Értekezés Az Élhang Témakörében

Budapest University of Technology and Economics
Department of Hydrodynamic Systems

Flow and acoustics of the edge tone conguration
Written by: Istv an Vaik
June 8, 2013
Supervised by: Gy orgy Pa al, Ph.D.
Ph.D. dissertation
Abstract
In this Ph.D. dissertation the well-known low Mach number edge tone conguration is investigated. As
described in Chapter 1 the edge tone has been the subject of research for more than one hundred years,
still the phenomena are not fully understood.
In Chapter 2 the ow of the edge tone is investigated by experimental and numerical means. Beside
several minor phenomena (e.g. the frequency drop of the rst stage in the multi stage coexistence mode)
the result of a detailed parametric study to explore the Reynolds number and dimensionless nozzle-to-
wedge distance dependence of the Strouhal number in the case of top hat and parabolic edge tones is
presented. Also the hysteresis, the mode switching and the jumps between the stages of the edge tone
are studied. Moreover, it is also shown that the phase of the jet disturbance between the nozzle and the
wedge does not vary linearly with the distance from the nozzle, thus the convection velocity of the jet
disturbance is not constant, as it is usually assumed in the theoretical models.
The edge tone is a planar ow that under certain circumstances generates an audible tonal sound
with three-dimensional dipole characteristics. To numerically calculate its sound production, a CFD
simulation has to be coupled to an acoustic simulation. Until now, in spite of the planar nature of the
ow the calculation of the 3D acoustic eld of a planar ow was only achievable by carrying out a 3D
CFD simulation that is highly inecient. In Chapter 3 a new method is proposed with which a 2D CFD
simulation can be coupled to a 3D acoustical simulation. Then the presented newly developed method is
applied to investigate the acoustic attributes of the edge tone.
At last, Chapter 4 opens a window to further research possibilities. A well known real-world
appearance of the edge tone, the ue organ pipe is investigated. Since for the subjective perception
of the sound the attack transient is crucial, that is mostly determined by the specialised edge tone
conguration formed by the jet blowing from the foot of the organ pipe and the upper lip (which acts
as a wedge in the system) only the foot model is investigated (i.e. without the resonator). It is shown
that the strongest and most stable edge tone oscillation occurs (and thus the strongest and most stable
sound production can be achieved) if the upper lip is placed exactly in the centreline of the jet. The main
outcome of this chapter is that the ow in the foot and mouth of an organ pipe can be reliably simulated
with a commercial CFD code which is of great value in organ pipe research.
i
ii
Kivonat
Ebben a Ph.D. disszertacioban a jol ismert alacsony Mach szam u elhang konguraciot vizsgalom. Amint
azt az 1. fejezetben bemutatom, az elhangot mar tobb mint szaz eve vizsgaljak, megis jelensegeit meg
mindig nem teljesen ertett uk meg.
A 2. fejezeteben az elhang aramlasi jelensegeit vizsgalom kserleti es numerikus eszkozokkel. Szamos
kisebb jelenseg (mint peldaul az elso modus frekvenciajanak lecsokkenese kevert modusok eseten) mellett
bemutatom az elhang Strouhal szamanak Reynolds szamtol es dimenziotlan f uvoka-ek tavolsagtol valo
f uggeset feltaro reszletes parametertanulmany eredmenyet. Az elhang modusaiban bekovetkezo hisztere-
zist, modusugrast es modus kapcsolasokat is tanulmanyozom. Tovabba megmutatom, hogy a szabadsugar
zavarasanak fazisa nemlinearis modon valtozik a f uvoka es az ek kozott, tehat a zavaras terjedesi sebessege
nem allando, mint ahogy azt altalaban feltetelezik a jelenseget lero elmeleti modellek.
Az elhang egy skaramlas ami bizonyos kor ulmenyek kozott haromdimenzios dipolus karakterisztikaj u
hallhato hangot hoz letre. Az elhang altal letrehozott hang numerikus kiszamtasahoz egy CFD szimulaciot
kell egy akusztikai szimulacioval osszekapcsolni. Eddig ehhez annak ellenere kellett 3D CFD szimulaciot
vegezni, hogy az aramlas skaramlas jelleg u, ami nagy mertekben megnoveli a szamtas eroforrasigenyet.
A 3. fejezetben egy uj modszert javaslok, amivel 2D CFD szimulaciot lehet 3D akusztikai szimulacioval
kapcsolni. Majd a bemutatott modszerrel vizsgalom az elhang akusztikai tulajdonsagait.
Vegezet ul a 4. fejezet uj lehetosegekre mutat ra a tovabbi kutatasok teren. Egyik jol ismert valos
eletbeli megvalosulasa az elhang jelensegnek az orgonaspban kialakulo aramlas. Mivel a szubjektv
hangerzekeles szempontjabol a kezdeti tranziens nagy jelentoseggel br, amit leginkabb az orgonasp
lababol kilepo szabadsugar es a felso ajak altal letrehozott specialis elhang konguracio hataroz meg,
ezert csak az orgonasp labanak modelljet vizsgalom (azaz rezonator nelk ul, csak az orgonasp labanak
es a felso ajaknak kornyezetet). Megmutatom, hogy a legerosebb es legstabilabb elhang konguraciot, es
ezaltal a legerosebb es legstabilabb hangkepzest akkor erheto el, ha a felso ajkat pontosan a szabadsugar
kozepvonalaba helyezz uk. A fejezet legfobb, az orgonakutatas szamara nagy jelentoseg u eredmenye, hogy
megmutatja, hogy kereskedelmi CFD szoftverrel megbzhatoan szimulalhato az orgonasp labanak es
szajreszenek kornyeken kialakulo aramlas.
iii
iv
Acknowledgements
I would like to express my deep and sincere gratitude to my supervisor, Gorgy Paal. His help, stimulating
suggestions and encouragement were invaluable for me during my research.
I am also indebted to Gisbert Stoyan, the supervisor of my MSc Thesis at the Eotvos Lorand
University. He was not only supervising my MSc thesis, but he was also the one who arose my interest
in numerical methods. I am really grateful for the consultations with him during my Ph.D. research.
The acoustic part of my dissertation would not have been possible without the acoustic solver (part
of CFS++) developed by Manfred Kaltenbacher and his group. Special thanks must go to Simon Trieben-
bacher who always had enough patience to solve the emerging problems I encountered during the coupled
simulations. I also would like to express my thanks to the sta at Erlangen (Stefan Becker, Irfan Ali and
Max Escobar) for their cooperative work.
This dissertation would not have been possible without my family. My wife, Sarolt and our sons, Vince
and Lorinc gave me encouragement day-by-day, and always cheered me up when I was a bit annoyed. I
also received indescribably much help from my parents who heartened me to start and nish the Ph.D.
study.
Last but not least I have to thank God for giving me always what I need, would it be strength,
courage, patience or wisdom. AMDG
v
vi
Nomenclature
Lists of abbreviations and symbols (with their SI units) used throughout the dissertation are collected
in the following pages. Nomenclatures that were only used during the linear algebraic deduction for the
necessary time step (pages 2023) are collected in a separate group.
Abbreviations
2D two-dimensional
3D three-dimensional
avr average (mean) value
CAA Computational AeroAcoustic
CFD Computational Fluid Dynamics
cmv cubic mean value
DES Detached Eddy Simulation
DNS Direct Numerical Solution
est.rel.err. estimated relative error
FFT Fast Fourier Transformation
LES Large Eddy Simulation
PML Perfectly Matched Layer
qmv quadratic mean value
rms root mean square
SAS Scale Adaptive Simulation
Nomenclature used only in pages 2023
j
see eq. (2.12)
j
see page 21
vii
A see eq. (2.12)
B
j
see eq. (2.12)
e
j
accumulated global error of the Second Order Backward Euler method after the j
th
time
step
f (t, y (t)) right-hand side of the ordinary dierential equation used in the time step study (eq. 2.4)
g
j
local error of the Second Order Backward Euler method in the j
th
time step
I identity matrix
L
f
Lipschitz constant of f (t, y (t))
N index of the last time step
S the matrix with which SAS
1
is the Jordan normal form of A
v amplitude of the velocity oscillation
v
j
see eq. (2.12)
y (t) solution of the initial value problem of the ordinary dierential equation (2.4)
y
0
initial value condition of equation (2.4)
y
j
numerical approximation of y (t
j
)
Symbols
/2 number of element layers in the source region of the acoustic mesh (half space only because
of the symmetry condition) -
width of the jet m
vc
width of the jet in the vena contracta point (in Section 4.2) m
phase lag in the feedback loop (eq. (1.17)) -
scaling factor during the 2D CFD 3D CAA simulations, when the sources are extruded,
= w
acou
/w
CFD
-
c
scaling factor during the 2D CFD 3D CAA simulations, when the sources are concentrated
on the symmetry plane,
c
= W/w
CFD
-
acoustical wavelength m
wavelength of jet disturbance m
dynamic viscosity
kg
/ms
kinematic viscosity
m
2
/s
viii
angular frequency
rad
/s
angle between the centreline of the jet and the y axis (in Section 4.2) rad
Phase delay relative to the reference point rad

/2 -
density of the uid
kg
/m
3
density perturbation
kg
/m
3
0
density of the uid at a reference state
kg
/m
3
time step s
ij
elements of the stress tensor Pa
angle between the position vector of the observation point and the y axis rad
A area of the cross section of the nozzle m
2
a
0
speed of sound
m
/s
b half-thickness of the jet m
c coecient of St (Re) in eq. (2.43) -
c
1
, c
2
, c
3
coecients of St (Re, h/) in eq. (1.16) or (2.48) -
d coecient of St (h/) in eq. (2.45) -
e
f
error of the frequency measurement Hz
e
error of the jet width measurment m

e
m
error of the mass ow rate measurement
kg
/s
e
A
error of the nozzle cross section measurment m
2
e
h/
error of the dimensionless nozzle-to-wedge distance measurement -
e
h
error of the nozzle-to-wedge distance measurement m
e
Re
error of the Reynolds number measurement -
e
St
error of the Strouhal number measurement -
F force acting on the wedge N
F in Lighthills analogy (eq. (3.2)), external force density
N
/m
3
f frequency of oscillation Hz
ix
f Frequency resolution of a spectrum Hz
F
y
amplitude of the force acting on the rst 7.2 mm of the wedge N
F
x
the x component of the force acting on the upper lip N
G strength of dipole sound source N

G(t) instantaneous strength of a dipole sound source (eq. (3.8)) N
h nozzle-to-wedge distance m
h the length of an element between the nozzle and the wedge m
h/ Dimensionless nozzle-to-wedge distance -
H
1
, H
2
, V sizes of the CFD domain of the 2D edge tone simulation (Figure 2.1) m
k exponent of h or h/ in the h dependence of f (eq. (1.15)) or h/ dependence of St
(eq. (1.16)) -
L cut-up length (organ pipe foot model) m
l number of layers in that the acoustical sources are extruded -
m order of accuracy -
m mass ow rate
kg
/s
m
in
mass ow rate that is injected into the system through the nozzle
kg
/s
M
z
torque Nm
Ma Mach number, Ma =
u
a
0
-
n ordinal number of the stage -
n
v
number of vortices passing at a xed spatial point next to the wedge -
p pressure Pa
amplitude of the acoustic pressure uctuation Pa

p
pressure perturbation Pa
p
0
pressure of the uid at a reference state Pa
Q mass source
kg
/m
3
s
q renement ration between meshes during the mesh study (Secton 2.1.1) -
R correlation coecient b
r distance from the dipole sound source m
x
r distance from the tip of the wedge m
R
2
coecient of determination -
Re Reynolds number, Re =
u
-
S dimensionless oscillation frequency as dened by Crighton (eq. (1.8)), S =
b
u
-
S
el
surface of an element in the mesh m
2
St Strouhal number, St =
f
u
-
St
coecient of St (h/) in eq. (2.45) -

St
coecient of St (Re) in eq. (2.43) -

T in eq. (2.34), temperature

C
T period time s
t time s
T size of the window during the sliding window Fourier transformation s
t time step of the signal s
t
j
time value of the j
th
time step, t
j
= j s
T
o
oscillating part of the simulation s
T
S
duration of simulation in simulated time s
t
s
elapsed time during the mass ow rate sensor calibration s
T
t
transient part of the simulation before the quasi-steady oscillation sets in s
t
v
elapsed time during the frequency measurement with the vortex counting method s
t
w
elapsed time between two subsequent photographs s
T
ij
elements of Lighthills stress tensor Pa
u mean exit velocity of the jet
m
/s
u
c
centreline velocity of the jet
m
/s
u
conv
convection velocity of jet disturbance
m
/s
u
vc
jet velocity in the vena contracta point (in Section 4.2)
m
/s
v velocity vector, v = [v
1
; v
2
; v
3
] [
m
/s;
m
/s;
m
/s]
v
1
, v
2
, v
3
velocity component in the x, y and z direction, respectively
m
/s
xi
V
el
volume of an element in the mesh m
3
W height of the ow m
w
acou
height of an element in the source region of the acoustic mesh m
w
CFD
thickness of the element layer in the 2D CFD simulation m
x the distance that the disturbance travels between two subsequent photographs
(in Chapter 2) m
x upper lip oset measured from the y axis (in Chapter 4) m
x
F
mean value of x
F
m
x,y,z Cartesian coordinate directions
x
0
the x position of the jet centreling at y = 0 (in Section 4.2) m
x
F
instantaneous point of force action m
x
r
the x position where the jet velocity is r times of the maximum (in Section 4.2) m
SPL Sound Pressure Level, SPL = 20 lg
_
/20 Pa
_
dB
xii
Contents
Abstract i
Kivonat iii
Acknowledgements v
Nomenclature vii
1 Introduction 1
1.1 What is the edge tone? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Literature overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.3 Numerical simulations and other research . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Frequency and phase characteristics of the edge tone . . . . . . . . . . . . . . . . . . . . . 10
1.4 Why the edge tone? Aims of the work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 The ow of the edge tone 15
2.1 The CFD setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 Two-dimensional simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.2 Three-dimensional simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2 The experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.1 Experimental system and instrumentation . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.2 Derived quantities and their error estimation . . . . . . . . . . . . . . . . . . . . . 28
2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.1 Varying the Reynolds number at a xed geometric conguration . . . . . . . . . . 35
2.3.2 Varying the nozzle-to-wedge distance . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.3 St (Re; h/) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.4 Stage jumps and mode switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.3.5 Miscellaneous CFD results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.4.1 Theses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
xiii
2.4.2 Tezisek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3 Acoustics of the edge tone 67
3.1 Hybrid Computational AeroAcoustic methods . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.1.1 Lighthills analogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.2 Basic ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.2.1 Computational algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3 Denition of test setups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3.1 Computational domains for the acoustic simulation . . . . . . . . . . . . . . . . . . 73
3.3.2 2D CFD with 3D acoustics - the treatment of sources . . . . . . . . . . . . . . . . 75
3.3.3 3D CFD with 3D acoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.4.1 2D-2D: acoustic mesh study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.4.2 3D-3D computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.4.3 2D-3D: source extrusion with dierent numbers of layers . . . . . . . . . . . . . . . 77
3.4.4 2D-3D: comparison with theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.4.5 2D-3D: dierent Reynolds numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4.6 The computational costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.5.1 Theses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.5.2 Tezisek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4 CFD simulations on an organ pipe foot model 87
4.1 Geometrical conguration, boundary conditions, mesh and solver settings . . . . . . . . . 87
4.2 Free jet simulation: velocity proles and jet centreline . . . . . . . . . . . . . . . . . . . . 91
4.3 Edge tone simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.4.1 Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.4.2 Tezis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Bibliography 101
Own publications 105
Appendix 107
xiv
Chapter 1
Introduction
1.1 What is the edge tone?
The edge tone is one of the simplest aero-acoustic ow congurations. It consists of a planar free jet that
impinges on a wedge-shaped object (traditionally called the edge). The main parameters of an edge tone
conguration are the mean exit velocity of the jet (u), the width of the jet () and the nozzle-to-wedge
distance (h) (Figure 1.1). Secondary parameters may also inuence the ow, such as the velocity prole
of the jet (top hat and parabolic proles are the most common ones), the oset of the wedge from the jet
center line, the shape of the nozzle or the angle of the wedge. Despite its geometric simplicity the edge
tone displays a remarkably complex behaviour. Under certain circumstances a self-sustained oscillation
evolves with a stable oscillation frequency. The oscillating jet creates an oscillating force on the wedge,
that generates a dipole sound source, that under certain circumstances creates an audible tone.
h
u
x
y
Figure 1.1: Snapshot from a CFD simulation of a rst stage edge tone ow with the main parameters of
the edge tone conguration ( width of the slit on the nozzle; h nozzle-to-wedge distance; u mean
exit velocity of the jet)
The oscillating jet can take dierent shapes, these are called the stages of the edge tone. Their ordinal
number corresponds roughly to the number of half waves between the nozzle and the wedge. Figure 1.1
shows a snapshot from the result of a computational uid dynamics (CFD) simulation of a rst stage
edge tone ow. The ow is visualised by virtual smoke introduced in the central part of the nozzle.
Figure 1.2(a) shows what happens with the oscillation frequency when the velocity is varied in a
xed geometrical conguration (at constant and h values). At low velocities the wedge cuts the jet in
half and a steady ow is formed. Increasing the velocity above a certain threshold velocity that value
is of course dependent on the geometric conguration the rst stage of the edge tone sets in (position
A). Further increasing the jet velocity the second stage comes into being with a sudden jump in the
frequency to a higher value (position B). Usually the rst stage still coexists with the new, second stage,
thus a multi-stage operation mode can be observed, but the second stage can be present purely as well.
1
CHAPTER 1. INTRODUCTION
u
f
A
A'
B'
B
Stage I
Stage II
Stage III
C
C'
(a) mean exit velocity of the jet
h
f
A
A'
B'
B
Stage I Stage II Stage III
C
C'
(b) nozzle-to-wedge distance
Figure 1.2: Characteristics of frequency variation as a function of the
At higher velocities the third stage of the edge tone is formed (position C) again with a sudden jump
in the frequency to a higher value either with some of the lower stages coexisting, or purely. Further
increasing the velocity, even higher stages may evolve, with a jump in the frequency to a higher value at
the onset of each new stage. Similar behaviour can be observed when the velocity of the jet is decreased.
The frequency of the oscillation decreases, and at a point the highest stage disappears (at positions C
and B the third and the second stage disappears, respectively) with a sudden drop in the frequency and
at last at low jet speeds the rst stage of the edge tone disappears and a steady ow is formed (position
A ). It can be that the velocity value at the point where a stage disappears during the decrease of the
jet velocity (position C, B or A ) may dier from the velocity value at the point where this stage rst
appeared when the velocity was increased (position C, B or A), so hysteresis may occur.
On the other hand, when the nozzle-to-wedge distance is varied while the velocity of the jet is kept
constant (Figure 1.2(b)), the following can be observed. At low distances no oscillation occurs. Increasing
the distance, at a certain lower limiting value the rst stage of the edge tone forms (position A). Further
increasing the distance, the frequency of the oscillation decreases. At a point the second stage sets in
with a sudden jump in the frequency to a higher value (position B). Further increasing the distance, the
frequency again decreases until the next stage forms with the sudden jump in the frequency again to a
higher value (position C). Now, if the nozzle-to-wedge-distance is decreased from a higher value, then
the frequency of oscillation increases until a certain position where the jet jumps back to a lower stage
where the frequency is also lower (position C and B ) and at last the oscillation disappears completely
(position A). Just as in the case when the velocity is varied hysteresis may occur, it can be that the
jump between the stages forth and back are at dierent positions (A = A
or B = B
or C = C
).
The following dimensionless numbers will be used throughout my dissertation:
- Reynolds number based on the mean exit velocity of the jet and the width of the jet will be used
as the dimensionless jet velocity: Re =
u
- Strouhal number based on the frequency of oscillation (f), the width of the jet and the mean
exit velocity of the jet will be used as the dimensionless oscillation frequency: St =
f
u
- h/ will be used as the dimensionless nozzle-to-wedge distance
Because of scaling laws, two edge tone congurations with dierent jet velocities and geometric sizes,
2
1.2. LITERATURE OVERVIEW
but at the same Reynolds number and h/ dimensionless nozzle-to-wedge distances produce oscillations
with dierent frequencies but with the same Strouhal numbers, therefore when comparing results from
dierent sources (theoretical and/or experimental and/or numerical) comparison of the Strouhal numbers
at same Reynolds numbers and at same h/ values will be carried out.
1.2 Literature overview
Brown [1, 2] gives a detailed overview on the early research on the edge tone phenomenon from the rst
80 years after it was rst noted by Sondhaus in 1854. Several researchers tried to explain theoretically
the mechanism of the edge tone production, others made extensive experimental investigations in the
eld. Without being exhaustive I shall give a short introduction to the most important studies. At rst,
I shall review the theories (in chronological order), then the experimental and numerical studies. There
are a couple of other experimental works carried out to investigate some specic aspects of the edge tone.
Some of them will be cited in the corresponding chapters.
1.2.1 Theories
Curle [3] published his purely hydrodynamic, vortex theory explaining the edge tone in 1953. He claims
that vortices of opposite circulation are produced at the nozzle (embryo vortex) and at the tip of the
wedge (secondary vortex) at the same time. The formation of the secondary vortex takes place when
the transverse velocity at the tip of the wedge is maximal, that occurs halfway between the alternate
vortices below and above the wedge. Thus the relationship between the nozzle-to-wedge distance and
the wavelength is: h =
_
n +
1
4
_
, where n denotes the ordinal number of the stage and the distance
between two consecutive vortices on the same side of the jet. Independently from this result he deduced a
semi-empirical formula for the velocity with which the vortices moves (u
conv
) in the case when h/ > 10:
u
conv
u
=
1
2
_
1

30
_
, (1.1)
thus the frequency of oscillation of the n
th
stage is:
f =
1
2
u
_
n +
1
4
h

1
30
_
(1.2)
He also emphasises that if Savics [4] result of
u
conv
u
= 1.024
_
is used instead of his semi-empirical

formula (1.1), then the frequency of oscillation becomes:
f =
u
1
2
h
3
2
c, (1.3)
where c = 1.43, 3.46, 6.00 and 8.98 for the rst four stages.
He suggests that n has a value such that =
h
n+
1
4
is near the wavelength for which an edgeless jet
is most sensitive, and after stage jumps gets closer to this value.
In 1954 Nyborg published his dynamic theory explaining the edge tone phenomena [5]. His theory assumes
that due to a kind of sources of hydrodynamic origin at the wedge transverse forces act on each particle
3
of the jet as it travels toward the wedge. He dealt with only the centreline of the jet, and supposed that
the vertical acceleration acting on any particle travelling towards the wedge depends only on two factors:
its actual horizontal distance from the wedge and the actual position of the jet displacement at the wedge.
With his theory, Nyborg was able to describe the shape of the centreline of the jet. Although he was not
able to determine the frequency of oscillation, he was able to determine the ratio of the frequencies of
the dierent stages (1 : 2.44 : 3.86 : 5.29 for the rst four stages). He also indicated that as h increases
higher modes becomes possible. He found the lower limit of the n
th
stage to be h/ > 2n, however the
theory is not capable to predict the position where the jet jumps from one stage to another. Because
of his dynamic theory only deals with the centreline of the jet it fails to predict that the frequency of
oscillation depends on the width of the jet.
Powell rst published his feedback loop theory in 1953 and then later, in 1961 he gave a detailed discussion
of it [6, 7]. He suggests that an innitesimal excitation at the nozzle exit grows along the jet via an
instability displacement wave. This distortion generates an oscillating force on the wedge that creates
a dipole sound source, which then closes the loop by exciting the jet at the nozzle exit. Despite of
the feedback loop being based on the dipole sound source, he stated that the feedback loop is purely
hydrodynamic, the sound radiation itself does not play an essential role in the mechanism. From this
feedback loop a phase criterion can be deduced inducing the oscillation frequency of the n
th
stage to be:
f =
_
n +
1
4
_
u
conv
h
, (1.4)
where u
conv
is the convection velocity of the disturbances. He emphasises that the frequencies of the
stages are not to bear ratios of
5
/4 :
9
/4 :
13
/4, because sinuosities of dierent wavelengths have dierent
convection velocities.
For the stages jumps he gave the following explanation: For any given u
conv
and h values a certain f
frequency can be calculated for each of the stages from equation (1.4). For low nozzle-to-wedge distances
these frequencies will be well above the region where the edge-less jet is sensitive to acoustic excitations
even for the rst stage, thus the edge tone phenomenon will not occur. As the nozzle-to-wedge distance is
increased to a point the frequency of the rst stage reaches the region of sensitivity, and the rst stage of
the edge tone sets in. Further increasing the nozzle-to-wedge distance the oscillation frequency decreases
and at a point it reaches the lower boundary of the sensitivity region and the rst stage disappears. At
this point the frequency of the second stage is already in the sensitivity region, thus by this point the
edge jumps to the second stage. Similar explanation can be given for the onset of the stages in the case
of a xed geometric conguration with varying jet velocity.
Powell notes that it can happen that the new stage is superimposed on the old stage, and the two
stages coexist. He also notes that it is more likely than not that the jumps between the stages will be
hysteretic.
He experimentally showed the dipole characteristics of the edge tone sound eld, and that the am-
plitude of the acoustic pressure is proportional to the third power of the jet velocity.
Holger et al. developed a vortex street theory in 1977 [8]. Their assumptions were that the wavelength
of the jet disturbance, the width of the vortex street and the propagation velocity of the vortices are
4
constant. While Curle based his vortex theory on the formation of secondary vortices at the edge, their
analysis does not depend on secondary vortices, and used an entirely dierent formula for describing the
oscillation frequency. Contrary to Powell, they assumed that the vortex street is fully formed by the time
it interacts with the edge. They found the frequency of oscillation to be:
f = 0.925
_
h
_1
2
u(n +
n
)
3
2
h
, (1.5)
where = 0.4, 0.35 and 0.5 for the rst, the second and the third stages, respectively.
In 1980 Holger et al. [9] extended their theory and gave an approximation on the vertical force acting
on the wedge. From this, they were able to calculate the acoustic pressure at an arbitrary point in the
far eld with Lighthills equation. They found that the integration length on the wedge should be chosen
as 2, and in this case the calculated force is
F 1.08Wu
2
, (1.6)
where is the density of the uid and W is the height of the ow. From this, the amplitude of the acoustic
pressure at a distance of r in the direction of maximum radiation is
|p
a
| =
fF
2ra
0
0.5
_
h
_3
2
(n +
n
)
3
2
u
3
W
ra
0
, (1.7)
where a
0
is the speed of sound. They also noticed that the vortex pair nearest to the tip of the wedge gives
the most signicant part of the force, and the instantaneous force has its maximum when the distance
between the tip of the wedge and the rst vortex downstream of it is 0.1.
In 1992 Crighton [10] created a linear analytical model to predict the frequency characteristics of the
edge tone oscillation. He dealt with a top hat jet impinging on a plate placed parallel in the center of the
jet. He assumed inviscid ow with vortex-sheet shear layers, and solved the problem asymptotically by
Wiener-Hopf methods. He found that the dimensionless oscillation frequency S =
b
u
, where = 2f
is the angular frequency and b =

2
is the jet half-thickness is
S =
_
b
h
_3
2
_
4
_
n
3
8
__3
2
, (1.8)
and the h/ ratio is
_
n
3
8
_
. He found that his Strouhal number (dened as at the beginning of the
introduction) is much larger than the values reported by Holger et al. For the relative convection velocity
of the disturbance he used the
u
conv
u
2S
1
3
formula, while Holger et al. used
u
conv
u
0.645S
1
3
. Without
essentially nding the cause of this large dierence, he concludes that his formula would give a better
prediction if
u
conv
u

