
0.0000
0.0000
0.0000
HCN
H2CN
HCNN
HCNO
HOCN
0.0000
0.0000
0.000
0.0000
0.0000
HNCO
NCO
N2
AR
C3H7
0.0000
0.0000
0.0067
0.0000
0.0000
C3H8
CH2CHO CH3CHO
0.0000
0.0000
0.0000
Table B2: The weight coefficients of each species, as obtained from PCA, to define
the progress variable for a premixed CH4 -air flame at = 1.9. The results are shown
to the fourth decimal. The entries for species with significant weight are in bold.
FIGURES OF APPENDIX B
194
0.14
0.12
2000
0.1
T[K]
YCO2
1500
0.08
0.06
1000
0.04
0.02
500
0
00
x [m]
x [m]
(a)
(b)
0.045
0.006
0.04
0.005
0.035
0.004
0.025
YOH
YCO
0.03
0.02
0.015
0.003
0.002
0.01
0.001
0.005
00
x [m]
(c)
00
x [m]
(d)
Figure B2: The structure of a freely propagating flame in a CH4 -air mixture for
= 0.85, T0 = 473 [K], and P = 1 [atm] in the physical space; solid line: physical
space solution; symbols: solution retrieved from the previously generated flamelet
table. (a): temperature; (b): CO2 mass fraction; (c): CO mass fraction; (d): OH
mass fraction.
FIGURES OF APPENDIX B
195
2200
0.14
2000
0.12
1800
0.1
1400
YCO2
T[K]
1600
1200
0.08
0.06
1000
0.04
800
0.02
600
400-0.2
-0.15
-0.1
-0.05
0.05
0-0.2
0.1
-0.15
-0.1
-0.05
0.05
0.1
0.05
0.1
(a)
(b)
0.045
0.006
0.04
0.005
0.035
0.004
0.025
YOH
YCO
0.03
0.02
0.015
0.003
0.002
0.01
0.001
0.005
0-0.2
-0.15
-0.1
-0.05
0.05
0.1
0-0.2
-0.15
-0.1
(c)
-0.05
(d)
Figure B3: The solution of the flame, shown in Fig. B2 versus the progress variable;
solid line: physical space solution; symbols: flamelet table. (a): temperature; (b):
CO2 mass fracion; (c): CO mass fracion; (d): OH mass fraction.
FIGURES OF APPENDIX B
196
2000
0.06
1800
0.05
1600
0.04
YCO2
T [K]
1400
1200
0.03
1000
0.02
800
0.01
600
400 0
0.01
0.02
0.03
0.04
0.05
0.06
00
0.07
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.06
0.07
x [m]
x [m]
(a)
(b)
0.12
4e-05
3.5e-05
0.1
3e-05
0.08
YOH
YCO
2.5e-05
0.06
2e-05
1.5e-05
0.04
1e-05
0.02
00
5e-06
00
0.01
0.02
0.03
0.04
0.05
0.06
0.01
0.02
0.03
0.05
x [m]
x [m]
(c)
0.04
0.07
(d)
Figure B4: The structure of a freely propagating flame in a CH4 -air mixture for
= 1.9, T0 = 473 [K], and P = 1 [atm] in the physical space; solid line: physical
space solution; circles: solution retrieved from the previously generated flamelet
table using the new definition of progress variable; triangles: solution retrieved from
the previously generated flamelet table using c = YCO2 + YCO + YH2O + YH2 . (a):
temperature; (b): CO2 mass fraction; (c): CO mass fraction; (d): OH mass fraction.
FIGURES OF APPENDIX B
197
2500
2000
1500
1000
500
0
0.15
0.15
0.2
0.1
0.05
0.25
0
0.05
0.3
0.1
0.15
c2
0.35
0.2
c1
(a)
Figure B5: The temperature distribution in the progress variable space; solid lines:
physical space solution for various equivalence ratios ; symbols: interpolated values
for equivalence ratio = 1.15.
