Sound Propagation in Lined Ducts With Parallel Flow: WWW - Tue.nl/taverne

Sound propagation in lined ducts with parallel flow
Citation for published version (APA):

Oppeneer, M. (2014). Sound propagation in lined ducts with parallel flow. [Phd Thesis 1 (Research TU/e /
Graduation TU/e), Mathematics and Computer Science]. Technische Universiteit Eindhoven.
https://doi.org/10.6100/IR775562
DOI:
10.6100/IR775562
Document status and date:

Published: 01/01/2014
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:

• A submitted manuscript is the version of the article upon submission and before peer-review. There can be
important differences between the submitted version and the official published version of record. People
interested in the research are advised to contact the author for the final version of the publication, or visit the
DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page
numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please
follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy

If you believe that this document breaches copyright please contact us at:
[email protected]
providing details and we will investigate your claim.
Download date: 17. janv.. 2024

Sound propagation in lined
ducts with parallel flow
Martien Oppeneer
Cover design: Geertje van de Weerdhof–Oppeneer.
A catalogue record is available from the Eindhoven University of Technology Library.
ISBN: 978-94-6259-217-9
Copyright © 2014 by M. Oppeneer, Amsterdam, The Netherlands.

All rights are reserved. No part of this publication may be reproduced, stored in a
retrieval system, or transmitted, in any form or by any means, electronic, mechanical,
photocopying, recording or otherwise, without prior permission of the author.
The work described in this thesis has

been supported by the Dutch National
Aerospace Laboratory (NLR), Amsterdam,
The Netherlands.
Sound propagation in lined
ducts with parallel flow
PROEFSCHRIFT
ter verkrijging van de graad van doctor aan de

Technische Universiteit Eindhoven, op gezag van de
rector magnificus, prof.dr.ir. C.J. van Duijn, voor een
commissie aangewezen door het College voor
Promoties in het openbaar te verdedigen
op maandag 30 juni 2014 om 16:00 uur
door
Martinus Oppeneer
geboren te Axel
Dit proefschrift is goedgekeurd door de promotoren en de samenstelling
van de promotiecommissie is als volgt:
voorzitter: prof.dr. E.H.L. Aarts

1e promotor: prof.dr. R.M.M. Mattheij
2e promotor: prof.dr.ir. B. Koren
copromotor: dr. S.W. Rienstra
leden: Univ.-Prof. Dr.-Ing. W. Schröder (RWTH Aachen)
prof.dr.ir. H.W.M. Hoeijmakers (Universiteit Twente)
prof.dr. J.J.M. Slot
adviseur: dr.ir. P. Sijtsma (NLR)
Contents
Notation iii
1 Introduction 1
1.1 Motivation: ramp noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Context: aircraft duct acoustics . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Problem description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Objectives and main results . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Basic equations and model 11

2.1 Governing equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.2 Conservation laws for a Newtonian fluid . . . . . . . . . . . . . . . 13
2.1.3 Mean flow equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.4 Acoustics—Linearized Euler Equations . . . . . . . . . . . . . . . . 18
2.1.5 Myers’ Energy Corollary . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 Duct modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Wall boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.1 Boundary condition for parallel slipping flow . . . . . . . . . . . . . 26
2.3.2 Sound propagation in porous material . . . . . . . . . . . . . . . . . 27
2.3.3 Bulk absorbing liners . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.4 Honeycomb liner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Pridmore-Brown equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.1 Preform for arbitrary cross-section . . . . . . . . . . . . . . . . . . . 31
2.4.2 Circularly cylindrical geometry . . . . . . . . . . . . . . . . . . . . . 32
2.5 High-frequency approximations . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.5.1 WKB-type solutions for the radial modes . . . . . . . . . . . . . . . 34
2.5.2 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3 Numerical approach 37
3.1 Collocation to solve the BVP . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Path-following based on linear extrapolation . . . . . . . . . . . . . . . . . 40
3.3 Finding bulk-absorber modes by using contour integration . . . . . . . . . 44
3.3.1 Eigenvalues as the roots of an analytic function . . . . . . . . . . . 45
3.3.2 Finding roots using contour integration . . . . . . . . . . . . . . . . 46
3.3.3 Implementation details . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3.4 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
i
ii Contents
4 Mode-matching for segmented ducts 51

4.1 Exact integrals of Pridmore-Brown eigenfunctions . . . . . . . . . . . . . . 52
4.1.1 Exact Integrals of Solutions of the Helmholtz Equation . . . . . . 52
4.1.2 Exact Integrals of Parallel-Flow Modal Eigenfunctions . . . . . . . 54
4.1.3 Exact Integrals of Radial Pridmore-Brown Modes . . . . . . . . . . 57
4.2 Mode-matching at an interface . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2.1 Construction of Matrix Equations (Classical Mode-Matching) . . . 59
4.2.2 Matching Conditions Based on the Bilinear Map . . . . . . . . . . 62
4.2.3 Scattering Matrix Formalism . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5 Asymptotic solutions for slowly varying impedance 75

5.1 WKB solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.2 Numerical comparison between WKB and mode-matching . . . . . . . . . 81
6 Application to APU exhaust duct 87

6.1 Effect of non-uniform mean flow . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2 Effect of non-uniform temperature . . . . . . . . . . . . . . . . . . . . . . . 91
6.3 Role of source in transmission loss . . . . . . . . . . . . . . . . . . . . . . . 97
6.4 Liner with varying depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
7 Concluding remarks 107
A Construction of bilinear map 109
B Approximate Blasius solution 111
Bibliography 120
Index 121
Summary 123
Samenvatting 125
Curriculum Vitae 129
Dankwoord / Acknowledgments 131

Notation
Coordinate systems
x, r, θ cylindrical coordinates
x, y, z Cartesian coordinates
u, v, w velocity components, cylindrical (or if men-
tioned explicitly: Cartesian)
Notation
T∞ , c ∞ , . . . reference temperature, sound speed, etc.
p 0 , v0 , . . . mean flow quantities
p 1 , v1 , . . . perturbations of mean flow
P, U, . . . , P , U , . . . modal shape functions (eigenfunctions)
a ‘physical’ vector or second order tensor
a ‘numerical’ vector
A matrix
Operators / maps
³ ´
d
dt
= ∂∂t + v ∇
³ ´
· total derivative
D ∂ ∂
Dt
= ∂t
+ u 0 ∂x
total derivative for parallel mean flow
∇⊥ , ∇⊥ · transverse gradient / divergence
〈〈F, F̃ 〉〉 bilinear map (arbitrary cross-section)
〈F, F̃ 〉 bilinear map (circular cross-section)
(·) y , (·) z y, z derivative
Latin symbols
b = d + dl [ m] liner outer wall radius
c [ m/s ] speed of sound
c p , cv [ J/(kg · K) ] specific heats at constant pressure, volume
d [ m] duct radius
dl [ m] liner depth
D [ J/(m3 · s) = kg/(m · s3 ) ] perturbation energy ‘dissipation’
e [ J/kg ] specific internal energy
E [ J/kg ] specific total energy
E [ J/m3 = kg/(m · s2 ) ] perturbation energy density
f [ 1/s ] frequency
h [ J/kg ] specific enthalpy
h j , h max continuation step size (Sec. 3.2)
(1) (2)
Hm , Hm Hankel functions of first and second kind
I [ J/(m2 · s) = kg/s3 ] perturbation energy flux vector
Jm Bessel function of the first kind
k [ 1/m ] axial wavenumber
L [ 1/m ] typical length scale / duct length
iii
iv Contents
M Mach number (dim. less axial mean velocity)

