Professional Documents
Culture Documents
The Use of Surface Integral Methods in Computational Jet Aeroacoustics
The Use of Surface Integral Methods in Computational Jet Aeroacoustics
The Use of Surface Integral Methods in Computational Jet Aeroacoustics
AEROACOUSTICS
A Thesis
of
Purdue University
by
FongLoon Pan
of
December 2004
ii
ACKNOWLEDGMENTS
TABLE OF CONTENTS
Page
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
NOMENCLATURE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 History and Development of Computational Aeroacoustics . . . . . . 1
1.2 Acoustic Analogy based methods . . . . . . . . . . . . . . . . . . . . 2
1.3 Surface Integral Methods . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Objective of the Present Research . . . . . . . . . . . . . . . . . . . 6
2 SURFACE INTEGRAL METHODS . . . . . . . . . . . . . . . . . . . . . 8
2.1 Formulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Kirchhoff Method . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 Porous Ffowcs Williams - Hawkings Method . . . . . . . . . . 10
2.2 Numerical Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Retarded Time Algorithm . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Forward Time Algorithm . . . . . . . . . . . . . . . . . . . . . 14
3 MEAN FLOW REFRACTION CORRECTIONS . . . . . . . . . . . . . . 16
3.1 Geometric Acoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Lilley’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4 VALIDATION OF SURFACE INTEGRAL METHODS FOR A MONOPOLE
SOURCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1 Grid Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 Kirchhoff Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2.1 Test 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
v
Page
4.2.2 Test 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Porous Ffowcs Williams - Hawkings Method . . . . . . . . . . . . . . 38
4.4 Assessment of the Forward Time Approach . . . . . . . . . . . . . . . 44
4.5 Frequency Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.5.1 Kirchhoff’s Method . . . . . . . . . . . . . . . . . . . . . . . . 48
4.5.2 Porous Ffowcs Williams-Hawkings Method . . . . . . . . . . . 49
4.6 Refraction Corrections . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5 SURFACE INTEGRAL METHODS APPLIED TO JET NOISE PREDIC-
TION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1 LES Jet Noise Calculation . . . . . . . . . . . . . . . . . . . . . . . . 55
5.2 Porous Ffowcs Williams - Hawkings Method . . . . . . . . . . . . . . 56
5.3 Refraction Corrections . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6 CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE WORK . 64
6.1 Surface Integral Methods . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.2 Assessment of Forward Time Approach . . . . . . . . . . . . . . . . . 65
6.3 Mean Flow Refraction Correction . . . . . . . . . . . . . . . . . . . . 66
LIST OF REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
vi
LIST OF TABLES
Table Page
4.1 Test 1: Accuracy studies for cubical control surface. R, φ and θ are
observer locations defined in Figure 4.2; N is the number of points
per wavelength; T is the number of temporal points . . . . . . . . . . 28
4.2 Test 2: Accuracy studies for curvilinear control surface. R, φ and θ
are observer locations defined in Figure 4.2; N is the number of points
per wavelength; T is the number of temporal points; . . . . . . . . . . 34
4.3 Accuracy studies for curvilinear control surface. R, φ and θ are ob-
server locations defined in Figure 4.2; N is the number of points per
wavelength; T is the number of temporal points . . . . . . . . . . . . 38
4.4 Accuracy studies for assessment of forward time approach. R, φ and θ
are observer locations defined in Figure 4.2; N is the number of points
per wavelength; T is the number temporal points . . . . . . . . . . . 44
5.1 Some jet noise experiments with Mach numbers similar to that of the
LES by Uzun et al. [35, 36] . . . . . . . . . . . . . . . . . . . . . . . . 59
vii
LIST OF FIGURES
Figure Page
3.1 Sound emission and propagation angle of a source due to refraction. . 17
3.2 |Gω /Gω | at (a) ∆ϕ = 0; (b) ∆ϕ = 30; (c) ∆ϕ = 90; (d) ∆ϕ = 120. . . 23
4.1 Y-X Control Surface Meshes. . . . . . . . . . . . . . . . . . . . . . . 26
4.2 Observer coordinates system. . . . . . . . . . . . . . . . . . . . . . . 29
4.3 Error Analysis A: Effect of temporal points and θ orientation with
observers at 1.75r. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.4 Error Analysis B: Effect of temporal points and φ orientation with
observers at 1.75r, θ = 90o . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.5 Error Analysis C: Effect of temporal points and φ orientation with
observers at 1.75r, θ = 0o . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.6 Error Analysis D: Effect of number of points per wavelength and φ
orientation with observers at 1.75r. . . . . . . . . . . . . . . . . . . . 32
4.7 Error Analysis E: Effect of temporal points and φ orientation with
observers at 17.5r. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.8 Error Analysis F: Effect of number of points per wavelength and φ
orientation with observers at 17.5r. . . . . . . . . . . . . . . . . . . . 33
4.9 Error Analysis G: Effect of number of points per wavelength for ana-
lytical and numerical Kirchhoff method with observers at 17.5r. . . . 33
4.10 Generalized Curvilinear Kirchhoff Control Surface . . . . . . . . . . . 35
4.11 Error Analysis A: Effect of number of points per wavelength and φ
orientation with observers at 17.5r. . . . . . . . . . . . . . . . . . . . 36
4.12 Error Analysis B: Effect of temporal points, number of points per
wavelength, and φ orientation with observers at 17.5r. . . . . . . . . . 36
4.13 Error Analysis C: Effect of number of points per wavelength for ana-
lytical and numerical Kirchhoff method with observers at 5r and 17.5r. 37
4.14 Error Analysis D: Effect of observer distance for curvilinear control
surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
viii
Figure Page
Figure Page
5.4 Overall sound pressure levels at 60rJ from the nozzle exit. . . . . . . 60
5.5 Jet mean-flow profile. . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
x
NOMENCLATURE
Roman Symbols
Ai Airy function
p Local pressure
ā
a Local sound speed normalized by its ambient value, a ≡ ao
St Strouhal number
ω
ko Streamwise wavenumber, ko = ao
L Wave operator
Greek Symbols
φ Acoustic variable
ω Cyclic frequency
ρ Fluid density
Other Symbols
av Average parameter
c Corrected parameter
m Measured parameter
′
Perturbation quantity (e.g.ρ′ = ρ − ρo )
·
Source time derivative
ˆ Unit vector
Abbreviations
AA Acoustic Analogy
GA Geometric Acoustics
SGS Subgrid-scale
xiv
ABSTRACT
1. INTRODUCTION
For the design process of any new airplane or helicopter, aerodynamics noise gen-
erated from fluids is one of the main concerns. There are many kinds of aerodynamics
noise including turbine jet noise, impulsive noise due to unsteady flow around wings
and rotors, broadband noise due to inflow turbulence and boundary layer separated
flow, etc. Accurate prediction of the noise mechanisms is essential in order to be
able to meet low-noise requirements for the design. Due to high expense, safety
risks, and atmospheric variability of the experimental tests, numerical approaches
are more appealing. Although complete noise models have not yet been developed,
numerical simulations with a proper model are increasingly being employed for the
prediction of aerodynamic noise as the availability of the computer resources is in-
creasing rapidly. This research has led to the emergence of a relatively new field:
Computational Aeroacoutics (CAA).
tion, and yet all of the methodologies developed up-to-date have their own strengths
and weakness. These methods used for CFD/ source modeling include Reynolds av-
eraged Navier-Stokes (RANS), Large Eddy Simulation (LES) and Direct Numerical
Simulation (DNS) based models. Here we will discuss further the methods used for
sound propagation.
We first introduce the Acoustic Analogy (AA) which has been used widely in
CAA. An acoustic analogy (AA) [1] is defined as an exact rearrangement of the
Navier-Stokes or Euler equations which leads to an equation for the propagation
of acoustic waves in a medium at rest or in a defined medium, with an equivalent
source term wherein the sources may move relative to the given mean motion of the
medium. The early work of AA was introduced by Lighthill. In his early work [2],
Lighthill had established some important basis for the jet noise prediction problem.
He derived a linear wave equation from the governing Navier-Stokes equations for a
stationary medium with a quadrupole-type source term. This approach is usually
referred to the Lighthill’s acoustic analogy. The strength and distribution of the
equivalent acoustic sources are determined from solutions of the exact Navier-Stokes
equations, numerical modeling, or experiments. Once the strength and distribution
of the equivalent acoustic sources are determined, the far-field noise reaching an
observer can be found from a weighted volume integration of the distribution of
sources. This is exact for a compact source distribution, especially at low Mach
numbers where the ratio of the characteristic length scale of the source region is
negligible compared to the typical acoustic wavelength. At higher Mach numbers the
acoustic source becomes non-compact with the result that flow-acoustic interaction
occurs within the flow. In the jet noise problem or when the observer is placed in
a uniform flow, Lighthill’s AA cannot predict the sound fields correctly due to the
mean flow effects on acoustic propagation.
3
Curle [8] has extended the Lighthill’s AA by introducing the effects of solid
walls at rest. Ffowcs Williams and Hawkings [12] further generalized the form of the
Lighthill’s AA to include moving surfaces. Although the original paper [12] simplified
the equations by restricting the surfaces to solid boundaries, the extensions of the
FW-H equation to permeable surfaces has been known since the original derivation.
However, the utilitization of the permeable surface form of the FW-H equation has
only recently been realized and will be discussed in the following Chapters.
More recently, Tam and Auriault [13] introduced a new approach based on the
use of the linearized Euler equations (LEE) to describe the acoustic propagation and
a phenomenological model for the noise sources. Morris and Farassat [14] made a
comparison of this method with the traditional acoustic analogy, and showed that
the results are similar if the same source distributions are assumed. Also, an AA
method recently described by Morris and Boluriaan [15] was able to predict noise
4
at all angles to the jet axis. This AA is formulated with the LEE as the sound
propagator. Flow-acoustic interaction effects are predicted using both high and low
frequency asymptotic solutions for the Green’s function to both Lilley’s equation and
the LEE. With their AA implementation, an increase in the radiated noise levels at
low frequencies was obtained. However, as mentioned by Morris and Boluriaan [15],
further improvements were desired to be sought to overcome the high frequency limit
of the asymptotic Lilley’s equation. One of the alternatives is to replace with the
adjoint Green’s function introduced by Tam and Auriault [16].