FIGURES OF APPENDIX B
198
H2
H
O
O2
OH
0.0066
0.0006
0.0030
-0.7767 0.0108
H2O
HO2
H2O2
C
CH
0.4345
-0.0002
-0.0000
0.0000
0.0000
CH2
CH(S)
CH3
CH4
CO
0.0000
0.0000
0.0008
-0.2191 0.1945
CO2
HCO
CH2O
CH2OH
CH3O
0.3491
0.0000
-0.0009
0.0000
0.0000
CH3OH
C2H
C2H2
C2H3
C2H4
-0.0002
0.0000
0.0011
0.0000
-0.0001
C2H5
C2H6
HCCO
CH2CO HCCOH
0.0000
-0.0010
0.0000
0.0001
0.0000
N
NH
NH2
NH3
NNH
0.0000
0.0000
0.0000
0.0000
0.0000
NO
NO2
N2O
HNO
CN
0.0003
0.0000
0.0000
0.0000
0.0000
HCN
H2CN
HCNN
HCNO
HOCN
0.0000
0.0000
0.0000
0.0000
0.0000
HNCO
NCO
N2
AR
C3H7
0.0000
0.0000
-0.0031
-0.0000
0.0000
C3H8
CH2CHO CH3CHO
0.0000
0.0000
0.0000
Table B3: The weight coefficients of each species, as obtained from PCA, to define
the first progress variable, c1 , for a premixed CH4 -air flame at [0.6, 1.5]. The
results are shown to the fourth decimal. The entries for species with significant
weight are in bold.
FIGURES OF APPENDIX B
199
H2
H
O
O2
OH
0.0486
0.0002
-0.0083
-0.0785 -0.0252
H2O
HO2
H2O2
C
CH
0.1581
0.0006
0.0000
0.0000
0.0000
CH2
CH(S)
CH3
CH4
CO
0.0001
0.0000
0.0093
0.1589 0.6553
CO2
HCO
CH2O
CH2OH
CH3O
-0.6395
0.0001
0.0066
0.0000
0.0000
CH3OH
C2H
C2H2
C2H3
C2H4
0.0009
0.0000
0.0105
0.0001
0.0096
C2H5
C2H6
HCCO
CH2CO HCCOH
0.0000
0.0077
0.0002
0.0028
0.0000
N
NH
NH2
NH3
NNH
-0.0001
0.0000
0.0000
0.0000
0.0000
NO
NO2
N2O
HNO
CN
0.0003
0.0000
0.0000
0.0000
0.0000
HCN
H2CN
HCNN
HCNO
HOCN
0.0000
0.0000
0.0000
0.0000
0.0000
HNCO
NCO
N2
AR
C3H7
0.0003
0.0000
-0.3190
0.0000
0.0000
C3H8
CH2CHO CH3CHO
0.0002
0.0000
0.0003
Table B4: The weight coefficients of each species, as obtained from PCA, to define
the second progress variable, c2 , for a premixed CH4 -air flame at [0.6, 1.5]. The
results are shown to the fourth decimal.The entries for species with significant weight
are in bold.
FIGURES OF APPENDIX B
200
2400
0.12
2200
0.1
2000
1800
0.08
YCO2
T [K]
1600
1400
0.06
1200
0.04
1000
800
0.02
600
400 0
0.005
0.01
0.015
0.02
0.025
00
0.03
0.005
0.01
0.015
0.02
0.025
0.03
0.02
0.025
0.03
x [m]
x [m]
(a)
(b)
0.07
0.006
0.06
0.005
0.05
0.004
YCO
YOH
0.04
0.03
0.003
0.002
0.02
0.001
0.01
00
0.005
0.01
0.015
x [m]
(c)
0.02
0.025
0.03
00
0.005
0.01
0.015
x [m]
(d)
Figure B6: The structure of a freely propagating flame in a CH4 -air mixture for
= 1.15, T0 = 473 [K], and P = 1 [atm] in the physical space; solid line: physical
space solution; symbols: solution retrieved from the previously generated flamelet
table. (a): temperature; (b): CO2 mass fraction; (c): CO mass fraction; (d): OH
mass fraction.
References
Alexander, R. (1977). Diagonally implicit Runge-Kutta methods for stiff ODEs.
SIAM Journal on Numerical Analysis, 14 (6), 10061021.
Alkislar, M. B., Krothapalli, A., & Butler, G. W. (2007). The effect of streamwise
vortices on the aeroacoustics of a Mach 0.9 jet. Journal of Fluid Mechanics, 578 ,
139169.