2S
1
3
1+
4
3
S
1
3
would be used, but (I cite, [10] p. 386) all such expressions would lead to the
same behaviour, namely preservation of essentially the form of equation (1.8), but with (4)
3
2
replaced
by a smaller coecient.
In 1996 and 1998 Kwon [11, 12] presented a theoretical model in which the jet-edge interaction was
modelled by an array of dipoles on the edge. By assuming the jet to be sinusoidally oscillating and
5
the convection velocity of the disturbances to be constant, his model can estimate the surface pressure
distribution on the wedge from that an array of acoustic dipoles on the wedge can be deduced. He found
that the peak value of the spatial pressure distribution on the wedge can be found approximately quarter
wavelength downstream from the tip of the wedge. He found that the phase criterion is:
h
+
h
= n
1
4
,
where is the wavelength of the upstream propagating disturbance (the acoustic eld of the dipole
sources). Thus, he claims that the point of the wedge surface where the pressure has its maximum
(quarter wavelength downstream from the tip) is the position of the eective acoustical source. He also
found that the convection velocity of the disturbance on the jet is approximately 60 % of the mean exit
velocity of the jet, and thus the Strouhal number of the oscillation can be approximated as:
St =
h
n
1
4
1.667 + Ma
, (1.9)
where Ma is the Mach number of the mean jet velocity (Ma =
u
/a
0
).
1.2.2 Experiments
In 1937 Brown [1, 2] investigated an edge tone setup of a = 1 mm wide, top hat jet with a wedge with
an angle of 20
experimentally. He found that the whole edge tone phenomenon occurs at frequencies for
that the edgeless jet is sensitive to sound and the frequency of the stages depends on the exit velocity of
the jet and the nozzle-to-wedge distance through the following formula:
f = 0.466 j (u 40)
_
1
h
0.07
_
, (1.10)
where u and h are measured in
cm
s
and in cm, respectively, and j = 1, 2.3, 3.8 and 5.4 for the rst,
the second, the third and the fourth stage, respectively. He claimed that the deviation between his
measurement and his formula for jet velocities u = 120 2000
cm
s
(that is in nondimensional values
Re = 75 1300) and frequencies f = 20 5000 Hz was maximum 6%. He found the limits of h to be
0.31 cm and 6 cm, so the nondimensional nozzle to wedge distance was between 3.1 and 60.
He found that for higher stages the rst stage could also be coexisting, and in this case the frequency
of the rst stage is about 7 % lower than the frequencies predicted by his formula. As the formula can
have as much as 6 % deviation from the measured values he concluded that this drop in the frequency
practically can be neglected.
In the case of higher stages he measured the wavelength of the jet disturbance as the distance between
two successive vortices on the same side of the stream (from the photographs he took of the visualised
ow), while for Stage I. he assumed that = h. With this and the measured oscillation frequency he
calculated the convection velocity of the vortices as u
conv
= f that resulted in values of about 40 % of
the jet exit velocity (
u
conv
u
0.4).
He also investigated how sound production eects the edge tone, and concluded that in some cases
acoustical excitation can control the stages of the edge tone.
In 1942 Jones [13] investigated an edge tone conguration with a 0.8 mm wide top-hat jet, at velocities up
to 50
m
s
, and with nozzle-to-wedge distances between 5 and 25 mm. In his experiments the wedge angle
was 25
. He reported two types of the edge tone: In the rst type that occurs at lower jet velocities
6
he found three stages, between which jumps occurs in the frequency of oscillation. In the second type of
the edge tone that occurs at higher jet velocities (above 37
m
s
) the jet is probably turbulent and
no jumps occurs if the parameters are varied, but the frequency changes continuously. He found that the
frequencies of the three stages of the rst type edge tone oscillation and also of that of the second type
can be described as:
f = j
u
h
k
, (1.11)
where u is measured in
cm
s
, h in mm. The values of j and k for the three stages of the rst type and for
the second type are: j = 3.9, 11.8, 24 and 6.8; k = 1, 1.14, 1.22 and 1.43, respectively.
In 1952 Nyborg et al. [14] made an extensive experimental research in mapping the stage boundaries in
the h q plane (where q is the volumetric ow rate of the air, thus in a given geometric conguration
proportional to the velocity of the jet) of small edge tones ( = 0.25 1.02 mm) with high frequency
oscillations (f up to 200 kHz). They used parabolic jets with dierent widths and several (in some cases
asymmetric) wedges that sometimes were placed with a transversal oset from the center of the jet.
They compared their measured frequencies to a somewhat simplied form of Browns semi-empirical
formula (equation (1.10)), namely:
f = 0.466 j u/h, (1.12)
and found that, the measured values agree well with the simplied formula at f 2.5 kHz but are a bit
lower than the formula below 2.5 kHz and are a bit higher than the formula above 2.5 kHz.
They found that the regions of the stages in the h q plane can overlap, indicating that hysteresis
may occur when changing the mean exit velocity of the jet or the nozzle-to-wedge distance, and the
overlapping regions are independent of the wedge angle if it is less than 40
.
They also made measurements on the directivity of the sound emitted by the edge tone, and found
that at f < 10 kHz a cosine law (dipole radiation pattern) ts their experimental results fairly well. Above
10 kHz they found a directivity pattern with double maxima that they were not able to explain.
Brackenridge and Nyborg [15, 16] reported experiments on top hat underwater edge tones with three
dierent jet widths ( = 0.51 mm, 1.02 mm and 2.04 mm) and a wedge with an angle of 28
. They found
that in the case of = 1.02 mm the frequencies of the stages can be described as:
f = 0.63 j (u 6)
_
1
h
0.23
_
. (1.13)
They noted that the value of 0.63 seems to be dependent of the jet width.
They found that at the formation of the second stage the rst one remains, and the two coexist. They
also found nonlinear interaction of the stages: sometimes frequencies equal to the sum and the dierence
of the frequencies of the two stages are also present in the spectrum. They also investigated the eect
when the edge has a blade attached to its tip that can vibrate at its own frequency, and they found that
this vibrating blade enforces a rst stage oscillation at the blade frequency even at higher jet velocities.
In 2004 Bamberger et al. [17] made an experimental and computational investigation on a parabolic edge
tone conguration with an asymmetric wedge. The Reynolds number range in their investigation was
7
approximately Re = 100 900. In their experiments they used two dierent nozzles (with = 0.5 mm
and 1 mm width) and nozzle-to-wedge distances between 2.2 mm and 8.7 mm. They concentrated on the
rst stage oscillation only, and found that the oscillation frequency in this stage is proportional to the
maximum jet velocity and inversely proportional to the nozzle-to-wedge distance:
f = c
d
u
max
h
, (1.14)
where as they claim c
d
depends on the width of the nozzle. From the experiments they concluded that
for the = 0.5 mm wide nozzle c
0.5
= 0.339 0.02 and for the = 1 mm wide nozzle c
1
= 0.344 0.02,
which in my opinion diers within the uncertainty of the values. However, from the CFD simulations
they obtained somewhat (about 13 15 %) lower values: c
0.5
= 0.29 0.04 and c
1
= 0.3 0.03.
1.2.3 Numerical simulations and other research
Numerical simulations have been carried out on the edge tone already in the last century. Although they
were able to catch certain typical characteristics of the edge tone ow, but they were still in early states,
and no detailed parameter study was carried out.
Ohring [18] in 1986 carried out CFD simulation on the edge tone. He used a nite dierence method
for solving the vorticity/stream-function formulation of the Navier-Stokes equations. He managed to
reproduce the basic features of the edge tone (oscillating ow, two stages) and certain results reported by
Lucas and Rockwell who did experimental research with a underwater parabolic edge tone conguration
in 1984 [19]. He carried out simulations at three dierent jet velocities (at dimensionless velocity values
of Re = 250, 450 and 650) and found dierent edge tone stages when Re = 450 was computed from the
result of the Re = 250 simulation with increasing the velocity or when Re = 450 was reached from the
result of the Re = 650 simulation with decreasing the velocity.
In 1994 Dougherty et al. [20] submitted a report to NASA about the numerical simulations of the edge
tone. They managed to reproduce Browns experimental data with a nite volume method based Navier-
Stokes solver (USA - Unied Solutions Algorithm) having rst order time and third order spatial accuracy.
They dealt with an edge tone conguration which geometric sizes matched Browns experimental setup.
They had a xed Reynolds number (Re = 1083), and made simulations at dierent nozzle-to-wedge
distances in the range of h/ = 3 16. In this nozzle-to-wedge range they managed to produce the
rst four stages of the edge tone. Although the duration of their simulations (in virtual time) was only
sucient to obtain four or ve periods of the lowest stage, they found it to be adequate for FFT analysis,
and the resulting frequencies of the stages agreed well with Browns experimental results.
In the last decade a couple of research were made about the edge tone that instead of carrying out a full
parameter study was investigating only certain aspects of the phenomenon. In the following I will give a
short overview on these new ideas and trends of the edge tone research in the new millennium.
In 2001 Lin and Rockwell [21] carried out experiments on high speed, turbulent underwater edge
tones. They had a Reynolds number approximately 5500 and varying nozzle to wedge distances up to
h/ = 7.5. They found two clearly identiable modes of instability in their experiments. The rst one is
a large scale, global mode, whose frequency corresponds well to the frequency of the rst stage measured
8
by previous researchers for low Reynolds number, laminar edge tones. The other one is a small scale, local
mode, in that vortical structures develop at the nozzle. The frequency of this mode is much higher
than the large scale, global mode, and the phase shift between the development of the vortices on the
two sides of the jet has no particular numerical value.
In 2003 Fujisawa and Takizawa [22] made a study on how the edge tone could be weakened by feedback
control. In their experiments they used a low speed underwater edge tone conguration (Re = 200,
h/ = 5) that resulted in an oscillation at a frequency of 0.26 Hz. They mounted two control nozzles at
the two sides of the jet next to the nozzle. An image processing system (processing pictures taken of the
visualised ow at 30 Hz) allowed them to precisely control the secondary nozzles to weaken the evolved
edge tone oscillation.
In 2004 Segoun et al [23] made an experimental study on how the geometry of the nozzle aects
the evolved ow. They tested four nozzle types: a nozzle creating a top hat jet (top hat nozzle); a nozzle
creating a parabolic jet (parabolic nozzle); a nozzle creating a parabolic jet with chamfered jet exit
(chamfered nozzle); and a nozzle created a parabolic jet with rounded jet exit (rounded nozzle). They
found that the rounded nozzle does not produce the edge tone phenomenon. Comparing the oscillation
frequencies with the top hat and the parabolic nozzle they found that with the same maximum velocity
values the oscillation frequency is about 50 % higher in the top hat case and also the oscillation sets in
at a lower velocity (by about a factor of 1.5).
In 2005 Devillers and Coutier-Delgosha [24] published their results on the investigation of the inuence
of the gas nature on the edge tone. They investigated both experimentally and numerically parabolic
edge tone congurations with three dierent nozzle-to-wedge distances h/ = 58.3, several jet velocities
from Re = 50 in some cases up to 1000 with three dierent gases (air, CO
2
, Neon), all of them injected
into air. Beside their extensive experimental and numerical work they also discussed a linear analysis of
the instability. From their investigations they concluded, that the density has a signicant eect on the
growth of the vortices on both sides of the edge. When the density of the gas used for the jet is higher
than the density of the environmental medium (CO
2
air,
CO
2
/
air
= 1.52), then the frequency of
oscillation is higher (in this case, by a ratio of St
CO
2
/St
air
= 1.1), and when the density of the gas of
the jet is lower (Neon air,
Neon
/
air
= 0.64), then the frequency is lower (in their case by a ratio of
St
Neon
/St
air
= 0.9).
Tsuchida et al. [25, 26] carried out incompressible three-dimensional (3D) CFD simulations on the
edge tone as a test for their numerical code. They only made simulations at a couple of Reynolds numbers
(Re 195, 325, 350 and 455) at a xed geometry (h/ = 6). They found that the Strouhal numbers of
their simulations is consistent with Browns experimental results.
Nonomura et al. [27, 28] investigated the eect of the Mach number on the edge tone phenomenon
numerically. In their simulations they used jets with Reynolds numbers equal to 208, 416 or 624 at Mach
numbers equal to 0.087, 0.174, 0.261, 0.348 or 0.435 on a top hat edge tone conguration with h/ = 5.
They found that when increasing the Mach number at a xed Reynolds number that was realised by
changing the size of the model the Strouhal number decreases, that agrees well with Powells feedback
loop theory for high speed edge tones [6]. They also investigated the phase lag in the feedback loop (
in equation (1.17)), and found that its almost constant ( between 0.17 and 0.21) in their Reynolds
9
number and Mach number regions. They found that the relative disturbance propagation velocity is
approximately 0.5 0.6.
In the last few years, a couple of authors made research in the eld of the compressible CFD simulations
on the edge tone. This eld is a step forward to the direct noise simulation of the edge tone, but holds
several challenges. For example, the typical boundary conditions of CFD simulations tends to generate
spurious reections of compressibility waves that totally disrupt the acoustical results.
In 2005 Kang and Kim [29] employed a nite dierence-based lattice Boltzmann method for the
direct numerical simulation of the sound of a two-dimensional (2D) parabolic edge tone conguration
with h/ = 6 for a couple of dierent jet velocities. They have found that the oscillation frequencies agrees
acceptably with the ones from the experiments of Bamberger et al. [17]. They succeeded in capturing
very small pressure uctuations resulting from the edge tone oscillation.
In 2010 Gao and Li [30] carried out compressible large eddy simulations (LES) on a high speed
edge tone conguration with Mach numbers of 0.18 and 0.23. They encountered reections from the
boundaries. They claim that these are because that their computational domain is not large enough,
and suggest that for further investigation a perfectly matched layer (PML) boundary condition should
be adopted in their solver. Nonetheless they drew the following conclusions from their simulations. They
found ve frequency peaks in the pressure spectrum, out of which only the third and fth are harmonically
related. Therefore they claimed that they had found four modes of the edge tone of which the second
one is dominant. Except for the frequencies of the third and fth peaks they are close to the frequencies
measured experimentally by Krothapalli and Horne [31] in a similar setup. Although the wavelength
of the sound at these frequencies is much higher than the dimensions of their CFD domain they also
investigated the directivity of the stages at a distance of approximately 0.59
2
measured from the tip of
the wedge, where
2
is the acoustic wavelength of their second mode. They found that the rst mode
is omni-directional while the second one radiates more to the downstream direction. The directivity
patterns they had computed are clearly not like that of a dipole that can be because of the reections
they encountered from the boundaries.
Takahashi et al. [32] published their results on the 2D and 3D compressible LES simulations with
OpenFOAM. In the 2D case they made a parameter study changing the mean exit velocity of the jet
with Reynolds numbers between 320 and 1940 on a xed geometry with h/ = 5 and made a single 3D
simulation at Re = 650 with the same nozzle-to-wedge distance. From the 2D CFD simulations they
obtained oscillation frequencies about 10% higher than those of Brown, but the 3D simulation agreed
well. They also investigated the sound sources computed from the simulated ow.
1.3 Frequency and phase characteristics of the edge tone
As has been shown, the literature is consequent in the proposition that the oscillation frequency is
roughly proportional to the mean exit velocity of the jet and inversely proportional to the k
th
power of
the nozzle-to-wedge distance:
f
u
h
k
. (1.15)
10
1.3. FREQUENCY AND PHASE CHARACTERISTICS OF THE EDGE TONE
Sometimes an additive constant in one or both of the relationships is also present (such as f
u+c
u
h
k
+c
h
).
About the value of the exponent k there has been a long debate. In the early phase of the research
rather k = 1 was favoured (Brown and other researchers before him [1], Curle [3]) later it became generally
accepted that k =
3
/2 (Curle using Savics results [3], Holger et al. [8], Crighton [10]). In 1942 Jones [13]
found a variety of exponents, all between 1 and
3
/2, depending on the stage number. Recent research
(Bamberger et al. [17]) and also the results of my experimental and numerical studies indicate that k = 1
is more correct.
In order to ensure comparability, the discussed frequency formulae were transformed to Strouhal
numbers and doing so it turned out that all of them can be described in the following form:
St
_
Re,
h
_
c
1
c
2
Re
_
_
1
(h/)
k
c
3
_
(1.16)
Table 1.1 shows the value of the coecients for the rst three stages, while Figures 1.3 and 1.4 show
the Strouhal number of the rst stage plotted as a function of the Reynolds number at h/ = 10 and
as a function of h/ at Re = 200, respectively. To avoid the overloading of the gures with curves only
three theoretical (Curles two formulae [3] and Holgers [8] formula) and three experimental (Browns [1],
Jones [13] and Brackenridges [16] formulae) results are plotted. The experiments of Brown and Jones
t acceptably well (for h/ = 10 above Re 150), the experiments of Brackenridge has a mentionable
but still not too high deviance from their results (above Re = 200). The formulae from the theoreti-
cal considerations tend to over-estimate the results of the measurements. It can be seen that for low
Reynolds numbers and/or low nozzle-to-wedge distances the curves separate and the dierence can easily
be more than 100 %. For higher Reynolds numbers or nozzle-to-wedge distances the dierences between
the formulae are somewhat more bounded, but still can reach 25 %.
All of the above mentioned theories can be summarised as the disturbances on the jet that born somewhere
near the nozzle have to travel to the wedge, where they somehow interact with it. As a result of their
interaction a signal is sent to the place where disturbances of the jet are born.
As Powell suggests: the oscillating jet creates an oscillating force on the wedge, that creates a dipole
sound source. The generated sound then excites the jet at low Mach numbers with no time delay and
a new disturbance is born that grows as it travels downstream to the wedge.
The phase relation of this loop can be summarised in the following equation:
h = (n + ) (1.17)
where is the wavelength of the disturbance, n is a whole number corresponding to the stage number,
and is a small number indicating that the eective resonance length of the edge tone system somewhat
diers from h.
There is no agreement in the literature about the value of , it may also depend on the details of the
conguration and on the stage number. The most often occurring value is 0.25 (Curle [3], Powell [7]),
Holger et al. [8] found values between 0.35 and 0.5 depending on the stage, but negative values are also
suggested 0.2 (Nonomura et al. [27, 28]), 0.25 (Kwon [11, 12]) or
3
/8 (Crighton [10]). One reason for
the uncertainty in the dependence of the frequency of oscillation on h is the uncertainty of . The exact
11
Table 1.1: Parameters of the St (Re, h/) relationships (equation (1.16)) by dierent authors
Stage Author c
1
c
2
c
3
k
Stage I
Brown [1] 0.4659 12.06 0.007 1
Jones [13] 0.39 0 0 1
Curle [3] 0.625 0 0.0267 1
Curle-Savic [3] 1.43 0 0
3
/2
Brackenridge [16] 0.6298 38.4 0.0235 1
Holger [8] 1.532 0 0
3
/2
Crighton [10] 2.477 0 0
3
/2
Kwon [12] (with Ma 0) 0.45 0 0 1
Bamberger et al. [17] (exp.) 0.513 0 0 1
Bamberger et al. [17] (CFD) 0.443 0 0 1
Stage II
Brown 1.072 27.74 0.007 1
Jones 1.217 0 0 1.14
Curle 1.125 0 0.0148 1
Curle-Savic 3.46 0 0
3
/2
Brackenridge 1.512 92.2 0.0235 1
Holger 3.332 0 0
3
/2
Crighton 10.385 0 0
3
/2
Kwon (with Ma 0) 1.05 0 0 1
Stage III
Brown 1.77 45.83 0.007 1
Jones 2.52 0 0 1.22
Curle 1.625 0 0.0103 1
Curle-Savic 6 0 0
3
/2
Holger 6.057 0 0
3
/2
Crighton 21.32 0 0
3
/2
Brackenridge 2.645 161.3 0.0235 1
Kwon (with Ma 0) 1.65 0 0 1
12
1.3. FREQUENCY AND PHASE CHARACTERISTICS OF THE EDGE TONE
0 100 200 300 400 500 600 700
0
0.01
0.02
0.03
0.04
0.05
Curle Curle-Savic Holger
Brown Jones Brackenridge
Re !-"
S
#

!
-
"
Figure 1.3: Dependence of the Strouhal number of the rst stage edge tone oscillation on the Reynolds
number at h/ = 10 in various results reported in the literature [1, 3, 8, 13, 16]
2 4 6 8 10 12 14 16
0
0.05
0.1
0.15
0.2
Curle Curle-Savic Holger
Brown Jones Brackenridge
h/ "-#
S
$

"
-
#
Figure 1.4: Dependence of the Strouhal number of the rst stage edge tone oscillation on the dimensionless
nozzle-to-wedge distance at Re = 200 in various results reported in the literature [1, 3, 8, 13, 16]
positions where the dipole source is located (i.e. at the tip of the wedge or at a certain distance away
downstream from the tip) and where the sound generated by the acoustic dipole source excites the jet
(directly at the nozzle, or somewhat further downstream) are still not explored.
Also the theories presented usually assumes that the wavelength and convection velocity of the distur-
bance do not change between the nozzle and the wedge, and thus the phase of the disturbance decreases
linearly in proportion to the distance, but this was found to not to be true (Stegen and Karamcheti [33],
Section 2.3.5).
Neglecting these minor problems, assuming that the disturbance has to travel the nozzlewedge
distance (Stage I with = 0, thus = h) with the mean speed of disturbance propagation that is about
40 % of the mean exit velocity of jet (Section 2.3.5), the period of one feedback loop is about T
h
0.4u
and so the frequency of the rst stage oscillation would be about f
0.4u
h
, which is very close to the
above formula of Jones for the rst stage. For the higher stages this heuristic model does not work.
13
1.4 Why the edge tone? Aims of the work
The edge tone is a very interesting aero-acoustic ow phenomenon. On the ow side, it is a planar
ow (as long as the depth (W) of the jet is much larger than its width()) while on the acoustic side,
under certain circumstances it generates an audible, tonal sound that has a true three-dimensional dipole
directivity at least in the far eld and as long as the depth of the jet is not comparable to the acoustic
wavelength (). Usually these conditions ( W, and W ) are fullled. Therefore the edge tone is a
perfect subject of testing a newly developed method of coupling a 2D ow simulation with a 3D acoustic
simulation that will be demonstrated in Chapter 3. At the same time, in spite of its geometric simplicity,
the edge tone produces ow phenomena that are interesting in themselves (Chapter 2) and not yet fully
understood. Despite its intensive research in the previous more than one hundred years the literature is
still not concordant even about its most basic attribute, its frequency characteristics.
The aims of this dissertation can be divided to three groups, that also makes the structure of the disser-
tation:
1. The aim of Chapter 2 is to investigate the ow of the edge tone phenomena with experimental
and numerical tools, verify one of the formulae for the Reynolds number and dimensionless nozzle-
to-wedge distance dependence of the Strouhal number published already and to point out a weak
point, an incorrect assumption of all the models that were published, namely that the convection
velocity of the disturbance on the jet is not constant.
2. The aim of Chapter 3 is to create a method of coupling a 2D ow simulation with a 3D acoustic
simulation, and to investigate the acoustic attributes of the edge tone with this newly developed
method.
3. At last, the aim of Chapter 4 is to investigate a real-world appearance of the edge tone phenomenon,
namely the ow inside the foot model of an organ pipe.
14
Chapter 2
The ow of the edge tone
The ow eld of the edge tone was investigated both by numerical and experimental methods. This
chapter will describe the Computational Fluid Dynamic (CFD) and the experimental setups and discuss
the results obtained by the two methods.
2.1 The CFD setup
ANSYS-CFX (Releases from CFX-5.7.1 to ANSYS-CFX v14; product of ANSYS Inc. Southpointe 275
Technology Drive Canonsburg, PA 15317, [34]) was used to simulate the ow. This solver is based on a
nite volume scheme, and uses an iterative method to solve the Navier-Stokes equation system. In our
case the iteration targets are the conservation of mass and momentum.
2.1.1 Two-dimensional simulations
The ow was assumed to be two-dimensional (2D). This assumption was justied: all the experimental
studies used a high aspect ratio nozzle and edge and no three-dimensional eects have been found. A
three-dimensional (3D) simulation was also performed, and in the central region the ow proved to be
almost perfectly two-dimensional and similar to the results of the 2D simulations (Section 2.3).
Although the software is only capable of calculating ows in 3D domains discretised with 3D elements,
it is still possible to calculate planar ows. For this the domain of the planar ow and the mesh discretising
it have to be extruded in the third direction with only one layer of elements. The height of this layer
can be chosen arbitrarily. As long as the aspect ratio of the elements is moderate, the magnitude of the
extrusion does not aect the result of the simulation. With symmetry boundary conditions prescribed on
the bottom and top surfaces of the extruded domain, the simulation leads to a planar ow.
Geometry, boundary conditions and solver settings
The geometry and the mesh for the CFD simulations were prepared using ANSYS ICEM CFD. The
geometric parameters were the following: the width of the slit on the nozzle () was 1 mm, the nozzle-
to-wedge distance (h) was varied between 3 and 15 mm and the angle of the wedge was 30
. This edge
tone conguration was placed inside a rectangular domain as shown in Figure 2.1 (the gure is not to
scale). It is advantageous for the ow development if the back boundary is placed somewhat behind
the nozzle exit (H
1
= 12.5 mm) otherwise non-physical vortices might appear in the ow. The other
boundaries were placed far enough from the region interest not to have any disturbing eect on the ow
(H
2
= V = 75 mm). The height of the domain in the z direction was 1 mm.
On the two x-y planes bordering the slice symmetry boundary conditions were prescribed. At the
solid walls of the wedge no slip wall boundary conditions whereas at the outer wall of the nozzle free
15
CHAPTER 2. THE FLOW OF THE EDGE TONE
h
H
1
H
2
V
V
x
y
30
u wedge nozzle
Figure 2.1: Sketch of the CFD domain (not to scale)
slip wall conditions were given. At the back wall a small inow was prescribed with an inlet velocity
of about 1% of the exit velocity. This was done in order to stabilize the ow, while not inuencing the
parameters studied. The spectral peaks appear at the same frequencies but they get sharper. Without
this, experience shows that there is an increased risk that non-physical vortices appear and remain in the
domain. At the nozzle exit inow boundary condition was prescribed with uniform or parabolic velocity
distribution. The mean exit velocity of the jet was determined from the required Reynolds number. All
other boundaries (dashed lines) were set to opening boundary condition, i.e. prescribed static pressure
without prescribed ow direction.
Air at 25
C ( = 1.185
kg
m
3
, = 1.831 10
5 kg
ms
) was used as uid. Because of the moderate Reynolds
number and low Mach number regions (Re = 60 2000, Ma = 0.003 0.09) the ow was assumed to be
laminar thus no turbulence model was used and incompressible. Second order accurate spatial (High
resolution scheme) and temporal (Second order backward Euler scheme) discretisations were used.
It has been tested to what extent the initial condition inuences the result. Simulations with initially
quiescent uid and initially steady state ow have been performed. It turned out that the initial condition
has no inuence on the nal character of the ow. No special measures had to be taken to initiate the
oscillation; the oscillation set in spontaneously after a short transient period.
The target root mean square (rms) residuum of the iteration was set to 10
5
. Some other values
(10
4
and 10
6
) were also tested. It was found that the permissive target is not sucient and the stricter
one is not necessary.
Mesh study
In order to increase the simulation accuracy a block-structured hexagonal mesh was used. First of all a
mesh convergence study has been performed on one conguration (h = 10 mm and Re = 200). Table 2.1
shows the main parameters of the three meshes that were tested: the number of elements (NoE), the
number of elements along the width of the nozzle exit (Nozzle res.) and the number of elements along
the nozzle exit - wedge tip distance (Nozzle-Wedge res.). Although the element sizes in the nozzle-wedge
region in both direction changed always by a factor of 2, the number of elements in the entire mesh
changed less than by a factor of 4 as the renement in the outer regions was a bit smaller. The two most
important attributes of the evolved ow the frequency of oscillation (f) and the rms of the force acting
on the wedge (F
rms
) are also presented in the table. Only the y component of the force was investigated
as that is the direction of maximum radiation of the acoustic dipole.
16
2.1. THE CFD SETUP
Table 2.1: Mesh parameters (NoE - Number of Elements; Nozzle res. - number of elements along the width
of the nozzle exit; Nozzle-Wedge res. - number of elements along the nozzle exit - wedge tip distance; f
- frequency of oscillation; F
rms
- rms of the force acting on the wedge; est.rel.err. - estimated relative
error)
Mesh no. NoE Nozzle res. Nozzle-Wedge res. f [Hz] F
rms
[mN] est.rel.err.
1 13 024 10 16 104 0.0255 4.0 %
2 36 300 20 32 112 0.0263 1.0 %
3 81 920 40 64 114 0.0265 0.25 %
With the generalised Richardson extrapolation [35] one can estimate the exact value of F
rms
from
extrapolating the values of F
rms
gathered from simulations with three dierent meshes. The renement
ratio between the 1
st
and the 2
nd
and the 2
nd
and the 3
rd
meshes is now q = 2. The m order of accuracy
can be estimated as:
m =
ln [(F
rms,1
F
rms,2
) / (F
rms,2
F
rms,3
)]
ln (q)
2 (2.1)
From that, the estimation of the exact value follows as:
F
rms,ex
= F
rms,3
+
F
rms,3
F
rms,2
q
m
1
0.0266 mN (2.2)
The last column of Table 2.1 shows the relative error of F
rms
comparing to the estimated exact value.
Figure 2.2 shows the monotonic convergence of the values of F
rms
to the F
rms,ex
estimated exact value (red
line) in the limit of h 0, where h is the length of an element between the nozzle and the wedge.
The dashed curve in the gure is the curve tted to the points during the process of the Richardson
extrapolation
_
F
rms
[mN] 0.0266 0.00272(h[mm])
2
_
.
Figure 2.3 shows snapshots of the velocity eld as vector plots with the three dierent meshes. It can
0 0.1 0.2 0.3 0.4 0.5 0.6
0.025
0.0255
0.026
0.0265
0.027
h [mm]
F
r
m
s

[
m
N
]
Figure 2.2: The y component of the rms of the force (F
rms
) acting on the wedge versus the
length of an element between the nozzle and the wedge (h); red curve: the estimated exact
value (F
rms,ex
); dashed curve: the curve tted to the points during the Richardson extrapolation
_
F
rms
[mN] 0.0266 0.00272(h[mm])
2
_
17
(a) Result with the 1
st
mesh
(b) Result with the 2
nd
mesh
(c) Result with the 3
rd
mesh
Figure 2.3: Velocity vector plots in case of the three dierent meshes; scale 3:1
18
2.1. THE CFD SETUP
be concluded from the mesh study that the coarser mesh was not satisfactory because certain important
ow structures were not well resolved (such as the vortex at the wedge wall near the tip, denoted by red
rectangle in Figure 2.3(a)) and also the oscillation frequency was nearly 10% lower than that on the nest
mesh. The results with the second and the third meshes dier negligibly, while the run time increased
dramatically for the third one.
Therefore the medium mesh was chosen as the best aordable. In this mesh the nozzle width was
resolved by 20 and the nozzle exit-wedge tip distance by 32 uniformly spaced elements resulting in element
sizes of about 0.3 mm x 0.05 mm (Figure 2.4(a)). When the distance h was varied, the number of elements
in this region was varied proportionally. At the wall of the wedge the ow domain was resolved by roughly
0.05 mm thick elements. Near the tip these elements had the same length as the elements between the
nozzle and the wedge and started to grow only after 50 elements (Figure 2.4(b)). In the outer regions of
the CFD domain the largest dimension of the elements was about 2.5 mm.
(a) Between the nozzle and the wedge (h = 10 mm case; scale 10:1)
(b) At the wedge wall near the tip (scale 10:1)
Figure 2.4: Snapshots of the CFD mesh
Time step study
Great care was also taken to determine the optimum temporal resolution. The optimum time step was
determined for the same reference case as in the mesh study. Simulations with time steps of = 0.05,
0.1, 0.2 and 0.4 ms were carried out to determine the optimum time step. Table 2.2 shows the oscillation
frequency and the rms of the force acting on the wedge in the case of the four simulations. Again only
the y component of the force was dealt with.
Using the generalised Richardson extrapolation this time for the time discretisation (with = 0.05, 0.1
19
Table 2.2: Results of the time step study ( - time step; f - frequency of oscillation; F
rms
- rms of the
force acting on the wedge; est.rel.err. - estimated relative error; CPU time - CPU time required for the
simulation using one core of an Intel Core i7-3770 CPU running at 3.4 GHz)
# [ms] f [Hz] F
rms
[mN] est.rel.err. CPU time
1 0.05 114 0.0259 0.8 % 11 h 57 min
2 0.1 112 0.0260 1.3 % 8 h 59 min
3 0.2 112 0.0263 2.3 % 6 h 5 min
4 0.4 110 0.0268 4.3 % 4 h 30 min
and 0.2 ms) the exact value of F
rms
can be estimated as:
F
rms,ex
0.0257 mN (2.3)
The fth column of Table 2.2 shows the relative error of F
rms
comparing to the estimated exact value.
The CPU time required for the simulation grew by 48% from the 3
rd
to the 2
nd
simulation, and by
another 33% to the 1
st
simulation. The estimated error of F
rms
and the error of oscillation frequency
were both around 2% in the = 0.2 ms case, therefore it was decided that = 0.2 ms is accepted as a
golden mean between computation resources and accuracy. The frequency of the oscillation here is 112 Hz,
which means about 45 time steps per cycle.
It is well known that the main source of the error in the numerical solution of a time-dependent partial
dierential equation is the accumulated error of the discretised time derivatives over many time steps.
Thus after the time step oering the best compromise was found, an analytical criterion to keep the error
of the temporal discretisation constant was derived as follows.
ANSYS CFX uses the Second Order Backward Euler time discretisation scheme. Let us analyse
this scheme via the numerical solution of an ordinary dierential equation (thereby assuming that the
error from the spatial derivatives remains constant):
y
t
= f (t, y (t)) , y (0) = y
0
(2.4)
where f (t, y (t)) is Lipschitz continuous with a Lipschitz constant L
f
_
i.e.
|f(t,y
1
)f(t,y
2
)|
|y
1
y
2
|
L
f
_
.
Let the time step, resulting a time discretisation of t
j
= j be small enough to satisfy the following
criterion:
3
2
L
f