m circumferential wavenumber
p [ N/m2 = Pa ] pressure
q [ N/m2 = Pa ] pressure in liner region
q [ J/(m2 · s) = W/m2 ] heat flux vector
R dimensionless mean flow density
R [ J/(kg · K) ] gas constant
s [ J/(kg · K) ] specific entropy
Ŝ, S, S̄ interface, segment, cumulative scattering matrix
t [ s] time
T [ K] temperature
v = 1/ρ [ m3 /kg ] specific volume
v [ m/s ] velocity vector
X = εx slow variable
Ym Bessel function of the second kind
Z, Z0 , Z c [ kg/(m2 · s) ] impedance, face sheet / characteristic (porous mat.)
Greek symbols
α [ m2 /s ] thermal diffusivity (Sec. 2.1)
α [ 1/m ] radial wavenumber
β s , βT [ m2 /N = 1/Pa ] adiabatic, isothermal compressibility
β, γ coefficients of Pridmore-Brown eqn.
γ = c p /c v ratio of specific heats
δ [ m] boundary layer thickness
ε, ² small parameter
² j , ²tol error in k-prediction (Sec. 3.2)
Θ dimensionless mean flow temperature
κ [ W/(m · K) ] thermal conductivity
λ [ m] wavelength
λ continuation parameter (Sec. 3.2)
Λ = ω − µM dimensionless Doppler-shifted frequency
µ, µ v [ Pa · s = kg/(m · s) ] dynamic / volume viscosity (Sec. 2.1)
µ, ν radial mode order
µp [ 1/m ] propagation constant (porous material)
ν [ m2 /s ] kinematic viscosity
Π dimensionless mean flow pressure
ρ [ kg/m3 ] density
ρe [ kg/m3 ] effective density (porous material)
σ [ kg/(m3 · s) ] resistivity (porous material)
τ [ N/m2 = Pa ] deviatoric stress tensor
Φ [ kg/(m · s3 ) ] viscous dissipation term in energy equation
ω [ rad/s ] radial frequency
ω := ∇ × v [ 1/s ] vorticity vector
Ω = ω − ku 0 [ 1/s ] Doppler-shifted frequency
Ω porosity (Sec. 2.3.2)
Dimensionless numbers
Re Reynolds number
Ec Eckert number
Pr Prandtl number
Pe Péclet number
Chapter 1
Introduction
1.1 Motivation: ramp noise

Have you ever had the experience that while boarding an aircraft you had to shout to
your fellow travelers to make yourself audible? Sometimes passengers board a plane
through a passenger boarding bridge, but boarding stairs are also frequently used,
which can normally be reached by walking from the gate in open air. Especially in this
last case a lot of noise can be heard coming from the aircraft while it is being prepared
for departure. If you recognize this experience you have probably also noticed that
the workers who are loading your luggage onto the plane are usually wearing gigantic
‘headphones’ for hearing protection.
The area where the airplane is parked so that boarding, the on- and offloading of
luggage, refueling, etc. can take place, is referred to as the airport apron or ramp. The
noise emitted by the aircraft in this area, while the main engines are still switched off,
is consequently called ramp noise. Ramp noise is caused mainly by the Auxiliary Power
Unit (APU) [62, 89].
More generally, an APU is an engine fitted on a transport vehicle that produces
the power used for other purposes than propulsion. In the modern aircraft context
the APU is a turbine engine in the tail of an aircraft (see Figures 1.1 and 1.2) that
produces power to operate the electrical systems (e.g. for air conditioning) when the
main engines are switched off. It also often produces bleed air—compressed air taken
from within the engine—that is used to start the main engines, to pressurize the main
cabin, for de-icing, and to operate pneumatic actuators.
In order to reduce the noise coming out of the APU exhaust duct, its inside is gener-
ally treated with an acoustically absorbing lining. Over the past decades aeroacoustic
research for lined flow ducts—generally referred to as duct acoustics—was primarily
aimed at reducing the noise levels in inlet and exhaust ducts of the main (turbofan) en-
gines. Recently ramp noise has been given more attention, as regulations have become
more stringent. The International Civil Aviation Organization (ICAO) has published
guidelines for noise certification regarding the maximum noise levels around the air-
craft during ground operations [50]. Also the European Union has defined limits to the
exposure of workers to ramp noise [38]. For some airports (e.g. Copenhagen [85]) the
use of the APU is even more restricted in order to reduce noise and pollution; external
power supplies have to be used instead.
1
2 Chapter 1. Introduction
Figure 1.1: Detail of the tail of an Airbus Figure 1.2: Schematic of an APU with its
A380 aircraft, where the exhaust duct of inlet and exhaust ducts. © Thomas Nodé-
the Auxiliary Power Unit (APU) can be rec- Langlois (Airbus S.A.S.) 2010.
ognized. © David Monniaux 2007.
As for basically any aircraft component, it is very important for an APU exhaust
duct to be light-weight. Moreover, geometrical constraints exist because of the coni-
cal shape of the aircraft fuselage tail. To find an optimum for the trade-off between
minimal weight and noise emissions there is a need for more insight in the main noise
source propagation mechanisms and accurate design tools. This was the main motiva-
tion for the research described in this thesis. Many results, however, are equally well
applicable to sound propagation in any other type of duct, like for example the turbo-
fan inlet and exhaust ducts. We therefore proceed by sketching the wider context of
aircraft duct acoustics [84].
1.2 Context: aircraft duct acoustics

The complete problem of noise radiating from an aircraft turbo-engine can be split in
three parts: the noise source mechanism, the propagation of the noise through the
duct, and the radiation emanating from it. This thesis concerns the second part: duct
acoustic propagation. Before we turn our attention to the specifics of an APU exhaust
duct, let us consider the wider context of the field of aircraft duct acoustics.
In order to be able to describe the relevant physical phenomena and some historical
context here, and to introduce the problem in the next section, it is required that we
first introduce some basic terminology that will be used henceforth. This introduction
is very concise—we will provide more details in the following chapters.
In the field of duct acoustics we are interested in small perturbations of a steady
mean flow. The acoustic problem is solved separately from the mean flow problem.
For aerodynamic reasons many ducts are by and large invariant in the direction of
the duct axis. Therefore, a common approach to model duct acoustics problems is to de-
scribe acoustic solutions in terms of duct modes. This approach will be used extensively
throughout this thesis. These duct modes are solutions that are self-similar in the axial
direction; they are characterized by a mode shape (eigenfunction), which is defined on
the cross-section of the duct, and an axial wavenumber (eigenvalue), which determines
the axial propagation of the mode. These mode shapes are oscillatory in nature; they
can be compared to the vibrations of the membrane of a drum. In general many modes
(or eigensolutions) exist. A discrete set of modes is generally ordered by the number
§ 1.2 Context: aircraft duct acoustics 3
of oscillations, e.g. a mode of higher order is more oscillatory then a lower order mode.
Some modes are more propagative than others; a mode which is (mainly) propagative
in character will be referred to as cut-on, whereas a cut-off mode exponentially decays
in the axial direction and does not propagate.
Classically, the field of duct acoustics was primarily concerned with musical in-
struments and air conditioning systems in buildings. For these applications the wave-
lengths are large with respect to the duct diameter, the mean flow speed is very small,
and the duct walls are rigid; hence models based on plane waves (the lowest order
mode) without mean flow could be used [37]. The field of duct acoustics significantly
broadened with the growth of commercial aviation after World War II.
Before the advent of the turbofan engine, the turbulent exhaust jet was the primary
source of aircraft noise. When the engine with bypass duct became prominent in the
sixties, noise from the compressor and the fan stages inside the engine became more
important. As the bypass ratio (the ratio of the mass flow rate through the annular
bypass duct to that of the engine core) increased, the importance of jet noise decreased
and fan noise became a relatively more important noise source. As a result, more atten-
tion was paid to sound propagating upstream through the inlet duct, and downstream
through the bypass ducts.
An important early development is the so-called Tyler-Sofrin rule [117]. Most of the
noise on the inlet side is due to rotor-stator interaction: rotor wakes are impinging on
stator vanes (which are present to recover energy from the swirl in the mean flow),
producing a specific set of interaction tones. By choosing the number of rotor blades
and stator vanes in a clever way it can be ensured that the first (few) harmonics are
cut-off. This greatly reduced the interaction noise.
Turbofans are very large with respect to the relevant wavelengths, so models based
on plane waves were not sufficient anymore, and higher order modes had to be taken
into account. Moreover, the velocity of the air flowing through the engine is high; even
at take off or landing the typical Mach number (the ratio of flow speed to sound speed)
is 0.7. Therefore, the Doppler-shift due to the convection of the sound waves can not be
neglected, and models based on uniform flow (sometimes referred to as plug flow) were
introduced [41].
The mean flow in the inlet duct is indeed almost uniform, with only a very thin
boundary layer. In the annular exhaust duct however, the flow is strongly sheared,
which requires models based on non-uniform mean flow velocities. For a fluid flowing
along a wall, the fluid velocity can be assumed to be approximately constant away
from the wall, and fall to zero in the boundary layer near the walls. As pointed out
by Pridmore-Brown [97], on a ray-acoustics picture, sound propagating downstream
through this fluid will be refracted towards the wall, while upstream sound rays will
be bent away from the wall. Pridmore-Brown derived an ordinary differential equation
for the acoustic duct modes in a parallel shear flow, which is usually referred to as the
Pridmore-Brown equation. This equation plays a central role in this thesis.
It is well known that sound waves refract towards the region with the lowest sound
speed, which is in stagnant flow the region with the lowest temperature. Consequently,
again on a ray-acoustics picture, it can be seen that cooling the duct wall leads to re-
fraction of the sound towards the walls for both upstream and downstream propagating
waves. The effect of transverse temperature gradients can be included in the Pridmore-
Brown equation [16, 56, 83]. The typical temperature gradients of the mean flow are
not very large in the inlet and exhaust ducts of a turbofan, yet they are important for
Figure 1.3: Honeycomb Helmholtz resonator array liner. Copied with

permission from [116].
an APU exhaust duct, as we will discuss in the next section.