The use of surface integral methods has increased substantially the last 10 years,
because of the development of reliable CFD methods that can be used for the evalu-
ation of the near-field. A review of the use of surface integral methods was given by
Lyrintzis [17]. Several researchers have used these methods for the jet noise problem.
Soh [18], Mitchel et al. [19], Zhao et al. [20], and Billson et al. [21] used the stationary
Kirchhoff method. Lyrintzis and Mankbadi [22], Chyczewski and Long [23], Morris
5
et al. [24], Gamet and Estivalezes [25], Choi et al. [26] and Kandula and Caimi [27]
used the uniformly moving formula. It should be noted that most of the above refer-
ences use Large Eddies Simulation (LES) code for the CFD data. However, a RANS
code can also be used, as shown in Ref. [27], where OVERFLOW [28] was used.
Lyrintzis and Mankbadi [22] also compared time and frequency domain formulations.
Mankbadi et al. [29] applied a modified Green’s function to avoid the evaluation of
normal derivatives. Balakumar [30] and Yen [31] used parabolized stability equations
for the jet simulation and a cylindrical (i.e. two-dimensional) Kirchhoff formulation
for the noise evaluation. Shih et al. [32] compared several Kirchhoff formulations
with the traditional acoustic analogy, extending the LES calculations and using a
zonal LES + LEE method. The results showed that the Kirchhoff method is much
more accurate than the acoustic analogy (for the compact source approximation
used) and much cheaper than extending the LES or performing a zonal LES + LEE.
Finally, Morris et al. [33, 34] and Uzun et al. [35, 36] used the porous FW-H method
and Hu et al. [37], used a two-dimensional formulation of the porous FW-H equation
to evaluate noise radiation from a plane jet. Uzun et al. [35,36] compared FW-H and
Kirchhoff for cold jets. Their comparison showed that the methods produced similar
results. Rahier et al. [38] compared the methods for numerical acoustic predictions
for hot jets. Both the Kirchhoff method based on pressure disturbance and the FW-
H equation gave good results, whereas the Kirchhoff method based on density gave
erroneous results.
Most of the above approaches have used an open control surface (i.e. without the
downstream end) in order to avoid placing the surface in a nonlinear region. Freund
et al. [39] showed a means of correcting the results to account for an open control
surface, for cases that the observer is close to the jet axis. More detailed discussion
of the effects of closing the surface can be found in Uzun et al. [36] and Rahier et
al. [38] Pilon and Lyrintzis [40–42] developed a method to account for quadrupole
sources outside the control surface. This approximation is based on the assumption
that all wave modes approximately decay in an exponential fashion. The volume
6
The equivalence of the porous FW-H equation and Kirchhoff formulation was
proven by Pilon & Lyrintzis [40] and Brentner & Farassat [45]. They showed that,
for a control surface placed in a linear region, the porous surface FW-H formula-
tion is equivalent to the linear Kirchhoff formulations, plus a volume integral of
quadrupoles. The Kirchhoff’s control surface is also restricted to a region where
the linear wave equation is satisfied. This leads to a higher sensitivity in the choice
of the control surface for the Kirchhoff’s method. Rahier et al. [38] compared the
methods for numerical acoustic predictions for hot jets. Both the Kirchhoff method
based on pressure disturbance and the FW-H equation gave good results, whereas
the Kirchhoff method based on density gave erroneous results.
The goal of the current study is to improve the state-of-the-art predictive tech-
niques, so that aircraft noise can be reduced. A previous set of Kirchhoff and Ffowcs-
Williams Hawkings (FW-H) equation subroutines in both time and frequency do-
main has been tested and extended to a generalized curvilinear control surface. The
accuracy of both methods has been studied. Forward time subroutines were also
developed. Mean flow refraction corrections have been investigated and included in
the subroutines in order to account for the zone of silence. The mean flow refraction
corrections have been studied using simple geometric acoustics (GA) approxima-
7
tion [44, 46]. This approximation is valid only at high-frequency region. However,
azimuthal variations between the sources in the sheared layer and the observers were
not taken into account and hence the accuracy of this correction may be limited.
It is worthwhile to attempt a more rigorous method in order to achieve more re-
liable prediction. A linearized homogeneous third-order convective wave equation,
namely Lilley’s equation [7], has been solved asymptotically using the stationary
phase method in the far-field [47]. The high-frequency limit solution [48] of Lilley’s
equation is re-visited and applied to refraction corrections. The frequency range in
which the approximation solution of Lilley’s equation holds, is also studied.
The outline of this thesis is as follows: After the introduction in Chapter 1, the
formulations and numerical algorithms of surface integral methods are described in
Chapter 2 followed by the mean flow refraction corrections in Chapter 3. In Chapter
4, the surface integral methods as applied to a monopole source are shown and the
effects of the geometric acoustics and the Lilley’s equation corrections are analyzed.
In Chapter 5, jet noise prediction using surface integral method is studied and the
effects of the geometric acoustics and the Lilley’s equation corrections are analyzed.
Conclusions are given in Chapter 6, where also some directions for future work are
proposed. Part of this thesis can be found in Ref. [49, 50].
8
2.1 Formulations
of light and in other electromagnetic problems, has also many applications in studies
of acoustic wave propagation.
Assume the linear, homogeneous wave equation
1 ∂ 2φ ∂ 2φ
− =0 (2.1)
a2◦ ∂t2 ∂xi ∂xi
is valid for some acoustic variable, φ, and sound speed, ao n the entire region outside
of a closed and bounded smooth surface. For stationary surfaces, the Kirchhoff
formula is Z · ¸ Z
1 1 ∂φ [φ]ret
4πφ(~x, t) = φ̇ cos θ − dS + 2
dS (2.2)
S R a◦ ∂n ret S R
where cos θ = R̂ · n̂, θ is the source emission angle, n̂ is the outward directed unit
normal vector of the closed control surface S, R is the distance between the observer
and the surface source. The surface integrals are over the control Kirchhoff surface,
S, and evaluated at retarded or emission time, τ described by the following equation
R
τ =t− (2.3)
ao
where t is observer or reception time. The dependent variable φ is normally taken to
be the disturbance pressure, but can be any quantity which satisfies the linear wave
equation.
With the use of a Fourier transform,
Z ∞
F{f (t)} = fb(ω) = f (t)e−iωt dt
−∞
Z ∞
1
F −1 {fb(ω)} = f (t) = f (ω)eiωt dt (2.4)
2π −∞
equation (2.2) can be expressed in the frequency domain (i.e, starting from Helmholtz
equation) as
" µ #
Z
1 iω ∂ b ¶ φb cos θ
φ
b x, ω) =
4π φ(~ e−iωR/a◦ cos θ φb − + dS (2.5)
S R a ◦ ∂n R2
where φb is the Fourier transform of φ, and ω is the cyclic frequency. Both time
and frequency expressions can be extended to a uniform rectilinear motion [42].
10
Equations (2.2) and (2.5) work well for aeroacoustic predictions when the control
surface is placed in a region of the flow field where the linear wave equation is
valid. However, this might not be possible for some cases. Additional nonlinearities
(i.e. quadrupoles) can be added outside the control Kirchhoff surface [54]. Kirch-
hoff’s method has also been extended to a moving deformable, piecewise smooth
surface [55], open surfaces [39] and to include mean flow refraction effects based on
the GA method [44].
A general version of the FW-H equation is needed because the usual practice
is to assume that the FW-H integration surface corresponds to a solid body and is
impenetrable. For the simple case of a stationary surface we can define (following
Brentner and Farassat [45] and Francescantonio [56]) new variables Ui and Li as
ρui
Ui = (2.6)
ρo
and
Li = Pij n̂j + ρui un (2.7)
where Z · ¸
∂ ρo Un
4πp′T (~x,
t) = dS (2.9)
∂t S R ret
Z · ¸ Z · ¸
′ 1 ∂ Lr Lr
4πpL (~x, t) = dS + dS (2.10)
a◦ ∂t S R ret S R2 ret
11
and the quadrupole noise pressure p′Q (~x, t) can be determined by any method cur-
rently available. The superscript ′
implies disturbances (e.g. ρ = ρ′ + ρo ). The
subscript r or n indicates a dot product of the vector with the unit vector in the
radiation direction r̂ or the unit vector in the surface normal direction n̂i , respec-
tively. It should be noted that the three pressure terms have a physical meaning for
rigid surfaces: p′T (~x, t) is known as thickness noise, p′L (~x, t) is called loading noise
and p′Q (~x, t) is called quadrupole noise. For a porous surface the terms lose their
physical meaning, but the last term p′Q (~x, t) still denotes the quadrupoles outside
surface S.
An alternative way [56] is to move the time derivative inside the integral: (For-
mulation II)
Z " #
ρo U̇n
4πp′T (~x, t) = dS (2.11)
S R
ret
Z " # Z · ¸
1 L̇r Lr
4πp′L (~x, t) = dS + dS (2.12)
a◦ S R S R2 ret
ret
The dot over a variable implies source-time differentiation of that variable. It appears
that Formulation I (Eqs. 2.8-2.10) has less memory requirements, because it does
not require storage of the time derivatives, and requires less operations per integral
evaluation. For our case of a stationary Kirchhoff surface the integral can be reused
at the next time step. Formulation I was used by Strawn et al. [57] for rotorcraft
noise predictions using a nonrotating control surface with very good results. On
the other hand taking the time derivative inside could prevent some instabilities for
moving control surfaces. Thus Formulation II (Eqs. 2.11-2.12) might be more robust
for a moving control surface, Formulation II was used for rotorcraft noise prediction
by Brentner and Farassat [45] with a rotating control surface with very good results.