1
c
(2.5)
where c > 0 is a constant.
The Second Order Backward Euler dierentiation scheme takes the following form:
y
t
(t
j
)
1
_
3
2
y (t
j
) 2y (t
j1
) +
1
2
y (t
j2
)
_
f (t
j
, y (t
j
)) (2.6)
With approximating the exact values y (t
j
) by the numerical values y
j
the algebraic equation to be solved
is the following:
1
_
3
2
y
j
2y
j1
+
1
2
y
j2
_
= f (t
j
, y
j
) (2.7)
20
2.1. THE CFD SETUP
where y
j1
and y
j2
are known from the previous time steps and y
j
is to be determined.
The local error (g
j
) of this discretisation is the dierence of the two sides of (2.6):
g
j
:= g (t
j
, ) =
1
_
3
2
y (t
j
) 2y (t
j1
) +
1
2
y (t
j2
)
_
f (t
j
, y (t
j
))
=
1
_
3
2
y (t
j
) 2
_
y (t
j
)
y
t
(t
j
) +

2
2
2
y
t
2
(t
j
)

3
6
3
y
t
3
(t
j
) +O
_
4
_
_
+
1
2
_
y (t
j
) 2
y
t
(t
j
) +
4
2
2
2
y
t
2
(t
j
)
8
3
6
3
y
t
3
(t
j
) +O
_
4
_
__
f (t
j
, y (t
j
))
=
1
y
t
(t
j
)

3
3
3
y
t
3
(t
j
) +O
_
4
_
_
f (t
j
, y (t
j
)) =
2
3
3
y
t
3
(t
j
) + O
_
3
_
(2.8)
subtracting (2.7) from (2.8) with the notation of e
j
= y (t
j
) y
j
we obtain:
1
_
3
2
e
j
2e
j1
+
1
2
e
j2
_
= f (t
j
, y (t
j
)) f (t
j
, y
j
) + g
j
(2.9)
Introducing the
j
=
f(t
j
,y(t
j
))f(t
j
,y
j
)
y(t
j
)y
j
notation equation (2.9) transforms into
e
j
_
3
2

j
_
= 2e
j1
1
2
e
j2
+ g
j
(2.10)
As
1
3
2
j
=
2
3
+
2
j
3(
3
2
j)
it yields to:
e
j
=
4
3
e
j1
1
3
e
j2
+

3
2

j
_
4
3
j
e
j1
1
3
j
e
j2
+ g
j
_
(2.11)
Introducing the following notations:
j
=
_
e
j1
e
j
_
; A =
_
0 1
1
3
4
3
_
; B
j
=
_
0 0
1
3
j
1
3
2
j
4
3
j
1
3
2
j
_
; v
j
=
_
0
1
3
2
j
g
j
_
(2.12)
(2.11) in a matrix notation becomes:
j
= A
j1
+ B
j
j1
+ v
j
(2.13)
Let S be a matrix with which SAS
1
is the Jordan normal form of A:
S =
_

1
2
3
2
1
2
1
2
_
, SAS
1
=
_
1 0
0
1
3
_

_
_
SAS
1
_
_
= 1 (2.14)
where P
:= max
i
j
|p
ij
|.
After multiplying both sides of (2.13) from the left side by S, with S
1
S = I we get:
S
j
= SAS
1
S
j1
+ SB
j
S
1
S
j1
+ Sv
j
(2.15)
with the norm
j
S
:= S
j
and the usual inequalities for norms:
S

_
_
SAS
1
_
_
j1
S
+ S
B
j
_
_
S
1
_
_
j1
S
+ S
(2.16)
S

j1
S
+ 2 B
j
4
j1
S
+ 2 v
(2.17)
21
Using the Lipschitz property of f:
B
j
= |
j
|
1
3
2

j
_
1
3
+
4
3
_
L
f
1
3
2

j
5
3

5
3
L
f
1
3
2
L
f
(2.18)
v
j
1
3
2

j
g
j
1
3
2
L
f
|g
j
| (2.19)
(2.17) takes the following form:
S

j1
S
+ 8
5
3
L
f
1
3
2
L
f
j1
S
+ 2
1
3
2
L
f
|g
j
| (2.20)
Since
3
2
L
f

1
c
,
S

_
1 +
40
3
L
f
c
_
j1
S
+ 2c |g
j
|
_
1 +
40
3
L
f
c
_
j
S
+ 2c
j
k=1
|g
k
|
_
1 +
40
3
L
f
c
_
jk
_
1 +
40
3
L
f
c
_
j
_
S
+ 2c
j
k=1
|g
k
|
_
e
40
3
jL
f
c
_
S
+ 2c
j
k=1
|g
k
|
_
(2.21)
The -norm of
j
can be limited from below and above with the S-norm:
S
= S
j
= 2
j
, and (2.22)
=
_
_
S
1
S
j
_
_

_
_
S
1
_
_
S
j
= 4
j
S
(2.23)
thus (2.21) can be written as:
4
j
S
4e
40
3
jL
f
c
_
S
+ 2c
j
k=1
|g
k
|
_
4e
40
3
jL
f
c
_
2
0
+ 2c
j
k=1
|g
k
|
_
(2.24)
Where
j
= max (|e
j1
| ; |e
j
|), thus |e
j
|
j
. Let us assume that the initial values are correct

and the initial error
0
is zero, so:
|e
j
| 8e
40
3
jL
f
c
_
c
j
k=1
|g
k
|
_
(2.25)
From (2.8):
g
j
=
2
3
3
y
t
3
(t
j
) +O
_
3
_
|g
j
|

2
3
max
3
y
t
3
+O
_
3
_
(2.26)
Thus at T
S
= N ,i.e. at the end of the simulation the global error will be:
|e
N
| 8e
40
3
T
S
L
f
c
cN
_
2
3
max
3
y
t
3
+O
_
3
_
_
= 8e
40
3
T
S
L
f
c
T
S
c
2
3
max
3
y
t
3
+O
_
3
_
(2.27)
|e
N
| const e
T
S
T
S
2
max
3
y
t
3
+O
_
3
_
(2.28)
22
2.1. THE CFD SETUP
where const is a constant, independently of the time step.
If we assume that the ow velocity at a xed spatial point is a harmonic function of time, v sin (t),
the maximum of the absolute value of the third derivative can be approximated with v
3
. The global
error of the simulation after the last time step can be estimated as (with f = /(2)):
|e
N
| const e
T
T
2
vf
3
+O
_
3
_
(2.29)
where T is the duration of the simulation, is the time step, v is the amplitude of the velocity oscillation
and f is the expected oscillation frequency.
For our reference case, Re = 200 a certain optimum time step was determined. Our task is to keep the
same error at other Reynolds numbers.
The frequency of oscillation as will be shown later is nearly proportional to the mean exit velocity
(u) of the jet in each stage. The duration of the simulations (in simulated time, T
S
) can be divided into
the transient part before the quasi-steady oscillation sets in (T
t
) and the oscillation part (T
o
): T = T
t
+T
o
.
This transient part was omitted from the signal when performing FFT. T
o
was mainly determined by
the required frequency resolution of the spectra (f = 1/T
o
). Of course, with decreasing time step, it
becomes increasingly dicult to get the same absolute frequency resolution in a reasonable time. At
higher Reynolds numbers it was decided that the duration of the simulation should ensure a frequency
resolution of 1 2% of the expected maximum frequency. Fortunately it turned out that the duration
of the transient part decreased linearly with increasing frequency. That means that the total duration
of the simulation (T
S
) could be reduced inversely proportionally with frequency (f) thus also with the
mean exit velocity of the jet (u) while keeping a constant relative frequency resolution. The amplitude
( v) of the velocity oscillation is also proportional to the mean velocity u. Finally, the exponential factor
can be ignored since then we are on the conservative side. This factor decreases anyway with increasing
velocity and tends to 1.
Putting all the information together, it can be concluded that, in order to keep the absolute error
constant, the time step has to be decreased more quickly than inversely proportional to the mean velocity;
it has to be proportional to u
3
/2
. Thus the number of time steps per cycle increases from about 34 at
Re = 150 to 272 at Re = 1800. Below Re = 150 smaller time steps were used in order to have the
period resolved ne enough. Although the simulations were carried out keeping this condition, it has to
be mentioned that if the relative error is to kept constant it is sucient if the time step is proportional
to u
1
resulting in a constant period resolution.
When a simulation is started the expected oscillation frequency and thus the required time step can
be estimated by linear extrapolation. This frequency and time-step estimation was always a posteriori
veried. When a higher stage appears with a sudden frequency rise then the simulation had to be repeated
with the time step adjusted accordingly.
The frequency of oscillation was determined by means of FFT from pressure and velocity histories
in several points of the ow eld or from the history of the force acting on the wedge. No signicant
dierences were found in the frequencies whether the force, velocity or pressure histories were used and
whether this or that point was used in case of the velocity or pressure signals. Further post processing of
the CFD simulations are described in Section 2.3.5.
23
2.1.2 Three-dimensional simulation
A 3D CFD simulation was also carried out at a Reynolds number of 225 to verify the planar nature of the
ow. Re = 225 was choose as that was the highest Reynolds number where a pure rst stage oscillation
was found among the 2D simulations. The width of the nozzle was the same as in the 2D simulations
( = 1 mm) and the nozzle-to-wedge distance was h = 10 mm. The nozzle and the wedge had dierent
heights: 25 mm and 70 mm, respectively. This was done to allow the jet to spread in the z direction and
to have the full eect on the wedge while minimize the end eects. This edge tone setup was placed in
a 90 mm x 151 mm x 70 mm rectangular domain. Again a block-structured hexahedral mesh was used
(Figure 2.5). In the central region (12.5 mm < z < +12.5 mm, where z = 0 is the middle plane) the
elements had a height of 1 mm. From the end of the nozzle in the following 7.5 mm the height of the
elements grew up to 3 mm from where the elements had a height of 3 mm. The mesh in any cross section
perpendicular to the z direction had the same resolution in the nozzle-wedge region as the mesh used
in the 2D simulations and was coarser at the boundaries. The mesh contained about 685 000 elements.
Comparison between the results of the 2D and the 3D simulations will be given later in Section 2.3.
Figure 2.5: Mesh of the 3D CFD simulation
24
2.2. THE EXPERIMENTAL SETUP
2.2 The experimental setup
2.2.1 Experimental system and instrumentation
The rig to produce the edge tone ow is depicted in Figure 2.6(a). Shop air with a pressure reduced
to 0.5 bar by a pressure reducing valve was led by
3
/4 reinforced exible plastic tubes to a cylindrical
pressure reservoir with a volume of 57 l.
A mass ow rate sensor (Sensortechnics, Honeywell AWM700, working on a heated element principle with
a voltage output) was built into the line between the pressure reducing valve and the reservoir tank to
determine the mean velocity of the jet. There was a long copper pipe section before the sensor to ensure
undisturbed inow. The sensor was placed between two throttle valves to keep the pressure at the sensor
constant as the calibration of the sensor showed that the sensor might be sensitive to the pressure inside.
This was examined, and it was found out that this is only crucial if the pressure inside the sensor exceeds
10 kPa. To be on the safe side the calibrations and the measurements were all done at an inside pressure
of 0.4 kPa.
The control mass ow rate for the calibration was generated by a sinking bell vessel. A bell vessel
was plunged into an oil bath upside down. The air beneath the vessel was lead out through a pipe
onto which the mass ow rate sensor was mounted. With two valves at both sides of the mass ow rate
sensor it was possible to keep the pressure inside the sensor at a constant 0.4 kPa value. The sinking
height (z = z
1
z
2
, where z
1
and z
2
are the two levels read at the beginning and at the end of the
measurement both with 0.5 mm uncertainty) was measured during a certain time (t
s
, measured by a
stopwatch having precision of 0.01 s). Although the value of t
s
always exceeded 60 s (thus the error in
the time measurement is negligible) still at extremely low mass ow rates the value of z was only
1015 mm, thus its uncertainty was to around 57 %. Fortunately these mass ow rates were below the
values used during the experiments on the edge tone. In the domain of interest the value of the error was
less than 3 %. From these values, the mass ow rate was determined as m = z/t
s
, where is the
area of cross section of the vessel ( = 0.398
dm
3
mm
0.5 %) and is the density of the air leaving the vessel
throttle valves
mass ow rate sensor
nozzle
reservoir
wedge
location of pressure tap
pressure
reducing valve
(a)
top hat
nozzle
reservoir
incense
stick
(b)
parabolic
nozzle
reservoir
(c)
Figure 2.6: (a) Sketch of the measurement system; (b) Details of the top hat nozzle; (c) Details of the
parabolic nozzle
25
which was calculated assuming an isothermal process by measuring the pressure inside the vessel. The
error of the density calculation was also investigated and found to be less than 0.2 %. The error of the
measurement of the control mass ow rate thus originates mainly from the error of the sinking height of
the vessel. Therefore it can be concluded that the measurement of mass ow rate with this sensor during
the edge tone measurement has an uncertainty of at most 3 %.
Two dierent nozzles were used to create a top hat or a parabolic velocity prole for the jet. They will
be referenced as the top hat nozzle and parabolic nozzle. The top hat nozzle was formed by two
quarter-cylinders ensuring a quick contraction (Figure 2.6(b)). In the parabolic nozzle (Figure 2.6(c))
two parallel 150 mm long plates with a 15 mm radius rounded entry ensured the development of the
parabolic velocity prole.
The shape of the cross section of the nozzles was nominally rectangular with an aspect ratio over 20
which ensured good two-dimensionality in the central region, but there were some deviations at the ends of
the nozzles. These deviations were taken into account when calculating the area of the cross section needed
to calculate the mean exit velocity of the jet from the mass ow rate but was neglected when measuring
the width of the slit on the nozzle (only averaging the values measured in the central two quarters). This
could be done, as the ow was two-dimensional and the ow eld was analysed always in the central
plane. To calculate the area of the cross section the width of the slit was measured at dierent heights,
and was integrated over the height of the nozzle. The width was measured by a standard gap-measuring
kit (Hungarian Standard MSZ 11151) with an accuracy of 0.05 mm. The estimated relative error of the
width of the nozzle considering both the accuracy of the gap-measuring kit and the deviations of the
width of the nozzle was around 1.8 % and 4.2 %, for the top-hat and parabolic nozzles, respectively. The
error of the area measurement was around 2 % for both nozzles. The main dimensions of the two nozzles
can be found in Table 2.3.
Table 2.3: Main dimensions of the nozzles with their estimated error: A - area of cross section; - width
of the slit
A[mm
2
] [mm]
top hat nozzle 249.7 4.84 3.31 0.058
parabolic nozzle 190.8 3.91 3.09 0.129
When using the top hat nozzle visualisation was possible with an incense stick inserted just before
the converging part of the nozzle. The smoke lament was illuminated with oodlight and the image was
recorded with a high speed digital camera (LaVision ImagerCompact) taking pictures with a maximum
frequency of 90 Hz with a spatial resolution of 320x240 pixels. No visualization was made in the parabolic
case.
The wedge was made of well-polished solid steel with an angle of 30
. Its height (z direction) was 150 mm

i.e. this dimension was twice as long as the slit of the nozzle to avoid end eects. The distance between
26
the nozzle and the wedge was adjustable in the range of 5 mm to 53 mm. To measure this distance, a
strip of a mm scale paper was attached on the mount under the wedge. With the help of a needle placed
on the wedge it was possible to read the position of the wedge with an accuracy of 0.5 mm. The lowest
nozzle-to-wedge distance for which the edge tone phenomenon was observed was 10 mm, so that the
relative error of the nozzle-to-wedge distance measurement can be in this case as much as 5 % but when
the Reynolds number dependence was examined, the nozzle-wedge distance was 34 mm (and 30 mm for
the parabolic jet), so that in that case the error was about 1.5 %.
A pressure transducer (Sensortechnics, 113LP01D-PCB) was built into the wedge to measure the surface
pressure at a distance of 26.2 mm from the tip of the wedge. The measuring range of this sensor is
[100 Pa; 100 Pa] and the output for this range is [1 V; 6 V]. Although calibration data from the vendor
was available it was further veried by means of a micromanometer in the [100 Pa, 100 Pa] range with
10 Pa resolution, and the linearity was veried. The weakness of the calibration is that data points were
not located densely enough in the relevant [4 Pa; 4 Pa] range. The dynamic behaviour of the sensor was
also not fully explored in the absence of any other reliable pressure measuring device that could provide
reference values.
The amplitude of the pressure uctuation at the tap for the lower range of Reynolds numbers is
less than 0.05 Pa while at higher Reynolds numbers it was still only about 4 Pa. This means that the
amplitude of the output voltage uctuation of the sensor was 0.00125 0.1 V which at the lower end
means an only four-step resolution when using a 16 bit National Instruments A/D converter with an input
voltage region of [10 V; 10 V]. Therefore in order to be able to digitalize such small pressures with the
mentioned A/D converter, the analogue signal of the pressure transducer was amplied with a special
amplier which transformed the [3.3 V; 3.7 V] voltage region into the [10 V; 10 V] region.
To check the characteristics of the amplier, true sinusoidal signals were produced with the help of
a signal generator that were then amplied. Both the original and the amplied signals were recorded.
First the time lag between the two signals was determined with cross correlation. The delay was between
0.14 and 0.2 ms when the signal frequency was less than 1 kHz. Above this frequency limit the delay was
even less. The correlation coecient of the two signals was above 0.999 for the signal frequencies less
than 1 kHz, and was always more than 0.94. Then the amplication was determined by calculating the
coecients of the linear relationship with the method of least squares. Under 100 Hz the amplication is
almost constant with a mean value of 50.3 and a standard deviation of 0.077. By 500 Hz the amplication
drops to about 43. At 1 kHz the amplication is around 34, and by 10 kHz it drops to 4. Although the
amplication factor drops above 100 Hz, this is not crucial as the main target of the measurement is
only the frequency without the amplitude, but it has to be mentioned that the amplitudes of dierent
frequency components can only be compared if their frequencies are both below 100 Hz.
Although for extremely low nozzle-to-wedge distances much higher frequencies can be achieved (e.g.
maximum measured frequency was 949 Hz during a third stage oscillation with parabolic prole at
Re = 1200 with a nondimensional nozzle wedge distance of h/ = 5.5), the typical values of the edge
tone frequencies during the measurements were less than 100 Hz.
27
2.2.2 Derived quantities and their error estimation
The error analysis of the dimensionless nozzle-to-wedge distance, the Reynolds number and the Strouhal
number derived from the measured quantities and from their error estimates are given below.
Dimensionless nozzle-to-wedge distance
The error of the dimensionless nozzle-to-wedge distance (h/) derives from the error of the nozzle-to-
wedge distance (e
h
) and from the error of the measurement of the width of the slit on the nozzle (e
).
Thus the (relative) error is:
e
h/

_
(h/)
h
_
2
e
2
h
+
_
(h/)
_
2
e
2
(2.30)
e
h/
h/

_
e
2
h
h
2
+
e
2
2
(2.31)
This error had its maximum
_
e
h/
h/
5.9 %
_
when the parabolic nozzle was used
_
e
4.2 %
_
and the
nozzle-wedge distance was the lowest
_
h 12 mm, thus
e
h
h
4.2 %
_
, but usual values were less than 3 %
in the top hat case and less than 5 % in the parabolic case.
Reynolds number
Re =
u
(2.32)
where u is the mean exit velocity of the jet, is the width of the nozzle and is the kinematic viscosity of
the air. The mean exit velocity of the jet is calculated from the measured mass ow rate ( m) by dividing
it by the area of the orice (A) and the density (). Using the dynamic viscosity () instead of the
product the Reynolds number can be calculated from the measured quantities with the following formula:
Re =
m
A
(2.33)
The measurement of m, and A have already been discussed. The dynamic viscosity of the air at
atmospheric pressure depends only on the temperature (T) [36] and in the [0
C; 40
C] temperature
region it can be linearly approximated as:
= 4.714 10
8
T + 1.718 10
5
(2.34)
where T is now the ambient temperature in

C and is in
kg
ms
. From this, the value of the dynamic viscosity
at 20
C is approximately 1.812 10
5 kg
ms
, while the slope of the linear formula is 4.714 10
8 kg
ms
C
, thus
it can be concluded that the error of the dynamic viscosity due to the erroneous temperature readings is
negligible.
The error of the Reynolds number derives from the error of the mass ow rate measurement (e
m
), the
error of the measurement of the width of the slit on the nozzle (e
) and the error of the area measurement

of the nozzle cross section (e
A
). The error of the viscosity calculation can be neglected. These errors are
28
collected in Table 2.4. From these values the error of the Reynolds number can be calculated as:
e
Re

_
Re
m
_
2
e
2
m
+
_
Re
_
2
e
2
+
_
Re
A
_
2
e
2
A
(2.35)
e
Re
Re

_
_
e
m
m
_
2
+
_
e
_
2
+
_
e
A
A
_
2
(2.36)
Table 2.4: Relative error of the components of the Reynolds number, and the derived error of the Reynolds
number
e
m
m
[%]
e
[%]
e
A
A
[%]
e
Re
Re
[%]
top hat 3 1.75 1.94 3.98
parabolic 3 4.19 2.05 5.55
Strouhal number
St =
f
u
(2.37)
where f is the frequency of oscillation, u is the mean exit velocity of the jet and is the width of the
nozzle. From the measured quantities, the Strouhal number can be calculated as:
St =
fA
m
(2.38)
Assuming that the air behaves as an ideal gas its density () can be calculated from the atmospheric
pressure and the ambient temperature.
Initially four methods were used to determine the frequency:
1) FFT of the time history of the pressure signal
2) Processing the photo-series taken of the smoke lament:
(a) The vortex counting method
(b) The wavelength method
3) Stroboscope principle
1) The amplied output voltage of the pressure transducer was captured for 10 100 seconds
depending on the lowest frequency component of the oscillation, with 4 10 kHz sampling frequency
depending on the highest frequency component of the oscillation. The output of the signal was stored on
the hard disk, and the pressure history and its spectrum was plotted. This way it was possible to have
a look at the spectra on the spot, and later a detailed processing of the signal with MatLab was also
possible.
Figure 2.7(a) shows a typical history of the output voltage of the pressure transducer. This was
recorded at Re 150. At such a low Reynolds number the signal to noise ratio was quite low, as seen
29
4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5
5.7
5.8
5.9
(a)
t [s]
U

[
V
]
0 5 10 15 20 25 30
0
0.005
0.01
0.015
0.02
(b)
f [Hz]
a
m
p
U

[
V
]
Figure 2.7: (a) A piece of the time history of the output of the pressure transducer (black: originally
recorded output voltage history, grey: after creating an moving-average of 475 adjacent values);
(b) Spectrum of the output voltage of the pressure transducer;
in the signal. The black graph represents the recorded, while the grey one represents the smoothed data.
Figure 2.7(b) shows the spectrum of this signal.
The frequency resolution of the spectrum was always between 0.01 0.1 Hz. The uncertainty of the
frequency determination came from the broadening of the spectral peaks, which was usually less than
4 % of the frequency of the peak for the most of the cases, but in some special cases it was as high as
10 15 %.
2) The frequency of oscillation can also be determined by processing the photo-series of the smoke
lament. There are two ways to do this: one of them is to count the n
v
number of vortices passing at
a xed spatial point next to the wedge and measuring the elapsed time t
v
. The n
v
/t
v
ratio gives the
frequency of oscillation. This method will be referenced as the vortex counting method. The exact time
of the exposure of each picture is recorded in s resolution. The time between two adjacent photos in the
series is not less than 10 000 s, thus the error of t
v
can be neglected. The error of n
v
can be estimated
as less than 0.25 vortices. From this the error of f can be estimated as e
f
< 0.25/t
v
, thus the relative
error e
f
/f < 0.25/n
v
. Typically this error was 0.7 %0.5 %.
The other method estimates the disturbance wavelength visually and measures the disturbance
propagation speed by comparing the position of two same phase points in two subsequent images (the
30
wavelength method). The frequency of oscillation can be calculated as the ratio of disturbance velocity
and the wavelength. In Figure 2.8 the propagation of the same phase point (peak of the wave) indicated
by an arrow can be observed. The coordinates of the arrow (in pixels or after calibration of the camera
in mm-s) can be acquired from the picture. The propagation speed then is the ratio of the distance (x
1
or x
2
) and the time elapsed between the two photographs. Figure 2.8 also shows the measurement of
the wavelength (). In most cases it was not possible to observe a full wave thus it is measured as the
double of the distance between the bottom and the peak points or two inection points of the wave.
Because of the uncertainty of the readings of the location of that special same phase point and the
uncertainty of the readings of the half wave the accuracy of frequency determination with this method
(c)
(b)
(a)
x
1
x
2
/2
/2
/2
Figure 2.8: Subgures (a), (b) and (c) are three subsequent images from a photo series (with t
w
time
elapsed between them); x
1
and x
2
are the distances that the peak points of the wave (indicated by
an arrow) travelled from (a) to (b) and from (b) to (c), respectively; /2 denotes how the (half of a)
wavelength can be measured from a single image; The frequency of oscillation is thus: f
x
1
/
2/2
or
f
x
2
/
2/2
31
was only about 1530 %. Also this method assumes that the wavelength of the disturbance and the speed
of disturbance propagation is independent of phase and of spatial location. This assumption was proved
to be wrong as demonstrated by the results of CFD simulations in Section 2.3.5. In spite of the the large
uncertainty and the wrong assumption this method is quite unique as it can determine the oscillation
frequency from only two pictures instead of from a long pressure history or a long photo-series.
Both techniques based on processing of the photo-series are only capable of determining the oscillation
frequency of the dominant stage. When there is a dominant second stage oscillation with the rst stage
also coexisting it is impossible to measure the frequency of the rst stage with these methods.
3) A fourth method of measuring the oscillation frequency is simply to use a stroboscope light. This
could be used only for low Reynolds numbers since with the appearance of higher modes the periodicity
disappears.
Comparison of dierent frequency measurement techniques: As the image processing
techniques are capable of measuring the frequency only of the dominant stage those are shown in the
results. Table 2.5 shows the results for the rst and for the second stages with the approximated error
limits calculated as given above. Each line of the table shows the frequency measured with the four
method of a certain edge tone ow. The upper and lower parts of the table show frequencies of Stage I
and of Stage II oscillations respectively.
As discussed before the so called wavelength method is very imprecise. The other three methods give
the same results within their accuracy. Although the best accuracy in the frequency measurement can be
achieved with the vortex counting method, the best way to measure the frequencies of the edge tone
oscillation is the FFT method, because this method is capable to measure the frequencies of the lower
Table 2.5: Comparison of the results of dierent frequency measurement techniques
f [Hz]
FFT Vortex counting Wavelength Stroboscope
(min ; peak ; max) method method
(3.5 ; 4 ; 4.13) 3.87 0.05 5.56 1.35 4.17
(6.85 ; 7 ; 7.15) 6.97 0.05 7.25
(9.35 ; 9.45 ; 9.75) 9.44 0.05 12.65 3.01 9.83
(12.3 ; 12.5 ; 12.7) 12.49 0.05 13.3
(28.3 ; 28.5 ; 28.7) 28.82 0.05 24.77 2.96 30
28.61 0.05 34.05 3.77
33.16 3.32
(34.8 ; 35 ; 35.4) 34.97 0.05 36.7
(49.2 ; 49.4 ; 49.6) 49.18 0.61 51.3
(62.2 ; 62.5 ; 62.9) 60.88 0.38 65.3
32
2.3. RESULTS
stages when several stages coexist superposed on each other and its accuracy is only slightly lower than
that of the vortex counting method.
The error of the Strouhal number derives from the error of the mass ow rate measurement (e
m
), the
error of the frequency measurement (e
f
), the error of the measurement of the width of the slit on the
nozzle (e
) and the error of the area measurement of the nozzle cross section (e
A
). The error of density
calculation can be neglected. Most of these errors were already collected in Table 2.4. From these values
the error of the Strouhal number can be calculated as:
e
St

_
St
f
_
2
e
2
f
+
_
St
_
2
e
2
+
_
St
A
_
2
e
2
A
+
_
St
m
_
2
e
2
m
(2.39)
e
St
St

_
e
f
f
_
2
+
_
e
_
2
+
_
e
A
A
_
2
+
_
e
m
m
_
2
(2.40)
e
St
St

_
_
_
_
e
f
f
_
2
+ 0.00158 3.98 % , in the top hat case
_
_
e
f
f
_
2
+ 0.00308 5.55 % , in the parabolic case
(2.41)
The relative error of the frequency measurement (e
f
/f) depends always on the used frequency measurement
technique and also on the actual measurement. With the FFT method which was used mostly during
the parametric studies the typical values of the relative error of the Strouhal number were around 4.5 %
for the top hat case and around 6.2 % for the parabolic case.
2.3 Results
Verication of the planar nature of the ow
Both the 3D CFD simulation and the experiments veried that the ow is indeed two-dimensional. In
the 3D CFD simulation, such as in the 2D case, after a short transient part during which the edge tone
oscillation evolves, a stable rst stage oscillation evolved with a frequency of 1300.5 Hz. The role of the
higher harmonics can be neglected. The frequency of oscillation is equal in the 2D and 3D simulations
and the structure of the ow is very similar in the two cases. Figures 2.9(a) and 2.9(b) show snapshots
of the velocity eld as vector plots in a same phase state for the 2D and 3D cases, respectively. In the 3D
case the velocity eld at the central (z = 0 m) plane is shown. In this plane there is no velocity in the z
direction at all.
Figure 2.10 shows
_
v
1
x
+
v
2
y
_
V
el
/ m
in
at the central plane, where V
el
is the volume of the nite
elements of the mesh, and m
in
is the mass ow rate injected into the system through the nozzle. This was
done because ( v) V
el