To alleviate the noise problem aircraft engine ducts are treated with an acoustically
absorbent lining. For turbofans these liners consists of a honeycomb structure (rect-
angular cavities have also been introduced) sandwiched between a solid backing plate
and a facing sheet, see Figure 1.3. The face sheet generally consists of a perforated
plate (for example punched or laser-drilled aluminum or mechanically-drilled carbon
fiber composites) or a sheet of woven metal wires (wire mesh). This configuration is of-
ten referred to as a single degree of freedom (SDOF) type of liner, in contrast to a double
degree of freedom (2DOF) liner, which adds an extra layer of honeycomb material and
a porous mid sheet or septum. The 2DOF liner can be designed to attenuate a broader
range of frequencies. The fact that these liners are light-weight, solid structures is an
important advantage for aircraft applications [77].
Honeycomb liners can be modeled as Helmholtz resonator arrays. The depth of
the individual cells of these honey-comb structures is typically a few centimeters1 , and
the diameter of the order of one centimeter. For the prevailing frequencies only plane
waves are therefore propagating in the direction along the cell axis perpendicular to
the liner wall, and no waves are permitted to propagate in the direction parallel to
the liner wall. Consequently, these liners can be described by a single number which
is independent of the position (the impedance, a complex-valued number at a fixed
frequency); they are referred to as locally reacting.
A disadvantage of Helmholtz resonator type liners lies in their limited frequency
range. When a broader range is to be attenuated also porous material may be used
between the solid backplate and the face sheet, because its attenuation properties are
less frequency dependent. This type of liner is referred to as a bulk absorber. For the
main engines bulk absorbers are currently not implemented, since they do not have the
required mechanical and thermal properties, and because of weight issues. This may
change as more advanced materials, such as metal or ceramic foam or fibers become
available.
As the porous material permits waves traveling in a direction parallel to the liner
wall, the acoustic fields in the liner and the duct regions are inherently coupled [106].
The liner can therefore not be modeled by means of a single point value; this is why
bulk absorber liners are also referred to as non-locally reacting. Also large hollow
liner cavities (as can be found in mufflers), and honeycomb liners with interconnecting
1 Approximately 1 λ = 1 c at f = 1500 − 4000 Hz.
4 4 f
§ 1.3 Problem description 5
drainage slots are non-locally reacting.

Finally, a more recent development is the zero-splice inlet liner. The inlet lining of
most turbofans consists of two to three sections that are fitted together, which leaves
gaps (non-treated wall) between the individual lining sections. These gaps are referred
to as splices. It was always assumed that the influence of these hard-wall patches is
minor because of their small surface area, so they were usually neglected during the
liner design process. However, a significant noise reduction could be achieved after
realizing that these splices break the circumferential symmetry of the inlet waveguide.
As a result, the spinning acoustic field connected to the rotor produces an interaction
field (similar to the rotor-stator interaction tones mentioned earlier) of weakly damped
other modes. Because of this, the effectiveness of the liner is improved significantly by
ensuring that there are no splices [11, 69].
1.3 Problem description

The problem of understanding and predicting the acoustic performance of aircraft en-
gines can be approached in several ways: through experiments, or with the aid of an-
alytical or numerical models. As the production of an aircraft engine is exceptionally
costly, the possibilities for performing experiments on real engines are very limited,
and in many cases experiments are even technically impossible to perform. During
the last decades, the increase of computing power and the continuous improvement
of numerical methods has enabled the study of ever more complex physical problems
without resorting to experiments. However, the general importance of simplified an-
alytical models of complex physical and technical problems can hardly be overstated
[30]. A solid theoretical framework is always required for developing, performing and
interpreting both computational and experimental investigations. Asymptotic results
may often be used to increase numerical efficiency and accuracy. On the other hand,
scientists and especially engineers are often primarily interested in the actual solu-
tion of a problem [67]. Whereas analytical models give mostly (although not always)
qualitative answers and can be applied to simplified geometries, numerical methods
are able to handle more complex (realistic) geometries and always give definitive nu-
merical answers. However, the computational cost of numerical methods is often still
prohibitively large for realistic problems, and the computations might produce unphys-
ical results if not performed carefully. An analytical approach may then often be used
in tandem with a numerical approach, which we loosely call ‘semi-analytical’, in order
to have the best of both worlds.
As mentioned before, the research described in this thesis is motivated by the prob-
lem of sound propagation through an APU exhaust duct. We will now describe its
typical properties, which are also schematically depicted in Figure 1.1, and mention a
number of challenges that we wish to address.
The duct is typically straight and circular cylindrical, has a typical length of 1 m
and a typical radius of 15 cm. Cool air enters the duct through the ventilation inlet
on the side. This results in a shear layer between the APU turbine exhaust gas flow
and the ventilation air flow with a strong thermal gradient. In reality the mean flow
velocity and temperature develop along the axial direction due to the existence of ther-
mal conduction and viscous forces. However, here we use the assumption (common for
acoustics) that the gas is ideal, inviscid and non-heat conducting; in view of the fact
that the boundary and shear layers develop slowly (on a length scale of the duct length,
cool air inlet

hard wall resistive sheet
exhaust
liner cavity
mean flow velocity temperature
profile u 0 ( r ) profile T0 ( r )
Figure 1.4: Schematic overview of a typical APU exhaust duct geometry.
which is typically one order of magnitude smaller than the duct radius), and because
the mean flow is free of swirl (i.e. it is not rotating in the plane perpendicular to the
duct axis), we assume a fully developed parallel mean flow. In other words: only the
axial component of the mean flow velocity is non-zero and varies in the transverse di-
rection, and also the mean temperature varies only in the transverse direction. Typical
Mach numbers (i.e. velocity–sound speed ratios) range from 0.1 to 0.3, and the temper-
ature difference between the hot and cold flows is typically a few hundred degrees. The
uniform mean (ambient) pressure has a typical value of 1 bar. The mean density then
follows from the transverse temperature profile.
Due to the conical shape of the aircraft fuselage tail, the lining inside the duct wall
has an axially decreasing depth ranging from more than ten to a few centimeters. It
is covered with a (mostly resistive) facing sheet consisting of a perforated plate and a
wire mesh. The lining typically consists of a number of relatively large annular liner
segments, which may be open or filled with porous material. The duct wall may also
consist of Helmholtz resonator arrays (honeycombs) with axially varying depth.
To compute the sound inside the duct we formulate the problem in terms of acoustic
duct modes, i.e. the infinitely many eigensolutions of the boundary value problem con-
sisting of the Pridmore-Brown equation for parallel mean flow and transverse temper-
ature gradients subject to suitable boundary conditions. We assume that all acoustic
solutions can be described in terms of these modes (which might not be the case for
shear flow [23]). For this purpose it is important that all relevant modes (for example
with axial wavenumbers in a specific part of the complex plane) are found.
The Pridmore-Brown equation reduces to Bessel’s equation for the specific case of
uniform mean flow and temperature. Consequently, the eigenfunctions are in this case
Bessel functions and the eigensolutions follow from the roots of a function that is avail-
able in closed form. For boundary conditions corresponding to a hard wall these roots
are even real-valued, so they are easily found numerically, since the real numbers are
ordered. For the case of locally or non-locally reacting boundary conditions these roots
lie in the complex plane, so there is no natural ordering. A Newton-based root-finder
may be used, provided that good initial guesses are available. Depending on the lo-
§ 1.4 Objectives and main results 7
cation of the initial guesses and the basins of attraction of the actual roots, Newton’s
method may not converge to all of the relevant roots. This problem is often mitigated
somewhat in practice by using an excess number of initial guesses, but this does not
fundamentally solve the problem, hence there is a risk that some modes are missed.
For the more general case of non-uniform mean flow and temperature solutions of
the Pridmore-Brown equation are not known in closed form. Hence, we have to solve
the boundary value problem numerically. Also in this case good initial guesses are
important to ensure that all relevant modes are found.
For modes to exist it is necessary that, at least locally, the duct geometry is invari-
ant in the axial direction. More specifically: the duct should have constant boundary
conditions. For an APU exhaust duct the liner depth is axially varying even though
this variation is small scale compared to the duct length. Consequently, the boundary
condition varies in the axial direction. If this variation is continuous, then duct modes
in a strict sense do not exist.
If the boundary conditions are step-wise continuous, i.e. if the duct is subdivided
into a number of segments, each individual segment having a uniform depth, then the
duct is axially invariant within each segment. At the interface between two liner seg-
ments the boundary conditions are discontinuous, and both segments have a different
set of duct modes. An incoming wave from one segment is partly reflected and partly
transmitted to the other segment. These scattering effects are studied by means of
the mode-matching method, which relates the modal amplitudes in two adjacent seg-
ments via a scattering matrix. The entries of this matrix are integral inner products
of duct eigenfunctions. For uniform flow and temperature the values of these integrals
can be computed analytically since the eigenfunctions are available in closed form; for
non-uniform flow and temperature we have to use the numerically computed eigen-
functions, so the integrals are not available in closed-form. Numerical quadrature
can be used, but this is computationally expensive due to the oscillatory nature of the
eigenfunctions, and it has only a finite numerical accuracy.
A common design criterion for liner design is the transmission loss, which is defined
as the difference between the acoustic power that is radiated from the duct and the
input acoustic power. Knowledge about the noise source characteristics is therefore
required for its computation. Little is known about the specific characteristics of the
noise coming from the APU turbine; it is typically characterized as broad-band, having
frequencies ranging from 100 Hz to 10 kHz. As this thesis considers only the duct
propagation, no specific knowledge of noise sources is assumed, so we choose a very
general form if required. Furthermore, we assume that the duct end is reflection free,
or in other words, that it has infinite length. This is in general not true, specifically
for a hard-walled duct at low frequencies (wavelength comparable to the duct radius).
However, for higher frequencies and for waves not too close to cut-off, the reflections
are very small. In addition, when the duct wall is treated, the amplitude of the sound
field is reduced, and the influence of the end reflection may be even smaller. [105, 108]
1.4 Objectives and main results