With the use of Fourier transform pair in Eq. (2.4), both formulations can be
written in the frequency domain for a stationary surface as
Z bn
ρo U
p′T (~x,
4πb ω) = iω e−iωR/a◦ dS (2.13)
S R
12
Z b Z br
iω −iωR/a◦ Lr L
p′L (~x,
4πb ω) = e dS + dS (2.14)
a◦ S R S R2
bn , and L
where pb′ , U br are the Fourier transforms of p′ , Un , and Lr , respectively.
and ω is the cyclic frequency. Both the above formulations provide a Kirchhoff-
like formulation if the quadrupoles outside the control surface (p′Q (~x, t) term) are
ignored. The equivalence of the porous FW-H equation and Kirchhoff formulation
was proven by Pilon & Lyrintzis [54] and Brentner & Farassat [45]. The reader
is referred to the above references for the details. The major difference between
∂p ∂p
Kirchhoff’s and FW-H formulation is that Kirchhoff’s method needs p′ , ∂n , ∂t as
input, whereas the porous FW-H needs p′ , ρ, ρui . Also, the porous FW-H method
puts less stringent requirements on the placement of surface in the linear region than
Kirchhoff’s method, as shown in Ref. [45] for a rotorcraft noise problem. However for
fine meshes, the results are similar [36]. The method has been extended to include
refraction corrections based on GA in Ref. [46] but no applications were given.
The crucial part in evaluating the surface integral equation is to account for the
time-lag between emission (source) and reception (receiver) times. Two approaches
can be used to achieve this: the retarded time approach, and the forward time
approach.
The retarded time approach has been the basis of several Kirchhoff codes, i.e.
Lyrintizis & Mankbadi [22], Strawn et al. [58], Lyrintizs et al. [59], Polacsec & Prier
[60]. Brentner & Farassat [45] and Strawn et al. [58] have also applied this algorithm
to the FW-H codes. As mentioned in several articles, i.e. Wissink et al. [61], and
Strawn et al. [62], this algorithm can be easily parallelized by partitioning the control
surface and distributing to different processors. Since the only communication is the
13
final global summation the parallel efficiency of the code is very high. Lockard [63]
have discussed parallelization of FW-H codes. Long and Brentner [64] proposed a
master-slave approach for load balancing.
The retarded time approach starts with a sequence of reception times. For each
source element, the corresponding emission time is determined at a given reception
time and depends on the distance from the corresponding source element to the
receiver. The far-field signal at a given reception time is then computed by taking
the integral over all the source strengths evaluated at the respective emission times.
One can write the retarded time algorithm simply from
|x − y|
τk = t − (2.15)
ao
where τk is the emission time at the k-th source element, for a given reception time,
t.
When the control surface is in motion, the emission and reception times are
implicitly related by
|x − y(τk )|
τk − t + =0 (2.16)
ao
Equation (2.16) involves nonlinearily, and hence cannot be solved analytically. There-
fore, root finding methods are used such as Newton-Raphson or divide and conquer.
Since the flow times at which the flow data is collected, in very rare situations
coincide with the emission times, interpolation is required. Suggested interpolations
are linear or spline methods. Storage of the source strength is needed at all surface
elements at all time steps, requiring extensive memory usage when fine control sur-
faces are used. Moreover, it is difficult to write a versatile code for various mesh
topologies used by current CFD codes, including unstructured grids.
14
The forward time calculation had not gained its popularity until recently. This
approach was suggested by Brentner [65] in 1994. Lyrintzis et al [66] have imple-
mented the forward time approach in Kirchhoff method to rotocraft noise predic-
tion. At the same time, Ozyoruk et al [67] have also computed the sound radiated
from engine inlets using this approach. Recently, Kim et al [68] have predicted
the sound pressure near the side-view mirror by FW-H method with this approach.
Casalino [69] has used the term ”advanced time approach” and used for his rotor
noise prediction code, Advantia.
The forward time algorithm starts with a sequence of emission times. One can
consider using flow times as emission times. In this approach, for a given emission
time, the source strengths at all source elements are computed, and are projected
forward into the reception times when each sound signal from each source element
reaches the receiver. The reception times can be computed from Eq. (2.17)
|x − y|
t=τ+ (2.17)
ao
Since the batch of sound signals emitted at the same emission time will arrive at
the receiver at different times, this algorithm results in a sequence of sound signals
arriving at the receiver at different times with unequal intervals. Computation of
sound signals at the sequence of desired reception times, therefore, would require
interpolations of the sound signals reaching the receiver.
Brentner [65] has written this algorithm symbolically as
N
X
′
4πp (x̄, t) = I[Ki (t), t∗ ] (2.18)
i=1
The surface integral methods assume that the sound propagation outside the control
surface is governed by the simple wave equation, and can efficiently and accurately
predict aerodynamically generated sound provided the control surface surrounds the
entire nonlinear source region [17, 46]. However, in jet noise prediction, the control
surface cannot enclose the entire aerodynamic generated source region. Also, there
is a mean shear flow downstream of the control surface. Thus, the linear wave
equation is invalid when the sound propagates through the region near the jet axis,
downstream of the control surface.
When acoustic waves propagate through the jet mean flow downstream of the
control surface, they cross velocity-shear interfaces, which deflect the radiated sound
away from the mean flow direction. This effect is called refraction [47]. For isothermal
or hot jets, the effect of refraction gives rise to a “zone of silence”, a region where a
substantial reduction of sound occurs. Early investigations [47,70,71] found that the
waves propagating in the zone of silence initially start as evanescent waves near the
source. The presence of the zone of silence was also measured experimentally [72–74].
Surface integral methods are not able to predict the refraction outside the control
surface and the resulting zone of silence.
The refraction problem has drawn attention, because of its importance in the
open jet wind tunnels experiments. Early development of solution of the refraction
by a thick cylindrical shear layer with a source on the jet axis was given by Tester
and Morfey [70], and for a plane shear layer by Amiet [75]. A recent extension for
the source off the jet axis was given by Morfey and Joseph [76]. These refraction
17
ϑο
θ U
source
Figure 3.1. Sound emission and propagation angle of a source due to refraction.
ao ao
=U+ (3.1)
cos ϑo cos θ
where U is the steady velocity at the downstream end of the control surface, θ is the
sound emission angle with respect to the jet axis, and ϑo is the propagation angle in
the stagnant, ambient air, as shown in Figure 3.1
18
On reaching the shear layer, the sound is refracted and hence both amplitude
and angle change. The refraction can be analyzed as a collection of point acoustic
sources acting at a radial location, and the parallel sheared mean flow is determined
at each location. Equation (3.1) can be rearranged to show that the critical angle is
defined by
1
cos θcr = (3.2)
1+M
Below the critical angle, θcr , no sound can reach the observer and this gives the zone
of silence.
This result is purely kinematic and a thick cylindrical shear layer is considered
here to capture more detailed information of the refraction. The amplitude correction
can be found from the following equation,
¯ ¯
¯ Pc ¯
¯ ¯ = (1 − M cos ϑ)2 (3.3)
¯ Pm ¯
where Pc is the corrected pressure, and Pm is the measured (i.e. uncorrected) pres-
sure. Note that the Mach number of the mean sheared flow, M varies along the
control surface in the radial direction. The correction should be valid provided the
radial distance is large compared to the shear layer’s wavelength and the source
dimensions.
An additional correction is needed, which converts the present source position to
the retarded source position; it can be done by a factor multiplication,
rret (1 − M cos ϑ)
=p (3.4)
rc (1 + M 2 ζ 2 )
p
where ζ = (1 − M cos ϑ)2 − cos2 ϑ
A final correction is necessary to accurately account for the mean flow refraction.
Imposing the local zone of silence condition described earlier allows a surface source
at a relatively large radial location to radiate sound into and through the shear flow.
This is because the local zone of silence decreases in size with the radial location
19
of the source. The simple correction is to set the source strength to zero if the
observation point is located closer to the jet axis than the source point on the FW-H
surface. It should be noted that it is questionable that the energy argument used to
derive the result for the large thickness case is still feasible for zero thickness case.
More details can be found in Ref. 23. The GA corrections can be applied to surface
integral methods [44,46] for the downstream portion of the control surface using the
local velocity distribution. The velocity is assumed to be parallel to the axis.
Another issue is the azimuthal variations between observer point and source,
which are not accounted using this simplified approach. This azimuthal variation
parameters will be introduced by using the alternative correction method described
in the following section.
Lilley’s equation [7] is derived from the inhomogeneous wave equation. The
equation describes explicitly the combined effects of sound convection and refraction
by a unidirectional transversely sheared mean flow. Lilley’s equation can be written
as
LΠ = Γ (3.5)
where Π denotes the acoustic pressure fluctuation normalized by ρ̄ā2 , ρ̄ is the mean
flow density, ā is the mean speed of sound, Γ represents the acoustic source distri-
bution. L is the wave operator originally derived by Pridmore-Brown (1958) [78].
µ ¶
D D2 ∂
L= 2
− ∇ā ∇ + 2ā2 ∇ū∇
2
(3.6)
Dt Dt ∂x
where
D ∂ ∂
= + ū
Dt ∂t ∂x
20
is the substantial derivative relative to mean flow. ū is the mean flow velocity.
In the past, many researchers [10,47,71] have tried to obtain the Greens’ function
expression of Lilley’s equation using high-frequency, far-field asymptotic approxima-
tion. This approximation is valid only when the acoustics wavelength of the aerody-
namic noise is much shorter than the characteristic length scale of the mean flow, and
also the mean velocity profile is horizontal. At the end of the jet potential core, for
example, the mean velocity profiles are significantly distorted, and the parallel mean
flow assumption would fail. Based on this assumption, three different closed-forms
of Green’s function solution with locally parallel and axisymmetric mean flow were
obtained. The first expression was developed by Goldstein [47] in which the source
was placed on several wavelengths off the jet axis. At the same time, Balsa [71] de-
veloped an expression with the source located near the jet centerline axis. The third
form was derived by Goldstein [79] with the implementation of ray-theory for paral-
lel round jets. If a monopole source distribution is considered, the Green’s function
associated to the Fourier transformed solution of Lilley’s wave equation, Gω (x|xs )
satisfies the following equation
D 2
L[Gω (x|xs )e−iωt ] = [ā δ(x − xs )e−iωt ] (3.7)
Dt
where ω is source frequency, xs is source position, δ is the Dirac delta function.