_
V
el
( v) dV =
_
S
el
v
n
dS that is the net mass ow out from the element
through its surface (S
el
). This is zero as the uid is incompressible. If
v
1
x
+
v
2
y
is used instead of ( v),
then information is gathered about the two-dimensionality of the ow. If it is (close to) zero, than there
is no change in the third direction, thus the ow is two dimensional. At last this value is normalised by
the mass ow rate injected into the system, to be able to judge its smallness. The gure shows that its
value in the nozzle-to-wedge region and near to the wedge tip is almost zero. Its value only reaches 0.1 %
33
(a) In the case of the 2D CFD simulation
(b) In the case of the 3D CFD simulation, at the z = 0 mm plane
Figure 2.9: Snapshots of the velocity eld; Re = 225, h/ = 10
Figure 2.10:
_
v
1
x
+
v
2
y
_
V
el
/ m
in
where V
el
is the volume of the nite elements of the mesh and
m
in
is the mass ow rate injected into the domain through the nozzle at z = 0 mm in the case of the
3D CFD simulation; Re = 225, h/ = 10; (scale 2 : 1)
34
2.3. RESULTS
Figure 2.11: Snapshot of streamlines starting at the height of the nozzle in the case of the 3D CFD
simulation; Re = 225, h/ = 10
more than 2h far away from the tip of the wedge. This again clearly shows that there is no 3D eect in
edge tone in the central plane of the jet and thus the 2D assumption is veried.
Figure 2.11 demonstrates that the streamlines that exit from the nozzle, are parallel between the
nozzle and the wedge, and only separate after a short distance from the tip of the wedge. The z = 0 mm
plane, as well as the streamlines are coloured by the velocity magnitude. Similar observations were
made during the experiments. The ow was visualised with several incense sticks at dierent heights
simultaneously. The trace lines at dierent heights oscillated together.
2.3.1 Varying the Reynolds number at a xed geometric conguration
The dimensionless nozzle-to-wedge distance was set to around h/ 10, while the Reynolds number of
the ow was varied by varying the mass ow rate. Table 2.6 shows the values of h/ and the investigated
Reynolds number ranges during the CFD simulations and during the experiments in the top hat and the
parabolic cases.
Table 2.6: Values of h/ and ranges of Reynolds numbers under that the Reynolds number dependence
studies were carried out
Top hat Parabolic
CFD Exp. CFD Exp.
h/ 10 10.26 0.23 10 9.72 0.44
Re range 0 1800 0 1200 0 2000 0 - 1400
35
Top hat prole
At rst measurements and CFD simulations were carried out at very low Reynolds numbers (Re < 60)
where no edge tone activity was found.
At Re 60 in the simulations and also in the experiments an intermittent rst stage edge tone oscillation
sets in. In the case of the CFD simulations this means that the simulations produce a steady ow eld but
with a small disturbance (the exit velocity of the jet was increased with a transversal component of 0.5
m
s
that is approximately 50 % of the mainstream velocity in the Re = 60 case for about 100 150 steps)
the oscillation could be initiated and it stayed. In the case of the experiments the edge tone phenomenon
was switching on and o, in some cases it could be observed. Above Re = 100 a stable rst stage edge tone
oscillation occurred without any external excitation both in the CFD simulations and in the experiments.
At Re = 250 in the simulations and Re = 200 in the experiments the second stage of the edge tone
came into being next to the rst stage. With the advent of the second stage the frequency and thus the
Strouhal number of the rst stage dropped by about 14.5 2 % in the simulations and 14 2 % in the
experiments compared to the value extrapolated from the pure rst stage using equation (2.43) with
coecients from Table 2.7 to be introduced later on in this subsection to the current inlet velocity. This
drop is in good agreement with the results of Jones [13] and of Stegen and Karamcheti [33].
Figure 2.12 shows the Reynolds number dependence of the Strouhal number of the rst stage in the
pure and in the multi-stage modes. Solid and dashed curves in the gure show the best t with formulae
to be introduced in equation (2.43) with coecients in Table 2.7 tted on the observed Strouhal numbers
of the rst stage in the pure (solid) and multi- (dashed) stage modes. The thick solid line gets thin at
the appearance of the second stage from where extrapolated values are used for the graph. Right of this
point the drop of the Strouhal number of the rst stage the dierence between the thin solid line and
the dashed line of the same colour can nicely be observed. Error bars are representing the width of
the spectral peak in the case of the simulations, and the uncertainty of the measurements in the case of
the experiments. The dierence between the red (experimental) and the yellow (CFD) curves is usually
around 7.5 8.5 % and only grows over 10 % at the appearance of the rst stage when the steepness of
the curves are quite high, thus small error in the Reynolds number results in a large error in the Strouhal
number or at the appearance of the second stage. The magnitude of the dierence is within the range
of the uncertainty of the Strouhal number determination. The rst stage was always present parallel to
the second stage, the question was only if the second stage is dominant or their strengths are comparable.
For example Figure 2.13 shows two parts of the same CFD simulation (Re = 600, h/ = 10, top hat
jet prole). Above is a short part of the time history of the force acting on the wedge and below is the
corresponding spectrum of a somewhat longer part of the signal. First the two stages were comparable
(left side) then later the second stage became dominant (right side).
Brown [1] found the second stage to set in around Re 220. He also mentioned to observe
simultaneous existence of stages. He took photos of the oscillating ow illuminated by light through a
stroboscopic disk. Therefore except for some special cases he was only able to measure the frequency
of the dominant stage when several stages coexisted. From these special cases he concludes that the
36
2.3. RESULTS
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.01
0.02
0.03
0.04
0.05
Re []
S
t

[
]

CFD
Exp
funCFDpure
funExppure
funCFDmulti
funExpmulti
Figure 2.12: Reynolds number dependence of the Strouhal number of the rst stage; top hat prole,
h/ 10; fun-CFD-pure, fun-CFD-multi, fun-Exp-pure and fun-Exp-multi are the functions described in
equation (2.43) and Table 2.7 in the case of CFD simulations (yellow lines) or Experiments (red lines)
and pure (solid lines) or multi -stage modes (dashed lines)
0.0596 0.063 0.0663
0.4
0.2
0
0.2
0.4
t [s]
F

[
m
N
]
0 500 1000 1500 2000 2500
0
0.05
0.1
0.15
0.2
0.25
f [Hz]
a
m
p
F

[
m
N
]
0.2182 0.2204 0.2226
0.4
0.2
0
0.2
0.4
t [s]
F

[
m
N
]
0 500 1000 1500 2000 2500
0
0.05
0.1
0.15
0.2
0.25
f [Hz]
a
m
p
F

[
m
N
]
Figure 2.13: Stage I & II coexistence edge tone oscillation in the CFD simulations at Re = 600, h/ = 10
with top hat jet prole. The two sides represents two parts of the same force signal: left side: the rst
and the second stages are comparable; right side: the second stage dominates
37
frequency of the rst stage, when coexisting with other stages are about 7 % lower than using the formula
deduced from the frequencies of pure rst stage oscillations. He states that his formula in some cases may
have an error of 6 %, thus he concludes that the frequency of the rst stage remains practically unaltered,
although it could mean also that the drop is even higher than 7 % which would agree with the results of
Jones [13] and also with my computational and experimental results.
With the advent of the third stage the rst two stages remained thus three stages coexisted superposed
on each other. In the CFD simulations the third stage rst appeared in the Re = 900 simulation, but it
was not yet stable, the third stage was observable only during a short period of the whole simulation.
From Re = 1400 the third stage was present in the whole simulation. In the experiments the third stage
appears surprisingly early, at around Re 250. This was only discoverable when using the pressure
sensor for frequency measurement, because when the third stage sets in, rst it is very weak, and from
the three stages that are superposed only the dominant second one can be seen from the visualised ow.
Figure 2.14 shows photos taken of the oscillating ow at around Re = 400 and 700 respectively, and the
spectrum of the belonging pressure signals. Three half waves can be observed in the Re = 700 case, and
only two half waves in the Re = 400 case, but the spectral peak for the third stage are present in both
spectra. Figure 2.14 shows a typical nonlinear eect too: often when two or more stages coexist, peaks
Re 400
0 20 40 60 80
0
0.02
0.04
0.06
0.08
0.1
f [Hz]
a
m
p
U

[
V
]
Re 700
0 50 100 150
0
0.05
0.1
0.15
0.2
f [Hz]
a
m
p
U

[
V
]
f
1
f
2
f
3
f
3
f
2
2f
1
f
2
f
1
f
1
Figure 2.14: Stage III edge tone oscillation in the experiments: left side: Re 400; right side: Re 700;
upper line: photo taken of the visualised ow; bottom line: the spectrum of the output signal of the
pressure transducer; f
n
is the frequency of the n
th
stage, n = 1, 2 or 3
38
2.3. RESULTS
appear in the spectra at the dierence or at the sum of the frequencies of these stages. At Re 400 in the
spectrum of the output signal of the pressure transducer a peak at the dierence of the frequencies of the
second and the rst stages (f
2
f
1
) can be observed. This was observed experimentally by Brackenridge
and Nyborg [15] and also by Lucas and Rockwell [19].
Within each stage the frequency of oscillation was found to be a linear function of the (mean) exit velocity
of the jet:
f a
1
+ a
2
u (2.42)
Thus the Strouhal number of each stage follows as:
St (Re) =
a
1

u
+ a
2
=
a
1

2
/
u /
+ a
2
=
c
Re
+ St
(2.43)
The values of c and St
for the dierent stages deduced by the method of least squares from the results
of the experiments and from the CFD simulations are collected in Table 2.7 for all the three stages
(separating the results of the pure rst stage oscillation and the results of the rst stage of a multi stage
coexistence mode).
Table 2.7: Coecients of equation (2.43) for the top hat edge tone
Stage I
Stage II Stage III
pure multi
c St
c St
c St
c St
CFD 0.7387 0.04010 1.079 0.03541 4.072 0.1034 13.76 0.1740

Exp. 1.150 0.04522 0.6008 0.03775 1.800 0.1001 1.841 0.1608
Figure 2.15 summarizes the results of the computational and experimental investigations. The
Strouhal numbers are plotted against the Reynolds numbers for the three stages of the edge tone with a
top hat jet. Browns semi-empirical formulae are also plotted in the gure. No error bars are plotted here
to avoid overcrowding the gure.
Because of the negative c values the rst term of the Strouhal number (c/Re) is negative and
hyperbolically tends to zero as Re grows. Thus as presented in Figure 2.15 the Strouhal numbers of
the stages rst increase in the low Reynolds number region then they are nearly constant. Although in
some cases the value of c for the CFD and for the experimental results diers signicantly, the dierence
in the Strouhal numbers are within accuracy as seen in Figure 2.15. This can be because the denominator
(Re) of the rst term is at least an order of magnitude larger than the dierence in the values of c. The
dierence of St
is also signicant in the pure rst stage mode, but there Re is in the range of 60 250,
thus c/Re is non-negligible part of the Strouhal number indeed.
Parabolic prole
When changing the mean exit velocity of the jet, the parabolic prole jet-edge conguration shows
a remarkably dierent behaviour from the top hat jet-edge conguration. At the beginning at low
39
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Re []
S
t

[
]

Stage I CFD
Stage II CFD
Stage III CFD
Stage I Exp.
Stage II Exp.
Stage III Exp.
Stage I Brown
Stage II Brown
Stage III Brown
Figure 2.15: Reynolds number dependence of the Strouhal number; top hat prole, h/ 10
Reynolds numbers similar behaviour can be observed: after reaching a lower Reynolds number limit
(Re = 50 in the CFD simulations and Re 85 in the experiments) the rst stage edge tone oscillation
sets in.
Then increasing the Reynolds number the second stage appears together with the rst stage at
Re = 150 in the simulations and Re 180 in the experiments, both of them coexisting. As in the top hat
case, the frequency of the rst stage drops. In this case the magnitude of the drop is about 9.7 2.2 %
that is somewhat smaller compared to the top hat case. This drop is only observable from the results of
the experiments as in the CFD simulations the rst stage disappeared soon after the onset of the second
stage, and thus there were not enough data for the frequency of the rst stage from multi stage operation
mode to reliably detect the frequency drop. Further increasing the Reynolds number a qualitatively
dierent behaviour from the top hat case can be observed: from Re = 300 in the CFD simulations and
Re 360 420 in the experiments the rst stage disappears and a pure second stage oscillation can be
found. This is the highest stage found with the help of CFD simulations.
Further increasing the Reynolds number in the experiments (at Re 650) the second stage disappears
and the third stage of the edge tone come into being, but at the same time the rst stage appears again,
both coexisting. At the highest Reynolds numbers a pure third stage oscillation was observed (from
Re 900).
Similar behaviour can be observed when decreasing the Reynolds number but having stage jumps at
40
2.3. RESULTS
dierent Reynolds numbers, and without a pure second stage oscillation. Figure 2.16 shows the above
described hysteresis in the stages of the edge tone. It can be observed that in the Reynolds number
350 600 region when increasing the Reynolds number there is a pure second stage oscillation (the
rst stage disappeared) but when decreasing the Reynolds number (coming from a third stage oscillation
superposed on the rst stage) the coexistence of the third and the rst stages lasts longer, and there is no
pure second stage oscillation at all. The results of the CFD simulations are also plotted in Figure 2.16(a).
To precisely allocate the locations of stage jumps and investigate the above-mentioned hysteresis the
measurements were repeated several times. Thus also the repeatability of the measurement was veried
by having the same Strouhal numbers for the same Reynolds number (within the accuracy of the mea-
0 200 400 600 800 1000 1200 1400 1600
0
0.05
0.1
0.15
0.2
(a)
Re []
S
t

[
]

Up Stage I
Up Stage I & II
Up Stage II
Up Stage I & III
Up Stage III
CFD Stage I
CFD Stage II
0 200 400 600 800 1000 1200 1400 1600
0
0.05
0.1
0.15
0.2
(b)
Re []
S
t

[
]

Down Stage III
Down Stage I & III
Down Stage I & II
Down Stage I
Figure 2.16: Hysteresis in the stages of the edge tone; parabolic prole, h/ = 10; (a) increasing Reynolds
number, (b) decreasing Reynolds number
41
surements). To avoid overcrowding the gure results from only one measurement series are presented in
Figure 2.16. The appearance of the rst and second stage was observed always around the same Reynolds
numbers (Re 85 and Re 180), but the values where the rst stage disappeared (at Re 360 420),
when the rst - third stage coexistence evolved (Re 550 650) and when the rst stage disappeared
again (Re 650 1000) had much more uncertainty.
Within each stage just as in the top hat case the frequency of oscillation is a linear function of
the mean exit velocity of the jet. Thus the Strouhal numbers can be written as in equation (2.43):
St (Re) = c/Re +St
. The values of c and St
deduced by the method of least squares for the dierent

stages can be found in Table 2.8. Contrary to the top hat case the value of c is not negative for all the
stages. Although in the experiments it was found that c is positive for the third stage, the rst term
(c/Re) of equation (2.43) is in this case at least two orders of magnitude smaller than the second term,
thus practically it means that the Strouhal number is constant.
In the case of CFD simulations the results of the rst stage were not separated to pure and multi
parts because the rst stage disappeared too early. Therefore the curve tted to these points are between
the curves of the experimental results for the pure rst stage and the rst stage of a multi stage coexistence
modes.
Table 2.8: Coecients of equation (2.43) for the parabolic edge tone
Stage I Stage II Stage III
c St
c St
c St
pure multi pure multi

CFD 0.7894 0.04740 3.094 0.1194 - -
Exp. 1.169 0.6269 0.05070 0.04435 3.012 0.1320 1.352 0.1854
Comparison: top hat vs. parabolic velocity proles
The following dierences were found in the behaviour of the two cases:
- In the parabolic case the rst and the second stages appear at around the same Reynolds number
as in the top hat prole case, but in the CFD simulations no third stage was found at all and in the
experiments it appears only at a much higher Reynolds number: the third stage comes into being
at Re 600 in the parabolic case and at Re 250 in the top hat case.
- In the case of the top hat prole, when the Reynolds number is increased and a new stage appears,
the old, lower stage exists further and in most of the cases the two or three stages are superposed
on each other. Since the stage frequencies are not harmonically related to each other, the ensuing
motion is strictly speaking not periodic. There is no sign at higher Reynolds numbers that the lower
stage would tend to disappear. Contrary to this, in the parabolic case the lower stages disappear
when a higher stage appears thus pure second and pure third stage oscillations can also be found.
- In the parabolic case almost everywhere a strong frequency component at
1
/3
rd
of the fundamental
can be detected and several of its multiples (e.g. Figure 2.26(b)). This nding agrees strikingly
42
2.3. RESULTS
with the experimental ndings of Lucas and Rockwell [19] and Kaykayoglu and Rockwell [37] who
reported exactly the same.
- For the parabolic prole the frequency of the oscillation (and thus the Strouhal number) is about
15 20 % larger relative to the top-hat prole. Segoun et al. [23] found experimentally that the
top hat prole produces about 50 % higher frequencies than the parabolic prole with the same
maximum velocity. Comparing their results of top hat and parabolic edge tones at equivalent
Reynolds numbers (based on the mean velocity) turns out that the frequency of oscillation is
almost equal at lower Reynolds numbers (when the oscillation sets in) but increases faster in the
parabolic case resulting in 10 20 % higher frequencies.
- In the parabolic case the rms value of the pressure uctuations rose about 35 % relative to the
top-hat prole case.
- With a constant pressure and zero velocity initial condition the initial transient (the time until the
steady state oscillation is reached) was about half as long for the parabolic as for the top hat case.
This was only investigated with CFD.
Some of these ndings (but not all) can be tentatively explained by the fact that about 20 % more
momentum and about 54 % more energy is injected into the system in the case of a parabolic prole for
the same Reynolds number than in the top hat prole case.
The momentum and the energy that is injected to the system is proportional to
_
/2
/2
u(y)
2
dy and
_
/2
/2
u(y)
3
dy, respectively. Therefore if the Reynolds number were based on the quadratic mean value
_
qmv, u
qmv
=
_
1
_
/2
/2
u(y)
2
dy
_
or the cubic mean value
_
cmv, u
cmv
=
3
_
1
_
/2
/2
u(y)
3
dy
_
of the
inlet velocity instead of the average (mean) velocity
_
avr, u
avr
=
1
_
/2
/2
u(y) dy
_
, the same amount of
momentum or energy would be injected into the system at the same Reynolds number independently of
the velocity prole.
The qmv of the u(y) =
3
2
u
_
2
4
y
2
_
4
2
parabolic prole that has an average velocity of u is
u
qmv
1.095u, therefore if based on that, compared to the conventional denition, the Reynolds number
would increase by a factor 1.095 and the Strouhal number would decrease by a factor of 1.095. The cmv
of the same parabolic prole is u
cmv
1.156 u, therefore using that the Reynolds number would increase
by a factor 1.156 and the Strouhal number would decrease by a factor of 1.156.
Figure 2.17 shows the Reynolds number dependence of the Strouhal numbers when the average
velocity, the qmv or the cmv are used for the Reynolds and Strouhal numbers. Without going into detail
for its reason it can be noted that it seems that if the Reynolds and Strouhal numbers would be based
on the cubic mean value of the exit velocity of the jet then a top hat and a parabolic edge tone with the
same Reynolds number would give nearly the same Strouhal number. This would indicate the importance
of the energy equivalence, rather than the momentum equivalence.
2.3.2 Varying the nozzle-to-wedge distance
The Reynolds number was kept constant while the nozzle-to-wedge distance was varied. Exact values
and intervals of the parameters can be found in Table 2.9. In all cases qualitatively the same behaviour
43
0 200 400 600 800 1000 1200 1400 1600 1800
0
0.05
0.1
0.15
0.2
Re []
S
t

[
]
(a) Re and St based on u
avr
=
1
/2
/2
u(y) dy
0 200 400 600 800 1000 1200 1400 1600 1800
0
0.05
0.1
0.15
0.2
Re []
S
t

[
]
(b) Re and St based on u
qmv
=
/2
/2
u(y)
2
dy
0 200 400 600 800 1000 1200 1400 1600 1800
0
0.05
0.1
0.15
0.2
Re []
S
t

[
]
(c) Re and St based on u
cmv
=
3
/2
/2
u(y)
3
dy

top hat Stage I
parabolic Stage I
top hat Stage II
parabolic Stage II
top hat Stage III
parabolic Stage III
Figure 2.17: Inuence of the velocity averaging on the Strouhal number Reynolds number relationship
44
2.3. RESULTS
Table 2.9: Values of the Reynolds numbers and and ranges of h/ values during the h/ value dependence
studies
top hat parabolic
CFD Exp. Exp.
Re 350 189, 326 and 380 192, 384, 586 and 911
h/ range 0 15 0 16 0 17
could be observed. The boundary of the stages are a bit dierent for dierent Reynolds numbers and
in the parabolic case higher stages can be present purely just like it was found in the Reynolds number
dependence study.
In the top hat case the following behaviour was observed (Figure 2.18)
- The rst stage appears at around h/ 3 4 (lower values at higher Reynolds numbers).
- At around h/ 711 the second stage appears together with the rst stage and at the same time
a slight drop of the trend of the Strouhal number of the rst stage can be observed.
- In the experiments for the higher Reynolds numbers (Re 326 and 380) increasing the distance at
h/ 11 13 the third stage appears parallel to the rst and the second stages.
- Further increasing the value of h/ the second stage disappears the rst and third stage coexists.
0 2 4 6 8 10 12 14 16 18 20
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
h/ []
S
t

[
]

Stage I pure CFD
Stage I multi CFD
Stage II CFD
Stage I pure Exp.
Stage I multi Exp.
Stage II Exp.
Stage III Exp.
Figure 2.18: Dependence of the Strouhal number on the dimensionless nozzle-to-wedge distance; top hat
prole, Re = 350 and Re 380 for the CFD and experimental results, respectively
45
Figure 2.18 shows the Strouhal numbers from the experiments at Re 380 and from the CFD simulations
at Re = 350. The agreement is excellent, especially if it is kept in mind, that the Strouhal number at a
xed h/ value still grows a bit in the Re = 350380 region. The experiments at the other two Reynolds
number result in quite the same trend.
In the parabolic case several kinds of stage constellation can be observed:
- The rst stage sets in at around the same h/ value as in the top hat case (h/ 4 5, lower
values at higher Reynolds numbers).
- At around h/ 6 10 the second stage appears next to the rst stage and at the same time a
slight drop of the trend of the Strouhal number of the rst stage can be observed.
- Increasing the distance between the nozzle and the wedge sometimes but not always a pure second
stage oscillation can be found.
- At the onset of the third stage (at around h/ 9 17) rst it coexists with the rst and the
second stages, then later the second stage disappears and the rst and third stages remain.
- Further increasing the value of h/ a pure third stage oscillation can be observed.
- For the highest distances and Reynolds numbers combination (at h/ 15 17 and Re 570
and 890) the fourth stage also appeared purely or coexisting with the third stage.
The presence of hysteresis was investigated in the parabolic case at Re 380. No hysteresis was found,
the stage jump occurred at the same wedge position when increasing and when decreasing the distance
between the nozzle and the wedge.
The results of all of the numerical and experimental investigations at dierent Reynolds numbers show
that the frequency of oscillation is a linear function of 1/h.
f b
1
1
h
+ b
2
(2.44)
Thus the Strouhal number (at a xed Reynolds number):
St
_
h
_
=
b
1

h u
+
b
2

u
= d
1
h/
+ St
(2.45)
Values of d and St
(deduced by the method of least squares) are presented in Table 2.10 for the results
of the top hat case with Re = 350 (CFD simulations) and Re 380 (experiments). Results are only
presented in these two cases as the results of the other cases (at dierent Reynolds numbers and/or with
parabolic proles) are similar to this and the aim of this subsection is to show the linear dependence of
the Strouhal number on the reciprocal of the dimensionless nozzle-to-wedge distance and the agreement
between the CFD simulations and the experiments. Figures with the h/ dependence of the Strouhal
numbers at other Reynolds numbers will be shown later in Figures 2.20 and 2.21 in the next subsection
with an universal St (Re; h/) function that ts all the results of the top hat or the parabolic cases.
The value of St
is not negligible as it is about 510 % of the rst term at h/ = 10. This somewhat
explains the uncertainty about the value of k described in the Introduction (Section 1.3). A pure
d
h/
function is not sucient to achieve a perfect t for St. In case of a negative St
value if a pure inverse

46
2.3. RESULTS
Table 2.10: Coecients of equation (2.45) for the top hat edge tone at Re = 350 for the CFD simulations
and Re 380 for the measurements
Stage I
Stage II Stage III
pure multi
d St
d St
d St
d St
CFD 0.4079 0.002681 0.4288 0.003141 0.9174 0.000980 - -

Exp. 0.4259 0.003732 0.4080 0.004638 0.9975 0.003446 1.755 0.01815
proportional function ts well the Strouhal numbers at medium h/ values then it underestimates the
Strouhal numbers at lower h/ and overestimates the Strouhal numbers at higher h/ values. This can
be somewhat balanced if the exponent of h/ in the denominator is higher than 1.
Figure 2.19 shows the third stage of the edge tone in the measurements at Re 380 with three
tted curves. Although all the three curves are within the measurement accuracy it still can be observed,
that the red one (that has the linear formula presented in equation (2.45)) follows best the trend of the
measured points. While the green curve that has a form of d/ (h/)
k
with k = 1 over/underestimates
at the two ends of the dataset with approximately 3 % while this is somewhat corrected with k = 1.22
(the value suggested by Jones [13] blue curve).
10 11 12 13 14 15 16 17
0.08
0.1
0.12
0.14
h/ []
S
t

[
]

Exp.
1.755/(h/)0.01815
1.51/(h/)
2.677/(h/)
1.22
Figure 2.19: Strouhal numbers of the third stage top hat edge tone at Re 380 and h/ = 11 16; Best
t curves of type d/(h/)
k
, k = 1, 1.22 and d/(h/) + St
are plotted for comparison

2.3.3 St (Re; h/)
It was found that the frequency of oscillation at a xed geometry (xed h value) is a linear function of
the mean exit velocity of the jet (u) (equation (2.42)) and at a xed u is a linear function of
1
h
(equation
47
(2.44)). These suggest that the frequency is a bilinear function of u and
1
h
. It was found that instead
of the most general bilinear form (with four parameters: f (u, h) = p
1
+ p
2
u + p
3
1
h
+ p
4
u
1
h
) the
following somewhat more specic form also ts perfectly with only three parameters:
f (u, h) = p
1
(u + p
2
)
_
1
h
+ p
3
_
(2.46)
Calculating the Strouhal number:
St (u, h) =
f
u
= p
1
_
1 +
p
2
u
_
_
1
h/
+ p
3

_
=
=
_
p
1
+
p
1
p
2
/
u /
__
1
h/
+ p
3

_
(2.47)
This formula has the same form as that suggested by Brown [1] or by Brackenridge [16].
Because of similarity rules, if the width of the slit on the nozzle changes but the dimensionless
nozzle-to-wedge distance and the Reynolds number are the same, the ow should behave similarly, i.e.
the Strouhal number should be the same. This was veried by CFD simulations with an increased jet width
( = 3.2 mm) but same dimensionless nozzle-to-wedge distance (h/ = 10) at a few Reynolds numbers
(Re = 150, 200, 250 and 300). The Strouhal numbers of these simulations are collected in Table 2.11
together with the Strouhal numbers at these Reynolds numbers of the = 1 mm case. It was found that
the second stage sets in at the same Reynolds number and the Strouhal numbers agree well.
Table 2.11: Strouhal numbers of the edge tone with nozzles that have = 1 mm or = 3.2 mm wide
slit on them but having geometric conguration with the same h/ = 10 dimensionless nozzle-to-wedge
distance
St [-]
Re [-] = 1 mm = 3.2 mm
Stage I Stage I
150 0.0345 0.0345
200 0.0362 0.0352
St [-]
Re [-] = 1 mm = 3.2 mm
Stage I Stage II Stage I Stage II
250 0.0375 0.0849 0.0369 0.0856
300 0.0328 0.0900 0.0361 0.0901
Therefore as long as the dimensionless nozzle-to-wedge distance is kept constant the width of the slit
on the nozzle itself should not have any inuence on the Strouhal number of the stages of the edge tone,
thus the Strouhal number can be written as the following:
St
_
Re,
h
_
=
_
c
1
c
2
Re
_
_
1
h/
c
3
_
(2.48)
The results of the computational and the experimental investigations dier within their accuracies
therefore curve tting was carried out on the joint dataset. In these cases the built-in nonlinear least
48
2.3. RESULTS
square method of MatLab was used for the curve tting. Table 2.12 shows the coecients of the best t
curves together with the coecient of determination (R
2
value) of the t. (The coecient of determination
of a y = f (x) function tted to a set of x
i
, y
i
i = 1, . . . , n values shows the goodness of the t. It is
dened as R
2
= 1
n
i=1
(y
i
f(x
i
))
2
n
i=1
(y
i
y)
2
, where y =
1
n
n
i=1
y
i
.) Its value except for the rst stage of the
multi-stage mode case is always higher than 0.95 thus the t is good enough. Even in that case it is still
not lower than 0.9, thus is acceptable. In this case the rms value of the relative dierence between the
measured Strouhal numbers and the tted curve is about 6 %, and can be explained as a consequence of
the dierence between the Strouhal numbers of the experiments and the CFD simulations found during
the Reynolds number dependence study (Figure 2.12) that is in all other cases much lower.
Table 2.12: Coecients of the St(Re, h/) formula equation (2.48)
c
1
[-] c
2
[-] c
3
[-] R
2
Top hat
Stage I pure 0.4837 12.31 0.005461 0.9941
Stage I multi 0.4167 0.2292 0.01426 0.9015
Stage II 1.066 27.11 0.004157 0.9614
Stage III 1.884 19.96 0.01261 0.9934
Brown [1]
Stage I 0.4659 12.06 0.007
Stage II 1.072 27.74 0.007
Stage III 1.77 45.83 0.007
Parabolic
Stage I pure 0.5230 11.08 0.004836 0.9953
Stage I multi 0.5029 6.6451 0.01417 0.9832
Stage II 1.177 37.15 0.002273 0.9786
Stage III 1.972 6.954 0.007792 0.9916
Stage IV 2.365 55.21 0.000999 0.9982
Brown used the same c
3
= 0.007 parameter for all the three stages, while in my case it varies from
stage to stage. The values of the parameters of the pure rst and the second stages agree well (Brown
did not publish results for the rst stage of a multi stage mode). Although the values for the third stage
seem to be a bit dierent, if it is kept in mind that Brown observed the third stage Reynolds numbers
above 900, then it can be concluded that the c
2
/Re part is almost negligible compared to c
1
and the
dierence in c
3
compensates the dierence in c
1
. At h/ = 10 both Browns and my formula gives 0.1646
if the c
2
/Re part is neglected.
In the case of the fourth stage of the parabolic edge tone, the three parameters of the curve are
determined from only four observations, therefore although it ts the results it should be treated with
caution. This also explains the qualitative dierence of this case compared to the other cases (i.e. only in
this single case c
2
is negative).
Figures 2.20 and 2.21 show the values of the Strouhal number from the CFD and the experimental
investigations together with the tted curves in the top hat and in the parabolic cases, respectively.
49
0
2
0
0
4
0
0
6
0
0
8
0
0
1
0
0
0
1
2
0
0
0
0
.
0
5
0
.
1
0
.
1
5
(
a
)

S
t
(
R
e
,

h
/

1
0
.
2
6
)
,

E
x
p
.