The main goal of this work is to provide semi-analytical solutions for the propagation
of sound in lined flow ducts with parallel flow and strong thermal gradients, and axi-
ally varying lining. We focus primarily on the applicability of the models for an APU
exhaust duct, although other applications like turbofan inlet and exhaust ducts are
equally well possible. For this purpose several ingredients are required. Firstly, we
need an eigenmode solver that is numerically efficient and robust; it is important that
all relevant eigensolutions are found. Secondly, we require methods to handle the axial
variation of liner properties.
Before we proceed by mentioning our specific objectives and results, we first men-
tion that the basic modeling steps that eventually lead to the Pridmore-Brown equation
are presented in Chapter 2. In this chapter we also describe how sound absorbing walls
can be modeled through the boundary conditions.
Path-following to find Pridmore-Brown eigensolutions
Our numerical approach to find Pridmore-Brown eigensolutions for non-uniform flow

and temperature in an efficient and robust manner is based on a combination of the
COLNEW code and a path-following (or continuation) approach (this is the topic of Sec-
tions 3.1 and 3.2). To ensure that we find all relevant eigensolutions, we start from an
easy solution (for example uniform flow and temperature) and trace the solution when
the relevant problem parameters are varied to the values of interest, tacitly assuming
that no other solutions appear during the path. Our path-following approach is refined
by using a prediction-correction scheme. The prediction is found by linear extrapo-
lation of the previous solutions. The correction step is then an updated solution by
COLNEW, with the prediction as the starting value. In addition, we take into account
that for specific choices of the impedance paths some modes may appear from infinity,
in which case they will be missed. We make sure that the impedance is varied such
that all modes start from a finite value and move towards infinity as the impedance
goes to a hard wall value. The effectiveness of the numerical method is illustrated by
comparing it with some asymptotic results (which are presented in Section 2.5).
Root finding by contour integration to find initial guesses
Our initial guesses of the eigensolutions are found with the aid of a robust and efficient
root-finding technique based on complex contour integration (this is the topic of Sec-
tion 3.3). Finding the solutions for a lined duct carrying uniform flow and temperature,
which serve as the initial guesses for the path-following algorithm, amounts to finding
the complex-valued roots of an analytic function. To compute these roots we first con-
struct a polynomial that has the same roots as the analytic function of interest by using
contour integration (i.e. numerical quadrature). Subsequently the roots of the polyno-
mial can be computed in a standard manner. The advantage of this method lies in the
fact that it is guaranteed that all roots inside a given area of the complex plane are
found (apart from finite numerical accuracy issues), whereas a Newton method only
converges to all of the roots if it is started from sufficiently close initial guesses. The
location of these initial guesses for the Newton method is not always self-evident, as for
example in case of surface waves and modes that exist only with non-locally reacting
lining.
Mode-matching based on closed-form integrals
One way to deal with axially varying liners is the mode-matching method, which is
essentially a projection method. For non-uniform flow, in the classical approach, the
necessary inner products need to be evaluated by numerical quadrature. We present
§ 1.4 Objectives and main results 9
in Chapter 4 a new mode-matching method, which consists of replacing the inner-

products with expressions that can be evaluated in closed form. Instead of the stan-
dard inner product we use a bilinear form that resembles an inner product; it is an
integral of a weighted combination of products of Pridmore-Brown modes that can be
evaluated in closed form. Apart from numerical efficiency, this approach features also
a higher accuracy, because it avoids the inherently inaccurate numerical quadrature of
oscillating functions. Numerical results of the new and the classical approach are com-
pared, and the agreement is excellent, with higher accuracy and greater computational
efficiency for the new approach.
Once we have the mode-matching system of equations available for each interface,
we compute the combined transmission and reflection effects at multiple interfaces
with the aid of the numerically stable scattering-matrix formalism. In contrast, the
transfer matrix formalism, which considers the propagation of cut-off modes in both
directions, can potentially lead to exponentially large and small values in the matrix
equations, which can result in numerically ill-posed problems. The scattering-matrix
formalism only considers decaying waves; it is therefore numerically stable, also for a
large number of segments.
The availability of the numerical Pridmore-Brown eigenfunctions enables us to also
consider asymptotic solutions in the form of slowly varying modes of WKB type. Solu-
tions of this type for an impedance that is continuously varying in the axial direction
are presented in Chapter 5. Additionally, we compare these asymptotic results with
mode-matching results. We illustrate that for a realistic APU exhaust duct config-
uration, for which the impedance is not slowly varying due to liner resonances, the
asymptotic results are not applicable.
Application to APU exhaust duct

Finally, we aim to illustrate the practical applicability of the developed methods and to
present results and insights that facilitate design studies of actual APU exhaust ducts.
Based on test cases that are typical of an APU exhaust duct geometry, we describe the
refraction effects due to non-uniform mean flow and temperature, and describe how
this influences the sound attenuation. Furthermore, we discuss an approach to model
the noise source for transmission loss calculations, when the exact source is not known.
For the test cases that we present especially the temperature non-uniformity is very
beneficial for the attenuation. It is also shown that modes of the lowest circumferential
order are the least attenuated, and consequently the most interesting ones for design
calculations.
Chapter 2
Basic equations and model
We start this chapter by formulating the equations that govern acoustics in Section
2.1. Based on the modal approach described in Section 2.2 we subsequently describe in
Section 2.3 the boundary conditions that model sound absorbing walls, and derive the
Pridmore-Brown equation in Section 2.4. We conclude this chapter by presenting some
asymptotic approximations of solutions of this equation in Section 2.5.
2.1 Governing equations