The fundamental solution of the Fourier transformed Lilley’s equation, which
is mathematically called the reduced Green’s function, describes the effect of the
surrounding mean flow on the radiated sound. Mean velocity gradients and tem-
perature gradients are known to refract the acoustic energy within the mean shear
layer. For simplicity, an isothermal jet condition is considered with no temperature
gradients. Hence, the refraction is caused only by non-uniform mean flow within the
shear layer, which varies radially at the outflow of the control surface. One can solve
the reduced Green’s function using a high-frequency asymptotic approximation.
As mentioned in Ref. [72], if the distance between the source and the jet centerline
axis is sufficiently large (several factors of 1/ko , where ko is streamwise wavenumber,
21
ko = ω/ao ), then the critical azimuthal wavenumber, n can be scaled to the order
of ko . This scaling was considered by Goldstein [47] and the resulting acoustic field
was referred to as the asymmetric, high-frequency approximation.
n = O(ko ) as ko → ∞
On the other hand, as the source moves closer to the jet centerline axis, the scaling
becomes
n = O(1) as ko → ∞
Here, the critical wavenumber, n scales like 1/rJ , where rJ is the jet nozzle radius.
This near-axis source problem was studied by Balsa [71] and the resulting acoustic
field was referred to as the quasi-symmetric, high-frequency approximation.
The following equation gives the ratio of the reduced Green’s function to the free-
space Green’s function, which was implemented in the refraction corrections codes
developed in this research.
Gω (x |xs ) Rω (x |xs )
∼ (3.8)
Gω (x |xs ) a(rs )(1 − M (rs ) cos θ)
It should be noted that this equation is valid only when both ko and R → ∞. The
first assumption, i.e. ko → ∞ needs to be well defined and the lowest wavenumber,
ko value should be correctly determined. An analysis was done for this evaluation in
Ref. [48]. The same results are obtained here, which show that the asymptotic high-
frequency approximation for solutions of Lilley’s equation remains accurate with
Strouhal number as small as 1/2.
Based on the asymmetric and quasi-symmetric behaviors of acoustic field, two
distinct Rω (x|xs ) functions from Eq. (3.8) are derived in separate approaches and
are used to describe the acoustic field accordingly. The asymmetric, high-frequency
solutions are used for off-axis source locations, whereas the quasi-symmetric solutions
are applied to near-axis locations.
22
n=∞
" p # 12
X 2 −η (r
n s )
Rω (x|xs ) ∼ Ai[ηn (rs )]ǫ (3.9)
n=−∞
ko rs Qn (rs )
where
2
ǫ = ein∆ϕ+iko (ζn (r)−R sin θ)
s µ ¶2
nk o
Qn (r) = q 2 (r) −
r
s· ¸2
1 − M (r) cos θ
q(r) = − cos2 θ
a(r)
· ¸ 23
3
ηn (r) = − ko ζn (r)
2
Z r
ζn (r) = Qn (r)dr, Q2n (rδ ) = 0
rδ
where a ≡ ā/ao is the local sound speed normalized by its ambient value, M = U/ao
is the Mach number based on the ambient speed of sound, and ∆ϕ is the azimuthal
difference between observer and source point, rs is the source location, Ai is the Airy
function. Note that cube root in the definition of ηn in Eq. (3.9) is taken such that
ηn < 0 for r > rδ and ηn > 0 for r < rδ ,
For the quasi-symmetric, far-field approximation, Rω (x|xs ) is defined by the fol-
lowing equation.
2
Rω (x|xs ) ∼ eiko (ξ(r)−R sin θ−ξ(rs ) cos ∆ϕ)
(3.10)
where Z r
ξ(r) = q(r)dr
0
It should be noted that the rδ defined in Eq. (3.9) does not appear in Eq.
(3.10). rδ is the solution for which the nonlinear equation, Qn is zero, and is called
turning point. rδ is determined by solving Qn numerically. Due to its dependency
on Mach number, which varies along the radial direction, the use of asymmetric
23
0.0018 0.0018
asymmetric asymmetric
quasi-symmetric quasi-symmetric
0.0015 0.0015
0.0012 0.0012
0.0009 0.0009
0.0006 0.0006
0.0003 0.0003
0 0
30 60 90 120 150 30 60 90 120 150
a) θ (degree) b) θ (degree)
0.0018 0.0018
asymmetric asymmetric
0.0015 quasi-symmetric 0.0015 quasi-symmetric
0.0012 0.0012
0.0009 0.0009
0.0006 0.0006
0.0003 0.0003
0 0
30 60 90 120 150 30 60 90 120 150
c) θ (degree) d) θ (degree)
Figure 3.2. |Gω /Gω | at (a) ∆ϕ = 0; (b) ∆ϕ = 30; (c) ∆ϕ = 90; (d) ∆ϕ = 120.
Figures 3.2 (a)-(b) show the comparisons of |Gω /Gω | at various azimuthal-angle
parameters, ∆ϕ for both asymmetric and quasi-symmetric approximations. The
asymmetric and quasi-symmetric approximations are indicated by the solid and
dashed lines, respectively. We used the case suggested in Ref. 15, i.e. a point
source at rs = 0.75 using a mean flow profile given by equation
24
ū(r) M (r)
= = MJ sech2 (2r) (3.11)
ā(r) a(r)
The Strouhal number is 0.5 and can be calculated as
ko rJ
St = (3.12)
π MJ
where rJ = 0.5 is the jet radius, MJ = 0.9 is the Mach number at jet centerline. The
results are identical to Ref. 15.
One can see how |Gω /Gω | varies with different azimuthal angle parameters, ∆ϕ.
This implies that simply ignoring the azimuthal variations between the source and
the observer points will result to inaccurate predictions.
25
We have developed the Kirchhoff and the Ffowcs Williams-Hawkings (FW-H) codes
in both time and frequency domain. For validation, we tested the codes for a
monopole source located at the system origin, for curvilinear and cubical control
surfaces. We first introduce the algebraic method used for the grid generation of
curvilinear Kirchhoff’s control surface.
y = Aex + B (4.1)
where
(Cr −Ct )
A= e−Cx −e(L−Cx )
B = Cr − Ae−Cx
L(2Cr +Ct )
Cx = L − 3(Cr +Ct )
L
8
0 ct Cr
y
-2
-4
-6
-8
-2 -1 0 1 2
x
where ξ and η are structured coordinates, and ymax denotes the y coordinate of the
control surface edge. In this case the values of x and y for a given ξ and η are
easily recovered. To convert the physical domain to the computational domain, the
following numerical transformation methods are used.
ξ = ξ(x, y, z)
η = η(x, y, z)
ζ = ζ(x, y, z)
ξx = J −1 (yη zζ − zη yζ )
ηx = J −1 (zξ yζ − yξ zζ )
ζx = J −1 (yξ zη − zξ yη )
ξy = J −1 (zη xζ − xη zζ )
ηy = J −1 (xξ zζ − zξ xζ )
ζy = J −1 (zξ xη − xξ zη )
ξz = J −1 (xη yζ − yη xζ )
ηz = J −1 (yξ xζ − xξ yζ )
ζz = J −1 (xξ yη − yξ xη )
∂(x,y,z)
J denotes the Jacobian of the inverse transformation ∂(ξ,η,ζ)
, given by
J = xξ (yη zζ − zη yζ ) − yξ (xη zζ − zη xζ ) + zξ (xη yζ − yη xζ )
1 rs
p(xs , t) = sin(ω(t − )) (4.2)
rs ao
The source distribution on the control surface are obtained from the exact solution
of the monopole source defined in Eq.(4.2).
The control surface is divided into N panels. The integrand in the surface integral
formula is evaluated at the center of each panel at that point’s retarded time. This
approximation works well as long as the panel size is sufficiently small. In this case,
sufficiently small means that the source strength variation is approximately linear
over the panel and the retarded time does not vary significantly over the panel.
28
4.2.1 Test 1
Several situations are considered in order to validate the Kirchhoff method code
with the retarded time approach using a cubical control surface. At first, we have
investigated the accuracy of the numerical solutions for various numbers of points
per period and points per wavelength. The observer coordinates are shown in Figure
4.2. The parameters of the different cases are listed in Table (4.1) and the results
are plotted in Figures 4.3 - 4.9.
Case Figure R φ θ N T
A 4.3 1.75r 0o 0o − 180o 13 10-20
B 4.4 1.75r 0o − 180o 90o 13 10-20
C 4.5 1.75r 0o − 180o 0o 13 10-20
D 4.6 1.75r 0o − 180o 90o 9-16 20
E 4.7 17.5r 0o − 180o 90o 13 10-20
F 4.8 17.5r 0o − 180o 90o 11-20 20
G 4.9 17.5r 90o 0o 0-30 20
Figures 4.3 - 4.5 show the root mean square (RMS) errors incurred by the nu-
merical solutions at different observer orientations employing Kirchhoff’s method.
The RMS errors for a collection of signal data, yi with an average value, ybi can be
obtained using
29
θ
0
x
φ
z
Figure 4.2. Observer coordinates system.
v
u N
u1 X
RM S = t (yi − ybi )2 (4.3)
n n=1
These figures show the results at different observer positions in order to ensure that
all parts of the Kirchhoff surfaces are taken into account during the computation.
As shown in these figures, the errors decrease when temporal points increase. The
variations of these results are almost identical due to the symmetric behavior of the
cubical control surface.
Figure 4.6 describes the effects of various numbers of points per wavelength on
the solutions while the number of temporal points is kept constant. As the number of
points per wavelength increases, the RMS errors decrease. Figure 4.7 and 4.8 repeat
the accuracy investigations with observer locations at far-field distance (R = 17.5r).
Figure 4.9 compares the solutions obtained from Kirchhoff method using analytical
and numerical expressions for the time and the normal derivatives. 20 temporal
points are chosen because our previous tests show satisfactory results choosing this
30
value. If we use less than 6 points per wavelength, the error increases substantially,
but above this value, the errors stay at an acceptable level.