0
5
1
0
1
5
2
0
0
0
.
0
5
0
.
1
0
.
1
5
(
b
)

S
t
(
R
e

1
8
9
,

h
/
)
,

E
x
p
.
0
5
1
0
1
5
2
0
0
0
.
0
5
0
.
1
0
.
1
5
(
c
)

S
t
(
R
e

3
8
0
,

h
/
)
,

E
x
p
.
0
5
1
0
1
5
2
0
0
0
.
0
5
0
.
1
0
.
1
5
(
d
)

S
t
(
R
e

3
2
6
,

h
/
)
,

E
x
p
.
0
5
0
0
1
0
0
0
1
5
0
0
2
0
0
0
0
0
.
0
5
0
.
1
0
.
1
5
(
e
)

S
t
(
R
e
,

h
/

=

1
0
)
,

C
F
D
0
5
1
0
1
5
2
0
0
0
.
0
5
0
.
1
0
.
1
5
(
f
)

S
t
(
R
e

=

3
5
0
,

h
/
)
,

C
F
D
S
t
a
g
e

I

p
u
r
e
S
t
a
g
e

I

m
u
l
t
i
S
t
a
g
e

I
I
S
t
a
g
e

I
I
I
F
i
g
u
r
e
2
.
2
0
:
S
t
r
o
u
h
a
l
n
u
m
b
e
r
s
o
f
t
h
e
t
o
p
h
a
t
e
d
g
e
t
o
n
e
.
C
r
o
s
s
e
s
w
i
t
h
e
r
r
o
r
b
a
r
s
d
e
n
o
t
e
t
h
e
n
u
m
e
r
i
c
a
l
(
C
F
D
)
o
r
e
x
p
e
r
i
m
e
n
t
a
l
(
E
x
p
.
)
r
e
s
u
l
t
s
.
S
o
l
i
d
l
i
n
e
s
a
r
e
t
h
e
b
e
s
t
t
c
u
r
v
e
s
d
e
s
c
r
i
b
e
d
i
n
e
q
u
a
t
i
o
n
(
2
.
4
8
)
w
i
t
h
c
o
e
c
i
e
n
t
s
i
n
T
a
b
l
e
2
.
1
2
50
2.3. RESULTS
0
5
0
0
1
0
0
0
1
5
0
0
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
(
a
)

S
t
(
R
e
,

h
/

9
.
7
2
)
,

E
x
p
.

0
5
1
0
1
5
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
(
b
)

S
t
(
R
e

1
9
2
,

h
/
)
,

E
x
p
.
0
5
1
0
1
5
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
(
c
)

S
t
(
R
e

3
8
4
,

h
/
)
,

E
x
p
.
0
5
1
0
1
5
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
(
d
)

S
t
(
R
e

5
8
6
,

h
/
)
,

E
x
p
.
0
5
1
0
1
5
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
(
e
)

S
t
(
R
e

9
1
1
,

h
/
)
,

E
x
p
.
0
5
0
0
1
0
0
0
1
5
0
0
2
0
0
0
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
(
f
)

S
t
(
R
e
,

h
/

=

1
0
)
,

C
F
D
S
t
a
g
e

I

p
u
r
e
S
t
a
g
e

I

m
u
l
t
i
S
t
a
g
e

I
I
S
t
a
g
e

I
I
I
S
t
a
g
e

I
V
F
i
g
u
r
e
2
.
2
1
:
S
t
r
o
u
h
a
l
n
u
m
b
e
r
s
o
f
t
h
e
p
a
r
a
b
o
l
i
c
e
d
g
e
t
o
n
e
.
C
r
o
s
s
e
s
w
i
t
h
e
r
r
o
r
b
a
r
s
d
e
n
o
t
e
t
h
e
n
u
m
e
r
i
c
a
l
(
C
F
D
)
o
r
e
x
p
e
r
i
m
e
n
t
a
l
(
E
x
p
.
)
r
e
s
u
l
t
s
.
S
o
l
i
d
l
i
n
e
s
a
r
e
t
h
e
b
e
s
t
t
c
u
r
v
e
s
d
e
s
c
r
i
b
e
d
i
n
e
q
u
a
t
i
o
n
(
2
.
4
8
)
w
i
t
h
c
o
e
c
i
e
n
t
s
i
n
T
a
b
l
e
2
.
1
2
51
2.3.4 Stage jumps and mode switching
As already discussed, at certain parameter values the edge tone jumps from a mode to another when
changing the Reynolds numbers or the h/ value. Near the boundaries a mode switch can also happen
without any external excitation or change in the parameters. The jump can be permanent (the new mode
stays stable) but it can also be temporary meaning that at the boundary the edge tone switches back
and forth randomly between the modes.
Figure 2.22 shows the measurement points on the Re h/ plane for the top hat and parabolic
cases. Each mark of the gure denotes a single measurement point and its colour corresponds to the
highest stage that was found in that point. The boundaries of the stages are often blurred and sometimes
hysteresis can be found as already shown in Section 2.3.1.
0 400 800 1200
0
5
10
15
20
Re []
h
/
]
(b)

Stage I Stage II Stage III Stage IV
0 400 800 1200
0
5
10
15
20
Re []
h
/
]
(a)
Figure 2.22: Measurement points on the Re h/ plane in the case of (a) top hat and (b) parabolic edge
tone; The colour of each point corresponds to the highest stage that was observed there.
With the help of a simple sliding window Fourier transformation technique it can be investigated if
the oscillation changes qualitatively in time (like mode switching during the observation). Only a T
long part starting at t
i
= i of the whole signal (e.g. time history of the force acting on the wedge in
the case of the CFD simulations or the amplied output signal of the pressure sensor in the case of the
experiments) is used to calculate the spectrum. This is done for i = 0, 1, . . ., and the spectrum of each
(i ; i + T) window is visualised in one contour plot (e.g. Figures 2.23(b) and 2.24). The colours in
a horizontal cut of the gure at an arbitrary t
i
= i height show the amplitude of the dierent frequency
components in the T long part of the force signal starting at t = t
i
.
Figures 2.23 and 2.24 show two examples of the many mode switches that were observed. Figure 2.23
shows the time history and the sliding window Fourier transformation of the pressure signal at a point on
the wedge surface of a CFD simulation (Re = 250, h/ = 10, top hat prole). The qualitative change of
52
2.3. RESULTS
(a) Time history of the pressure signal in a point on the wedge surface
0.3 0.35 0.4 0.45
4
2
0
2
4
t [s]
p

[
P
a
]
(b) Sliding window Fourier transformation of the pressure signal in a point on the wedge surface
f [Hz]
t

[
s
]
0 50 100 150 200 250 300 350 400
0.1
0.2
0.3
0.4
0.5
Figure 2.23: Mode switch from Stage I & II coexistence mode to a pure rst stage in a CFD simulation
at Re = 250, h/ = 10 with top hat jet prole
f [Hz]
t

[
s
]
0 50 100 150 200 250 300
4
5
6
7
8
Figure 2.24: Mode switch from a pure second stage oscillation to a Stage I & III coexistence mode in a
measurement at Re 790, h/ 9.72 with parabolic jet prole; Sliding window Fourier transformation
of the amplied output signal of the pressure sensor
53
the spectrum of the pressure signal in the t 0.3 0.4 s interval can clearly be observed. Before t < 0.3 s
the jet oscillates in a rst and second stage coexistence and then it jumps into a pure rst stage mode.
At the same time the frequency of the rst stage increases a bit. Figure 2.24 shows the sliding window
Fourier transformation of the pressure signal of a measurement (Re = 790, h/ 9.72, parabolic prole)
when the edge tone jumps from a pure second stage oscillation to a Stage I & III coexistence mode (at
around t 6 s).
2.3.5 Miscellaneous CFD results
The previous sections showed that the experimental and computational results agree well, thereby
validating the results of the CFD simulations. Thus further detailed investigation of the ow can be
carried out with the help of these simulations. E.g. the pressure distribution on the wedge wall or the
phase of the propagating disturbance between the nozzle and the wedge can be easily and cost-eectively
investigated.
Figures 2.25 and 2.26 (and gures in the Appendix) show miscellaneous results of simulations with
top hat and parabolic jet velocity proles, respectively at dierent Reynolds numbers at a xed geometric
conguration (h/ = 10). These gures hold the same subgures for the dierent simulations, namely:
- subgures (a) show a piece of the history of the y component of the force acting on the wedge;
- subgures (b) show the spectrum of this force history. (The initial transient part of the signal was
always omitted from the FFT and in the cases of higher stages an appropriate part of the signal
was chosen for FFT.);
- subgures (c) show the instantaneous point of force action;
- subgures (d) show the distribution of the rms of the pressure on the wedge wall;
- subgures (e) show the phase;
- subgures (f) show the development of the rms of the transversal velocity between the nozzle and
the wedge.
Development of the rms of the transversal velocity
The growth of the disturbance along the jet in the nozzle-wedge distance can be described with the
transversal velocity uctuation. Therefore the rms of the transversal velocity was investigated (subgure
(f) of Figures 2.25 and 2.26 or the gures in Appendix), and it was found that the pure rst stage
oscillation behaves somewhat dierently than the higher stage modes. In the former case a linear rise at
the beginning, a short constant region (from about x/h 0.4 to x/h 0.6) and then a much sharper rise
at the end can be observed. Whereas in the latter cases the rise at the beginning is rather exponential than
linear and much longer (ends at x/h 0.7) . The reason for the sharp rise near the edge is the acceleration
of uid between the vortex near the edge and the edge. Similar trends were observed experimentally by
Stegen and Karamcheti [33] and by Lucas and Rockwell [19].
54
2.3. RESULTS
0.26 0.2779 0.2957
0.04
0.02
0
0.02
0.04
t [s]
F
[
m
N
]
(a)
0 100 200 300 400
0
0.01
0.02
0.03
f [Hz]
a
m
p
F
[
m
N
]
(b)
0.26 0.2779 0.2957
11
13
15
17
19
21
t [s]
x
t
[
m
m
]
(c)
0 10 20 30
0
0.5
1
1.5
2
2.5
3
distance along the wedge surf. [mm]
p
r
m
s
[
P
a
]
(d)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/h [-]
/
2
[
-
]
(e)
0 1 2 3 4 5 6 7 8 9 10
0
0.2
0.4
0.6
0.8
1
x [mm]
v
s
t
d
[
m
s
1
]
(f )
Figure 2.25: Miscellaneous CFD results; Stage I at Re = 200, h/ = 10 with top hat prole; for description
of subgures see Section 2.3.5
55
0.06 0.0614 0.0628 0.0643
1.5
1.25
1
0.75
0.5
0.25
0
0.25
0.5
0.75
1
1.25
1.5
t [s]
F
[
m
N
]
(a)
0 1000 2000 3000 4000 5000
0
0.1
0.2
0.3
0.4
0.5
0.6
f [Hz]
a
m
p
F
[
m
N
]
(b)
0.06 0.0614 0.0628 0.0643
10
12
14
16
18
20
22
t [s]
x
t
[
m
m
]
(c)
0 10 20 30
0
10
20
30
40
50
60
70
80
p
r
m
s
[
P
a
]
(d)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/h [-]
/
2
[
-
]
(e)
0 1 2 3 4 5 6 7 8 9 10
0
1
2
3
4
5
6
7
x [mm]
v
s
t
d
[
m
s
1
]
(f )
Figure 2.26: Miscellaneous CFD results; Stage II at Re = 800, h/ = 10 with parabolic prole; The rst
peak in the spectrum (subgure (b)) are exactly at
1
/3
rd
of the second peak that corresponds to Stage II;
for description of subgures see Section 2.3.5
56
2.3. RESULTS
Instantaneous point of force action (subgure (c)), pressure distribution on the wedge wall
(subgure (d)), rms of the force acting on the wedge
Subgure (d) shows the distribution of the standard deviation of the pressure on the wedge wall. All
Reynolds numbers display a similar pressure distribution except for the magnitude. The sharp peak in
the immediate vicinity of the wedge tip is followed by gradual drop, a second, atter peak, and nally a
slowly decreasing long region. In the rst stage some spatial oscillation of the pressure can be observed
in the decreasing region which smooths out in the higher stages. If we take a look at the time-resolved
distribution of pressure along the wedge surface (not shown here), we see a regular wave pattern running
along the wedge over the period but with the same characteristic peak at each instant near the tip. This
distribution is basically similar to that Kaykayoglu and Rockwell [38] found experimentally. Kaykayoglou
and Rockwell found that after a medium value the pressure reaches a sharp maximum close to the wedge
tip and afterwards the pressure decreases roughly as x
1/2
. They measured the pressure only in a few
points. The basic features are reproduced with the sharp peak close to the edge tip. Since the spatial
resolution of these CFD simulations are much higher, more details can be seen in these subgures than
in Kaykayoglu and Rockwell [38]. The distribution can be represented rather with a piecewise linear than
with a power function.
Another dierence is that there the total streamwise length of the wedge was 0.8 h, whereas here
much longer, 7.5 h. Since the pressure is non-negligible along almost the whole length of the wedge, the
point of force action (x
F
) is much further behind the tip in our case than in theirs. This point is important
because this point should be related to the eective location of the acoustic dipole. The instantaneous
value of x
F
is calculated as:
x
f
(t) =
M
z
(t)
F
y
(t)
(2.49)
whereas its mean value (x
F
) is calculated as:
x
F
=
M
z,rms
F
y,rms
(2.50)
The temporal history of x
F
is shown in subgure (c) and we can see that the point of force action remains
in a narrow range most of the cycle. Every half cycle we see singularities that are physically not realistic;
they indicate that the y component of the force gets zero twice in a cycle. The location of x
F
denoted
with dashed line in subgure (d) is remarkably stable both as a function of the Reynolds number and
of h (lies always between 16 and 20 mm from the tip), so that is largely independent of the details of
the ow as long as the wedge geometry is unchanged. For comparison: Kaykayoglou and Rockwell found
roughly x
F
/h = 1/4 with a much shorter wedge, whereas here it is 1.6 2 measured from the tip (in the
h = 10 mm case).
The dependence of F
rms
on the mean exit velocity of the jet (or Reynolds number) is perfectly
parabolic for both of its components. This can be explained in the following way. The force is proportional
to the dynamic pressure on the surface; this is proportional to the square of the velocity perpendicular
to the wedge surface. This again must scale with the inlet velocity.
57
Convection velocity and wavelength of the disturbance
It is crucial in the understanding of the exact mechanism of the edge tone oscillation to determine
the velocity of the disturbance propagation along the jet. In the literature a theoretical value of 0.5
times the mean exit velocity is given [39]. The theory is with an assumption of an inviscid parallel jet and
corresponds to the phase velocity of the most unstable frequency disturbance. Experimental values scatter
around 0.3 0.6 times the mean exit velocity (e.g. Curle [3] found values around 0.3 0.4, Brown [1]
reported values around 0.4, Kwon [11] found 0.5 0.6). The expression phase velocity is meaningful
only if there is only one pure sinusoidal stage present or if the system is non-dispersive; when several
modes are superposed on each other, each mode propagates with a dierent velocity or the various modes
might interact with each other in an unknown way. In a multi-stage operation a disturbance convection
velocity can be identied as a group velocity rather then a phase velocity; or, alternatively the phase
velocity can be determined for each mode separately. Here only the single mode case will be considered.
The convection velocity of the disturbance was determined with a cross-correlation technique. The
instantaneous transversal component of the velocity at several points between the nozzle and the tip of
the wedge were compared to one of them chosen as the reference point (usually approximately at 0.6 h).
The signal of the rst point was shifted by i time steps, and then the correlation coecient (R)etween
the shifted signal of the rst point and the reference signal was calculated. The phase delay (relative to
the reference point) is determined as = i
max
t
2
T
, where i
max
is the shift value where R reaches
its maximum, t is the time step of the two signals and T is the period time. Figure 2.27 shows the
transversal component of the velocity at the reference point (blue line) and at a distance 0.25 h away
from the nozzle (red line). The dashed red line corresponds to the latter signal shifted by the time lag
(determined as described above) between the two signals.
Subgure (e) in Figures 2.25 and 2.26 (and in the gures in the Appendix) shows the phase delay
0.2 0.205 0.21 0.215 0.22 0.225 0.23 0.235 0.24 0.245
0.2
0
0.2
t [s]
u
y
[
m
s
1
]

x/h = 0.25 x/h = 0.25, shifted reference
Figure 2.27: Transversal velocity signals used for the cross correlation of a top hat edge tone ow with
Re = 150 and h/ = 10
58
2.3. RESULTS
divided by 2 (
=

2
). The absolute numbers on the vertical axis are not important thus the starting
point of the curves were adjusted to 0. As becomes apparent, these curves are universal in the sense that
they apply for every rst stage single-mode Reynolds number (only results of simulations with h = 10 mm
were used for this investigation). The phase delay is parabolic, the tted functions dier negligibly in the
case of dierent Reynolds numbers. Fitting a parabola to all the points together results in:
_
x
h
_
=
_
x
h
_
2
0.6036
_
x
h
_
2
+ 0.2781
x
h
(2.51)
This agrees almost perfectly with the power function

2
= 0.9
_
x
h
_
1.63
suggested by Stegen and
Karamcheti [33] after measuring the phase at a Re 950 and h/ 5.58. Figure 2.28 shows the two
curves together with the result of the CFD simulations at Re = 200, h/ = 10 with a top hat jet.
In the nozzle-to-wedge distance the phase drops by almost a full period. The acoustic wave of the
dipole sound source that excites the jet and generates the next instability wave reaches the nozzle
immediately, therefore the fact that the phase drop does not reach a full period in the nozzle-to-wedge
distance means that the dipole sound source is somewhat behind the tip of the wedge.
The derivative of the inverse function of T
(where T is the period) multiplied by h yields the

phase velocity that is actually the convection velocity of the disturbance:
u
con
=
h
T
(2.52)
Thus, its relative value follows as:
u
con
u
=
h
T u

1
= St
h
(2.53)
CFD simulations of rst stage edge tone oscillation with h/ = 10 were investigated only, but both with
top hat and parabolic proles. The former one results in Strouhal numbers from 0.028 to 0.038 while
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.2
0.4
0.6
0.8
1
x/h [-]
/
2
[
-
]
Figure 2.28: Phase drop of a rst stage edge tone oscillation; Results from CFD simulation at Re = 200,
h/ = 10, top hat prole (X); Fitted parabola (equation (2.51), solid line) and power function suggested
by Stegen and Karamcheti [33] (dashed line)
59
the latter from 0.032 to 0.044. Therefore the relative values of the convection velocities are very high
(1 1.58) at the orice and they continuously and rapidly decrease further downstream to values of
0.19 0.3. The initially high disturbance convection velocities can be explained so that the disturbances
have not developed there yet instead the jet moves rather like a solid stick. Since there is a continuous
change of the convection velocity it makes no sense to talk about wavelength since within one wavelength
the wavelength changes. The average relative convection velocity values are between 0.32 0.41 and
0.43 0.5 in the top hat and parabolic cases, respectively. This agrees well with the value found by
Brown [1] but is also not far from the theoretical value of Mattingly and Criminale [39].
2.4 Summary
The ow of the edge tone have been investigated both by numerical and experimental means. In the
experiments four methods were used to determine the oscillation frequency. Criteria to the value of the
time step were deduced analytically to keep the absolute or the relative numerical error at a constant
value of a reference case (Thesis #1). The planar nature of the ow was veried by comparing the results
of the 2D and the 3D CFD simulations and also experimentally by ow visualization. Parametric stud-
ies were carried out to determine how the Strouhal number of the oscillation depends on the Reynolds
number and on the dimensionless nozzle-to-wedge distance. The results of the CFD simulations agree
well with that of the experiments, the formulae describing the St (Re, h/) relationships in case of top
hat and parabolic edge tones were determined (Thesis #2). Several modes of the edge tone have been
observed in that stages of the edge tone are present purely or several stages coexist at the same time. In
some cases mode switching and hysteresis were also found at the boundary of the stages (Thesis #3). The
phase and convection velocity of the jet disturbance was investigated, and it was found that the phase
varies nonlinearly with the distance from the nozzle, thus the convection velocity is not constant as it is
assumed in all of the theoretical considerations (Thesis #4).
60
2.4. SUMMARY
2.4.1 Theses
Thesis #1
The edge tone phenomenon was produced both experimentally in the laboratory and by a commercial
Computational Fluid Dynamics (CFD) software (ANSYS-CFX) numerically.
1.1 The following four dierent methods for the measurement of the oscillation frequency of the edge
tone were compared:
- the vortex counting method;
- the wavelength method;
- stroboscope principle;
- the Fourier analysis of the signal of the pressure transducer;
I showed that except for the wavelength method the other three yields the same result within
their accuracy. In spite of its inaccuracy, the wavelength method is woth mentioning, because
it is able to measure the frequency of the edge tone oscillation from only two subsequent
photographs.
1.2 I have carried out a detailed mesh- and time step-dependence study in the Re = 200 , h/ = 10
case. I have deduced a formula with which the necessary time step to keep the absolute error of
the simulation on the reference level can be determined for other Reynolds numbers. As a result
I showed that in order to keep the absolute error at a constant value, the time step () has to
be changed in proportion to u
3
/2
. Furthermore I showed that if the relative error is to be kept
constant then it is sucient if u
1
, which taking into consideration that the f is nearly
proportional to u results in a constant period resolution.
Related publications: WoS journal: [55]; Non-WoS publications: [5962]
Thesis #2
Several formulae can be found in the literature describing the relationship between the dimensionless
parameters (Strouhal number (St), Reynolds number (Re) and dimensionless nozzle-to-wedge distance
(h/ value)) of the edge tone. The dierent formulae results in dierent Strouhal numbers at the same
Reynolds number and h/ values; at certain parameter values the relative dierence between the
Strouhal numbers can be up to 100 %.
The experimentally measured and numerically calculated Strouhal numbers (dimensionless
oscillation frequencies) at the same Reynolds number and dimensionless nozzle-to-wedge distance
agree within accuracy. My results verify Browns experimental results. The dependence of the Strouhal
number of the edge tone on the Reynolds number and the h/ value can be described by the following
formula, similar to that of Brown [1]:
St
_
Re,
h
_
=
_
c
1
c
2
Re
_
_
1
h/
c
3
_
,
where the values of c
1
, c
2
and c
3
constants are:
61
c
1
[-] c
2
[-] c
3
[-] R
2
Top hat
velocity prole
Stage I pure 0.4837 12.31 0.005461 0.9941
Stage I multi 0.4167 0.2292 0.01426 0.9015
Stage II 1.066 27.11 0.004157 0.9614
Stage III 1.884 19.96 0.01261 0.9934
Browns [1]
experimental results
top hat velocity prole
Stage I 0.4659 12.06 0.007
Stage II 1.072 27.74 0.007
Stage III 1.77 45.83 0.007
Parabolic
velocity prole
Stage I pure 0.5230 11.08 0.004836 0.9953
Stage I multi 0.5029 6.6451 0.01417 0.9832
Stage II 1.177 37.15 0.002273 0.9786
Stage III 1.972 6.954 0.007792 0.9916
Stage IV 2.365 55.21 0.000999 0.9982
I showed that if the Reynolds number is based on the cubic mean value of the velocity prole (that
guarantees energy-ux equivalence) instead of the mean velocity (that guarantees mass ow rate
equivalence) then the dierence between the Strouhal numbers of the top hat and parabolic edge
tones diminishes. From this it can be concluded that the attributes of the edge tone are determined
by the energy-ux.
Thesis #3
It is known from the literature that the appearance of the edge tone ow at some not well dened
parameter value changes qualitatively, the jet switches to another stage. Pure and coexisting stages
are both reported in the literature and sometimes also hysteretic behaviour in the stage jumps can be
found. I completed the observations of the literature with the following remarks:
3.1 I showed by experimental and numerical means that with top hat jets at the appearance of
the higher stages of the edge tone the lower stages do not disappear, the stages are coexisting
superposed on each other, while in the case of parabolic jets the second stage can be present
purely as well.
I showed by experiments in detail that in the case of parabolic jet prole the second stage rst
coexists with the rst stage then the rst stage disappears and the second stage is present purely.
Then at the advent of the third stage the rst stage again appears but the second stage disappears,
while at last the rst stage again disappears and the third stage is present purely.
3.2 I showed by experimental and numerical means that at certain parameter settings (at the stage
boundaries) the edge tone jumps from one stage to another without any external excitation. This
stage jump can be permanent (one-way) or it can be temporary meaning that the jet jumps back
and forth randomly between the two stages (mode switching).
62
2.4. SUMMARY
3.3 I showed by experiments that in case of parabolic jets increasing and then decreasing the Reynolds
number the position of the stage jumps changes, hysteretic behaviour can be observed. No
hysteresis can be observed with top hat jets.
Related publications: WoS journal: [55, 56]; Non-WoS publications: [59, 60, 62, 65]
Thesis #4
Each existing theoretical model of the edge tone assumes that the disturbance of the jet travels with
a constant convection velocity. The value of this velocity is dierent in dierent models, typically
0.3 0.6, in most cases 0.4 0.5 times of the mean exit velocity of the jet is used.
I showed by numerical simulations that the phase of the jet disturbance () changes nonlinearly
with the distance measured from the nozzle, thus the convection velocity of the jet disturbance is not
constant between the nozzle and the wedge. I found that in the case of a rst stage edge tone the
phase changes independently of the Reynolds number and can be described as:
_
x
h
_
=
_
x
h
_
2
0,6036
_
x
h
_
2
+ 0,2781
x
h
From that the convection velocity of the disturbance follows as:
u
conv
_
x
h
_
u
=
h
T u

1
_
x
h
_ = St
h
_
x
h
_
The Strouhal number of a rst stage edge tone is between 0.028 0.044 , from that at h/ = 10
the convection velocity of the disturbance relative to the mean exit velocity of the jet near the nozzle
is 1 1.58, which decreases to a value of 0.2 0.3 as it reaches the wedge.
The mean value of the disturbance convection velocity relative to the mean exit velocity of the jet is
between 0.32 0.5.
Related publications: WoS journal: [55]; Non-WoS publications: [59, 60]
63
2.4.2 Tezisek
1. Tezis
Laboratoriumi kserleti berendezesen illetve kereskedelmi forgalomban elerheto numerikus aramlasszi-
mulacios (CFD Computation Fluid Dynamics) szoftverrel (ANSYS-CFX) eloalltottam az elhang
jelenseget.
1.1

Osszehasonltottam negy k ulonbozo modszert az elhang lengesi frekvenciajanak kserleti beren-
dezesen valo meresere. Ezek a kovetkezok voltak:
-

Orvenyszamolos modszer
- Hullamhosszos modszer
- Stroboszkop elv
- Nyomasjel Fourier analzise
Megallaptottam, hogy a negy modszer koz ul a ,,hullamhosszos modszert leszamtva minde-
gyik hibahataron bel ul ugyanazt az eredmenyt adja. Pontatlansaga ellenere a ,,hullamhosszos
modszer emltesre melto, mivel ezzel a modszerrel csupan ket egymast koveto fenykepbol merheto
a lenges frekvenciaja.
1.2 A CFD szimulaciokhoz reszletes halo- es idolepes- erzekenyseg vizsgalatot vegeztem a Re = 200,
h/ = 10 esetre.
Kidolgoztam egy kepletet, aminek segtsegevel tovabbi Reynolds szamok eseten meghataroz-
hato a numerikus szimulaciohoz sz ukseges idolepes nagysaga () ugy, hogy a szimulacio ab-
szol ut numerikus hibaja ne legyen nagyobb mint a referencia esete. Ennek eredmenyekeppen
megallaptottam, hogy az abszol ut hiba allando erteken tartasahoz az idolepest u
3
/2
arany-
ban kell valtoztatni. Tovabba megallaptottam, hogy a relatv hiba referencia erteken valo tartasa-
hoz a u
1
arany megtartasa is elegseges, ami gyelembe veve, hogy a lengesi frekvencia kozel
egyenesen aranyos a sebesseggel egy cikluson bel ul konstans szam u osztopontot eredmenyez.
Kapcsolodo publikaciok: WoS-os folyoirat: [55]; Nem-WoS-os publikaciok: [5962]
2. Tezis
A szakirodalomban szamos keplet talalhato az elhang dimenziotlan jellemzoinek (Strouhal szam,
Reynolds sz am, dimenziotlan f uvoka-ek tavolsag (h/ ertek)) kapcsolatara. A k ulonbozo kepletek
azonos Reynolds szamok es h/ ertekek eseten k ulonbozo Strouhal szamokat eredmenyeznek; bizonyos
parameter ertekeknel akar 100 %-os elteres is elofordulhat az adodo Strouhal szamok kozott.
Az azonos Reynolds szam es dimenziotlan f uvoka-ek tavolsag mellett a kserletekben mert es a nu-
merikus szimulacioval szamolt Strouhal szamok (dimenziotlan lengesi frekvenciak) hibahataron bel ul
megegyeznek. Eredmenyeim megerostik Brown 1937-es meresi eredmenyeit.
Az elhang Strouhal szamanak a Reynolds szamtol, illetve a h/ ertektol valo f uggeset a kovetkezo
Brown kepletehez [1] hasonlo keplet rja le:
St
_
Re,
h
_
=
_
c
1
c
2
Re
_
_
1
h/
c
3
_
,
ahol c
1
, c
2
es c
3
az elhang modusara jellemzo konstansok, ertekei:
64
2.4. SUMMARY
c
1
[-] c
2
[-] c
3
[-] R
2
Egyenletes
sebessegprol
Tiszta 1 modus 0,4837 12,31 0,005461 0,9941
,,Kevert 1. modus 0,4167 0,2292 0,01426 0,9015
2. modus 1,066 27,11 0,004157 0,9614
3. modus 1,884 19,96 0,01261 0,9934
Brown [1] meresi
eredmenyei,
egyenletes prol
1. modus 0,4659 12,06 0,007
2. modus 1,072 27,74 0,007
3. modus 1,77 45,83 0,007
Parabolikus
sebessegprol
Tiszta 1. modus 0,5230 11,08 0,004836 0,9953
,,Kevert 1. modus 0,5029 6,6451 0,01417 0,9832
2. modus 1,177 37,15 0,002273 0,9786
3. modus 1,972 6,954 0,007792 0,9916
4. modus 2,365 55,21 0,000999 0,9982
Megmutattam, hogy ha a Reynolds szamot a terfogataram-egyenloseget garantalo atlagsebesseg
helyett az energiaaram-egyenloseget garantalo kobos kozepertek alapjan denialjuk, ugy az egyenletes
es a parabolikus sebessegprol u elhang Strouhal szamai kozotti k ulonbseg minimalisra csokken, amirol
arra lehet kovetkeztetni, hogy a kialakulo elhang jelenseg tulajdonsagait az energiaaram hatarozza
meg.
Kapcsolodo publikaciok: WoS-os folyoirat [55]; Nem-WoS-os publikaciok: [5964]
3. Tezis
A szakirodalomban ismert jelenseg, hogy az elhang aramlasi kepe bizonyos nem jol denialt
parameter ertekeknel kvalitatve megvaltozik, a szabadsugar modust valt. A k ulonbozo modusok tiszta
illetve kevert osszetetele a szakirodalom szerint egyarant elofordul, illetve a modusvaltasokban neha
hiszterezis gyelheto meg. Ezeket a szakirodalomban fellelheto eszreveteleket a kovetkezo megallap-
tasokkal egesztettem ki:
3.1 CFD szimulacioval es meresekkel megmutattam, hogy egyenletes sebessegprol u szabadsugar
eseten a magasabb modus megjelenesevel az alacsonyabb modusok nem t unnek el, a modusok
szuperpozcioja valosul meg, mg parabolikus sebessegprol u szabadsugar eseten a masodik modus
tisztan is elofordul.
Meresekkel reszletesebben megmutattam, hogy parabolikus sebessegprol u szabadsugar eseten a
masodik modus eloszor az elso modussal egy utt jelenik meg, majd az elso modus elt unik es a
masodik tisztan is elofordul. Ezt kovetoen a harmadik modus megjelenesekor az elso modus ujra
megjelenik am a masodik elt unik, veg ul az elso modus ismet elt unik es a harmadik modus tisztan
is elofordul.
3.2 Numerikus szimulacioval es meresekkel egyarant megmutattam, hogy bizonyos bealltasok eseten
(modusvaltasok hatarainal) az elhang k ulso beavatkozas nelk ul ugrik egyik modusbol a masikba.
Ez a modusvaltas lehet permanens (egyirany u), de elofordul olyan eset is, amikor a modusvaltas
csak atmeneti es az elhang veletlenszer uen valtogat a ket modus kozott (,,mode switching).
65
3.3 Meresekkel megmutattam, hogy a parabolikus sebessegprol u szabadsugar eseten a Reynolds
szamot novelve, majd csokkentve a modusvaltasok hatara valtozik, hiszterezis gyelheto meg.
Egyenletes prol u szabadsugar eseten nem gyelheto meg hiszterezis.
Kapcsolodo publikaciok: WoS-os folyoirat: [55, 56]; Nem-WoS-os publikaciok: [59, 60, 62, 65]
4. Tezis
Az elhangjelenseg eddig felalltott osszes elmeleti modellje feltetelezi, hogy a szabadsugaron letrejott
zavaras konstans sebesseggel halad az ek fele. Ennek a sebessegnek az erteke modellenkent eltero,
tipikusan a szabadsugar atlagsebessegenek 0,3 0,6-szorosaval, legtobbszor 0,4 0,5-szorosevel sza-
molnak.
Numerikus szimulaciokkal kimutattam, hogy a szabadsugaron bekovetkezett zavaras fazisa nem-
linearis modon valtozik a f uvokatol mert tavolsag f uggvenyeben, ezert a zavaras terjedesi sebessege a
f uvoka es az ek kozott nem allando. Megallaptottam, hogy az elhang elso modusaban a fazis valtozasa
a Reynolds szamtol f uggetlen ul lerhato a kovetkezo keplettel:
_
x
h
_
=
_
x
h
_
2
0,6036
_
x
h
_
2
+ 0,2781
x
h
Amibol a zavaras relatv terjedesi sebessege:
u
conv
_
x
h
_
u
=
h
T u