As acoustics is often defined as small dynamic perturbations of a steady fluid flow we
first recapitulate the basic equations of thermodynamics and the conservation laws of
fluid mechanics. We then consider the governing equations and state our basic assump-
tions for both the mean flow and the acoustic problem. It will be argued that viscous
and heat conductive effects can be neglected for both problems. We subsequently show
that the perturbations can be described by the Linearized Euler Equations (LEE), since
they are very small. Since any solution of the LEE satisfies Myers’ energy corollary,
this can be used to validate numerical results.
2.1.1 Thermodynamics
In thermodynamics [109] a distinction is made between thermodynamic state prop-
erties, like for example the specific (i.e. per unit mass) internal energy e, and other
variables, like for example the amount of specific heat q that is transferred to the sys-
tem and the specific work w that is done by the system. Changes in thermodynamic
properties are described by exact differentials d·, since when they are integrated, the
result will be independent of the thermodynamic path1 . The specific heat q and work
w however, are not state properties, so changes are described by an inexact differential
δ·, and the value of an integral of δ q or δw depends on the specific path.
The first law of thermodynamics, which describes the conservation of energy, can
be expressed as
d e = δ q − δw. (2.1)
1 A differential form d q = a( x, y)d x + b( x, y)d y is called exact if the value of the integral
R
d q is path-
independent.
³ This
´ is³true ´ if a and b are the components of a conservative vector field with potential q, and
∂q ∂q
hence d q = ∂ x d x + ∂ y d y.
11
12 Chapter 2. Basic equations and model
The absolute temperature T , specific entropy s, pressure p, density ρ and specific vol-
ume v = 1/ρ are thermodynamic state functions as well. One formulation of the second
law of thermodynamics, which describes the concepts of entropy and reversibility, is
δq
ds Ê , (2.2)
T
where equality only holds for reversible processes. For a simple compressible system
that can only do pressure-volume work δw = pdv it follows that
p
d e = T d s − pdv = T d s + dρ . (2.3)
ρ2
This is the fundamental thermodynamic relation for a closed single-component sys-

tem (a gas for instance). The fundamental thermodynamic relation holds for either
reversible or irreversible processes, since all quantities are state functions. However,
for irreversible processes, T d s and pdv can not be associated with heat and work any
more.
As we will see later sound propagation is an isentropic (i.e. with constant s) process;
the sound speed c, which is also a thermodynamic state property, is defined2 as
µ ¶
∂p
c 2 := . (2.4)
∂ρ s
Complying with the common notation in thermodynamics literature, the subscript s

is used here to express that p is considered to be a function of ρ and s, and that s
is kept constant in the partial derivative. For a simple compressible system, accord-
ing to the state postulate, the thermodynamic state is completely determined by two
thermodynamic
³ ´ ³ state
´ functions, so we have for example: ρ = ρ ( p, s). Consequently
∂ρ ∂ρ
dρ = ∂p s
d p + ∂ s d s, so we can write:
p
µ ¶
1 ∂ρ
dρ − d p = d s. (2.5)
c2 ∂s p
The heat capacity, which describes the capacity of a material to store energy as the
temperature is increased, can generally be defined as
δq
c heat = . (2.6)
δT
At constant pressure it follows from the fundamental thermodynamic relation that
d h = T d s, where the specific enthalpy is defined as h = e + pv. For a reversible process
we have T d s = δ q, so d h = δ q at constant pressure. Similarly, at constant volume we
have d e = δ q. This motivates the definition of the specific heat capacity at constant
volume c v and the specific heat capacity at constant pressure c p , and the ratio of heat
capacities as µ ¶ µ ¶
∂e ∂h cp
c v := , c p := , γ := . (2.7)
∂T v ∂T p cv
For air γ ≈ 1.4. Generally the state functions c p and c v depend on two of the other
thermodynamic properties, e.g. pressure and temperature. For an ideal gas this is not
the case, as we will show next.
2 We use := to denote a definition.
§ 2.1 Governing equations 13
³ ´ ¡ ∂s ¢
∂p
Differentiating (2.3) and using one of the so-called ‘Maxwell relations’ ∂T v
= ∂v T
(which follow from the fact that for exact differentials mixed partial differentials com-
mute) yields
µ ¶ µ ¶
∂e ∂s
=T −p
∂v T ∂v T
µ ¶ µ ¶
∂p ∂( p/T )
=T − p = T2 . (2.8)
∂T v ∂T v
In this research we assume that we can use the equation of state of an ideal gas
p = ρ RT, (2.9)
where R is the specific

³ gas´ constant (for air: R = 287 J/(kg · K)). It follows that for an
¡ ∂e ¢ 2 ∂(R/v)
ideal gas ∂v T = T ∂T
= 0, so e is a function of T only. It can also be shown that
v
h = h(T ) in a similar fashion. Thus for an ideal gas the relations
d e = c v dT, d h = c p dT (2.10)
can be used, from which it follows that d h = d( e + pv) = d e +d(RT ) so c p dT = c v dT +RdT

and consequently
R = c p − cv. (2.11)
Using these relations as well as d h = T d s + vd p it can also be found that for an ideal
gas
cv cp
d s = d p − dρ . (2.12)
p ρ
For constant entropy, i.e. d s = 0, it then follows from this relation that the sound speed
of an ideal gas can be computed as
µ ¶
2 ∂p γp
c = = = γRT. (2.13)
∂ρ s ρ
For a perfect gas c p and c v are constant so (2.12) can be integrated to find
s = c v log p − c p log ρ + ( s init − c v log p init + c p log ρ init ), (2.14)
where s init , p init , and ρ init denote the initial entropy, pressure and density of the fluid.3
2.1.2 Conservation laws for a Newtonian fluid

We shall next recapitulate the conservation laws of fluid dynamics [60]. We consider
a fluid with a velocity field v, specific density ρ , pressure p, deviatoric stress tensor τ
(the non-isotropic part of the Cauchy stress tensor), and heat flux vector (due to heat
conduction) q. The specific total energy E is the summation of the specific kinetic and
3 This particular form of the integration constant between brackets makes clear that the dimensions are
³ ´ ³ ´
p ρ
correct (i.e. the arguments of log p and log ρ are dimensionless).
init init
the specific internal energy: E = e + 12 |v|2 . The conservation laws in differential form
describing conservation of mass, momentum and total energy are respectively
∂ρ
∂t
·
+ ∇ (ρ v) = 0, (2.15a)
∂
∂t
·
(ρ v) + ∇ (ρ vv) = −∇ p + ∇ τ, · (2.15b)
∂
∂t
· · ·
(ρ E ) + ∇ (ρ Ev) = −∇ q − ∇ ( pv) + ∇ (τ v). · · (2.15c)
Often these equations are referred to as the compressible Navier-Stokes equations. The
fundamental thermodynamic relation (2.3) can be used, together with (2.15a) and the
inner-product of (2.15b) with v to obtain a different form of the energy conservation
law (2.15c), namely µ ¶
∂
ρT + v ∇ s = −∇ q + τ : ∇v,
· · (2.15d)
∂t
X
where the double dot product ( : ) of two tensors can be computed as a : b = a i j b ji .
i, j
Constitutive equations are required to solve this system; there are more unknowns
than equations in (2.15). Here we choose Fourier’s law of heat conduction and the
stress relation for a Newtonian fluid. For a fluid with thermal conductivity κ, dynamic
viscosity µ and volume viscosity µv (also called second viscosity coefficient) these are
q = −κ∇T, (2.16a)
³ ´
τ = µ ∇v + (∇v)T + µv (∇ v)δ, · (2.16b)
where δ is the unit tensor4 . For many applications the Stokes assumption µv = − 32 µ is
found to be sufficiently accurate.
For a fluid element that is carried with the flow along a stream line we can use the
convective derivative, which is defined as
µ ¶
d ∂
dt
:=
∂t
+v ∇ . · (2.17)
We also assume that the physical parameters µ, µv and κ are constant. The conserva-
tion laws can now be formulated as
dρ
dt
+ ρ ∇ v = 0, · (2.18a)
dv
ρ
dt
+ ∇ p = µ∇2 v + (µv + µ)∇(∇ v), · (2.18b)
ds
ρT = κ∇2 T + Φ, (2.18c)
dt
where
1 ³ ´ ³ ´
Φ = µ ∇v + (∇v)T : ∇v + (∇v)T + µv (∇ v)2 (2.19)·
2
is the viscous dissipation term. The system of (2.18) together with the ideal gas law
(2.9) and the thermodynamic relation (2.12) now form a complete system.
4δ
i j = 1 if i = j and 0 otherwise.
In acoustics we are interested in small unsteady perturbations of a steady mean

flow. The two separate problems (mean flow and acoustics) are very different in nature,
and need to be handled separately. The results presented in this thesis are based on
the assumption that the effects of viscosity and thermal conductivity can be neglected,
both for the mean flow and the acoustic propagation—this will be motivated in the next
sections. Consequently, the complete field can be described by the Euler equations
∂ρ
∂t
·
+ ∇ (ρ v) = 0 (2.20a)
dv
ρ + ∇p = 0 (2.20b)
dt
ds
= 0. (2.20c)
dt
An alternative for the energy equation can be found by substituting (2.20c) into (2.5):
dp dρ
= c2 . (2.21)
dt dt
Note that the ideal gas law is not used in the derivation of this equation, only the
assumption that the total flow is isentropic ( ddst = 0). By using the equation for conser-
vation of mass (2.20a) the energy equation can also be written as
dp
dt
+ γ p∇ v = 0, · (2.22)
where γ p = ρ c2 is used, or as
dT
dt
+ T (γ − 1)∇ v = 0, · (2.23)
by using the ideal gas law p = ρ RT . Generally p = p(ρ , s), but for homentropic flow we
have ³d s =´ 0 so s = s 0 is uniformly constant, and consequently p = p(ρ ). It follows that
∂p
∇p = ∂ρ s
∇ρ , so we have
∇ p = c2 ∇ρ . (2.24)
This relation holds generally for homentropic gas flows ( c p and c v not necessarily con-
stant).
2.1.3 Mean flow equations