RMS error
temp. pts=20
0.005
0.0025
0.0075 0
0 50
(degree)
100 150
φ
RMS error
0.0075
RMS error
0.005
0.005
0.0025
0.0025 0
0 50
(degree)
100 150
0.0025
0
0 50
φ (degree)
100 150
0
0 50 100 150
θ (degree)
RMS error
0.005
0.0025
0.0075 0
0 50
θ (degree)
100 150
RMS error
0.0075
RMS error
0.005
0.005
0.0025
0
0 50
φ (degree)
100 150
0
0 50 100 150
φ (degree)
φ waveleng
# of points per θ wavelen temp. pts=10
# of points per temp. pts=10
0.01
temp. pts=12
temp. pts=14
temp. pts = 10
temp. pts = 12
temp. pts = 14
temp. pts=12
temp. pts=20
0.01 temp. pts = 20
temp. pts=14
0.0075
0.01 temp. pts=20
RMS error
0.0075
RMS error
0.005
0.005
0.0025
0.0025
0.0075
0
0 50
θ (degree)
100 150
0 (degree)
RMS error
0 50 100 150
φ
θ
#npts/wave=9
#npts/wave=11
0.01
0.005 #npts/wave=13
#npts/wave=16
0.0075
RMS error
0.005
0.0025 0.0025
0
0 50
φ (degree)
100 150
0
0 50 100 150
φ (degree)
RMS error
0.0075
RMS error
0.005
0.005
0.0025
0.0025
0.0075
0
0 50
θ (degree)
100 150
0
0 50
(degree)
100 150
RMS error
0.0075
RMS error
0.005
0.0025
0.0025
0
0 50
(degree)
100 150
φ
0
0 50 100 150
φ (degree)
0.006
RMS error
0.005
0.004
0.005 0.003
0.002
0.001
RMS error
(degree)
0
0 50 100 150
φ
R =φ35,
θ
0.0375 #npts/wave=31
o o
0.0025
0.035
0.0325
= 90 , =0 , #npts/wave=39
#npts/wave=49
#npts/wave=59
0.03 0.015 #npts/wave=79
0.0275 Kirchhoff Analytic.
RMS error
0.0225
0.02 0.01
0.0175
0.015
0.0125
0.01 0.005
0.0075
0.005
0.0025
0
0
10 15 20
#npts/wave
25 30
0 20 40
φ (degree)
60 80
0
0 50 100 150
φ (degree)
RMS error
0.0070.0025
0.006
RMS error
0
0 50
φ (degree)
100 150
0.005
# of temporal points = 20, R
θ #npts/wave=21
#npts/wave=26
0.004 0.04 0.02 #npts/wave=29
R =φ35,
θ
0.0375 #npts/wave=31
o o
0.035
0.0325
= 90 , =0 , #npts/wave=39
#npts/wave=49
#npts/wave=59
0.03
0.003 0.0275 Kirchhoff Analytic.
0.015 #npts/wave=79
RMS error
0.025 Kirchhoff Numeric.
RMS error
0.0225
0.02 0.01
0.002 0.0175
0.015
0.0125
0.01 0.005
0.001 0.0075
0.005
0.0025
0
0
0
10 15 20
#npts/wave
25 30
0 20 40
φ (degree)
60 80
0 50 100 150
φ (degree)
0.04
# of temporal points = 20
0.0375 θ wavele
# of points per θ
R = 17.5r, φ = 90 o, θ =0 o, # of temporal points = 20
#npts/wave=11
temp. pts=10 0.009 #npts/wave=14
temp. pts=14 #npts/wave=17
0.035 temp. pts=20
0.008
#npts/wave=20
0.007
0.0325 0.005
0.006
RMS error
RMS error
0.03 0.005
0.004
0.0025
0.0275 0.003
Kirchhoff Analytic.
0.002
0.001
0
0.0225
0 50
φ (degree)
100 150
0
0 50
φ (degree)
100 150
0.02
# of temporal points = 20, R
0.0175 θ #npts/wave=21
#npts/wave=26
0.02 #npts/wave=29
#npts/wave=31
0.015 #npts/wave=39
#npts/wave=49
#npts/wave=59
0.0125 0.015 #npts/wave=79
RMS error
0.01 0.01
0.0075
0.005 0.005
0.0025 0
0
0 20 40
φ (degree)
60 80
10 15 20 25 30
#npts/wave
4.2.2 Test 2
After comparing the accuracy of the numerical solutions of Kirchhoff’s method for
a cubical control surface, a curvilinear control surface was employed. It is worthwhile
to investigate the effect of a curvilinear control surface because the grid is sometimes
non-Cartesian in the jet aeroacoustic problems. Figure 4.10 shows a generalized
curvilinear Kirchhoff’s surface with length, L = 4 and tip height, H = 17. Several
cases were analyzed and listed in Table (4.2). These results are plotted in Figure
4.11 to 4.14.
Case Figure R φ θ N T
A 4.11 17.5r 0o − 90o 90o 21-79 20
B 4.12 17.5r 0o − 90o 90o 59 10-36
C 4.13 17.5r 45o /75o 90o 0-80 32
D 4.14 0-35 0o 0o 69 32
Figure 4.11 shows the RMS errors incurred by the numerical solutions at different
far-field observer locations (R = 17.5r) using the curvilinear control surface. As
expected, the errors decrease as the number of points per wavelength increases.
Figure 4.12 shows the errors of the numerical solutions at different far-field observer
locations for various temporal points. It is shown that smaller errors can be achieved
with a higher number of temporal points. Therefore, we use next 32 temporal points
and check for the accuracy of numerical solutions using different numbers of points
per wavelength. Figure 4.13 compares both the analytical and numerical derivatives
to obtain the Kirchhoff solutions. We set the observer point at the far-field locations
(R = 17.5r) and varied the number of points per wavelength. Figure 4.14 shows
the RMS errors at various observer distances from the system origin. As observer
35
lies inside the Kirchhoff’s surface, the errors grow rapidly, which is expected and the
signal is close to zero.
X
Y
#npts/wave=21
# of temporal points = 20
θ wavele
# of points per θ #npts/wave=11
#npts/wave=26
0.02 temp. pts=10
temp. pts=14
0.009 #npts/wave=14
#npts/wave=17
#npts/wave=29
temp. pts=20 #npts/wave=20
0.008
#npts/wave=39
0.007
0.005
0.006
#npts/wave=49
RMS error
RMS error
0.005 #npts/wave=59
0.0025
0.004 #npts/wave=79
0.015 0.003
0.002
0.001
RMS error
0
0 50
φ (degree)
100 150
0
0 50
φ (degree)
100 150
0.01
R =φ 17.5r,
θ
0.04
o
0.0375
0.035
= 90
0.0325
0.03
0.0275 Kirchhoff Analytic.
0.025 Kirchhoff Numeric.
RMS error
0.0225
0.02
0.0175
0.015
0.0125
0.005 0.0075
0.01
0.005
0.0025
0
10 15 20 25 30
#npts/wave
0
0 20 40 60 80
φ (degree)
θ = 90 o, # of temporal po
0.07
# of temporal points = 32
# of points per wavelength
0.0075 θ φ= = 0o
0.16
0.14
0.12
RMS error
0.1
0.005 0.08
0.06
0.04
0.02
0.0025 0
10
R
20 30
0
0 20 40 60 80
φ (degree)
# pts/wavelength = 59, R =
θ
θ = 90 o, # of temporal points = 32
temp. pts=10(59)
temp. pts=14(59)
temp. pts=18(59)
0.015 temp. pts=24(59)
temp. pts=32(59)
0.0125
0.07 temp. pts=36(59)
0.01 0.065
RMS error
0.0075 0.06 Analy. Kirchhoff (R=17.5r,phi=75 )
o
o
0.005
0.055 Numeric. Kirchhoff (R=17.5r,phi=75 )
0.0025 Analy. Kirchhoff (R=17.5r, phi=45 o)
0.05 Numeric. Kirchhoff (R=17.5r,phi=45 )
o
0
0
0.045
20 40
φ (degree)
60 80
RMS error
0.025
RMS error
0.1
0.08
0.06 0.02
0.04
0.015
0.02
0
0.01 10
R
20 30
0.005
0
10 20 30 40 50 60 70
#npts/wave
o
temp. pts=10(59)
θ = 90 , # of temporal p
o
θ
temp. pts=18(59)
temp. pts=24(59)
temp. pts=32(59)
0.015 temp. pts=36(59)
temp. pts=10(79) 0.07
temp. pts=14(79)
temp. pts=18(79) 0.065
0.0125 temp. pts=24(79)
temp. pts=32(79) 0.06 Analy. Kirchhoff (R=35,phi=75 o)
0.16
RMS error
θ= φ =0
Numeric. Kirchhoff (R=10,phi=45 o)
RMS error
0.0075
0.04
0.035
0.005
0.03
0.14 0.025
0.0025
0.02
0.015
0
0
0.12
20 40
φ (degree)
60 80 0.01
0.005
0
10 20 30 40 50 60 70
#npts/wave
RMS error
0.1
0.08
0.06
0.04
0.02
0
10 20 30
R
∂u
ρ = ∇p (4.4)
∂t
A curvilinear control surface is modeled as shown in Figure 4.10 with length, L = 4,
and tip height, H = 17. With this control surface, we investigate the accuracy of
the FW-H numerical solutions for various numbers of points per period and points
per wavelength. A collection of different cases is listed in Table (4.3). The results
are plotted in Figures 4.15 - 4.18.
Case Figure R φ θ N T
A 4.15 17.5r 0o − 90o 90o 14-57 20
B 4.16 17.5r 0o − 90o 90o 14-29 32
C 4.17 17.5r 0o − 90o 90o 21 10-32
D 4.18 5r/17.5r 20o /45o 40o /90o 0-40 32
Figure 4.15 shows the root mean square (RMS) errors incurred by the numeri-
cal solutions at different far-field observer locations. As the number of points per
wavelength increases, the errors decrease and become more uniform at all orienta-
tions. Figure 4.16 shows the RMS errors at different orientations employing a higher
number of temporal points, (i.e. 32) than the previous case. As shown here, the
39
overall errors are smaller than that shown in Figure 4.15. Similarly, as the number
of points per wavelength increases, the errors become more uniform at all orienta-
tions. Figure 4.17 compares RMS errors of numerical solutions at different far-field
observer locations for various temporal points. It is shown that smaller errors can be
achieved with more temporal points, which is expected. Figure 4.18 shows the errors
obtained from 32 temporal points and various numbers of points per wavelength for
2 observer locations. Both observer locations show the same trend, i.e. the RMS
errors decrease as the number of points per wavelength increases.