1
_
x
h
_ = St
h
_
x
h
_
Mivel a Strouhal szam az elso modus u elhang eseten 0,028 es 0,044 kozotti erteket vesz fel, amibol
h/ = 10 mellett a zavaras relatv terjedesi sebessege a f uvoka kozeleben 1 1,58-re adodik, es ez
az ek kozeleben 0,2 0,3-ra csokken.
A zavaras atlagos terjedesi sebessegenek relatv erteke 0,32 0,5 kozott van.
Kapcsolodo publikaciok: WoS-os folyoirat: [55]; Nem-WoS-os publikaciok: [59, 60]
66
Chapter 3
Acoustics of the edge tone
Resource requirements for the calculation of sound production in a planar ow are often unnecessarily
high due to a full, three-dimensional CFD simulation. However, planar ows can be simulated in two
dimensions thus reducing the time and memory requirement of the CFD part of the coupled aeroacoustic
computation. This chapter will present a newly developed hybrid Computational AeroAcoustic (CAA)
method to calculate the sound generation of a planar ow by coupling a 2D CFD simulation with a 3D
acoustic simulation. The basic idea of this method is to calculate the acoustic source terms from the 2D
CFD simulation with Lighthills analogy and perform a 3D acoustic simulation after extruding them in
the third direction. It is also shown to be sucient to properly scale the sources on the surface of the
2D CFD mesh and carry out the 3D acoustic simulation with only a surface source distribution without
extruding it. The proposed method is applied to the edge tone to investigate its acoustic eld numerically.
3.1 Hybrid Computational AeroAcoustic methods
The source of a noise can be of several origins, e.g. structural vibrations, ow elds. Aeroacoustics
deals with the generation and propagation of ow induced noise. The propagation of acoustic waves is
basically propagation of pressure waves with very low amplitudes and therefore that can be viewed as
a ow phenomenon. Hence for a certain aeroacoustic problem both the generation and the propagation
of the noise can be directly simulated by solving the compressible Navier-Stokes equations (direct noise
simulation). In practice direct noise simulations cannot be used for engineering applications because of
some basic dierences between acoustics and uid dynamics. Some of these dierences and challenges in
direct noise simulations are:
- Low Mach number ows are often treated as incompressible, while sound propagation is per
denition compressible.
- There are several magnitude dierences in the aerodynamic and acoustic variables. E.g. a 1
m
s
velocity uctuation is not extreme in ows (the rms value of the transversal velocity uctuation of
the edge tone at the tip of the wedge at Re = 200 (Figure 2.25(f)) is approximately 1
m
s
), while the
amplitude of the acoustic velocity uctuation at a particle velocity level of 80 dB is only 0.5
mm
s
. The
human hearing threshold (for good ears) at 1 kHz is around 0 dB sound pressure level, corresponding
to 20 Pa, which is again several orders of magnitude smaller than an average pressure uctuation
in aerodynamics.
Therefore special care has to be taken by employing numerical schemes with high order of accuracy
to ensure that the numerical error of a direct simulation is much smaller than the magnitude of
the acoustic waves, otherwise the numerical noise may totally disrupt the acoustic solution.
- For aerodynamic problems the spatial resolution is determined by the smallest scale of motion,
while in acoustics the length scale is determined by the wavelength of the sound wave. Usually this
67
CHAPTER 3. ACOUSTICS OF THE EDGE TONE
latter is much larger than the former. For example, the organ pipe investigated by Paal et al. [40]
has a ue width size of 0.7 mm and cut-up length of 19.7 mm. This means that the smallest scale of
motion is in the order of 1 mm or smaller, while the frequency of the produced sound by this organ
pipe is 161 Hz, thus its wavelength is 2.13 m.
Therefore in the source region a very ne spatial resolution is needed to resolve the ow structures
that is unnecessary to resolve the acoustic phenomena, while in the far eld, where there is no ow
the necessary grid size is determined by the wavelength of the sound generated by the ow. This
means that if the mesh is such that it can resolve both the aerodynamic and acoustical phenomena,
then there will be dierences of several order of magnitudes in the element sizes.
- The permissible time step to satisfy the Courant-Friedrichs-Lewy criterion is determined by the
smallest elements in the source region, and of course it is far too low for the larger elements in the
acoustical far eld.
- To achieve uniform propagation properties of the acoustic waves in the whole domain, the numerical
accuracy (e.g. dispersion and dissipation errors) should also be uniform, which with highly dierent
cell sizes can again cause diculties.
- Due to the low amplitude of the acoustic variables, the normal CFD boundary conditions usually
not suitable for CAA simulations because they can generate spurious reections because of their
insucient accuracy.
Direct noise simulations, inter alia, because of these reasons are computationally too demanding, and
therefore not commonly used. Nevertheless the dierent behaviour of aeroacoustic noise generation and its
free eld propagation makes it possible to decompose the phenomena into two parts: a CFD simulation of
the source region in that the ow-induced sound sources are also determined and an acoustic propagation
simulation. These two can be solved separately with dierent numerical techniques and meshes. These
methods are collectively referred to as hybrid CAA methods. There are a couple of acoustic analogies
(e.g. Lighthills analogy [41] and its extensions by Curle [42] and Ffowcs-Williams and Hawkings [43];
Powells vortex sound theory [44] and its extension by Howe [45]) for calculating the acoustic source
terms from the CFD simulations and to couple them to an acoustic simulation. In his dissertation, Wim
DeRoeck [46] gives a nice and detailed overview on these analogies, hereby I shall only write about the
one that I have used, namely Sir James Lighthills analogy [41].
3.1.1 Lighthills analogy
The continuity and momentum equations of the Navier-Stokes equation can be written as:
t
+
3
i=1
v
i
x
i
= Q, (3.1)
v
i
t
+
3
j=1
v
i
v
j
x
j
= F
p
x
i
+
3
j=1
ij
x
j
(i = 1 . . . 3) , (3.2)
where is the density, v
i
is the i
th
velocity component, p is the pressure, Q is a mass source (mass per unit
volume, per unit time), F is an external force density (force per unit volume) acting on the uid (such
68
3.1. HYBRID COMPUTATIONAL AEROACOUSTIC METHODS
as the gravitational force) and
ij
denotes the elements of the stress tensor. Let us suppose a uniform
reference state where the density is
0
, the pressure is p
0
and the speed of sound is a
0
and let us notate
the perturbations dened as the deviation from this reference state as
=
0
and p
= p p
0
. Now,
let us take
(3.1)
t

3
i=1
(3.2)
i
x
i
a
2
0
3
i=1
x
2
i
, thus:
t
2
a
2
0
3
i=1
x
2
i
i=1
3
j=1
v
i
v
j
x
i
x
j
=
Q
t

3
i=1
F
i
x
i
+
3
i=1
2
p
x
2
i
i=1
3
j=1
ij
x
i
x
j
a
2
0
3
i=1
x
2
i
(3.3)
As
0
and p
0
are independent of time and space, thus with the notation of:
T
ij
= v
i
v
j
+
_
p
a
2
0
ij

ij
, (3.4)
that is called the Lighthills stress tensor, where
ij
is the Kronecker delta, we conclude to Lighthills
famous equation:
t
2
a
2
0
3
i=1
x
2
i
=
Q
t

3
i=1
F
i
x
i
+
3
i=1
3
j=1
2
T
ij
x
i
x
j
(3.5)
This is actually the wave equation completed with source terms on the right-hand side and it is exact since
it is deduced from the Navier-Stokes equations without any assumptions. The terms on the right-hand
side are called monopole, dipole and quadrupole source terms, respectively.
However, for isentropic ows in which p
/p
0
and
/
0
are very small (that is usually fullled),
_
p
a
2
0
_
can be neglected. Moreover, the eects of viscosity and heat conduction are expected to cause
only a slow damping due to the conversion of acoustic energy into heat and to have a signicant eect
only for very large distances. Under these assumptions, it is possible to neglect
ij
entirely. Therefore, the
resulting approximation of Lighthills stress tensor is given by T
ij
v
i
v
j
which at low Mach-numbers
when the ow is incompressible can be further simplied to
T
ij

0
v
i
v
j
. (3.6)
When neither mass source nor external force is present in the system (Q = 0 and F = 0) only T
ij
acts as
a sound source, thus the non-homogeneous wave equation that describes the generation and propagation
of the sound is:
t
2
a
2
0
3
i=1
x
2
i
=
3
i=1
3
j=1
0
v
i
v
j
x
i
x
j
(3.7)
Thus from the result of the CFD simulation the right-hand-side of Lighthills wave equation can be
determined, and the sound propagation can be simulated with an acoustic solver separately from the
CFD simulation. The steps of the hybrid CAA method (using Lighthills analogy) that was used for the
computational aeroacoustic simulations of the edge tone are the following:
Step 1: the ow eld was calculated with ANSYS-CFX
Step 2: the right-hand side of the inhomogeneous wave equation (the nodal source terms) was calculated
using the approximation of Lighthills stress tensor (equation (3.6)) on the CFD grid
Step 3: the nodal source terms were interpolated onto the acoustic grid with a conservative interpolation
Step 4: the propagation of the acoustic waves was calculated with CFS++
69
CFS++ [47] is an in-house code developed by Manfred Kaltenbacher and his group currently at Technische
Universitat Wien. It uses a nite element method to solve the two- or three-dimensional inhomogeneous
wave equation in time or frequency domain. It also has a subroutine to read the transient result les of
ANSYS-CFX and to calculate the nodal source terms from them.
The acoustic eld of a fully planar ow (i.e. an innitely thick ow in that there is no ow in the
3
rd
direction) is also two-dimensional. In this case a 2D CFD simulation can be coupled to a 2D acoustic
simulation, but the interpretation of the result of a 2D acoustic simulation is not entirely clear. In reality
there are no truly planar ows because of end eects and three-dimensional instabilities, but some ows
can be well approximated as planar, especially for moderate Reynolds numbers, because their essential
features are captured. Although the ow itself can be approximated as planar, their acoustic (far)eld is
usually (when the height of the ow is not much larger than the acoustic wavelength) three-dimensional.
Normally, even in the case of a planar ow, to calculate the 3D acoustic eld generated by the ow a
full 3D CFD simulation has to be carried out which is unnecessarily expensive in terms of computational
resources.
Since the result of a 2D CFD simulation of a planar ow agrees acceptably with a slice of the 3D
simulation of the same ow, the question arises: is it possible to carry out a 3D CAA computation from
a 2D CFD simulation and thus avoid an expensive 3D CFDCAA coupling? It is, as this chapter shows
on the example of a rst stage edge tone oscillation at moderate Reynolds numbers.
The aim of this chapter is to examine and compare various ways to couple a planar ow simulation
with a 3D acoustic simulation. Hence, the simplest mode of the edge tone, a pure rst stage oscillation,
was chosen as a test case.
3.2 Basic ideas
As discussed earlier above ANSYS-CFX is only capable of calculating in a 3D domain. Planar ows
meaning a nearly perfectly two-dimensional ow of W thickness; like the edge tone at moderate Reynolds
numbers can be calculated using a one-cell-thick mesh with symmetry boundary conditions on the top
and bottom surfaces. The result is a w
CFD
(the thickness of the element layer) thick slice of an innitely
thick fully planar ow with no change in the 3
rd
direction. The velocity eld at the bottom (or the top)
plane (with symmetry boundary condition) that is the velocity eld of this fully planar ow at an
arbitrary plane perpendicular to the 3rd direction can be coupled to a 2D acoustic simulation.
The computed acoustic sources correspond to this slice, and the upper layer of nodes holds exactly
the same source strength as the bottom one. If W is not large (comparable to or smaller than the acoustic
wavelength ()) but neither too small (the ow is still nearly perfectly two-dimensional) then the acoustic
eld generated by the ow is not 2D and the 3D eects become stronger as W decreases.
In this case there is no need to calculate the ow in its full height but it is sucient to carry out a
2D CFD simulation, calculate the acoustic sources generated by the w
CFD
slice of the ow, and create a
virtual extrusion of this source distribution. The acoustic eld of a W = w
CFD
( = 1, 2, . . .) high
virtual extrusion of the ow can then be calculated if this layer of elements with their nodal sources is
copied over itself times and then the overall source of each node is determined as the sum of the sources
70
3.2. BASIC IDEAS
from the elements below and above it. Thus the bottom and the top layer will have the same sources as
from the 2D CFD simulation, while the intermediate layers will have double that.
Figure 3.1 shows the schematics for one element only: from the CFD simulation the element with a
height of w
CFD
holds s
i
i = 1, . . . , 4 nodal sources in its bottom four corners and the same sources in its
top four corners. A virtual conguration with a height of W = w
CFD
(in Figure 3.1 has a value
of 3) will have s
i
sources on the bottom and top layer and 2 s
i
sources in the intermediate layers.
s
1
s
1
s
4
s
4
s
3
s
3
s
2
s
2
s
4
s
3
s
4
+ s
4
s
4
+ s
4
s
3
+ s
3
s
3
+ s
3
s
1
s
1
s
1
+ s
1
s
1
+ s
1
s
4
s
3
s
2
s
2
+ s
2
s
2
+ s
2
s
2
Figure 3.1: Extruded element with acoustic nodal sources
The resource requirement of the interpolation can be reduced if this source extrusion is done after
the interpolation of the acoustic nodal sources from the CFD to the acoustic mesh. The bottom layer of
sources from the CFD mesh is interpolated onto the bottom layer of the acoustic mesh and afterwards
these sources are copied onto each further layer. For this, the acoustic mesh has to be an extruded
mesh in the source region. Furthermore, the case when CFD and acoustic layers have dierent heights
can be considered. On the CFD mesh, one layer of nodes holds sources for a w
CFD
/2 high slice of the
ow. Assuming that the acoustic mesh has an element height of w
acou
, the interpolated acoustic sources
have to be multiplied by a factor of = w
acou
/w
CFD
. When copying the interpolated sources on top of
each other, the internal layers will again have twice as strong sources as the bottom and the top layers
(Figure 3.2(a)-(c)).
The computational resources can be further reduced if the symmetry is taken into account. Because
of the extruded sources, the source eld is symmetric with respect to the central plane of the extrusion,
so it is sucient to carry out simulations for the half-space only. The element height ratio remains the
same, thus with the same factor for the bottom and top layer and 2 factor for the intermediate
layers, the whole virtual extrusion will contain sources for a domain of W/2 height. This is then doubled
by the reecting boundary condition (Figure 3.2(d)).
If W/2 , the source region is acoustically compact and the resource requirement can be even
further reduced. For W/2 , the sources of the whole W/2 high planar ow are summed up and
concentrated onto the bottom surface of the acoustic mesh (which is assumed to be the symmetry plane).
In this case the sources interpolated onto the bottom layer of the acoustic mesh have to be multiplied by
c
= W/w
CFD
that will provide sources for a ow of
c
w
CFD
/2 = W/2 height (Figure 3.2(e)), that is
71
z = w
CFD
z = 0
s
f
s
f
s
a
s
a
2s
a
z = W = 4w
acou
z = w
acou
interp.
2s
a
2s
a
s
a
s
a
2s
a
s
a
(a) (b) (c) (d)
c
s
a
(e)
Figure 3.2: Scaling for dierent heights of CFD and acoustic source region
again doubled by the reecting boundary condition.
3.2.1 Computational algorithm
A 2D CFD simulation is to be carried out with one layer of w
CFD
high elements to calculate a planar ow.
The acoustic source region mesh has /2 layers of elements, each having a height of w
acou
= (W/2)/(/2).
Then the steps of the nal computational algorithm to compute the 3D acoustic eld of the planar ow
with a height of W is as follows:
Step 1: 2D CFD simulation
Step 2: Calculation of the acoustic nodal sources on the CFD mesh
Step 3: Conservative interpolation of the acoustic nodal sources onto the bottom layer of the acoustic
mesh.
Step 4: Copying and scaling the nodal sources by a factor of = w
acou
/w
CFD
for the bottom and the
top layers and 2 for each intermediate layer.
Alternatively: concentrating all the sources onto the bottom layer by scaling the sources on the
bottom layers with a factor of = W/w
CFD
without copying it to further layers.
Step 5: If harmonic analysis is sucient, then the acoustic nodal sources are Fourier transformed and
a harmonic acoustic simulation is performed.
Alternatively: if necessary, a full time domain transient acoustic simulation is carried out.
3.3 Denition of test setups
The same edge tone conguration was used as in the 3D CFD simulation in Section 2.1.2: a = 1 mm
wide and W = 25 mm high top hat jet impinging on a wedge with and angle of 30
placed at a distance
of h = 10 mm far away from the nozzle. The Reynolds number was 225 during the development of the
methodology, while it was varied between 60 and 225 in the Reynolds number dependence study. In the
Re = 225 case 2D3D and 3D3D couplings were both carried out, while at other Reynolds numbers
only 2D3D coupling was done. At Re = 225 the frequency of oscillation was around 130 Hz leading to
a wavelength of around = 2.6 m. At lower Reynolds numbers the wavelengths were even higher. The
72
3.3. DEFINITION OF TEST SETUPS
height of the ow (W = 25 mm) is much larger than the width of the jet ( = 1 mm) therefore the ow
is 2D, but it is much smaller than the acoustical wavelength ( = 2.6 m) therefore the acoustic eld is
3D, thus this is a perfect subject for validating the methodology to couple 2D ow simulation with 3D
acoustic simulation.
The rst step is to calculate the ow eld. This is already discussed in Chapter 2. The ow has a well-
dened oscillation frequency, hence there is no need to carry out a full time domain acoustic simulation, a
harmonic simulation at this frequency is sucient. At Re = 225, which was used during the development
of the methodology, the last 13 periods of the whole simulation were used to calculate the sources for the
acoustic simulation. This part of the simulation contained the fully evolved periodic edge tone oscillation.
These 13 periods took approximately 0.1 s virtual time, thus giving a 10 Hz resolution on the frequency
spectrum, which is more than adequate to capture the oscillation frequency of 130 Hz. Similar proportions
were used at the other Reynolds numbers.
3.3.1 Computational domains for the acoustic simulation
The 3D acoustic simulation domain was a cube with an edge size of about 3. There are two contradictory
requirements for the element size in the acoustic mesh: (i) a rule of thumb for the acoustic simulations is
that there is no need to have smaller elements than /20 and (ii) the sources produced by the ow have
to be resolved properly as some regions of the ow could have sources of dierent phase located very close
to each other. So, an interpolation of such sources onto one element of the acoustic source region mesh
could lead to a dramatic amplitude loss and in extreme cases to qualitatively wrong, unrealistic acoustic
elds.
The second rule leads to the conclusion that a part of the source region (that is smaller than a /15
sized cube located in the middle of the whole acoustic simulation domain) should have as small elements
as in the CFD mesh. For the case considered here, this corresponds to an element size of about /52 000.
To eectively handle this huge element size ratio (2 600 : 1), an unstructured source region mesh
and two nonmatching interfaces were used, thus producing the following region structure for the acoustic
simulation domain:
(a) Acoustic source region: (Figure 3.3(a))
The innermost region is the acoustic source region corresponding to the computational domain of
the CFD simulation. An unstructured mesh can be used here with element sizes between /52 000
(at the tip of the wedge) and /100. In the case of a 2D simulation, this is a triangular mesh. In the
case of a 3D acoustic simulation, the mesh of the acoustic source region had a special topology as
displayed in Figure 3.4(a). The top and bottom part of the mesh was an unstructured tetrahedral
mesh, while the central part consisted of wedge elements in 10 layers. This had two reasons: (i) the
acoustic sources are located in this central part because the jet was only 25 mm high and the CFD
domain and the wedge was 70 mm; (ii) the sources at the tip of the wedge are very closely located
to each other with dierent phases, thus a very ne resolution is needed there in the acoustic source
mesh. This resulted in elements with very dierent sizes in the central part. With an unstructured
mesh above and below it was possible to have nearly equally sized elements on the top and bottom
of the nonmatching grid interface.
73
When coupling a 2D CFD simulation with a 3D acoustic simulation with the method described
in Section 3.2 it is enough to consider a half space, shown in Figure 3.4(b). In this test setup, the
acoustic source region mesh contained 24 layers of extruded triangular, i. e. prismatic elements at
the bottom for the H/2 = 12.5 mm height.
(b) Propagation - intermediate region: (Figure 3.3(b))
The second region is the propagation (intermediate) mesh, and it is connected to the source region
by a nonmatching grid interface. This has roughly the same element size as the outer parts of the
source region (/100), but while the source region has triangular/tetrahedral elements this region is
(a) (b) (c)
Figure 3.3: Bottom plane of the mesh for the acoustic simulation
(a) (b)
Figure 3.4: Mesh of the 3D acoustic source region: (a) surface mesh the elements are much larger than
the elements of the CFD mesh; (b) cut of the mesh at z = 0 mm the mesh density at the tip of the
wedge reaches the density of the CFD mesh
74
3.4. RESULTS
an equidistant Cartesian quadrilateral/hexahedral mesh. The nonmatching grid interface is needed
to handle the transition from the unstructured into the structured mesh.
(c) Propagation - acoustic far eld and PML (Perfectly Matched Layer) region: (Figure 3.3(c))
The acoustic far eld is connected by a second nonmatching grid interface to the grid of the propa-
gation (intermediate) region. This mesh is again an equidistant Cartesian quadrilateral/hexahedral
mesh but now with an element size of about /20. The outermost layers of this region are the
elements of the PML that eciently models the free eld radiation (for details of the used PML
technique see [48, 49]).
3.3.2 2D CFD with 3D acoustics - the treatment of sources
The method described in Section 3.2 was used to simulate the acoustic eld of the edge tone ow based
on the w
CFD
= 1 mm high 2D CFD simulation. The acoustic source region mesh contained 24 layers
of extruded triangular, i.e. prismatic elements (that is 25 layers of nodes) at the bottom (i.e. next to
the reecting boundary). First, the sources were extruded to all the layers (acoustic source region has
/2 = 24 layers of elements, thus 25 layers of sources, w
acou
= (W/2)/(/2) = 12.5/24 0.52 mm,
= w
acou
/w
CFD
0.52), then to every second one (/2 = 12, thus 13 layers of sources, w
acou
1.04 mm,
1.04), then to every third one (/2 = 8, thus 9 layers of sources, w
acou
1.56 mm, 1.56), and so
on. At the end there were 8 simulations with 25, 13, 9, 7, 5, 4, 3 and 2 layers of sources all having the
same strength in sum. Also a ninth simulation with only one layer of sources concentrated on the bottom
plane was carried out. These simulations will be referred to as the l layer case (l = 25, 13, 9, 7, 5, 4, 3, 2 or
1).
3.3.3 3D CFD with 3D acoustics
This is the most realistic and most expensive case. Until now this was the only way to calculate the acoustic
eld of a planar ow: to carry out a full 3D CFD simulation and couple it with a 3D sound propagation
computation. This kind of coupling was carried out to demonstrate the eciency and accuracy of the
method described in Section 3.2.
3.4 Results
The main result of an aeroacoustic simulation is the acoustic pressure eld generated by the ow, but as
the result of an intermediate step it is useful to have a look at the results of the CFD simulation too. This
was already done in Chapter 2, therefore here only the results of the acoustic simulations are discussed.
3.4.1 2D-2D: acoustic mesh study
A 2D acoustic simulation without an appropriate rescaling to t the real 3D behaviour results in an
acoustic pressure with a unit [Pa/m]. Such a rescaling was not studied, and therefore only a qualitative
comparison of the dierent results for the 2D acoustic computations was made.
75
Figure 3.5: Directivity of the normalised acoustic pressure in the 2D-2D coupling
Figure 3.5 shows the normalised amplitude of the acoustic pressure for points on a circle with a center
C = [0.0225 m; 0 m] (that is the tip of the wedge) and a radius r = 2.6 m (that is approximately one
wavelength). A dipole radiation pattern is obtained.
Since 2D acoustic simulations are computationally ecient, a detailed acoustic mesh study was carried
out and the experience gained here was used for the more complex 3D acoustic simulations. The results
of the mesh study are the following:
- The source region mesh should resolve the region of the tip of the wedge as ne as the CFD mesh.
An even ner mesh would not make any sense, as the acoustic source terms are calculated on this
mesh. Results obtained by using coarser meshes were compared with this one. A strategy to reduce
the number of nite elements could be a structured mesh which is the same as the CFD mesh near
the tip, and coarser at the corners of the region. However, in a structured mesh the element size
cannot be coarsened in only some parts of the domain without producing elements with bad aspect
ratio at other parts of the domain, so that an unstructured mesh is a better choice. In this way the
element number in the source region could be reduced to less than 4 000 in contrast to 36 300 in
the reference case with the dierence in the results remaining below 0.2%. This mesh - named as
the best aordable - had a very ne resolution in a circle around the tip of the wedge with a radius
of 1 mm. The triangles in this circle had a size of 0.05 mm (the corresponding elements in the CFD
mesh were 0.3125 mm 0.05 mm). Without this ne resolution the acoustic sources at the tip of
the wedge are blurred and they cancel out each other, resulting in a decreased acoustic pressure
compared to the reference case.
- The mesh generation of the propagation-intermediate region can be performed independently of the
source region, since the interface between these two regions can be nonconforming (Figure 3.3(a)).
To resolve the wave propagation precisely, about 20 nite elements with linear basis functions per
acoustic wavelength are required. However, a too large change between the mesh size from source
to propagation region along the nonmatching grid interface could generate spurious reections.
Therewith, we keep the increase of the mesh size in the propagation region below a multiplicative
76
3.4. RESULTS
factor of ten. This results in a mesh, which has at least six nite elements along each side of the
nonmatching interface corresponding to an element size of /100.
- The far eld mesh and the mesh in the PML region do not play a signicant role for the results.
The far eld mesh can be as coarse as /5 without a noticeable change in the result. However, the
element size of /20 has been kept to have the phase of the acoustic wave resolved ne enough.
3.4.2 3D-3D computations
For the 3D-3D coupling, several simulations were performed with dierent acoustic source region mesh
sizes, resulting in 12 000 to 55 000 nodes. The computational time did not dier too much with the
dierent meshes (just a few minutes relative to the about 60 minutes of overall time of all the steps). The
dierence in results between the 55 000 and 37 000 node cases was less than 1%. Further decreasing the
number of nodes to 12 000 increased the dierence to about 5%. Thus the mesh with about 37 000 nodes
(resulting in 140 000 elements) was chosen. The whole acoustic domain had about 373 000 elements and
383 000 nodes.
3.4.3 2D-3D: source extrusion with dierent numbers of layers
The results of the present investigation clearly show that the number of the source layers has almost no
inuence on the computed acoustic pressure. Figure 3.6 shows the acoustic eld for the one layer case.
The directivity plots are shown in Figure 3.7 on a circle around the tip of the wedge with a radius
of r = 2 m in the z = 0 m plane (used for all cases in the 2D-3D coupling and also for the 3D-3D
coupling). The sound pressure level SPL = 20 lg
_
/20 Pa
_
dB is plotted against the angle, where

p
is the amplitude of the pressure uctuation. The dierence in acoustic pressure between the results of
the 2D-3D simulations and the real 3D-3D coupling is less than 1%, even for the one layer case. With
these negligible dierences there is no need to carry out full 3D-3D coupling, but it is sucient to use
the sources from a 2D CFD simulation. Assuming that the jet is acoustically compact, the sources from
a 2D CFD simulation can be applied in a single layer without the need to extrude them. This also means
Figure 3.6: Acoustic pressure plot over the complete domain (at the outer region, the PML damps the
acoustic waves without reections
77
10
5
0
5
10
15 dB
0
o
30
o
60
o
90
o
120
o
150
o
180
o
210
o
240
o
270
o
300
o
330
o

3D3D
2D3D: 1 layer
2D3D: 2 layers
2D3D: 3 layers
2D3D: 4 layers
2D3D: 5 layers
2D3D: 7 layers
2D3D: 9 layers
2D3D: 13 layers
2D3D: 25 layers
Figure 3.7: Comparison of 2D-3D coupling with dierent numbers of layers and 3D-3D coupling.
Directivity patterns of the SPL in the z = 0 plane around the tip of the wedge with a radius of 2 m
that there is no need to prepare a complicated mesh structure with extruded elements at the bottom of
the source region mesh, saving not only resources when running the simulation, but also user time.
Thus it can be concluded that whenever the ow conguration allows it, the 2D-3D coupling method
is much cheaper numerically and also much faster than the full 3D-3D coupling method. Computational
results discussed in the following sections are all from one layer 2D-3D coupled aeroacoustic simulations.
3.4.4 2D-3D: comparison with theory
According to [50], the time dependent acoustic pressure (with t denoting the time) in the near and far eld
of an acoustic dipole at the origin, oriented in the y direction with a strength of G(t) can be approximated
as
p
= cos
_
G(t r/a
0
)
4r
2
+
G(t r/a
0
)
4ra
0
_
, (3.8)
where is the angle between the position vector of the observation point and the y axis, r is the distance
of the observation point from the origin. Supposing that G(t) has only one harmonic component (at the
frequency where the simulation was carried out) i. e. G(t) =