Arakeri, V. H., Krothapalli, A., Siddavaram, V., Alkislar, M. B., & Lourenco, L.
(2003). On the use of microjets to suppress turbulence in a Mach 0.9 axisymmetric
jet. Journal of Fluid Mechanics, 490 , 7598.
Avital, E., & Sandham, N. (1997). A note on the structure of the acoustic field
emitted by a wave packet. Journal of sound and vibration, 204 (3), 533539.
Bailly, C., Bogey, C., & Candel, S. (2010). Modeling of sound generation by turbulent
reacting flows. International journal of aeroacoustics, 9 (4 & 5), 461490.
Balsa, T. F., Gliebe, P. R., Kantola, R. A., Mani, R., Strings, E. J., & Wong II, J.
C. F. (1978). High velocity jet noise source location and reduction. Tech. Rep.
FAA-RD-76-79, Federal Aviation Administration.
Becker, H., & Yamazaki, S. (1978). Entrainment, momentum flux and temperature
in vertical free turbulent diffusion flames. Combustion and Flame, 33 , 123149.
Berkooz, G., Holmes, P., & Lumley, J. (1993). The proper orthogonal decomposition
in the analysis of turbulent flows. Annual Review of Fluid Mechanics, 25 (1),
201
FIGURES OF APPENDIX B
202
539575.
Berland, J., Bogey, C., & Bailly, C. (2006). Low-dissipation and low-dispersion
fourth-order rungekutta algorithm. Computers & Fluids, 35 (10), 14591463.
Blokhintsev, D. (1956). Acoustics of a nonhomogeneous moving medium. Tech. Rep.
TM-1399, NACA.
Bodony, D. J. (2004). Aeroacoustics Prediction of Turbulent Shear Flows. Ph.D.
thesis, Stanford University.
Bogey, C., & Bailly, C. (2004). A family of low dispersive and low dissipative explicit schemes for flow and noise computations. Journal of Computational Physics,
194 (1), 194214.
Bogey, C., & Bailly, C. (2005). Effects of inflow conditions and forcing on subsonic
jet flows and noise. AIAA journal , 43 (5), 10001007.
Bogey, C., & Bailly, C. (2006). Computation of a high Reynolds number jet and its
radiated noise using large eddy simulation based on explicit filtering. Computers
& fluids, 35 (10), 13441358.
Bogey, C., & Bailly, C. (2007). An analysis of the correlations between the turbulent
flow and the sound pressure fields of subsonic jets. Journal of Fluid Mechanics,
583 (-1), 7197.
Bogey, C., & Bailly, C. (2010). Influence of nozzle-exit boundary-layer conditions on
the flow and acoustic fields of initially laminar jets. Journal of Fluid Mechanics,
663 , 507538.
Bogey, C., Bailly, C., & Juve, D. (2000). Numerical simulation of sound generated
by vortex pairing in a mixing layer. AIAA journal , 38 (12), 22102218.
FIGURES OF APPENDIX B
203
Bogey, C., Bailly, C., & Juve, D. (2003). Noise investigation of a high subsonic, moderate Reynolds number jet using a compressible large eddy simulation. Theoretical
and Computational Fluid Dynamics, 16 (4), 273297.
Bogey, C., Barre, S., Fleury, V., Bailly, C., & Juve, D. (2007). Experimental study
of the spectral properties of near-field and far-field jet noise. International Journal
of Aeroacoustics, 6 (2), 7392.
Bogey, C., Marsden, O., & Bailly, C. (2011). Large-eddy simulation of the flow
and acoustic fields of a reynolds number 100,000 subsonic jet with tripped exit
boundary layers. Physics of Fluids, 23 , 035104.
Bose, S., Moin, P., & You, D. (2010). Grid-independent large-eddy simulation using
explicit filtering. Physics of Fluids, 22 , 105103.
Brentner, K. (1986). Prediction of helicopter discrete frequency noise-a computer
program incorporating realistic blade motions and advanced acoustic formulation.
Tech. Rep. NASA TM - 87721, NASA.
Brentner, K., & Farassat, F. (2003). Modeling aerodynamically generated sound of
helicopter rotors. Progress in Aerospace Sciences, 39 (2-3), 83120.