As mentioned above, we are interested in small unsteady perturbations of a steady
mean flow, so that the field variables can be expressed as
v( x, t) = v0 ( x) + v1 ( x, t), p = p 0 ( x) + p 1 ( x, t), etc.. (2.25)
From the conservation laws (2.18) we find that the steady mean flow field satisfies
·
∇ (ρ 0 v0 ) = 0, (2.26a)
·
ρ 0 (v0 ∇)v0 + ∇ p 0 = µ∇2 v0 + (µv + µ)∇(∇ v0 ), · (2.26b)
·
ρ 0 T0 (v0 ∇) s 0 = κ∇2 T0 + Φ0 . (2.26c)
In this section we will motivate why it is reasonable to assume that the mean flow is
inviscid and non-heat conducting [60].
We start by considering the effect of viscosity in the momentum equation. Because
the viscosity is very small the dominant balance in the flow is between the pressure
2
and inertia forces, so we scale the pressure on ρ ∞ v∞ , where ρ ∞ and v∞ are the typical
mean flow density and velocity. Thus we use
v0 ρ0 ∆ p0 x
ṽ0 = , ρ̃ 0 = , ∆ p̃ 0 = 2
, x̃ = , (2.27)
v∞ ρ∞ ρ ∞ v∞ L
where we also introduced the typical length scale L. The conservation laws for mass
and momentum then take the form (tildes omitted)
·
∇ (ρ 0 v0 ) = 0,
· ¸
(2.28a)
1 µv + µ
·
ρ 0 (v0 ∇)v0 + ∇ p 0 =
Re
∇2 v0 +
µ
∇(∇ v0 ) , · (2.28b)
where the Reynolds number is defined as
ρ ∞ v∞ L
Re := . (2.29)
µ
For ρ ∞ = 1.25 kg/m3 (the density of air), v∞ = 100 m/s, L = 1 m, and µ = 1.8·10−5 kg/(m · s)
(the dynamic viscosity of air) this results in a Reynolds number of the order of 106 .
Thus we conclude that for the free mean flow viscous forces can be neglected.
Close to the wall, however, a very thin boundary layer develops [60]; its typical
thickness δ is much smaller than the length scale L. Since the inviscid equations have
no solutions that satisfy the no-slip boundary condition at a surface, viscosity does play
a role here. As a simple prototypical example, consider a uniform flow that impinges
on a flat surface that is oriented in the direction of the flow. We use a two-dimensional
Cartesian coordinate system with x along the surface and y normal to it. The longitudi-
nal and transverse velocity components are u and v. Assume that compressibility plays
no role in a small region of interest along the boundary, i.e. ∇ v0 = 0. The longitudinal ·
component of the momentum equation is then
µ 2 ¶
∂u ∂u 1 ∂p ∂ u ∂2 u
u +v =− +ν + , (2.30)
∂x ∂y ρ ∂x ∂ x 2 ∂ y2
where ν = µ/ρ is the kinematic viscosity. The boundary layer changes in the x-direction
on a length scale L, and in the y-direction on a length scale δ. Furthermore it follows
from ∂∂ux + ∂∂vy = 0 that v ∼ v∞ δ/L, where ∼ means ‘of the order of’. Hence we can estimate
the magnitude of the advective terms as
2
∂u v∞ ∂u
u ∼ ∼v . (2.31)
∂x L ∂y
A measure of the viscous term is
∂2 u ν v∞
ν ∼ . (2.32)
∂ y2 δ2
If the viscous and advective forces are of equal importance we can estimate the devel-
opment of the boundary layer thickness as
L Lv∞
δ= p , with ReL = . (2.33)
ReL ν
We see that the boundary layer develops very slowly; its thickness is typically of the
order of millimeters at one meter from the edge of the surface. We therefore consider
the mean flow to be parallel.
Next we consider the effects of viscosity and thermal conductivity in the energy
equation. In order to make it dimensionless we first rewrite this equation as
· ·
ρ 0 c p v0 ∇T0 − v0 ∇ p 0 = κ∇2 T0 + Φ0 . (2.34)
This formulation can be found by substituting T d s = c p dT − d p/ρ (the fundamental

thermodynamic relation for an ideal gas) in (2.18c) and assuming steady state. Next to
the scaling already mentioned in (2.27) we use
dT0 Φ0
dT̃0 = , Φ̃0 = 2 /L2
, (2.35)
∆T µ v∞
where ∆T is a characteristic temperature difference. The energy equation then be-

comes (tildes omitted)
1 2 Ec
·
ρ 0 v0 ∇T0 − Ec v0 ∇ p 0 = · Pe
∇ T0 +
Re
Φ0 , (2.36)
with dimensionless Eckert and Péclet numbers Ec and Pe, which will be described next.
The Eckert number, which is defined as
2
v∞
Ec := , (2.37)
c p ∆T
expresses the relation between the kinetic energy and the enthalpy of a flow. Here we
assume that Ec / Re ¿ 1. Before we introduce the Péclet number we define the Prandtl
number Pr and the thermal diffusivity α as
ν c pµ κ
Pr := = , α := . (2.38)
α κ ρ∞ c p
The Prandtl number is a measure for the ratio of momentum diffusivity to thermal
diffusivity. It is a property of a fluid; for most gases it is of order one. The Péclet
number for the diffusion of heat is defined as
Lv∞ Lv∞ ρ ∞ c p
Pe := = = Pr Re . (2.39)
α κ
It measures the ratio of the advective and the diffusive transport rate of heat, so ther-
mal diffusion plays a small role for large Péclet numbers. Because the Prandtl number
is of order one we can conclude that the Péclet number is large for gases. This leads us
to conclude that the heat conduction and viscous dissipation terms in the energy equa-
tion are small so that they can be neglected. It follows that the mean flow is isentropic
along stream lines, i.e. ddst0 = 0.
To summarize, the steady mean flow satisfies
·
∇ (ρ 0 v0 ) = 0, (2.40a)
·
ρ 0 (v0 ∇)v0 + ∇ p 0 = 0, (2.40b)
·
(v0 ∇) s 0 = 0. (2.40c)
Moreover, from (2.21) we have as a different form of the energy equation
· ·
v0 ∇ p 0 = c20 v0 ∇ρ 0 . (2.41)
We also have the ideal gas law

p 0 = ρ 0 RT0 (2.42)
and the thermodynamic relations
cv cp γ p0
ds0 = d p 0 − dρ 0 , c20 = = γRT0 . (2.43)
p0 ρ0 ρ0
This research concerns sound propagation through a straight circularly symmetric

duct. Hence it is natural to use a cylindrical coordinate system ( x, r, θ ), with axial,
radial and circumferential flow velocity components denoted by u, v and w. We assume
that the flow is swirl free, i.e. w0 = 0, and distinguish the following types of solutions
of the steady state Euler equations:
• parallel flow with v0 = u 0 ( r ) e x , with e x denoting the unit vector in the x direction;
• uniform flow with v0 = u 0 e x ;
• no mean flow, i.e. v0 = 0.
(Note that parallel flow is not rotation free, which means that it is not possible to
formulate a non-uniform flow in a straight duct in terms of a velocity potential.) For
these types of flows it follows from the momentum equation (2.40b) that the mean
pressure p 0 is constant. The density and the entropy then follow from the temperature.
We assume one of the following:
• constant mean temperature T0 ;
• a mean temperature that varies only in the radial coordinate: T0 = T0 ( r )
The temperature profile T0 ( r ) can be chosen independently of the mean flow velocity
profile. If the temperature T0 (and consequently the density ρ 0 ) is constant, then the
mean flow is homentropic, meaning that it has constant entropy s 0 .
2.1.4 Acoustics—Linearized Euler Equations