The comparison between Kirchhoff and FW-H methods is studied at different
observer locations in Figure 4.19. The error incurred by the Kirchhoff method is
significantly higher compared to the FW-H method. This error may be due to
the insufficient points employed to compute the normal derivatives in Kirchhoff’s
method. Figure 4.20 compares the Kirchhoff’s and FW-H methods using various
numbers of points per wavelength. Both methods demonstrate a decrease in the
RMS error as numbers of points per wavelength increase. However, FW-H method
shows a better accuracy of the solutions. An increase in RMS error for Kirchhoff
method with numerical normal derivatives is due to the insufficient points employed
to compute the normal derivatives. This can be verified with using analytical normal
derivatives as shown in the figure.
40
0.015
0.0035 0.0125
RMS error
0.01
0.003 0.0075
0.005
0.0025
0.0025
RMS error
0
0 20 40
φ (degree)
60 80
#npts/wave=19
o
R=35, phi=45 , theta=90
0.04
R=10, phi=20 o, theta=40
#npts/wave=21
RMS error
0.0015
RMS error
0.001 #npts/wave=26
0.03
0.001 #npts/wave=31
0.02
#npts/wave=57
0.0005
0.0005 0.01
0
0 0 0 10 20
#npts/wave
30
0
0 20 40
φ 20 60
(degree)
80
40 60 80
φ (degree)
# of temporal points
# of temporal = 32,= R = 17.5r, θ = 90
points θ
o
θ
temp. pts=10
temp. pts=12
0.02
0.0025 0.004 #npts/wave=14
temp. pts=14
temp. pts=18
0.0175 temp. pts=24
0.0035 #npts/wave=16
temp. pts=32
0.003
0.015
#npts/wave=19
0.0025
0.0125
#npts/wave=21
RMS error
RMS error
0.002
0.002 0.01
#npts/wave=29
0.0015 #npts/wave=14 0.0075
#npts/wave=19
#npts/wave=21
0.001 #npts/wave=26 0.005
#npts/wave=31
#npts/wave=57
0.0005
0.0025
0
0 20 40
φ (degree)
60 80
0
0 20 40
(degree)
60 80
RMS error
0.0015 φ
0.06
# of temporal points
0.001 0.05
o
R=35, phi=45 , theta=90
0.04
R=10, phi=20 o, theta=40
RMS error
0.03
0.0005
0.02
0.01
0
0 0 10
#npts/wave
20 30
0 20 40 60 80
φ (degree)
0.002
0.0015 #npts/wave=14
0.015 0.001
#npts/wave=19
#npts/wave=21
#npts/wave=26
#npts/wave=31
#npts/wave=57
0.0005
0
0.0125 0 20 40
φ (degree)
60 80
RMS error
0.01 θ
0.0025 #npts/wave=14
#npts/wave=16
#npts/wave=19 0.06
# of temporal points
#npts/wave=21
0.002 #npts/wave=29
0.0075 0.05
o
R=35, phi=45 , theta=90
RMS error
0.0015
0.04
R=10, phi=20 o, theta=40
RMS error
0.005 0.001
0.03
0.0005
0.02
0.0025
0 0.01
0 20 40
φ (degree)
60 80
0
0 0 10 20 30
0 20 40 60 #npts/wave 80
φ (degree)
θ # of temporal points = 32 θ
# of temporal points = temp. pts=10
temp. pts=12
0.06 0.004
0.02
temp. pts=14
temp. pts=18
0.0175 temp. pts=24
0.0035 temp. pts=32
0.003 0.015
0.0025 0.0125
RMS error
RMS error
0.05 0.002 0.01
0.0015 #npts/wave=14
#npts/wave=19 0.0075 o o
0.001
#npts/wave=21
#npts/wave=26
0.005
R=5r, phi=20 , theta=40
#npts/wave=31
o o
0.04
0.0005
#npts/wave=57
0.0025
R=17.5r, phi=45 , theta=90
0
0 20 40
φ (degree)
60 80
0
0 20 40
(degree)
60 80
RMS error
0.03 θ
0.0025 #npts/wave=14
#npts/wave=16
#npts/wave=19
#npts/wave=21
0.002 #npts/wave=29
0.02
RMS error
0.0015
0.001
0.01 0.0005
0
0 20 40
φ (degree)
60 80
0
0 10 20 30
#npts/wave
0.018
Kirchhoff
0.016 FW-H
0.014
RMS error
0.012
0.01
0.008
0.006
0.004
0.002
20 40 60 80
φ (degree)
FW-H
Kirchhoff (Numerical normal derivatives)
Kirchhoff (Analytical normal derivatives)
0.03
RMS error
0.02
0.01
0
0 10 20 30 40
#npts/wave
The forward time calculation is studied using a point source located at the origin
of the cylinder as shown in Figure 4.21. The cylindrical control surface used has
length, Lk = 20 and rk = 5. The observer is located at distance, R = 12rk from
the center of the cylinder. Figure 4.22 compares the FW-H numerical solutions with
the analytical solutions, which display a sinusoidal wave pattern. Further studies
are desired to check the feasibility of FW-H method using forward time approach.
Therefore, we have investigated three cases that listed in Table (4.4). The results
are plotted in Figures 4.23 - 4.25.
Case Figure R φ θ N T
A 4.23 12rk 0o − 90o 90o 10-30 52
B 4.24 12rk 0o − 90o 90o 6-20 32
C 4.25 12rk 0o − 90o 90o 20 32-72
Figure 4.23 compares the root mean square (RMS) errors at various observer
orientations evaluated with different number of points per wavelength, i.e. 10, 20
and 30. As expected, the errors decrease as the number of points per wavelength
increases. However, unlike the retarded time calculation, the decrease shown here is
almost trivial and the errors are significantly higher. This may be due to the binning
errors, i.e. the signals go to the wrong bin. Figure 4.24 shows similar results with 32
points per period. As shown here, the RMS errors drop significantly as the number
of points per wavelength varies from 6 to 10 but remains about the same beyond
that. Figure 4.25 shows the RMS errors variations at different observer orientations
evaluated at different temporal points, i.e. 32, 42, 52, 62, and 72. Consistently, the
RMS errors decrease as the time step decreases.
45
n^ Observer
S R (X,t)
rk
Point Source
x
Lk
Figure 4.21. Point source location and control surface geometry.
From the current studies of the FW-H solutions using forward time approach, we
conclude that the accuracy of the numerical prediction is not significantly affected by
the number of points per wavelength. However by altering the number of temporal
points, a significant improvement is seen. Therefore, a sufficient number of temporal
points must be provided in order to achieve an accurate noise prediction. When ap-
plied to a practical problem, e.g. jet noise prediction, the forward time formulations
allows one to accumulate progressively the time trace of the radiated acoustic signal
by using aerodynamic data. Hence, the traditional concept of a postprocess acoustic
prediction is partially bypassed. This new methodology has reduced the memory
burden dramatically. The practical advantages are also including:
(1) the inherently parallel behavior
(3) no retarded time equation must be solved
46
0.75
0.5
0.25
P’ 0 FW-H solutions
Analytic solutions
-0.25
-0.5
-0.75
-1
1 2 3 4 5 6
time
0.045
0.025
0.02
0.015
0.01
0.005
0
0 20 40 60 80
φ (degree)
0.07
#npts/wave = 6 (fw)
0.065 #npts/wave = 10 (fw)
#npts/wave = 20 (fw)
0.06
#npts/wave = 6 (ret)
0.055 #npts/wave = 10 (ret)
#npts/wave = 20 (ret)
0.05
0.045
0.035
RMS error
0.03
0.025
0.02
0.015
0.01
0 20 40 60 80
φ (degree)
A simple monopole is used to study the Kirchhoff’s and porous Ffowcs Williams-
Hawkings (FW-H) method in frequency domain. A cylindrical control surface (Lk =
20, rk = 5) as shown in Figure 4.21, was used in this validation study. The point
source is placed at the middle of the surface and the surface pressure is defined as
1 −iω ars
pb(xs , ω) = e o (4.5)
rs
Figure 4.26 shows the Kirchhoff and the exact solutions of the monopole source
at different far-field observer locations. A normalized amplitude of 1 is expected
for the monopole source case. Figure 4.27 demonstrates the errors incurred by the
Kirchhoff’s method using different numbers of points per wavelength. A decrease in
the errors in shown as number of points per wavelength increases.
1.1
0.9
Exact
0.8 Kirchhoff
Normalized amplitude
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 50 100 150
θ (degree)
Figure 4.26. Comparisons of the Kirchhoff’s method and the exact solutions.
49
R = 20r, φ = 0 o, k = 1
0.005
#npts/wave=6
#npts/wave=10
0.004 #npts/wave=15
#npts/wave=20
#npts/wave=25
0.003
Error
0.002
0.001
0 50 100 150
θ (degree)
Figure 4.28 shows the Ffowcs-Williams Hawkings (FW-H) and the exact solu-
tions of the monopole source at different far-field observer locations. A normalized
amplitude of 1 is expected for the monopole source case. Figure 4.29 demonstrates
the errors incurred by the FW-H method using different numbers of points per wave-
length. A decrease in the errors is shown as number of points per wavelength in-
creases. Figure 4.30 compares the Kirchhoff’s and the FW-H methods at different
observer locations. In general, the FW-H solutions are more accurate than the
Kirchhoff’s method under the same conditions.
50
1.1
0.9
Exact
0.8
Normalized amplitude
FW-H
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 50 100 150
θ (degree)
Figure 4.28. Comparisons of the FW-H method and the exact solutions.
R = 20r, φ = 0 o, k = 1
0.005
#npts/wave=6
#npts/wave=10
0.004 #npts/wave=15
#npts/wave=20
#npts/wave=25
Error
0.003
0.002
0.001
0 50 100 150
θ (degree)
0.0009
0.0008
Kirchhoff
FW-H
0.0007
0.0006
Error
0.0005
0.0004
0.0003
0.0002
0.0001
0
0 50 100 150
θ (degree)
1 −iω ars
pb(xs , ω) = e o (4.6)
rs
The acoustics wavenumbers, ko of 1.0 and 1.5 are considered. We are using 20
points per wavelength. The distance between the midpoint of the cylinder’s axis
and the observer is taken to be 60rJ , where rJ is 1. The result with no refraction
correction is shown in Figure 4.31. A normalized amplitude of 1 is expected with no
mean flow outside the FW-H surface.