Bui, T., Ihme, M., Meinke, M., Schroeder, W., & Pitsch, H. (2007). Numerical
investigation of combustion noise and sound source mechanisms in a non-premixed
flame using LES and APE-RF. AIAA Paper , 3406 .
Butcher, J. (2008). Numerical Methods for Ordinary Differential Equations, Second
Edition.. John Wiley & Sons Ltd.
Cabra, R., Chen, J., Dibble, R., Karpetis, A., & Barlow, R. (2005). Lifted methaneair jet flames in a vitiated coflow. Combustion and flame, 143 (4), 491506.
FIGURES OF APPENDIX B
204
FIGURES OF APPENDIX B
205
Di Francescantonio, P. (1997). A new boundary integral formulation for the prediction of sound radiation. Journal of Sound and Vibration, 202 (4), 491509.
Ekaterinaris, J. (1999). Implicit, high-resolution, compact schemes for gas dynamics
and aeroacoustics. Journal of Computational Physics, 156 (2), 272299.
Farassat, F. (1975). Theory of noise generation from moving bodies with an application to helicopter rotors. Tech. Rep. NASA Tech. Rep. R-451, Langley Research
Center.
Farassat, F. (1994). Introduction to generalized functions with applications in aerodynamics and aeroacoustics. Tech. Rep. TP-3428, NASA.
Farassat, F. (2000). Introduction to generalized functions with applications in aerodynamics and aeroacoustics. Journal of Sound and Vibration, 230 , 460462.
Farassat, F., & Succi, G. P. (1982). The prediction of helicopter rotor discrete frequency noise. In In: American Helicopter Society, Annual Forum, 38th, Anaheim,
CA, May 4-7, 1982, Proceedings.(A82-40505 20-01) Washington, DC, American
Helicopter Society, 1982, p. 497-507., (pp. 497507).
Ffowcs Williams, J. E. (1960). Some thoughts on the effect of aircraft motion and
eddy convection on the noise from air jetspressure. Tech. Rep. 155, University of
Southampton Aero. Astr.
Ffowcs Williams, J. E. (1963). The noise from turbulence convected at high speed.
Royal Society of London Philosophical Transactions Series A, 255 , 469503.
Ffowcs Williams, J. E., & Hawkings, D. L. (1969). Sound generation by turbulence
and surfaces in arbitrary motion. Philosophical Transactions of the Royal Society
of London. Series A, Mathematical and Physical Sciences, 264 (1151), 321.
FIGURES OF APPENDIX B
206
Fiorina, B., Baron, R., Gicquel, O., Thevenin, D., Carpentier, S., & Darabiha, N. (2003). Modelling non-adiabatic partially premixed flames using flameprolongation of ildm. Combustion Theory and Modelling, 7 (3), 449470.
Flemming, F., Sadiki, A., & Janicka, J. (2007). Investigation of combustion noise
using a LES/CAA hybrid approach. Proceedings of the Combustion Institute,
31 (2), 31893196.
Fleury, V., Bailly, C., Jondeau, E., Michard, M., & Juve, D. (2008). Space-time correlations in two subsonic jets using dual particle image velocimetry measurements.
AIAA journal , 46 (10), 24982509.
Gaitonde, D., & Visbal, M. (2000). Pade-type higher-order boundary filters for the
navier-stokes equations. AIAA journal , 38 (11), 21032112.
Germano, M., Piomelli, U., Moin, P., & Cabot, W. (1991). A dynamic subgrid-scale
eddy viscosity model. Physics of Fluids A: Fluid Dynamics, 3 , 1760.
Gicquel, O., Darabiha, N., & Thevenin, D. (2000). Laminar premixed hydrogen/air
counterflow flame simulations using flame prolongation of ildm with differential
diffusion. Proceedings of the Combustion Institute, 28 , 19011908.
Goldstein, M. E. (2003). A generalized acoustic analogy. Journal of Fluid Mechanics,
488 , 315333.
Goldstein, M. E., & Howe, M. S. (1973). New aspects of subsonic aerodynamic noise
theory. Tech. Rep. TN D-7158, NASA.
Golub, G., & Van Loan, C. (1996). Matrix computations. Johns Hopkins Univ Pr.
Goodwin, D. (2001). An object-oriented software toolkit for chemical kinetics, thermodynamics, and transport processes. California Institute of Technology.