In this section we describe the equations that govern the small (acoustic) perturbations,
which are denoted by the subscript 1 . First we will show that these perturbations are
very small compared to the mean flow quantities [108], and that for this reason we
can neglect the nonlinear terms that arise after substitution of (2.25) into (2.18). (This
linearization matches with our common experience: we can distinguish multiple con-
versations in a room—each linear combination of solutions of the governing equations
is a solution itself, so the individual sound fields do not influence each other [39].) Sub-
sequently we will motivate why we neglect viscosity and thermal conductivity also for
the acoustic problem [17, 94].
In the following we again use ∞ to denote reference values that are used to make
the equations dimensionless; these are the typical values of the mean flow variables.
For simplicity we allow the reference speed v∞ to be identical to the reference sound
speed c ∞ . We then have
v0 ρ0 p0 T0 s0
ṽ0 = , ρ̃ 0 = , p̃ 0 = , T̃0 = , s̃ 0 = , (2.44)
c∞ ρ∞ p∞ T∞ s∞
with the definitions

c p ∆T
p ∞ := ρ ∞ c2∞ , s ∞ := . (2.45)
T∞
The definition for s ∞ is motivated by taking p 0 constant in (2.43) from which we find
d s 0 = c p dT /T0 . As before, ∆T is the typical temperature difference of the mean flow.
In acoustics the Sound Pressure Level (SPL) is defined as
µ ¶
prms
1
SPL = 20 log , pref = 2 · 10−5 Pa, (2.46)
pref
where prms
1 is the root mean square of the acoustic fluctuation. The sound pressure
level corresponding to the threshold of pain is about 140 dB, which corresponds to a
p
pressure signal with prms
1 = 200 Pa. Consequently, the ratio p10 is (for our purposes) of
the order of 10−3 or smaller. As this ratio is small we can linearize the thermodynamic
relation for an ideal gas (2.14) to find
cv cp cv ¡ ¢
s1 = p1 − ρ 1 = p 1 − c20 ρ 1 (2.47)
p0 ρ0 p0
ρ1 p1
if we can assume that ρ0
is of the same order as p 0 . It will be shown later that
s 1 = 0 for uniform temperature, in which case p 1 = relation can be used c20 ρ 1 . This
p ρ
to find p10 = γ ρ 10 , which justifies our previous assumption. For the temperature per-
turbations, differentiate the logarithm of the ideal gas law and use γ p = ρ c2 to obtain
dp c 2 dρ T1
= dTT . Based on this result we can estimate T
ρ
p − γp = (γ − 1) ρ 1 . To estimate the size
0 0
of the velocity perturbations we consider a plane wave traveling through a medium at
rest having uniform temperature. The acoustic particle velocity is then u 1 = p 1 /(ρ 0 c 0 ).
Dividing this relation by c 0 and using again p 1 = ρ 1 c20 yields uc01 = ρ 10 . Finally, dividing
ρ
³ ´
p1
(2.47) by s ∞ results in ss∞1 = T
ρ1
∆T γ p 0 − ρ 0 , where ∆T /T∞ is of order unity.
∞
ρ1
Based on these estimates we therefore use the ratio ρ0
as a small parameter ² and
use the scaled variables
p1 ρ1 v1 T1 s1
p̃ 1 = , ρ̃ 1 = , ṽ1 = , T̃1 = , s̃ 1 = . (2.48)
² p∞ ²ρ ∞ ² c∞ ²T∞ ²s∞
Furthermore, we scale time on the acoustic frequency f , and distances on the acoustic
wavelength λ = c ∞ / f . Upon substituting the scaled variables in the conservation laws
(2.18), subtracting the mean flow terms that are of zeroth order in ², and neglecting
the terms of order ²2 , we obtain
∂
∂t
·
ρ 1 + ∇ (v0 ρ 1 + v1 ρ 0 ) = 0, (2.49a)
³∂ ´
ρ0
∂t
· · ·
+ v0 ∇ v1 + ρ 0 (v1 ∇)v0 + ρ 1 (v0 ∇)v0 + ∇ p 1 = µ∇2 v1 + (µv + µ)∇(∇ v1 ), (2.49b) ·
³∂ ´
ρ 0 T0
∂t
· ·
+ v0 ∇ s 1 + ρ 0 T0 v1 ∇ s 0 = κ∇2 T1 + Φ1 , (2.49c)
¡ ¢ ¡ ¢
with the linearized viscous dissipation term Φ1 = µ ∇v0 + (∇v0 )T : ∇v1 + (∇v1 )T +
· ·
2µv (∇ v0 )(∇ v1 ). Note that gradients of mean flow and perturbation variables do not
necessarily have the same length scale. Suppose that the mean flow changes over a
length scale L and the perturbations over the distance of one wavelength λ. Terms
of the order ²λ/L then arise in the first order equations. However, even for very low
frequencies ²λ/L ¿ 1, so the linearization is still warranted.
To illustrate why viscosity does not play a role at the frequencies of interest we
consider a plane wave propagating in the x-direction without mean flow and at uniform
mean temperature. The linearized mass and momentum equations then reduce to
∂ρ 1 ∂u1
+ ρ0 = 0, (2.50a)
∂t ∂x
∂u1 ∂ p1 ∂2 u 1
ρ0 + = µ Nv , (2.50b)
∂t ∂x ∂ x2
where the viscosity number Nv defined as
Nv := (µv + 2µ)/µ (2.51)
has order of magnitude one. Together with the relation p 1 = c20 ρ 1 this leads to a viscous
wave equation
∂2 u 1 1 ∂2 u 1 µ Nv ∂3 u 1
− + = 0. (2.52)
∂ x2 c20 ∂ t2 ρ 0 c20 ∂ x2 ∂ t
By substituting a plane-wave solution of the form e−iω t+i kx we find the dispersion rela-
tion Ã " #!
2 ωµ Nv ω2
k 1−i 2
= 2. (2.53)
ρ 0 c0 c0
The term between square brackets, which is not present in the dispersion relation for
inviscid wave propagation, can be interpreted by defining an acoustic Reynolds number
ρ 0 c0 λ ωµ Nv 2π Nv
Reac := , so that = . (2.54)
µ ρ 0 c20 Reac
Consequently the importance of viscous effects increases with frequency. At frequen-

cies in the audible range viscosity can be neglected: for a frequency of 10 kHz we have
Reac of the order of 109 .
We now also motivate why we neglect heat conduction for sound propagation. For
simplicity we again consider plane-wave propagation in the x-direction and assume
no mean flow and uniform mean temperature (and consequently constant s 0 ). The
linearized energy equation (2.49c) then becomes
∂s1 ∂2 T 1
ρ 0 T0 =κ . (2.55)
∂t ∂ x2
By using the linearized versions of the fundamental thermodynamic relation (2.62) and
the ideal gas law (2.61) we find
Ã ! Ã !
∂ p1 ∂2 γ p 1
ρ0 c p − ρ1 = κ 2 − ρ1 . (2.56)
∂ t c20 ∂x c20
From (2.50) it can easily be seen that the inviscid mass and momentum equations can
be combined to find
∂2 ρ 1 ∂2 p 1
− = 0. (2.57)
∂ t2 ∂ x2
Together with (2.56) this leads to
Ã ! Ã !
∂ ∂2 1 ∂2 2π ∂2 ∂2 γ ∂2
− p1 = − p1 , (2.58)
∂ t̃ ∂ x2 c20 ∂ t2 Peac ∂ x̃2 ∂ x2 c20 ∂ t2
where the acoustic Péclet number is defined as
ρ 0 c p c0 λ
Peac := . (2.59)
κ
We also introduced dimensionless numbers t̃ = ω t and x̃ = xω/ c 0 in order to compare
the two terms between brackets. The bracketed term on the left hand side describes
adiabatic wave propagation with sound speed c20 , whereas the bracketed term on the
right describes isothermal wave propagation with sound speed c20 /γ. The acoustic Pé-
clet number is of the same order of magnitude as the acoustic Reynolds number since
they are related as Peac = Pr Reac , where the Prandtl number Pr is of the order unity
for most gases. We therefore conclude that heat conduction can be neglected for sound
within the audible frequency range.
Very close to walls, viscosity and thermal conductivity give rise to thin acoustic
boundary layers. We assume here that the wavelength is much larger than the thick-
ness of these boundary layers, so that they can be neglected. For more details we refer
to [94] and [17].
To summarize, it follows that the perturbations satisfy the Linearized Euler equa-
tions (LEE)
∂
∂t
·
ρ 1 + ∇ (v0 ρ 1 + v1 ρ 0 ) = 0, (2.60a)
³∂ ´
ρ0
∂t
· ·
+ v0 ∇ v1 + ρ 0 (v1 ∇)v0 + ρ 1 (v0 ∇)v0 + ∇ p 1 = 0,· (2.60b)
³∂ ´
∂t
·
+ v0 ∇ s 1 + v1 ∇ s 0 = 0. · (2.60c)
We also include here the linearized versions of the ideal gas law and the thermody-
namic relation:
p 1 = R(ρ 0 T1 + ρ 1 T0 ), (2.61)
cv cp cv ¡ ¢
s1 = p1 − ρ 1 = p 1 − c20 ρ 1 . (2.62)
p0 ρ0 p0
If the mean flow is homentropic (when we have constant temperature T0 ) it follows