The refraction corrections are then applied to the right base of the cylinder using
the simple GA method and Lilley’s equation. The shear flow is modeled with velocity
profile shown in Figure 4.32 using Eq.(4.7).
2
M (r) = MJ eAr (4.7)
where MJ = 0.5, M (r) is the Mach number at a distance r, A is a constant, i.e. -0.21.
The mean flow velocity profile is applied only to the right base of the cylindrical
control surface. At surface radius of 5, the corresponding Mach number is 0.002,
while the centerline Mach number is 0.5. This is a reasonable distribution for jet
velocities [46]. Since velocity becomes negligible near the radial distance of 5, the
refraction correction is applied only for r < 5.
When solving asymptotic solutions for the Lilley’s equation, which is based on
the high-frequency assumption, a minimum Strouhal number must be considered.
The Strouhal number can be determined from Eq. (3.12). As demonstrated in
53
1.1
0.9
normalized amplitude
0.8
0.3
0.2
0.1
0
0 10 20 30 40 50 60 70
θ (deg)
the last section, at Strouhal number of 1/2, the solutions are still very close to the
exact solutions of Lilley’s equation. Below this Strouhal number limit, asymptotic
approximation does not apply and the refraction correction must be done by an
alternative way, which we choose to be the simple GA method. Since the asymmetric
solution is more accurate, it is used for all source elements except for a near-axis
source, where the asymmetric solution breaks down and a quasi-symmetric solution
is used. Brent’s method is used to compute the turning point, rδ in Eq.(3.9).
As shown in Figure 4.31, the zone of silence is predicted in a region where there
is a substantial reduction in the normalized pressure amplitude. For the monopole
case, Lilley’s equation results to a smooth amplitude decrease in the zone of silence,
whereas the simple GA method shows an abrupt decrease at this region. A similar
trend was obtained from the Kirchhoff method, and hence only the FW-H result is
shown here.
54
0
0 0.1 0.2 0.3 0.4 0.5
Mach number
The azimuthal variation between the source and observer points has been ignored
in the simple GA method, whereas the variation parameters can be evaluated with
Lilley’s equation, which has been discussed in the previous section. The azimuthal
variations between the sources and observer points are important because the FW-H
control surface utilized is three-dimensional.
It is worthwhile to note that the limitation of the high-frequency applications of
Lilley’s equation to jet noise can be overcome by adding the axial length of the con-
trol surface to capture more of the low frequency noise. Also, the refraction effects
inside the control surface are captured better with a longer surface. However, higher
computational cost will result from the increase of the surface length.
55
In this chapter, we will look at some far field aeroacoustics results obtained by
coupling the near field LES data with a Ffowcs Williams-Hawkings code which has
the capability to work on any general control surface geometry.
The surface integral methods can extend the near-field CFD flow data to far-
field acoustic properties, e.g. acoustic pressure. The near-field input for the surface
integral methods will be obtained from a Large Eddy Simulation (LES) calculation
[35, 36].
In this study, the LES calculation is performed for a turbulent isothermal round
jet at a Mach number of 0.9 and Reynolds number of 400,000. A sixth order compact
difference is used for all interior points. Third-order one-sided compact scheme
equations are applied on the left and right boundaries. For the interior points next
to the boundaries, fourth-order central compact scheme formulations are used. The
standard fourth-order explicit Runge-Kutta scheme is used for time advancement.
The LES calculation employs an implicit sixth-order tri-diagonal spatial filter [80] as
the implicit subgrid-scale (SGS) model. The jet centerline temperature is chosen to
be the same as the ambient temperature, To and set to 286K. A fully curvilinear grid
consisting of approximately 16 million grid points is used in the simulation. The grid
has 390 points along the streamwise x direction and 200 points along both the y and
z directions. The physical portion of the domain in this simulation is extended to
35 jet radii in the streamwise direction and 30 jet radii in the transverse directions.
56
The coarsest resolution in this simulation is estimated to be about 400 times the
local Kolmogorov length scale.
The boundaries of the computational domain are treated using Tam and Dong’s
3-D radiation and outflow boundary conditions [81], as illustrated in Figure 5.1.
A sponge zone [82] near the downstream surface is used in which grid stretching
and artificial damping are applied to suppress the unwanted reflections from the
outflow boundary. Since the actual nozzle geometry is not included in the present
calculations, randomized perturbations in the form of a vortex ring are added to the
jet velocity profile at a short distance downstream of the inflow boundary in order
to excite the 3-D instabilities in the jet and cause the potential core of the jet to
break up at a reasonable distance downstream of the inflow boundary. The inflow
forcing used here was proposed by Bogey and Bailly [83]. There are 16 azimuthal
modes in total and the first 4 modes are not included in the vortex ring forcing. A
more detailed study of the inflow vortex ring forcing can be found in Ref. 34. The
mean shear layer imposed on the inflow boundary has a streamwise velocity profile
· · µ ¶¸¸
UJ r
ū(r) = 1 + tanh 10 1 − (5.1)
2 rJ
where UJ is the jet centerline velocity at nozzle exit, rJ is the jet nozzle radius.
There are about 14 grid points in the initial jet shear layer.
of 25,000 time steps during LES run. The initial transients exit the domain over the
first 10,000 time steps. The total acoustic sampling period corresponds to a time
scale in which an ambient sound wave travels about 14 times the domain length in
the streamwise direction. Assuming at least 6 points per wavelength are needed to
accurately resolve an acoustic wave, it is found that the maximum frequency resolved
with the grid spacing around the control surface corresponds to a Strouhal number
of approximately 3.0. This is the grid cutoff frequency used in computation domain.
Based on the data sampling rate, the number of temporal points per period in this
maximum frequency is 8.
Figure 5.2. Schematic of the control surface surrounding the jet flow.
br at each frequency. These terms are calculated using the retarded time algorithm,
L
e.g. Eq. (2.3), which can be solved by performing a simple linear interpolation.
It should be noted that the refraction corrections evaluated on the downstream
control surface are employing the frequency domain FW-H equation. These results
are added to the time domain FW-H results for an open surface [36]. Thus, an
inverse FFT is needed to recover the signal for the downstream refraction surface in
the time domain.
Numerical results of far-field overall sound pressure level (OASPL) are calculated
using Eq. (5.2) along an arc of radius 60rJ from the jet nozzle.
µ ¶
E
OASPL = 20log10 (5.2)
Pref
where Pref is the non-dimensional reference pressure calculated from the ambient
temperature, To = 286K, ambient pressure, po = 1.01KPa, M = 0.9 at jet centerline
nozzle exit, and dimensional reference pressure, 2x10−5 Pa. E is the total energy
59
Figure 5.3. Schematic showing the center of the arc and how the
angle θ is measured from the jet axis.
R
contained in the spectrum, defined as < Pav (f )df >, where Pav is calculated from
the average acoustic pressure signals over the equally spaced 36 azimuthal points on
a full circle at a given θ location on far-field arc of radius 60rJ from the jet nozzle.
The center of the arc is chosen as the jet nozzle exit and the angle θ is measured
from the jet axis as illustrated in Figure 5.3.
Table 5.1 Some jet noise experiments with Mach numbers similar to
that of the LES by Uzun et al. [35, 36]
M ReD Reference
0.9 540,000 Mollo-Christensen et al. [84]
0.88 500,000 Lush [73]
0.9 3,600 Stromberg et al. [85]
0.9 400,000 SAE ARP 876C [86]
0.9 400,000 Uzun 3-D LES [35]
60
Simple GA refraction
Lilley’s refraction
120 Open control surface
Closed control surface
118 SAE ARP 876C prediction
Mollo-Christensen et al. data (cold jet)
116 Lush data (cold jet)
Stromberg et al. data (cold jet)
114
OASPL (dB)
112
110
108
106
104
102
100
0 10 20 30 40 50 60 70 80 90 100 110 120
θ (deg)
Figure 5.4. Overall sound pressure levels at 60rJ from the nozzle exit.
One should remove the low frequencies resulting from the outflow boundary which
may not be damped out effectively by the sponge zone. In Uzun et al. [35, 36]
jet case, this frequency with the corresponding Strouhal number of 0.05 was found
by the authors. Hence, the OASPL spectra integration only includes the resolved
frequencies range of 0.05 ≤ St ≤ 3.0.
OASPL result using open control surface with no refraction correction at observer
location along a far-field arc of x = 60rJ . Table (5.1) summarizes the Mach numbers
and Reynolds numbers of the experiments and the ARP database [86] that we are
comparing to. It should be noted that the experimental jets were cold jets, whereas
our jet is an isothermal jet. However, as indicated in the ARP database, the effect
of the temperature is very small, i.e. about 0.3 dB.
From the OASPL plot for an open surface, it is shown that the prediction from
ARP database for a cold jet at ReD of 400,000 is slightly lower (about 1dB) than
the LES calculations for high angles, i.e. θ ≥ 40o . Also the OASPL are comparable
to the experimental values at these angles. However, at shallow angles, i.e. θ ≤ 40o ,
the OASPL from the LES drops significantly due to the omission of the signal from
the downstream surface. For the closed control surface, the OASPL is generally
higher than that of the open control surface. The results increase significantly for
lower angles. However, there is a significant overprediction at angles θ ≤ 20o , as
refraction effects outside the control surface are not included. For higher angles, i.e.
θ ≥ 70o , there is also an overprediction maybe due to spurious line of dipoles on the
downstream surface area, as the quadrupoles move downstream and exit the control
surface. [80]
2
M (r) = MJ eAr + B (5.3)
where MJ = 0.46, A, B are constants, i.e. -0.14 and 0.0044, respectively. Equation
(5.3) is represented with dash-line in Figure 5.5.