FIGURES OF APPENDIX B
207
URL http://code.google.com/p/cantera/
Greska, B., Krothapalli, A., & Arakeri, V. (2003). A further investigation into the
effects of microjets on high speed jet noise. In the 9th AIAA/CEAS Aeroacoustics
Conference, Hilton Head, SC , AIAA-2003-3128.
Greska, B., Krothapalli, A., Burnside, N., & Horne, W. (2004). High-speed jet noise
reduction using microjets on a jet engine. In 10th AIAA/CEAS Aeroacoustics
Conference, Manchester, Great Britain, AIAA-2004-2969.
Hairer, E., & Wanner, G. (2010). Solving ordinary differential equations II: Stiff and
differential-algebraic problems. Springer.
Howe, M. (2003). Theory of vortex sound , vol. 33. Cambridge University Press.
Howe, M. S. (1998). Acoustics of fluid-structure interactions. Cambridge University
Press.
Hu, F., Hussaini, M., & Manthey, J. (1996). Low-dissipation and low-dispersion
runge-kutta schemes for computational acoustics.
Journal of Computational
FIGURES OF APPENDIX B
208
FIGURES OF APPENDIX B
209
Laufer, J., Schlinker, R., & Kaplan, R. (1975). Experiments on supersonic jet noise.
In American Institute of Aeronautics and Astronautics, Aero-Acoustics Conference, 2 nd, Hampton, Va, U. S. Department of Transportation, vol. 24.
Le Diz`es, S., & Millet, C. (2007). Acoustic near field of a transonic instability wave
packet. Journal of Fluid Mechanics, 577 (1), 123.
Lee, Y., & Pope, S. (1995). Nonpremixed turbulent reacting flow near extinction.
Combustion and flame, 101 (4), 501528.
Lele, S. (1992). Compact finite difference schemes with spectral-like resolution. Journal of Computational Physics, 103 (1), 1642.
Lew, P. (2009). A study of sound generation from turbulent heated round jets using
3-D Large Eddy Simulation. Ph.D. thesis, Purdue University.
Lew, P., Najafi-Yazdi, A., & Mongeau, L. (2010). Unsteady numerical simulation of a
round jet with impinging microjets for noise suppression. In 48th AIAA Aerospace
Sciences Meeting, January 4-7 2010, Orlando, Florida, USA, AIAA-2010-0018.
Lighthill, M. J. (1952). On sound generated aerodynamically. I. General theory.
Proceedings of the Royal Society of London. Series A, Mathematical and Physical
Sciences, 211 (1107), 564587.
Lighthill, M. J. (1954). On sound generated aerodynamically. II. Turbulence as a
source of sound. Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences, 222 (1148), 132.
Lilley, G. M. (1974). On the noise from jets. AGARD CP-131 , (pp. 13.113.12).
Lockard, D. (2002). A comparison of ffowcs williams-hawkings solvers for airframe
noise applications. AIAA Paper , 2580 (8).
FIGURES OF APPENDIX B
210
Lund, T. (2003). The use of explicit filters in large eddy simulation. Computers &
Mathematics with Applications, 46 (4), 603616.
Lush, P. A. (1971). Measurements of subsonic jet noise and comparison with theory.
Journal of Fluid Mechanics, 46 (03), 477500.
Lyons, R. (2004). Understanding digital signal processing. Prentice Hall PTR.
Maas, U., & Pope, S. (1992a). Implementation of simplified chemical kinetics based
on intrinsic low-dimensional manifolds. In Symposium (International) on Combustion, vol. 24, (pp. 103112). Elsevier.
Maas, U., & Pope, S. (1992b).
FIGURES OF APPENDIX B
211
Michalke, A. (1970). A wave model for sound generation in circular jets. Tech. rep.,
Deutsche Luft- und Raumfahrt.
URL http://elib.dlr.de/63062/1/DLR_FB70-57.pdf
Michalke, A. (1972). An expansion scheme for the noise from circular jets. Zeitschrift
f
ur Flugwissenschaften, 20 , 321367.
Mitchell, B. E. (1995). Direct computation of the sound generated by subsonic and
supersonic axisymmetric jets. Ph.D. Dissertation, Stanford University.