G sin(t), the formula simplies to
p
= cos
_

G sin (t r/a
0
)
4r
2
+

G cos (t r/a
0
)
4ra
0
_
. (3.9)
Applying trigonometric identities, the amplitude of the pressure uctuation can be approximated as
= cos
G
4r
1
r
2
+

2
a
2
0
. (3.10)
Figure 3.8 shows the pressure changes along the y direction away from the tip of the wedge ( = 0
,
in the direction of strongest radiation). The solid line is the best t curve using equation (3.10) with
78
3.4. RESULTS
0 0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8 2 2,2
0
0,001
0,002
0,003
0,004
r [m]
a
c
o
u
s
t
i
c

p
r
e
s
s
u
r
e

[
P
a
]
Figure 3.8: The change of the acoustic pressure along the y direction away from the tip of the wedge;
(blue) computational result, (red) theoretical dipole with a strength of

G (equation (3.10))
0 0,5 1 1,5 2 2,5 3
0
5E-5
10E-5
15E-5
20E-5
r [m]
a
c
o
u
s
t
i
c

p
r
e
s
s
u
r
e

[
P
a
]
Figure 3.9: Qualitative change in the trend of the acoustic pressure; (blue) computational result, (red)
theoretical dipole with a strength of

G (equation (3.10))
G = 0.692 mN. The range r < /20 has been excluded from the curve tting, because equation (3.8) is
only valid in the near and far eld but not in the proximal eld. A qualitative change in the trend of
the acoustic pressure curve was found near the PML (Figure 3.9) that was considered to be a numerical
error, and thus r > 2 m was also excluded from the curve tting.
There is a periodic force acting on the uid, that creates a source behaving like a dipole source of
strength

G in the far eld. Figure 3.10 shows the directivity plot for the Re = 225 case and the theoretical
dipole directivity with the same

G dipole strength.
In the CFD simulation the wedge itself was much longer than it would have been needed for the
edge tone oscillation. Thus the force acting on the wedge is larger than the eective dipole-creating
force. Figure 3.11 shows the distribution of the acoustic nodal sources. The amplitude of the harmonic
component at the oscillation frequency is visualised in each point relative to the highest acoustic nodal
source. The strongest sources are located at the tip of the wedge. Farther away from the tip of the wedge
the sources are much weaker, and usually not located next to the wedge, but rather correlated with the
ow structure. Note that any source larger than 20% of the strongest source is plotted with red, so the
scale is just from 0% to 20%.
The question arises how much of the wedge surface from the tip is included in the eective sound
production. To decide this the forces acting on the wedge surface as a function of the distance included
79
10
5
0
5
10
15 dB
0
o
30
o
60
o
90
o
120
o
150
o
180
o
210
o
240
o
270
o
300
o
330
o
Figure 3.10: Directivity plot of the SPL: (blue) computational result, (red) the theoretical dipole with a
strength of

G
Figure 3.11: Acoustic nodal source term visualization: values are normalised to the maximum acoustic
nodal source
0 10 20 30 40 50 60 70 80
0
0.2
0.4
0.6
0.8
1
x [mm]
F

[
m
N
]

3D CFD full wedge 3D CFD middle 1 mm 2D CFD
x 7.2 mm
Figure 3.12: Amplitude of the force acting on the rst xmm of the wedge; 3D CFD - full wedge: the whole
height of the rst xmm of the wedge is taken into consideration; 3D CFD - middle 1 mm: the pressure
is only integrated on the middle 1 mm strip of the wedge, and the integral is multiplied by 25; 2D CFD:
the force acting on the 1 mm thick wedge is multiplied by 25
80
3.4. RESULTS
in the pressure integration for 2D and 3D are compared in Figure 3.12.
In the 3D case there are two curves: one is simply the pressure integrated on the whole height of the
wedge surface, the other one is integrated on the central 1 mm strip and multiplied by 25. This latter
was to check the two-dimensionality of the pressure distribution and we see that the agreement between
the two curves is good. The 2D curve was produced by simply multiplying the 2D result corresponding
to a 1 mm slice by 25. When comparing the 2D and 3D curves we see that up to about 7.2 mm the two
curves run perfectly together and from that point the curves clearly separate. After carefully inspecting
the respective velocity elds we found that while in the 2D case the jet remains close to the wedge with
vortices swimming periodically along the surface, in the 3D case after the rst vortex, the further vortices
separate from the surface and ow away into free space. This discrepancy in the velocity elds occurs
roughly at the same location, i. e. at about 7.2 mm.
Since the far eld acoustic pressures agree very well in the 2D-3D and the 3D-3D cases we can deduce
that the part that is important for the sound production process is exactly this rst 7.2 mm. Using this
distance the force amplitudes are

F
y
= 0.495 mN and 0.501 mN for the 2D CFD simulation (for a virtually
25 mm high ow) and for the 3D CFD simulation integrated on the full wedge height, respectively.
3.4.5 2D-3D: dierent Reynolds numbers
A pure rst stage edge tone oscillation can be found in our conguration from Re = 60 to Re = 225. 2D
CFD - 3D acoustic coupled simulations with concentrated sources on the symmetry plane were carried
out for Reynolds numbers Re = 60, 70, 80, 100, 150, 175, 200 and 225. Varying Reynolds numbers were
produced by varying the inlet velocity of the jet. For each case the same acoustic domain and mesh
detailed in Section 3.3 was used. The edge tone oscillation frequencies (with their wavelength) for these
Reynolds number can be found in Table 3.1. As the acoustic domain was only about 7.8 m7.8 m3.9 m,
it is clear that the conditions for the far eld approximation were not fullled.
For all the Reynolds numbers the simulation produced a dipole sound eld, and the results were
similar to those shown for the Re = 225 case (Figures 3.8 and 3.10). Figure 3.13 shows the directivity of
the sound produced by the edge tone for Re = 150, 175, 200 and 225 at a distance of 2 m around the tip
of the wedge on the z = 0 plane.
Comparing the force amplitudes deduced from the acoustic pressure (
G) with the ones calculated on

the wedge surface (
F
y
) for various Reynolds numbers we nd that the theoretical one is consequently
larger and the ratio between the two increases with the frequency (Table 3.1).
This can be explained in the following way. The two force amplitudes should agree if the edge tone
were a perfect dipole source. However it is a dipole only if observed from the far eld. Near the wedge
wall it is rather like two unconnected monopoles of opposite phase, separated by the wedge. If we go to
the far eld, this separation becomes unimportant but in the proximal and near eld it is important and
it increases the amplitude relative to the perfect dipole because of the lack of connection. The fact that
in the lower frequency cases the two values dier less, can be explained by the larger diraction eects
around the wedge, so that the separation eect and thus the distortion of the perfect dipole behaviour is
less.
In the Re = 225 case the dierence in the sound pressure is only about 1 % between the 2D-3D and
81
Table 3.1: Frequencies (f), wavelengths (), relative distance (r/(/2)) of the monitoring point from
the tip of the wedge, amplitude of force uctuation acting on the rst 7.2 mm of the wedge (
F
y
) and the
strength of the dipole sound source (
G) for edge tone oscillations at dierent Reynolds numbers. Results

are from 2D-3D coupled simulations except the last column, that is the result of the 3D-3D coupled
simulation at Re = 225.
Re [-] 60 70 80 100 150 175 200 225
225
(3D3D)
f [Hz] 26 32 38 49 80 96 111 130 130
[m] 13.1 10.6 8.9 6.9 4.3 3.5 3.1 2.6 2.6
r/(/2) [-] 0.96 1.18 1.4 1.81 2.96 3.55 4.1 4.8 4.8
F
y
[mN] 0.00881 0.0312 0.0513 0.0787 0.206 0.318 0.424 0.495 0.501
G[mN] 0.0108 0.0389 0.0666 0.105 0.299 0.429 0.58 0.692 0.664
G
/
F
y
[-] 1.226 1.247 1.298 1.334 1.451 1.349 1.368 1.398 1.325
15
10
5
0
5
10
15 dB
0
o
30
o
60
o
90
o
120
o
150
o
180
o
210
o
240
o
270
o
300
o
330
o

Re = 150
Re = 175
Re = 200
Re = 225
Figure 3.13: Directivity plots for dierent Reynolds numbers at a distance of 2 m from the tip of the
wedge; (black) Re = 150, (blue) Re = 175, (green) Re = 200, (red) Re = 225
3D-3D cases. This dierence increases to about 4 % for the theoretical dipole strength (
G). This is due

to the fact that the near eld plays a larger role in the determination of the dipole strength than in
that of the pressure. To demonstrate the relative importance of the far and near eld, the ratio of the
two terms in equation (3.8) were calculated. In our monitor point that is strictly speaking not far eld,
corresponding to only about 0.75, the near eld term contributes only about 1 % to the total acoustic
amplitude.
82
3.5. SUMMARY
In Figure 3.14 the amplitude of the acoustic pressure uctuation at r = 2 m away from the tip of the
wedge in the y direction is plotted as a function of the Reynolds number. The best t curve in the sense
of the method of least squares is
[Pa] 6.08 10
6
[Pa] Re
3
. (3.11)
The same

p
Re
3
relationship was found experimentally by Powell [7]. Although the observation points
were in none of the cases in the real far eld, in most cases a far eld-like behaviour can be observed.
This results coincides with the know relationship that the acoustic power of a dipole sound source scales
as the sixth power of the ow speed [51] while thus the intensity also scales as the sixth power of the ow
velocity. And as in the far eld the particle velocity and the acoustic pressure are in phase, the acoustic
intensity scales as the square of the acoustic pressure. Therefore, the acoustic pressure in the far eld of
a dipole scales as the third power of the ow speed and thus the Reynolds number.
25 50 75 100 125 150 175 200 225 250
0
10
20
30
40
50
60
70
Re [-]
a
m
p
l
i
t
u
d
e

o
f

t
h
e
a
c
o
u
s
t
i
c

p
r
e
s
s
u
r
e

[
P
a
]
Figure 3.14: Amplitude of the acoustic pressure (
) at 2 m away from the tip of the wedge at 90
. The
solid line is the best t c Re
3
curve with c = 6.08 10
6
[Pa]
3.4.6 The computational costs
The 2D CFD computations were about eight times faster than the 3D CFD computations. Furthermore,
the memory consumption reduced by a factor of 10 for the 2D CFD computations. The CPU times for
both the 2D and 3D acoustic simulations can be neglected compared to the CFD simulation times. The
percentage of the acoustic computation time compared to the overall time for the CFD-CAA computation
was about 2 %. Hence, the 3D CFD computation is the most expensive one and the overall computational
time for CFD-CAA investigations can be greatly decreased by performing 2D CFD computations instead.
3.5 Summary
It has been shown that it is not necessary to carry out an expensive 3D CFD simulation for a model ow
which can be well approximated as planar to calculate its sound production. It is sucient to calculate
the acoustic nodal sources from a 2D CFD simulation with high grid resolution, and the 3D acoustic eld
83
can then be calculated after the extrusion of the sources. At the price of an acceptable error in the nal
results, which was in our case below 1 %, the calculation time can be reduced by a factor of 8. It has also
been shown that the acoustic sources can be concentrated in one layer without committing a signicant
error (Thesis #5). In addition, we have shown that our numerical results compare well with the analytical
formula of an acoustic dipole generated by a periodic force acting on the uid. It is also shown that only
the rst 7.2 mm long part of the wedge is important for the sound production, and the amplitude of
the force acting on this part of the wedge is lower than the strength of the dipole. Furthermore, the
scaling of the acoustic pressure with the third power of the Reynolds number could be well demonstrated
(Thesis #6).
84
3.5. SUMMARY
3.5.1 Theses
Thesis #5
Several ow phenomena can be well approximated as planar ow although the sound generated by
them is three-dimensional. Until now for a three-dimensional Computational AeroAcoustic (CAA)
simulation a three-dimensional CFD simulation was needed (3D-3D coupling).
I have developed a method with that the three-dimensional acoustic eld generated by a (nite W
thick) planar ow can be determined without the need of the three-dimensional CFD simulation of the
full W thick ow domain by modelling only a thin (w thick) slice of the planar ow (2D-3D coupling).
The basic idea is that the acoustic sources are computed from the two-dimensional ANSYS-CFX
simulation, then they are interpolated onto the acoustic mesh where they are extruded in the third
direction and there a proper scaling (given in the dissertation) is applied. Further simplication can
be achieved if the acoustic sources are not extruded, only a W/w scaling is applied, thus the acoustical
sources are concentrated onto a plane.
I showed that if the thickness of the planar ow is much smaller than the acoustical wavelength
i.e. the sound source is acoustically compact, then the acoustical far eld resulting from a 2D-3D
coupling negligibly depends on the resolution of the extrusion (including the case when the sources
are concentrated on a plane without extrusion). Furthermore I showed that the acoustic far elds
resulting from a 3D-3D or 2D-3D coupling dier negligibly.
Thesis #6
I investigated the acoustic eld of the edge tone by the method described in Thesis #5.
I showed that from the sound generation point of view only a short, 0.72h long part of the wedge
plays a role. I investigated the amplitude of the uid force acting on this part of the wedge
_
F
_
and
the strength of the sound source generated by the ow
_
G
_
. I found that contrary to the theoretical
dipole in the edge tone conguration the generated acoustic sources at the two sides of the wedge are
separated by the wedge, thus reducing the interaction between the two theoretical monopoles of the
dipole, therefore

G >

F. The ratio of the two force
_
G/
F
_
depends on the frequency of the generated
sound. At lower frequencies because the diraction eects at the edges of the wedge is greater the
ratio of the forces is lower.
Related publications: WoS journal: [57]
85
3.5.2 Tezisek
5. Tezis
Sok aramlasi jelenseg jo kozeltessel skaramlasnak tekintheto, am az altaluk keltett hang megis
haromdimenzios. Haromdimenzios numerikus aero-akusztikai (CAA - Computational AeroAcoustics)
szimulaciohoz ezidaig harom dimenzios CFD szimulaciot kellett vegezni (3D-3D kapcsolas).
Kidolgoztam egy eljarast, amivel (egy veges W magassag u) skaramlas altal keltett haromdimen-
zios akusztikai ter meghatarozhato a teljes W magas aramlasi ter haromdimenzios CFD szimulacioja
nelk ul, pusztan a skaramlas egy vekony (w vastagsag u) szeletenek modellezesevel (2D-3D kapcsolas).
A modszer lenyege, hogy a w vastag szeleten elvegzett kvazi-ketdimenzios ANSYS-CFX szimulaciobol
szamolt akusztikai forrastagokat az akusztikai halora interpolaljuk, ott a harmadik iranyban extru-
daljuk es a modszerben megadott modon (a CFD es az akusztikai halotol f uggoen) skalazzuk. Tovabbi
egyszer ustest jelent, ha a kvazi-ketdimenzios CFD szimulaciobol szamolt forrastagokat az akusztikai
halora valo interpolalas utan nem extrudaljuk, csak egy W/w faktorral skalazzuk. Ezaltal az akusztikai
forrastagokat egy skra koncentraljuk.
Megmutattam, hogy amennyiben az aramlas (W) magassaga sokkal kisebb, mint az akuszti-
kai hullamhossz (), azaz a hangforras akusztikailag kompakt, akkor a 2D-3D kapcsolassal kapott
akusztikai t avolter az extrudalas felbontas atol az extrudalas nelk uli, egy skra koncentralt forrastagok
esetet is ideertve elhanyagolhato mertekben f ugg. Tovabba megmutattam, hogy a 3D-3D illetve a
2D-3D kapcsolassal kapott akusztikai tavolterek elhanyagolhato mertekben k ulonboznek.
Kapcsolodo publikaciok: WoS-os folyoirat: [57]; Nem-WoS-os publikacio: [6668]
6. Tezis
Az 5. Tezisben vazolt modszerrel vizsgaltam az elhang akusztikai teret.
Megallaptottam, hogy a hangkepzes szempontjabol az eknek csak egy rovid, a cs ucstol mert 0,72h
hossz u darabja jatszik szerepet. Megvizsgaltam az erre az ekszakaszra az aramlas altal letrehozott
oszcillalo er o amplit udojat
_
F
_
es az aramlas altal letrehozott hangforras erosseget
_
G
_
.
Megallaptottam, hogy szemben az elmeleti dipolussal az elhang konguracioban az ek ket oldalan
letrejovo akusztikai forrasokat az ek szetvalasztja ezaltal csokkentve az elmeleti dipolus ket monopolu-
sanak egymassal valo kolcsonhatasat, ami miatt

G >

F. A ket ero aranya
_
G/
F
_
f ugg az letrehozott
hang frekvenciajatol. Alacsonyabb frekvencia eseten, mivel a hullamok elhajlasa az ek eleinel nagyobb
mertek u, az erok aranya alacsonyabb.
Kapcsolodo publikaciok: WoS-os folyoirat: [57]
86
Chapter 4
CFD simulations on an organ pipe
foot model
A well-known realization of the edge tone phenomenon is the ow in a ue organ pipe. The air leaves
the foot of the organ pipe through the ue forming a planar free jet. This jet interacts with the upper
lip and initially starts to oscillate with a frequency determined by the edge tone characteristics. During
the attack transient this oscillation frequency very quickly transforms into the resonator fundamental
frequency. Since the attack transient is crucial for the subjective perception of the sound, and the acoustic
features of the attack transient are determined by the exact way of transition (sometimes through a
higher harmonic), it is very important for the understanding of the acoustics of the organ pipe to study
the behaviour of the ue-upper lip area that is in fact a specialised edge tone system.
Auerlechner et al. [52, 53] built an experimental pipe foot model (i.e. an organ pipe without a
resonator) to investigate how the jet leaves the foot of the organ pipe, and how it interacts with the upper
lip. First they made free jet experiments without the upper lip. They found that near the ue the jet has
a Nolle prole [54] and at larger distances (> 5 mm) the velocity prole becomes approximately Gaussian.
They determined the centreline and the width of the jet as parameters of the Gaussian probability density
function describing the velocity prole and found that the jet contracts asymmetrically to about 7887%
of the width of the slit as it leaves the foot and that the centreline of the jet is almost a straight line.
After the free jet measurements they made edge tone measurements. They found a mixed mode of the
edge tone with the rst, second and third stages coexisting. By changing the spatial position of the upper
lip placed at 10 mm above the ue they found that the maximum edge tone amplitude and thus the
strongest produced sound can be achieved when the upper lip is placed in the centreline of the jet.
This chapter shows that these experiments can be reproduced with remarkable delity with the help
of CFD simulations opening the possibility to investigate other congurations numerically.
4.1 Geometrical conguration, boundary conditions, mesh and
solver settings
The ow at the central z-plane of the foot can be well approximated as two dimensional, thus planar ow
simulations were carried out with ANSYS CFX for a 1 mm high slice of the ow just as in the case of
the edge tone (Section 2.1.1). The Mach number of the real ow was low (Ma 0.1, based on the vena
contracta velocity of the jet) so that in the computations incompressible ow was assumed.
The geometrical conguration of the main part of the foot model was the following (Figure 4.1(a)):
the ue width () was 1.3 mm, the angle of the languid () was 45
. The upper lip - when present - had

a rectangular shape with a width of 1 mm and a height of 40 mm. It was always parallel with the y axis,
87
CHAPTER 4. CFD SIMULATIONS ON AN ORGAN PIPE FOOT MODEL
(a) Area of main interest
: ue width
L: cut-up
length
upper lip
lower lip
x
y
(b) Full CFD domain; Boundary conditions: thick
black line - wall; grey line - inlet; dotted line -
opening
1
5
0

m
m
90 mm
201 mm
2
0
7

m
m
124 mm
Figure 4.1: Geometric conguration of the CFD simulation of the organ pipe foot model (not to scale)
the cut-up length (L) was 10 mm, and its oset from the y axis (x) varied from 2.05 mm to 6.05 mm.
The origin of the coordinate system was the inner corner of the lower lip. This conguration is the same
as the experimental setup of Auerlechner et al. [52, 53]. Although the dimensions of the organ pipe foot
are much larger than those of the ue and the cut-up, and the ow inside is comparatively slow and
steady, it was still simulated, so that the correct, non-uniform exit velocity prole of the jet can evolve.
A large, rectangular domain was simulated around the upper lip, to have the boundary conditions far
enough from the nozzle - upper lip region and thus to prevent any boundary inuences on the evolved
oscillating ow. Figure 4.1(b) shows the whole CFD domain in which the ow was simulated.
The mesh had a block-structured conguration. The inner blocks at the region of interest (close to
the nozzle-upper lip region) contained hexahedral elements in a fully structured layout. The elements
between the nozzle and the upper lip had sizes of about 0.1 mm by 0.25 mm (Figure 4.2). The block
structure had an O-grid layout around the upper lip thus it was possible to rene the mesh near the
upper lip and fully resolve the boundary layer of the ow. The outer blocks of the mesh were swept
blocks (containing wedges and hexahedral elements in an unstructured layout) to minimize the number
of elements. After some test runs it was found that a mesh with less than 20 000 elements was enough to
calculate the ow eld with sucient accuracy.
The boundary conditions for the simulation were the following. The wall of the foot of the organ pipe
and the upper lip (denoted by thick black lines in Figure 4.1(b)) were set to no-slip wall. At the bottom
of the foot (denoted with a grey line in Figure 4.1(b)) an inlet boundary condition was prescribed with a
uniform velocity prole. The inlet velocity at the foot hole was set to 2.6
m
s
that results in a 36.4
m
s
vena
contracta velocity in the jet, adjusted to correspond to the measurements of Auerlechner et al. [52]. The
other boundaries (at the far eld, denoted by dotted lines in Figure 4.1(b)) were set to opening, that is a
prescribed uniform pressure distribution allowing ow normal to the boundary surface in both directions.
88
4.1. SETTINGS
.
ICEM CFD 14.0; vaik@Pitot
X
Y
Z
Figure 4.2: Mesh in the nozzle - upper lip region; x = 4.05 mm (scale 5:1)
When using one of the turbulence models, a low intensity (1 %) turbulence was set for the inow and
medium intensity (5 %) for the far eld boundaries.
The Reynolds number (based on the vena contracta velocity, and jet width there that is not equal to
the ue width due to the contraction) is moderate (Re 2350), thus it was not clear a priori whether
any turbulence modelling was needed. SAS (Scale-Adaptive Simulation), LES (Large Eddy Simulation)
and DES (Detached Eddy Simulation) turbulence models were tested, as well as the DNS model (Direct
Numerical Solution i.e. without any turbulence modelling). I am aware that these turbulence models
were developed for 3D CFD simulations, but for example Takahashi et al. [32] successfully used the LES
model for edge tone simulations, thus it is not illegitimate to try these models for 2D cases as well. The
mesh was ne enough to resolve not only the large ow structures but also the smaller ones, in order to
reach targets of the iterations in each time step even in the case of the DNS simulation.
Turbulence modelling with the SAS or the DES models means that in addition to the momentum
and mass conservation equations two additional equations of the turbulence model have to be solved,
thus with the same spatial and temporal discretisation as in the DNS case the resource requirement of
the simulation is larger. In the case of the LES model a low-pass ltering operator is applied on the
Navier-Stokes equations thus eliminating small scales of the solution and modelling rather than resolving
them. This approach (with the same spatial and temporal discretisation) requires about the same amount
of resources as the DNS simulation.
Figure 4.3 shows the spectrum of the x component of the force acting on the upper lip in the
optimum case (x = 4.05 mm, when the upper lip is placed in the centreline of the jet issuing from the
foot see Section 4.3.) for the dierent turbulence models. The following observations can be made: the
frequencies of the stages (when present) are nearly the same for all the models: f
1
1.1 kHz, f
2
2.9 kHz
and f
3
4.6 kHz for the rst, second and third stage, respectively. However, with the dierent turbulence
models dierent relative weight of the stages can be found. In the DNS simulation the rst three stages
coexist and the rst one is dominant. With the SAS model a clear pure rst stage oscillation is present
89
(a) DNS
0 1 2 3 4 5 6 7 8
0
50
100
150
f [kHz]
F
x
[
m
N
]
(b) SAS turbulence model
0 1 2 3 4 5 6 7 8
0
100
200
300
f [kHz]
F
x
[
m
N
]
(c) DES turbulence model
0 1 2 3 4 5 6 7 8
0
50
100
150
f [kHz]
F
x
[
m
N
]
(d) LES turbulence model
0 1 2 3 4 5 6 7 8
0
50
100
150
f [kHz]
F
x
[
m
N
]
Figure 4.3: Spectrum of the x component of the force acting on the upper lip. x = 4.05 mm
without higher stages. In the case of the DES model the rst stage is absent and the frequencies of
the second and third stages are considerably lower compared to the other cases, and the
1
/3 and
2
/3
frequencies of the second stage are also present. The presence of the
1
/3 and
2
/3 frequencies in the edge
tone conguration was also reported by Kaykayoglu and Rockwell [37] and was found also in the CFD
simulation of the edge tone (Section 2.3.1) but with a parabolic jet prole, whereas here the jet prole is
rather top hat than parabolic. The LES model disproportionately amplies the second stage of the edge
tone. Direct quantitative comparison with Auerlechner et al. [52] is not possible since they measured
sound by a microphone and here the force acting on the upper lip surface is presented. The relationship
between these two is not straightforward, yet it can be expected that the harmonic composition of the
two signals exhibits a qualitative similarity.
Figure 4.4 shows the spectrum in the case when the upper lip is 2 mm o its optimum position
(x = 2.05 mm). It can be observed that the SAS model produces the same single peak as in the
x = 4.05 mm case but the laminar model produces a much noisier spectrum. Thus it can be concluded
that the SAS model forces a remarkably clear, pure rst stage edge tone oscillation not typical in this
Reynolds number range (the second stage of the edge tone appears at a much lower Reynolds number
(Section 2.3.1)) even when the upper lip oset is so large, that instead of the edge tone oscillation a
free jet instability oscillation may evolve.
Summarizing the result of the simulations with dierent turbulence models it can be concluded that
no turbulence modelling is necessary, as the stop conditions of the iteration were fullled in each time
90
4.1. FREE JET SIMULATION
(a) DNS
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
(b) SAS turbulence model
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
Figure 4.4: Spectrum of the x component of the force acting on the upper lip. x = 2.05 mm
step even in the DNS case and the result of the DNS simulation is the only one which is consistent with
the experimental result of Auerlechner et al. [52] that three stages of the edge tone coexist, the rst one
being the strongest. Thus for the forthcoming investigations DNS simulations will be carried out.
The oscillation frequency of the rst stage of the evolved edge tone ow at this jet velocity is
approximately 1.1 kHz. The time step for the simulation was set to 1.5 10
5
s, corresponding to about
60 steps in each expected period. Based on the experience gathered from the edge tone simulations
(Section 2.1.1), this time resolution is more than sucient. The duration of the simulation was set to
0.3 s resulting in 20 000 time steps corresponding to about 330 oscillation periods in order to have a
long enough transient simulation to investigate not only the main frequency component of the ow, but
also the change of the spectrum in time.
Even with this relatively small mesh, an average simulation took up to 3 days with an Intel Core i7
950 CPU running at 3.07 GHz because of the large number of time steps.
4.2 Free jet simulation: velocity proles and jet centreline
Free jet simulation without the upper lip was carried out to investigate the evolved velocity proles,
the position of the centreline and the contraction of the jet. The evolved ow is unsteady and after
an initial transient part, the jet oscillates due to its natural instability. Figure 4.5 shows a snapshot of
the simulated oscillating ow. The sinuous instability of the jet can be nicely observed. Because of the
strongly asymmetric nozzle, and the sharp edges of the languid, the velocity prole of the jet in the ue
is asymmetric and the velocity vectors in the ue are not even parallel. By a height of y = 0.5 mm the
ow nearly fully evolves into an ordered jet ow. Figure 4.6 shows the velocity eld of the jet at the ue.
The horizontal line in the gure shows the y = 0.5 mm position.
The jet centreline at an arbitrary y > 0.5 mm height (below y = 0.5 mm the jet is strongly asymmetric)
was determined from the mean velocity prole as the arithmetic mean of the two x1
/2
coordinates (where
x
r
is the x position where the jet velocity is r times of the maximum). In case of symmetric velocity
proles (at y > 0.5 mm) the result of the determination of the centreline position is not sensitive to the
value of r. It was found that the jet centreline is almost a straight line starting from x
0
= 0.6 mm at
91
Figure 4.5: Velocity eld of the free jet near the nozzle. The centreline of the jet (the diagonal line) and
the monitoring lines (the horizontal lines) are also denoted (scale 3.5:1)
Figure 4.6: Velocity eld of the free jet with streamlines near the nozzle (scale 16:1)
92
4.2. FREE JET SIMULATION
the jet exit (y = 0 mm) with an angle of = 23.1
(measured from the y axis) indicated by the black