62
LES data
5 Analytical data
r 3
0
0 0.1 0.2 0.3 0.4 0.5
M
This profile is used for both the simple GA method and Lilley’s equation calcu-
lations. This sheared mean flow has an influence on the signals emitted from the
surface elements on the downstream outflow FW-H surface. Refraction corrections
are performed on the surface elements inside the sheared flow, i.e. on the downstream
surface. The variation of the velocities across the radial direction of the shear layer
is taken into account at each surface element location. The refraction corrections
schemes result in the damping of far-field signal contributed by the surface element.
Within the high-frequency region of 0.50 ≤ St ≤ 3.0, this refraction corrections are
performed at each frequency and on each surface element.
As shown in Figure 5.4, the OASPL with refraction corrections starts to decrease
as θ goes below 20o . This gives a zone of silence as expected from the experimental
results. The slight decrease of the results with no refraction corrections is due to
the existence of refraction inside the FW-H control surface. With the refraction
corrections outside the control surface using the GA method, the results show a
63
decrease on the OASPL making it closer to the experiments. The results from the
Lilley’s equation method exhibit a smoother decrease on the OASPL inside the zone
of silence, which is consistent with the results for a monopole source shown in Figure
4.31.
With the application of Lilley’s equation, the computational walltime required
increases to about 180 hours for each arc location inside the zone of silence using eight
375 MHz processors run in parallel. The maximum time ratio for Lilley’s equation
method to the GA method is 60. This significant increase in computational time is
expected when solving Lilley’s equation, because the mean flow velocity dependent
turning point, rδ in Eq. (3.9) needs to be evaluated numerically.
64
In this thesis, the application of surface integral methods in computational jet aeroa-
coustics is both demonstrated and analyzed numerically. Although in the past, the
surface integral methods have proven to be an efficient and accurate tool for the
prediction of jet noise, a new improvement to account for the mean flow refraction
effects is developed in this study. Geometric acoustics (GA) is revisited and a new
approach, Lilley’s equation is also investigated due to its more realistic assumption
for acoustic propagation. Moreover, the assessment of the forward time approach is
also described and analyzed numerically. The summary of our findings and results
is described below.
determine the mid-field and far-field sound. If the source region is non-compact, as in
most important problems, this volume integration leads to prohibitive computation
times and memory requirements.
With this motivation, we have chosen the surface integral methods to perform
jet noise prediction. We first evaluated the methods with a simple monopole source
enclosed by different control surfaces, i.e. cubical and curvilinear shape. For each
of the control surfaces, the solutions computed from the surface integral methods
are analyzed qualitatively at different situations. The results are improved as the
number of points per wavelength and the number of temporal points increase.
Surface integral methods fail to predict the zone of silence accurately due to the
lack of the refraction effect. In order to account for the refraction effect, two methods
are employed: geometric acoustics (GA) and Lilley’s equations. Lilley’s equation
is approached using high-frequency, far-field asymptotic solutions. The mean flow
refraction effect has been successfully demonstrated using a simple monopole source
enclosed by a cylindrical control surface. A smooth decrease of OASPL in side the
zone of silence can be seen for Lilley’s method while an abrupt decrease occurred for
GA method.
The methods were applied to the noise calculation of an isothermal jet of Reynolds
number 400,000. The solution on the FW-H control surface was provided using LES.
The above refraction corrections were applied on the downstream control surface
to account for the mean shear flow outside the control surface. Both refraction
corrections improve the OASPL results dramatically at shallow angles near the axis.
The comparisons between the two methods were also made and it was shown
that the simple GA method is much simpler and requires less computational time
compared to Lilley’s equation. On the other hand, the high-frequency asymptotic
solutions for Lilley’s equation has the potential of being more realistic as it allows
azimuthal variations between source and observer. The use of the simple GA method
and Lilley’s equation have demonstrated a successful prediction of the zone of silence,
which cannot be done by the traditional surface integral methods.
LIST OF REFERENCES
67
LIST OF REFERENCES
[16] C. K. W. Tam and L. Auriault. Mean Flow Refraction Effects on Sound Radi-
ated from Localized Sources in a Jet. Journal of Fluid Mechanics, 370, 1998.
[17] A. S. Lyrintzis. Surface Integral Methods in Computational Aeroacoustics-From
the (CFD) Near-Field to the (Acoustic) Far-Field. Aeroacoustics, 2(2):95–128,
2003.
[18] W. Y. Soh. Unsteady Jet Flow Computation-Towards Noise Prediction. AIAA
Paper No. 1994-0138, January 1994.
[19] B. E. Mitchell, S. K. Lele, and P. Moin. Direct Computation of the Sound Gen-
erated by Vortex Pairing in an Axisymmetric Jet. Journal of Fluid Mechanics,
383:113–142, 1999.
[20] W. Zhao, S. H. Frankel, and L. Moungeau. Large Eddy Simulations of Sound
Radiation from Subsonic Turbulent Jets. AIAA Journal, 39(8):1469–1477, Au-
gust 2001.
[21] M. Billson, L. Eriksson, and L. Davidson. Jet Noise Prediction Using Stochastic
Turbulence Modelling. AIAA Paper No. 2003-3282, May 2003.
[22] A. S. Lyrintzis and R. R. Mankbadi. On the Prediction of the Far-Field Jet
Noise Using the Kirchhoff Method. AIAA Journal, 34:413–416, 1996.
[23] T. S. Chyczewski and L. N. Long. Numerical Prediction of the Noise Produced.
AIAA Paper No. 96-1730, May 1996.
[24] P. J. Morris, L. N. Long, T. Scheidegger, Q. Wang, and A. R. Pilon. High Speed
Jet Noise Simulations. AIAA Paper No. 98-2290, June 1998.
[25] L. Gamet and J. L. Estivalezes. Application of Large-Eddy Simulations and
Kirchhoff Method to Jet Noise Prediction. AIAA Journal, 36(12):2170–2178,
December 1998.
[26] D. Choi, T. J. Barber, L. M. Chiappetta, and M. Nishimura. Large eddy sim-
ulation of high-Reynolds number jet flows. AIAA Paper No. 99-0230, January
1999.
[27] M. Kandula and R. Caimi. Simulation of Jet Noise with Overflow CFD Code
and Kirchhoff Surace Integral. AIAA Paper No. 2002-2602, June 2002.
[28] P. G. Buning, D. C. Jesperson, T. H. Pulliam, W. M. Chan, J. P. Stotnick,
S. E. Krist, and K. J. Renze. OVERFLOW User’s manual Version 1.8g. NASA
Langley Research Center, March 1999.
[29] R. R. Mankbadi, S. H. Shih, R. Hixon, J. T. Stuart, and L. A. Povinelli. Ex-
tension of Near Field to Far Field Noise Prediction. AIAA Paper No. 96-2651,
July 1996.
[30] P. Balakumar. Prediction of Supersonic Jet Noise. AIAA Paper No. 98-1057,
January 1998.
[31] C. C. Yen and N. L. Messersmith. The Use of Compressible Parabolized Stability
Equations for Prediction of Jet Instabilities and Noise. AIAA Paper No. 99-
1859, May 1999.
69
[46] A. S. Lyrintzis and A. Uzun. Integral Techniques for Jet Aeroacoustics Calcu-
lations. AIAA Paper No. 2001-2253, May 2001.
[47] M. E. Goldstein. Aeroacoustics. New York: McGraw-Hill, 1976.
[48] D. W. Wundrow and A. Khavaran. On the Applicability of High-Frequency
Approximations to Lilley’s Equation. Technical report, NASA/CR 2003-212089,
2003.
[49] F. L. Pan, A. Uzun, and A. S. Lyrintzis. Refraction Corrections for Surface
Integral Methods in Jet Aeroacoustics. AIAA Paper No. 2004-2873, May 2004.
[50] F. L. Pan, A. Uzun, and A. S. Lyrintzis. Surface Integral Methods in Jet
Aeroacoustics - Refraction Corrections. submitted to AIAA Journal, 2004.
[51] G. R. Kirchhoff. Zur Theorie der Lichtstrahlen. Annalen der Physik und Chemie,
18:663–695, 1883.
[52] D. G. Crighton, A. P. Dolwing, J. E. Ffowcs Williams, M. Heckl, and F. G.
Leppington. Modern Methods in Analytical Acoustics: Lecture Notes. Springer–
Verlag, London, 1992.
[53] D. L. Hawkings. Noise Generation by Transonic Open Rotors. Westland Re-
search Paper, 1979.
[54] A. R. Pilon and A. S. Lyrintzis. Development of an Improved Kirchhoff Method
for Jet Aeroacoustics. AIAA Journal, 36(5):783–790, 1998.
[55] F. Farassat and M. K. Myers. Extension of Kirchhoff’s Formula to Radiation
from Moving Surfaces. Journal of Sound and Vibration, 123(3):451–461, 1988.
[56] P. Di Francescantonio. A New Boundary Integral Formulation for the Prediction
of Sound Radiation. Journal of Sound and Vibration, 202(4):491–509, 1997.
[57] R. C. Strawn, J. Ahmad, and E. P. N. Duque. Rotorcraft Aeroacoustics Cal-
culations with Overset-Grid CFD Methods. In Proc. AHS 54th Annual Forum,
volume I, pages 230–240, May 1998.
[58] R. C. Strawn, R. Biswas, and A. S. Lyrintzis. Helicopter Noise Predictions
Using Kirchhoff Methods. Journal of Computational Acoustics, 4(3):321–339,
September 1996.
[59] A. S. Lyrintzis, E. K. Koutsavdis, C. Berezin, J. Visintainer, and M. Pollack.
An Evaluation of a Rotational Kirchhoff Methodology. Journal of the American
Helicopter Society, 43(1):48–57, January 1998.
[60] C. Polacsek and J. Prieur. High-Speed Impulsive Noise Calculations in Hover
and Forward Flight Using a Kirchhoff Formulations. In Proc. of the 1st Joint
CEAS/AIAA Aeroacoustics Conference (16th AIAA Aeroacoustics Conference),
Munich Germany, volume II, pages 973–978, June 1995.
[61] A. M. Wissink, A. S. Lyrintzis, R. C. Strawn, L. Oliker, and R. Biswas. Efficient
Helicopter Aerodynamic and Aeroacoustic Predictions on Parallel Computers.
AIAA Paper No. 96-0153, January 1996.
71