Moin, P., Squires, K., Cabot, W., & Lee, S. (1991). A dynamic subgrid-scale model
for compressible turbulence and scalar transport. Physics of Fluids A: Fluid Dynamics, 3 , 2746.
Mollo-Christensen, E. (1967). Jet noise and shear flow instability seen from an
experimenters viewpoint. ASME Journal of Applied Mechanics, 34 (1), 17.
Mollo-Christensen, E., Kolpin, M. A., & Martucelli, J. R. (1964). Experiments on
jet flows and jet noise far-field spectra and directivity patterns. J. Fluid Mech,
18 , 285301.
Morfey, C. (1973). Amplification of aerodynamic noise by convected flow inhomogeneities. Journal of Sound and Vibration, 31 (4), 391397.
Morgans, A. S., Karabasov, S. A., Dowling, A. P., & Hynes, T. P. (2005). Transonic
helicopter noise. AIAA journal , 43 (7), 1512.
Morino, L. (1974). A general theory of unsteady compressible potential aerodynamics. Tech. Rep. NASA CR-2464, NASA.
FIGURES OF APPENDIX B
212
Systematic definition of
http://www.sciencedirect.com/science/article/pii/
S0010218011003087
Najafi-Yazdi, A., & Mongeau, L. (2010a). A low-dispersion and low-dissipation
implicit runge-kutta scheme. In the 16th AIAA/CEAS Aeroacoustics Conference,Stockholm, Sweden, AIAA-2010-3938.
Najafi-Yazdi, A., & Mongeau, L. (2010b). Repor of Invention: McGill Computational
Acoustic Analogy Package (MCAAP). Tech. Rep. 11042, McGill University.
FIGURES OF APPENDIX B
213
Nguyen, P., Vervisch, L., Subramanian, V., & Domingo, P. (2010). Multidimensional
flamelet-generated manifolds for partially premixed combustion. Combustion and
Flame, 157 (1), 4361.
Nocedal, J., & Wright, S. (1999). Numerical optimization. Springer Verlag.
Obrist, D. (2009). Directivity of acoustic emissions from wave packets to the far
field. Journal of Fluid Mechanics, 640 (1), 165186.
Obrist, D. (2011). Acoustic emissions from convected wave packets. Physics of
Fluids, 23 (2), 6101.
Oppenheim, A. V., & Schafer, R. W. (2010). Discrete-time signal processing. Prentice
Hall PTR.
Panda, J., & Seasholtz, R. G. (2002). Experimental investigation of density fluctuations in high-speed jets and correlation with generated noise. Journal of Fluid
Mechanics, 450 (-1), 97130.
Panda, J., Seasholtz, R. G., & Elam, K. A. (2005). Investigation of noise sources in
high-speed jets via correlation measurements. Journal of Fluid Mechanics, 537 (1), 349385.
Peters, N. (1983). Local quenching due to flame stretch and non-premixed turbulent
combustion. Combustion Science and Technology, 30 (1), 117.
Peters, N. (2000). Turbulent Combustion. Cambridge University Press.
Phillips, O. M. (1960). On the generation of sound by supersonic turbulent shear
layers. J. Fluid Mech, 9 (1), 128.
Pierce, C., & Moin, P. (2004). Progress-variable approach for large-eddy simulation
of non-premixed turbulent combustion. Journal of Fluid Mechanics, 504 , 7397.
FIGURES OF APPENDIX B
214
FIGURES OF APPENDIX B
215
Smagorinsky, J. (1963). General circulation experiments with the primitive equations. Monthly weather review , 91 (3), 99164.
Smith, G. P., Golden, D. M., Frenklach, M., Moriarty, N. W., Eiteneer, B., Goldenberg, M., Bowman, C. T., Hanson, R. K., Song, S., William C. Gardiner, J.,
Lissianski, V. V., & Qin, Z. (1999). GRI-30 Mechanism. Available online at
http://www.me.berkeley.edu/gri mech.
URL http://www.me.berkeley.edu/gri_mech/version30/text30.html
Smith, M. J. (2004). Aircraft noise. Cambridge Univ Pr.
Sripakagorn, P., Mitarai, S., Kosaly, G., & Pitsch, H. (2004). Extinction and reignition in a diffusion flame: a direct numerical simulation study. Journal of Fluid
Mechanics, 518 , 231259.