line in Figure 4.5.
In comparison, Auerlechner et al. [52] determined the centreline and width of the jet (at heights
of y = 5, 10, 15, 20, and 25 mm) as the parameters of a best t Gaussian function
_
a e
(x)
2
2
_
: the
width of the jet as 2 and the centreline position as . They found that the jet centreline starts from
x
0,A
= 0.435 mm with an angle
A
= 25.16
.
Near the ue where the velocity prole is not yet Gaussian, but symmetrical, the former method to
determine the centreline is still accurate and theoretically the two methods give the same result for the
centreline in the case of Gaussian velocity proles. Auerlechner et al. [52] used a Nolle prole to describe
the velocity of the jet at a height of y = 0.5 mm.
After transforming the coordinate system as = (x x
0
)cos ysin and = (x x
0
)sin +ycos
the jet direction in the new coordinate system will be the direction. Now, investigating an arbitrary
=
0
slice, the width of the jet can be determined as the distance between the two
r
coordinates where
r
is the position where the jet velocity is r times of the maximum. Usually r = 0.5 was used to be in
harmony with the determination of the centreline position.
It was found that near the ue the jet contracts because of the separation at the sharp languid edges
(also known as the vena contracta eect) and parallel to this the maximum value of the mean velocity
prole grows. With r = 0.5 the jet narrows to a width of about 1 mm by the y 1 mm height and the
width stays constant in the y 1 2 mm region. Thus the width of the jet in the vena contracta point is
about
vc
1 mm. The vena contracta velocity of the jet is u
vc
36.4
m
s
. The simulation was performed
so that this value is in agreement with Auerlechner et al. [52]. Thus the contraction coecient dened as
the ratio of the width of the jet at the vena contracta point and the width of the slit was approximately
0.77. In comparison, the value of the contraction coecient in Auerlechner et al. [53] was 0.78. Above
y 2 mm the jet slowly widens reaching a width of 1.1 mm at a height of about y 7 mm where the
maximum of the velocity prole is still 35.8
m
s
.
Instead of the mean velocity the time-varying velocity eld can be used to determine the centreline of the
jet in each time step, and thus, beside the mean, the standard deviation of the centreline position can
also be estimated. The mean of the time-varying centreline position of the jet at a height of y = 10 mm
was found to be at x = 3.62 mm. The mean position is almost identical with that obtained from the mean
velocity eld (x = 3.66 mm) with a standard deviation of 0.34 mm. At this height the width of the jet is
about 1.26 mm, thus the deviation of the centreline position is about 27 % of the width of the jet and is
due to the oscillation of the free jet.
Auerlechner et al. [52] measured the velocity proles of the jet at y = 0.5 mm, 5 mm, 10 mm, 15 mm,
20 mm and 25 mm (horizontal lines in Figure 4.5). Figure 4.7 shows the mean value of the simulated
unsteady velocities with their temporal root-mean-square deviation (dash-dotted lines) and the curve
tted to the velocities measured by Auerlechner et al. (dashed lines). The velocity proles from the
numerical simulations agree acceptably with the results of Auerlechner et al. [52]. Dierences are only
93
2 0 2 4 6 8 10 12 14 16 18 20
0
25
0
25
0
25
0
25
0
25
0
25
x[mm]
j
e
t
v
e
l
o
c
i
t
y
[
m
s
-
1
]
(f)
(e)
(d)
(c)
(b)
(a)
Figure 4.7: Velocity proles of the jet at dierent heights above the ue:(a) y = 0.5 mm, (b) y = 5 mm,
(c) y = 10 mm, (d) y = 15 mm, (e) y = 20 mm, (f) y = 25 mm. The solid and dash-dotted lines are the
mean velocities with their standard deviation around it from the results of the authors while the dashed
lines are the results of Auerlechner et al. [52].
that the measured velocity proles are somewhat wider and thus their maximum values are a bit lower
than the numerically simulated ones.
Having determined the position of the centreline, the centreline velocity (u
c
(y)) can be investigated.
Because of the contraction of the jet the mean velocity at the centreline increases until it reaches the
vena contracta cross section, then it remains almost constant until a height of y 7 mm and from there
an u
c
(y) C/
y function ts acceptably with C 96.6
mm
m
s
and R
2
= 0.94 . The corresponding value
in [52] was C
A
= 81.87
mm
m
s
.
4.3 Edge tone simulations
CFD simulations with the upper lip present were also carried out. The upper lip was placed at a height
of y = 10 mm above the ue parallel to the y axis, and its x oset value was varied (x = 2.05, 3.05,
3.55, 4.05, 4.55, 5.05 and 6.05 mm were investigated). This procedure corresponds to the experimental
procedure in [52]. When placed in a proper position, the jet hits the upper lip and oscillates around it,
94
4.3. EDGE TONE SIMULATIONS
thus creating an oscillating force. This oscillating force generates a dipole sound source with a strength
proportional to the amplitude of the force oscillation [50]. Thus the strongest and most stable far eld
sound is produced by the organ pipe foot model when the force oscillation acting on the wedge is the most
stable and its amplitude is the largest. When the upper lip is o its optimum position then a weaker and
less stable edge tone oscillation occurs. At higher oset distances it is even possible that the jet blows by
one side of the upper lip only oscillating due to the natural jet instability.
The spectrum of the x component of the force acting on the wedge was analysed. As the simulations
were two-dimensional and the CFD domain had a height of 1 mm, the force acting on the wedge (the
integral of the pressure on the surface of the upper lip) corresponds to a 1 mm high slice of the planar
ow. Figure 4.8 shows the spectrum for dierent x upper lip osets. The following observations can be
made:
- The clearest spectrum with a relatively strong rst stage, a weaker second stage and a much weaker
third stage oscillation was observed with the x = 4.05 mm oset of the upper lip (Figure 4.8(d)).
- At 0.5 mm away from this position (x = 3.55 mm) the second stage is stronger than the rst
one, and at +0.5 mm away (x = 4.55 mm) the spectrum is less clear and the rst stage is about
16% weaker than in the x = 4.05 mm case. (Figures 4.8(c) and 4.8(e))
- Farther away from the x = 4.05 mm position, the oscillation is rather due to the natural instability
of the free jet than the edge tone phenomenon. Figures 4.9(a) and 4.9(b) show the velocity eld as
vector plot for the x = 4.05 mm and x = 6.05 mm cases. In Figure 4.9(b) vortices can be seen
only on the right hand side of the upper lip due to the fact that the jet does not oscillate around it.
Thus the optimum conguration can be found at x = 4.05 mm upper lip oset so that the lower left
corner of the upper lip is at x = 3.55 mm. In Section 4.2 it was found that in the free jet case the jet at
10 mm has its centreline position at the x = 3.62 mm coordinate with a standard deviation of 0.34 mm.
These facts lead to the conclusion that the edge tone conguration is optimum when the lower left corner
of the upper lip is at the centreline of the jet which agrees with the conclusion of Auerlechner et al.
(numerically not specied there) [52].
With the help of the same sliding window Fourier transformation technique that is described in
Section 2.3.4 it can be investigated if the oscillation changes qualitatively in time. Figure 4.10 shows
the result of the sliding window Fourier transformation for x = 4.05, 4.55 and 5.05 mm with the same
colour scale. Similarly to the former analysis on the spectrum it can be observed that the strongest and
most stable rst stage edge tone oscillation evolves at x = 4.05 mm (i.e. when the upper lip is in the
centreline of the jet). It also can be seen that in the x = 4.05 mm case at about 0.1 s the second stage
of the edge tone becomes dominant for a short time but in most of the time the rst stage dominates. In
the x = 4.55 mm case the the contour looks similar to the x = 4.05 mm case, but the peaks are lower,
thus the stages are weaker than in the x = 4.05 mm case. Farther away from the optimum position at
x = 5.05 mm it can be seen that the spectrum is not that clear, the rst stage peak has split into two,
the amplitude of the second stage reaches the amplitude of the rst stage and the third stage is, albeit
not clearly, but present during the whole simulation. The above-described mode switching phenomenon
is a clear sign of mode coupling in a nonlinear system.
The frequencies of the three stages of the edge tone oscillation in the organ pipe foot model were
95
(a) x = 2.05 mm
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
(b) x = 3.05 mm
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
(c) x = 3.55 mm
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
(d) x = 4.05 mm
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
(e) x = 4.55 mm
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
(f) x = 5.05 mm
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
(g) x = 6.05 mm
0 1 2 3 4 5 6 7 8
0
20
40
60
f [kHz]
F
x
[
m
N
]
Figure 4.8: Spectrum of the x component of the force acting on the upper lip
96
4.3. EDGE TONE SIMULATIONS
(a) Edge tone type oscillation; x = 4.05 mm (b) Natural oscillation of the jet blowing by one side
of the upper lip; x = 2.05 mm
Figure 4.9: Computed velocity elds in the foot model of the organ pipe (scale 2.5:1)
around f
1
1100 Hz, f
2
2940 Hz and f
3
4600 Hz respectively, agreeing well with the experimental
result of Auerlechner et al. [53]: f
1,A
1105 Hz, f
2,A
2670 Hz and f
3,A
4600 Hz. Only the second
stage displays a slight deviation of about 10%. Thus the Strouhal number (based on the vena contracta
velocity u
vc
= 36.4
m
s
and width
vc
= 1 mm of the jet) St =
f
vc
u
vc
: St
1
0.0302, St
2
0.0808 and
St
3
0.1264 for the rst, the second and the third stages, respectively. Table 4.1 shows the Strouhal
numbers of the stages of the edge tone in the CFD simulation of the organ pipe foot model compared to
the top hat edge tone results (Section 2.3.3), the results of Brown [1] and those of Jones [13] at the same
dimensionless parameters (Re 2350 and h/
vc
= 10.9, where h = 10.9 mm is the distance between the
jet centreline at y = 0 mm and at y = 10 mm). Keeping in mind that the Strouhal number of the rst
stage drops to about the 85% of its former value with the advent of the second stage (Section 2.3.1) and
both Brown and Jones measured the frequencies of pure rst stage oscillations while the present results
are from the coexistence of stages, it can be concluded that the agreement between the Strouhal numbers
of the current study and former edge tone research is excellent.
Table 4.1: Strouhal numbers of the stages of the edge tone. Re 2350, h/ 10.9
Stage Organ pipe foot Edge tone Brown Jones
model CFD (Section 2.3.3) [1] [13]
1 0.0302 0.0323 0.0391 0.0358
2 0.0808 0.0924 0.0898 0.0799
3 0.1264 0.1484 0.1484 0.1367
97
(a) x = 4.05 mm
f [kHz]
t
0
[
s
]

0 1 2 3 4 5
0
0.05
0.1
0.15
0.2
0.25
0
0.5
1
1.5
2
x 10
4
(b) x = 4.55 mm
f [kHz]
t
0
[
s
]

0 1 2 3 4 5
0
0.05
0.1
0.15
0.2
0.25
0
0.5
1
1.5
2
x 10
4
(c) x = 5.05 mm
f [kHz]
t
0
[
s
]

0 1 2 3 4 5
0
0.05
0.1
0.15
0.2
0.25
0
0.5
1
1.5
2
x 10
4
Figure 4.10: Sliding window Fourier Transformation of the x component of the force acting on the upper
lip. T 0.02 s, = 0.005 s
4.4 Summary
It has been shown that the ow in the model of the foot of an organ pipe can be eectively simulated
with a commercial CFD code. Experimental results of Auerlechner et al. [52] were reproduced in many
respects. Good agreement was found in the: direction of the jet, the velocity proles, the jet centreline,
the optimum position of the upper lip (at the mean centreline position of the jet), and the frequencies
of the rst, second and third stages of the edge tone. In addition, the Strouhal numbers have been
compared to that of the edge tone (Section 2.3.3, [1, 13]) and the agreement is excellent (Thesis #7).
98
4.4. SUMMARY
The work presented here opens the possibility to investigate the inuence of other parameters in organ
pipe foot models, such as the ratio of cutup and ue width, the jet direction (by changing the position of
the languid), the sharpness of the upper lip, etc. Some numerical experiments with slight changes of the
relative position of the languid and the lower lip exerted a large inuence on the jet angle and contraction
(i.e. loss). If the upper lip was adjusted again in its optimum position, the frequency remained the same.
This indicates that the organ builder has to nd an optimum combination for the languid, lower lip and
upper lip position, matched with the pipe supply pressure. Modelling the full organ pipe would require
more work since the interaction of ow and acoustics is still not fully understood but having a reliable
tool in hand that can reproduce the edge tone ow accurately is of great value in organ pipe research.
99
4.4.1 Thesis
Thesis #7
A real-world appearance of the edge tone is the ow in an organ pipe. As the air exits the foot of the
organ pipe a jet is formed an it impinges on the upper lip.
I modelled the self-sustained oscillation in the foot of an organ pipe with the help of a commercial
CFD code (ANSYS-CFX). I investigated how the position of the upper lip (that plays the role of the
wedge in this specialized edge tone conguration) aects the evolved edge tone phenomenon. I showed
that the strongest and most stable edge tone ow and thus the strongest and most stable sound
production is achieved when the upper lip is placed in the centreline of the jet blowing from the
foot of the organ pipe. My numerical results and the experimental results of Auerlechner et al. [52]
in a similar conguration agree well.
Related publication: WoS journal: [58]
4.4.2 Tezis
7. Tezis
Az elhang egy gyakorlati elofordulasa az orgonasp lababol kilepo es a felso ajakra fel utkozo aramlas.
Kereskedelmi CFD koddal (ANSYS-CFX) modelleztem az orgonasp labreszeben kialakulo ongerjesz-
tett oszcillaciot.
Megvizsgaltam, hogy hogyan befolyasolja a kialakulo elhang jelenseget az orgonasp felso ajkanak
(ami az elhang konguracioban az ek szerepet tolti be) helyzete. Megallaptottam, hogy a legerosebb
es legstabilabb elhang jelenseget es ezaltal a legerosebb es legstabilabb hangkepzest akkor kapjuk,
amikor a felso ajak az orgonasp lababol kiaramlo szabadsugar kozepvonalaban van.
Numerikus eredmenyeim es Auerlechner et al. [52] hasonlo konguracion elert meresi eredmenyei
megegyeznek.
Kapcsolodo publikacio: WoS-os folyoirat: [58]
100
BIBLIOGRAPHY
Bibliography
[1] G. B. Brown, The vortex motion causing edge tones, P Phys Soc Lond, vol. 49, pp. 493507, 1937.
[2] G. B. Brown, The vortex motion causing edge tones, P Phys Soc Lond, vol. 49, pp. 508521, 1937.
[3] N. Curle, The mechanics of edge-tones, P Roy Soc Lond A Mat, vol. 216, no. 1126, pp. 412424,
1953.
[4] P. Savic, On acoustically eective vortex motion in gaseous jets, Philos Mag, vol. 32, no. 212,
pp. 245252, 1941.
[5] W. L. Nyborg, Self-Maintained Oscillations of the Jet in a Jet-Edge System. I, J Acoust Soc Am,
vol. 26, no. 2, pp. 174182, 1954.
[6] A. Powell, On edge tones and associated phenomena, Acustica, vol. 3, pp. 233 243, 1953.
[7] A. Powell, On the Edgetone, J Acoust Soc Am, vol. 33, no. 4, pp. 395 409, 1961.
[8] D. K. Holger, T. A. Wilson, and G. S. Beavers, Fluid mechanics of the edgetone, J Acoust Soc
Am, vol. 62, pp. 11161128, 1977.
[9] D. K. Holger, T. A. Wilson, and G. S. Beavers, The amplitude of edgetone sound, J Acoust Soc
Am, vol. 67, no. 5, pp. 15071511, 1980.
[10] D. Crighton, The edgetone feedback cycle. Linear theory for the operating stages, J Fluid Mech,
vol. 232, pp. 361391, 1992.
[11] Y.-P. Kwon, Phase-locking condition in the feedback loop of low-speed edgetones, J Acoust Soc
Am, vol. 100, pp. 30283032, November 1996.
[12] Y.-P. Kwon, Feedback mechanism of low-speed edgetones, J Acoust Soc Am, vol. 104, pp. 2084
2089, October 1998.
[13] A. T. Jones, Edge Tones, J Acoust Soc Am, vol. 14, no. 2, pp. 131139, 1942.
[14] W. L. Nyborg, M. D. Burkhard, and H. K. Schilling, Acoustical characteristics of jetedge and
jetedgeresonator systems, J Acoust Soc Am, vol. 24, no. 3, pp. 293304, 1952.
[15] J. B. Brackenridge and W. L. Nyborg, Acoustical characteristics of oscillating jet-edge systems in
water, J Acoust Soc Am, vol. 29, no. 4, pp. 459463, 1957.
[16] J. B. Brackenridge, Transverse Oscillations of a Liquid Jet. I, J Acoust Soc Am, vol. 32, no. 4,
pp. 12371242, 1960.
[17] A. Bamberger, E. Baensch, and K. G. Siebert, Experimental and numerical investigation of edge
tones, Z Angew Math Mech, vol. 84, no. 9, pp. 632646, 2004.
[18] S. Ohring, Calculations of self-excited impinging jet ow, J Fluid Mech, vol. 163, pp. 6998, 1986.
101
BIBLIOGRAPHY
[19] M. Lucas and D. Rockwell, Self-excited jet: upstream modulation and multiple frequencies, J Fluid
Mech, vol. 147, pp. 333352, 1984.
[20] N. S. Dougherty, B. L. Liu, and J. M. OFarell, Numerical simulation of the edge tone phenomenon.
NASA Contractor report 4581, 1994.
[21] J.-C. Lin and D. Rockwell, Oscillations of a turbulent jet incident upon an edge, J Fluid Struct,
vol. 15, pp. 791829, 2001.
[22] N. Fujisawa and Y. Takizawa, Study of feedback control of edge tone by simultaneous ow visal-
ization, contro and PIV measurement, Meas Sci Technol, vol. 14, pp. 14121419, 2003.
[23] B. F. C. Segoun and L. de Lacombe, Experimental investigation of the ue channel geometry
inuence on edge tone oscillations, Acta Acust United Ac, vol. 90, pp. 966975, 2004.
[24] J. F. Devillers and O. Coutier-Delgosha, Inuence of the nature of the gas in the edge-tone phe-
nomenon, J Fluid Struct, vol. 21, pp. 133149, 2005.
[25] J. Tsuchida, T. Fujisawa, and G. Yagawa, Direct numerical simulation of aerodynamic sounds by
a compressible CFD scheme with node-by-node nite elements, Comput Method Appl M, vol. 195,
pp. 18961910, 2006.
[26] J. Tsuchida, T. Fujisawa, and G. Yagawa, Computational uid dynamics on sounding mechanism
in air-reed instruments, Eng App Comp Fluid Mech, vol. 2, no. 3, pp. 375 381, 2008.
[27] T. Nonomura, H. Muranaka, and K. Fujii, Computational Analysis of Mach Number Eects on
Edgetone, in AIAA Fluid Dynamics Conference and Exhibit, (San Francisco, California), June
2006. no. 2006-2876.
[28] T. Nonomura, H. Muranaka, and K. Fujii, Computational Analysis of Mach Number Eects on the
Edgetone Phenomenon, AIAA J, vol. 48, no. 6, pp. 1248 1251, 2010.
[29] H. Kang and E. Kim, On Implementation of the Finite Dierence Lattice Boltzmann Method with
Internal Degree of Freedom to Edgetone, J Mech Sci Technol, vol. 19, no. 11, pp. 20322039, 2005.
[30] J. H. Gao and X. D. Li, Large-Eddy Simulation of the Noise from a Subsonic Jet-Edge System,
in 16th AIAA/CEAS Aeroacoustics Conference, (Stockholm, Sweden), June 2010. no. 2010-4014.
[31] A. Krothapalli and W. C. Horne, Recent observations on the structure of an edge-tone ow eld,
in AIAA/NASA 9th Aeroacoustics Conference, (Williamsburg, Virginia), October 1984. no. 84-2296.
[32] K. Takahashi, M. Miyamoto, Y. Ito, T. Takami, T. Kobayashi, A. Nishida, and M. Aoyagi, Numer-
ical analysis on 2D and 3D edge tones in terms of aerodynamic sound theory, in 20th Internationall
Congress on Acoustics, (Sydney, Australia), August 2010. no. 621.
[33] G. R. Stegen and K. Karamcheti, Multiple tone operation of edgetones, J Sound Vib, vol. 12,
no. 3, pp. 281284, 1970.
102
BIBLIOGRAPHY
[34] ANSYS Inc., Southpointe 275 Technology Drive Canonsburg, PA 15317, ANSYS-CFX.
http://www.ansys.com/Products/Simulation+Technology/Fluid+Dynamics/Fluid+Dynamics+
Products/ANSYS+CFX. (last visited June 8, 2013).
[35] J. Franke, A. Hellsten, H. Schl unzen, and B. Carissimo, Best practice guideline for the cfd
simulation of ows in the urban environment. http://www.mi.uni-hamburg.de/fileadmin/
files/forschung/techmet/cost/cost_732/pdf/BestPractiseGuideline_1-5-2007-www.pdf,
2007. (last visited June 8, 2013).
[36] R. B. Bird, W. E. Stewart, and E. N. Lightfoot, Transport Phenomena. John Wiley & Sons, 1960.
[37] R. Kaykayoglu and D. Rockwell, Unstable jet-edge interaction. Part 2. Multiple frequency pressure
elds, J Fluid Mech, vol. 169, pp. 151172, 1986.
[38] R. Kaykayoglu and D. Rockwell, Unstable jet-edge interaction. Part 1. Instantaneous pressure elds
at a single frequency, J Fluid Mech, vol. 169, pp. 125149, 1986.
[39] G. Mattingly and W. Criminale, Disturbance characteristics in a plane jet, Phys Fluids, vol. 14,
no. 11, p. 22582264, 1971.
[40] G. Paal, J. Angster, W. Garen, and A. Miklos, A combined LDA and ow-visualization study on
ue organ pipes, Exp Fluids, vol. 40, pp. 825835, 2006.
[41] M. J. Lighthill, On Sound Generated Aerodynamically. I. General Theory, P Roy Soc A-math Phy,
vol. 211, no. 1107, pp. 564587, 1952.
[42] N. Curle, The Inuence of Solid Boundaries on Aerodynamics Sound, P Roy Soc Lond A Mat,
vol. 231, pp. 505514, 1955.
[43] J. Ffowcs Williams and D. Hawkings, Sound Generation by Turbulence and Surfaces in Arbitrary
Motion, Philos T Roy Soc A, vol. 264, pp. 321342, May 1969.
[44] A. Powell, Theory of Vortex Sound, J Acoust Soc Am, vol. 36, pp. 177195, 1964.
[45] M. S. Howe, Contributions to the Theory of Aerodynamic Sound with Application to Excess Jet
Noise and the Theroy of the Flute, J Fluid Mech, vol. 71, pp. 625673, 1975.
[46] W. De Roeck, Hybrid methodologies for the computational aeroacoustics analysis of conned subsonic
ows. PhD thesis, Department of Mechanical Engineering, Katholiek Universiteit Leuven, Belgium,
May 2007. http://people.mech.kuleuven.be/
~
wdroeck/docs/thesis.pdf, (last visited June 8,
2013).
[47] M. Kaltenbacher, Advanced simulation tool for the design of sensors and actuators, in Proceedings
Eurosensors XXIV, pp. 597 600, 2010.
[48] M. Kaltenbacher, Numerical Simulation of Mechatronic Sensors and Actuators. Berlin: Springer,
2. ed., 2007.
103
BIBLIOGRAPHY
[49] B. Kaltenbacher, M. Kaltenbacher, and I. Sim, Stability of a Reduced Perfectly Matched Layer
Method for the Second Order Wave Equation in Time Domain, in Proceedings of the 10th Interna-
tional Conference on the Mathematical and Numerical Aspects of Waves (N. Nigam, ed.), pp. 455
458, The Pacic Institute for the Mathematical Sciences, July 2011.
[50] M. J. S. Lighthill, Waves in uids. Cambridge University press, 1978.
[51] M. J. Crocker, Handbook of Acoustics. A Wiley-Interscience Publication, Wiley, 1998.
[52] H. J. Auerlechner, T. Trommer, J. Angster, and A. Miklos, Experimental jet velocity and edge
tone investigations on a foot model of an organ pipe, J Acoust Soc Am, vol. 126, no. 2, pp. 878
886, 2009.
[53] H. J. Auerlechner, Stromungsakustische Untersuchungen des Schneidentons und Visualisierungen
des Freistrahls mithilfe eines Orgelpfeifenfumodells. PhD thesis, Universitat Stuttgart, Lehrstuhl
f ur Bauphysik, 2010.
[54] A. W. Nolle, Sinous instability of a planar air jet: Propagation parameters and acoustic excitation,
J Acoust Soc Am, vol. 103, pp. 36903705, June 1998.
104
OWN PUBLICATIONS
WoS journal publications
[55] G. Paal and I. Vaik, Unsteady phenomena in the edge tone, Int J Heat Fluid Fl, vol. 28, no. 4,
pp. 575 586, 2007. IF: 1.927.
[56] I. Vaik and G. Paal, Mode switching and hysteresis in the edge tone, J Phys Conf Ser, vol. 268,
no. 1, p. No. 012031, 2011.
[57] I. Vaik, G. Paal, M. Kaltenbacher, S. Triebenbacher, S. Becker, and I. Shevchenko, Aeroacoustics of
the edge tone: 2D-3D coupling between CFD and CAA, Acta Acust, vol. 99, pp. 245259, Mar.-Apr.
2013. IF: 0.569.
[58] I. Vaik and G. Paal, Flow simulations on an organ pipe foot model, J Acoust Soc Am, vol. 133,
pp. 11021110, Feb. 2013. IF: 1.55.
Non-WoS publications
[59] I. Vaik and G. Paal, Numerical simulations of the edge tone, in Forum Acusticum Budapest 2005:
4th European Congress on Acustics, (Budapest, Hungary), pp. 635639, 2005.
[60] I. Vaik and G. Paal, Unsteady ow phenomena in the edge tone, in Proceedings of Conference on
Modelling Fluid Flow (CMFF06), (Budapest, Hungary), pp. 332338, Sept. 2006.
[61] I. Vaik and G. Paal, Frequency and phase characteristics of the edge tone, in Acoustics08, (Paris,
France), pp. 753758, Jun. 2008.
[62] I. Vaik and G. Paal, Experiments on the edge tone, in Conference on Modelling Fluid Flow
(CMFF09), (Budapest, Hungary), pp. 711 718, Sept. 2009.
[63] I. Vaik and G. Paal, Az elhang aeroakusztikai szimulacioja, in OG
ET 2007: XV. Nemzetkozi Gepesz

Talalkozo, (Kolozsvar, Romania), pp. 403406, Apr. 2007. In Hungarian.
[64] I. Vaik and G. Paal, Geometrical and velocity prole inuences on the edge tone frequency, in
Proceedings of ISMA 2007, (Barcelona, Spain), Sept. 2007. No. 645.
[65] I. Vaik and G. Paal, Stage jumps in the edge tone, in 17th International Congress on Sound and
Vibration (ICSV17), (Cairo, Egypt), Jul. 2010. No. 854.
[66] I. Ali, I. Vaik, M. Escobar, M. Kaltenbacher, G. Paal, and S. Becker, Coupled ow and acoustic
simulations in the case of an edge tone conguration and a square cylinder, in Proceedings of
Conference on Modelling Fluid Flow (CMFF06), (Budapest, Hungary), pp. 340347, Sept. 2006.
[67] I. Vaik, I. Ali, M. Escobar, M. Kaltenbacher, S. Becker, and G. Paal, Computational noise prediction
of the edge tone, in Proceedings of ACOUSTICS High Tatras 06, (Strbske Pleso, Slovakia), Oct.
2006. No. 87.
105
OWN PUBLICATIONS
[68] I. Vaik, I. Ali, M. Escobar, M. Kaltenbacher, S. Becker, and G. Paal, Two- and three dimensional
coupling in the noise prediction of the edge tone, in Proceedings of 14th International Congress on
Sound and Vibration, (Cairns, Australia), Jul. 2007.
106
APPENDIX
Appendix
0.4114 0.4364 0.4614
0.02
0.01
0
0.01
0.02
t [s]
F
[
m
N
]
(a)
0 50 100 150 200 250 300
0
0.005
0.01
0.015
f [Hz]
a
m
p
F
[
m
N
]
(b)
0.4114 0.4364 0.4614
12
14
16
18
20
t [s]
x
t
[
m
m
]
(c)
0 10 20 30
0
0.25
0.5
0.75
1
1.25
1.5
p
r
m
s
[
P
a
]
(d)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/h [-]
/
2
[
-
]
(e)
0 1 2 3 4 5 6 7 8 9 10
0
0.2
0.4
0.6
0.8
1
x [mm]
v
s
t
d
[
m
s
1
]
(f )
Figure A.1: Miscellaneous CFD results; Stage I at Re = 150, h/ = 10 with top hat prole; for description
of subgures see Section 2.3.5
107
APPENDIX
0.2182 0.2204 0.2226
0.4
0.2
0
0.2
0.4
t [s]
F
[
m
N
]
(a)
0 500 1000 1500 2000 2500
0
0.05
0.1
0.15
0.2
0.25
f [Hz]
a
m
p
F
[
m
N
]
(b)
0.2182 0.2204 0.2226
12
14
16
18
20
t [s]
x
t
[
m
m
]
(c)
0 10 20 30
0
5
10
15
20
25
30
35
p
r
m
s
[
P
a
]
(d)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/h [-]
/
2
[
-
]
(e)
0 1 2 3 4 5 6 7 8 9 10
0
1
2
3
4
x [mm]
v
s
t
d
[
m
s
1
]
(f )
Figure A.2: Miscellaneous CFD results; coexistence of Stage I & II at Re = 600, h/ = 10 with top hat
prole; for description of subgures see Section 2.3.5
108
APPENDIX
0.0289 0.0299 0.0309 0.0319 0.0329 0.0339
6
4
2
0
2
4
6
t [s]
F
[
m
N
]
(a)
0 1000 2000 3000 4000 5000 6000
0
0.5
1
1.5
2
f [Hz]
a
m
p
F
[
m
N
]
(b)
0.0289 0.0299 0.0309 0.0319 0.0329 0.0339
8
10
12
14
16
18
20
22
24
t [s]
x
t
[
m
m
]
(c)
0 10 20 30
0
50
100
150
200
250
300
p
r
m
s
[
P
a
]
(d)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/h [-]
/
2
[
-
]
(e)
0 1 2 3 4 5 6 7 8 9 10
0
2
4
6
8
10
12
x [mm]
v
s
t
d
[
m
s
1
]
(f )
Figure A.3: Miscellaneous CFD results; coexistence of Stage I & II & III at Re = 1800, h/ = 10 with
top hat prole; for description of subgures see Section 2.3.5
109
APPENDIX
0.1449 0.1595 0.1742
0.06
0.04
0.02
0
0.02
0.04
0.06
t [s]
F
[
m
N
]
(a)
0 100 200 300 400 500
0
0.01
0.02
0.03
0.04
0.05
f [Hz]
a
m
p
F
[
m
N
]
(b)
0.1449 0.1595 0.1742
12
14
16
18
20
t [s]
x
t
[
m
m
]
(c)
0 10 20 30
0
0.5
1
1.5
2
2.5
3
3.5
p
r
m
s
[
P
a
]
(d)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
x/h [-]
/
2
[
-
]
(e)
0 1 2 3 4 5 6 7 8 9 10
0
0.2
0.4
0.6
0.8
1
1.2
x [mm]
v
s
t
d
[
m
s
1
]
(f )
Figure A.4: Miscellaneous CFD results; Stage I at Re = 200, h/ = 10 with parabolic prole; for
description of subgures see Section 2.3.5
110

Értekezés Az Élhang Témakörében

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Értekezés Az Élhang Témakörében

Transféré par

Droits d'auteur :

Formats disponibles

Budapest University of Technology and Economics

Department of Hydrodynamic Systems

error of the jet width measurment m

G strength of dipole sound source N

amplitude of the acoustic pressure uctuation Pa

coecient of St (h/) in eq. (2.45) -

coecient of St (Re) in eq. (2.43) -

is used instead of his semi-empirical

and the usual inequalities for norms:

. Let us assume that the initial values are correct

. Its height (z direction) was 150 mm

) and the error of the area measurement

CFD 0.7387 0.04010 1.079 0.03541 4.072 0.1034 13.76 0.1740

. The values of c and St

deduced by the method of least squares for the dierent

pure multi pure multi

value if a pure inverse

CFD 0.4079 0.002681 0.4288 0.003141 0.9174 0.000980 - -

are plotted for comparison

(where T is the period) multiplied by h yields the

G) with the ones calculated on

G) for edge tone oscillations at dierent Reynolds numbers. Results

G). This is due

) at 2 m away from the tip of the wedge at 90

. The upper lip - when present - had

(measured from the y axis) indicated by the black

y function ts acceptably with C 96.6

ET 2007: XV. Nemzetkozi Gepesz

Vous aimerez peut-être aussi