Stolz, S., & Adams, N. (1999). An approximate deconvolution procedure for largeeddy simulation. Physics of Fluids, 11 , 1699.
Strahle, W. (1971). On combustion generated noise. Journal of Fluid Mechanics,
49 (2), 399414.
Strahle, W. (1978). Combustion noise. Progress in Energy and Combustion Science,
4 , 157176.
Tacke, M., Linow, S., Geiss, S., Hassel, E., Janicka, J., & Chen, J. (1998). Experimental and numerical study of a highly diluted turbulent diffusion flame close to
blowout. In Symposium (International) on Combustion, vol. 27, (pp. 11391148).
Elsevier.
Tam, C. (1995). Supersonic jet noise. Annual Review of Fluid Mechanics, 27 (1),
1743.
FIGURES OF APPENDIX B
216
Tam, C., & Burton, D. (1984a). Sound generated by instability waves of supersonic
flows. part 1. two-dimensional mixing layers. Journal of Fluid Mechanics, 138 ,
249272.
Tam, C., & Burton, D. (1984b). Sound generated by instability waves of supersonic
flows. part 2. axisymmetric jets. Journal of Fluid Mechanics, 138 (1), 273295.
Tam, C., Golebiowski, M., & Seiner, J. M. (1996). On the two components of
turbulent mixing noise from supersonic jets. AIAA Paper , (96-1716).
Tam, C., & Morris, P. (1980). The radiation of sound by the instability waves of
a compressible plane turbulent shear layer. Journal of Fluid Mechanics, 98 (2),
349381.
Tam, C., Viswanathan, K., Ahuja, K., & Panda, J. (2008). The sources of jet noise:
experimental evidence. Journal of Fluid Mechanics, 615 (1), 253292.
Tam, C., & Webb, J. (1993). Dispersion-relation-preserving finite difference schemes
for computational acoustics. Journal of Computational Physics, 107 (2), 262281.
Tanna, H. K. (1977). An experimental study of jet noise Part I: Turbulent mixing
noise. Journal of Sound and Vibration, 50 (3), 405428.
The CGAL Project (2010). CGAL User and Reference Manual . CGAL Editorial
Board, 3.7 ed. Available online at http://www.cgal.org.
Thompson, K. (1987). Time dependent boundary conditions for hyperbolic systems.
Journal of Computational Physics, 68 (1), 124.
Uzun, A. (2003). 3-D Large Eddy Simulation for Jet Aeroaocoustics. PhD dissertation, School of Aeronautics and Astronautics, Purdue University.
FIGURES OF APPENDIX B
217
Van Oijen, J., & De Goey, L. (2000). Modelling of premixed laminar flames using
flamelet-generated manifolds. Combustion Science and Technology, 161 (1), 113
137.
Vichnevetsky, R., & Bowles, J. (1982). Fourier analysis of numerical approximations
of hyperbolic equations. SIAM Studies in Applied Mechanics. SIAM.
Visbal, M., & Gaitonde, D. (2002). On the use of higher-order finite-difference
schemes on curvilinear and deforming meshes. Journal of Computational Physics,
181 (1), 155185.
Viswanathan, K. (2002). Analysis of the two similarity components of turbulent
mixing noise. AIAA Journal , 40 , 17351744.
Weasoky, H. L. (1998). Environment Goals: Emissions and Noise Aeronautics &
Space Transportation Technology: Three Pillars for Success. In The 3rd Workshop
on NASAas Environmental Compatibility Research.
URL http://www.aeronautics.nasa.gov/events/encompat/ws3rdmpg.pdf
Wells, V. L., & Han, A. Y. (1995). Acoustics of a moving source in a moving medium
with application to propeller noise. Journal of Sound and Vibration, 184 (4), 651
663.
Williams, F. A. (1985). Combustion theory. Perseus Books.
Wu, X. (2005). Mach wave radiation of nonlinearly evolving supersonic instability
modes in shear layers. Journal of Fluid Mechanics, 523 , 121159.
Zaman, K. (1986). Flow field and near and far sound field of a subsonic jet. Journal
of Sound and Vibration, 106 (1), 116.
FIGURES OF APPENDIX B
Zaman,
K. (1998).
218