from (2.60c) that the entropy perturbations are constant along streamlines, which
amounts to s 1 = 0. Only in that case the commonly used relation p 1 = ρ 1 c20 follows
from (2.62), but we emphasize here that for non-uniform temperature this relation is
not valid. From (2.21) follows an alternative to the energy equation
µ ¶ µ ¶
∂ ∂ ¡ ¢ p1 ρ 1
∂t
· ·
p 1 + v0 ∇ p 1 + v1 ∇ p 0 = c20
∂t
· ·
ρ 1 + v0 ∇ρ 1 + v1 ∇ρ 0 + c20 v0 · ∇ρ 0 −
p0 ρ 0
,
(2.63)
which will be used later in Section 2.4. Yet another alternative to the energy equation
based on (2.23) is
µ ¶
∂
∂t
· · ·
+ v0 ∇ T1 + v1 ∇T0 + T0 (γ − 1)∇ v1 + T1 (γ − 1)∇ v0 = 0, (2.64) ·
which will be used in Section 5.1.
Three types of wave-like solutions to the LEE can be distinguished, namely entropy,
vorticity and acoustic waves, although in general they are coupled. It has been shown
[58, 114] that for uniform mean flow and temperature the three different waves are
decoupled. Pure entropy and vorticity waves are convected with the mean flow. En-
tropy waves carry only density perturbations (no pressure and velocity perturbations),
vorticity waves carry only velocity perturbations, and acoustic waves carry density,
pressure and velocity perturbations. In case of non-uniform mean flow or temperature
however, and in the presence of boundaries, the three waves interact [13, 14], and it
is less clear how to distinguish them. Therefore, in this case we cannot strictly speak
about ‘acoustic’ perturbations, as any disturbance may consist of acoustic, entropic and
vortical components. In this thesis we will often refer to small perturbations as ‘acous-
tics’, even though they may contain entropic and vortical components.
In this thesis we investigate time-harmonic perturbations at a fixed wave frequency,
and assume that turbulence does not play a role. Because we assume that the pertur-
bations are linear it follows that time-harmonic and turbulent perturbations can be
considered independently. Moreover, sound induced by a free turbulent flow is gen-
erally very quiet [31]. In other cases, such as a turbulent flow near the edge of a fan
blade, turbulence is more efficient in radiating sound. However, in this thesis we are
interested in sound propagation; we assume that the sound source is given, and do not
investigate sound source mechanisms.
2.1.5 Myers’ Energy Corollary

It is often useful to evaluate the transport or radiation of acoustical energy, or to use the
principle of conservation of energy as a check for the internal consistency of numerical
results. For this purpose an exact corollary of the energy equation for fluid dynamics
can be formulated for arbitrary steady flows, which is commonly referred to as Myers’
Energy Corollary [79, 81].
Let us assume that all quantities are expanded as a sum of perturbations of in-
creasing order q( x, t; ²) = q 0 ( x) + ² q 1 ( x, t) + ²2 q 2 ( x, t) + · · · in a small parameter ². When
this expansion is substituted in the Euler equations we find the zeroth order equations
that govern the mean flow, the Linearized Euler equations which govern the first order
§ 2.2 Duct modes 23
perturbations, and consecutively higher order systems of equations. In the same way
the zeroth, first, second, and higher order systems of equations can be found for the
energy conservation law. An energy conservation law for the linear perturbations q 1
is necessarily of the order ²2 . It is shown in [81] that it is possible to express this sec-
ond order part of the general energy conservation law exactly in zeroth and first order
quantities only, i.e. the perturbations of second order and higher are not required to be
known.
When we denote the perturbation energy density by E , the energy flux by I , and
the dissipation term by D, Myers’ Energy Corollary can be formulated as
∂E
∂t
·
+ ∇ I = −D (2.65)
where
p21 1 ρ 0 T0 s21
E= +
2ρ 0 c20 2
ρ 0 | v 1 | 2
+ ρ 1 v ·
0 v 1 +
2c p
, (2.66a)
µ ¶
p1
I = (ρ 0 v1 + ρ 1 v0 )
ρ0
·
+ v0 v1 + ρ 0 v0 T 1 s 1 , (2.66b)
D = −ρ 0 v0 · (ω1 × v1 ) − ρ 1 v1 · (ω0 × v0 ) + s 1 (ρ 0 v1 + ρ 1 v0 ) · ∇T0 − s 1 ρ 0 v0 · ∇T1 , (2.66c)
and the vorticity is denoted by ω := ∇ × v. This relation is valid for arbitrary steady
flows, i.e. also for flows with non-uniform mean velocity and temperature. These equa-
tions show that the energy of the linear perturbations is not strictly conserved—a
‘source’-term D is required, which is non-zero for cases with non-zero vorticity and en-
tropy perturbations. Note that for homentropic flows (constant temperature) s 1 = 0, in
which case Myers’ Energy Corollary reduces to the energy relation reported by Morfey
[75] and Goldstein [46]. If the fluid is quiescent (v0 = 0) as well, it is equivalent to
Kirchhoff’s acoustic energy relation [94], the more classical result.
Taking the time average of (2.65) for time-harmonic perturbations yields
·
∇ I = −D . (2.67)
By taking the volume integral of (2.67) over a volume V with boundary ∂V and applying
the divergence theorem we find
Z Z
·
I n d A + D dV = 0. (2.68)
∂V V
This relation will be used to assess the validity and internal consistency of numerical
results.
2.2 Duct modes

A very suitable way to approach the duct acoustic propagation problem is to describe
the (acoustic) perturbation field in terms of duct modes. In this section we briefly
discuss some basic concepts and definitions related to the modal approach, which will
be used in the next sections.
It is useful to investigate linear wave propagation problems with constant coeffi-

cients in the frequency domain, i.e. we look for time-harmonic solutions of the form
³ ´
p 1 ( x, y, z, t) = Re p̌ 1 e−iω t . (2.69)
with fixed radial frequency ω = 2π f . This prevents us from investigating transient

(e.g. switching) behavior and possible instabilities [33], but for our purposes this is not
a restriction. More general time-dependent solutions can be constructed via Fourier
synthesis if desired.
We can furthermore exploit the fact that the geometries of interest show some sym-
metry. The coefficients of the wave equation (which will be discussed in Section 2.4)
and the boundary conditions are independent of the axial coordinate x for the case of
parallel flow in a cylindrical duct with uniform boundary conditions. Since the prob-
lem is invariant in the axial direction we can look for wave-like solutions that remain
self-similar as they travel along the duct; it is easily shown that this amounts to duct
modes of the form P ( y, z) e−iω t+i kx , where P ( y, z) is a transverse mode shape, and k is a
complex-valued axial wavenumber.
The mode shapes and corresponding axial wavenumbers follow as eigensolutions of
a boundary value problem defined on the duct cross section, with suitable boundary
conditions at the duct wall. For flow through an axisymmetric cylindrical duct having
uniform mean velocity and temperature the transverse eigenvalue problem is formu-
lated as Bessel’s equation with suitable boundary conditions. When the temperature
and flow are sheared (non-uniform) and can be characterized by radial flow and tem-
perature profiles, the transverse differential equation becomes the Pridmore-Brown
equation. Hence, we are interested in the eigensolutions of this equation, which can be
seen as a generalized form of Bessel’s equation. The Pridmore-Brown equation will be
derived in Section 2.4.
For uniform or quiescent flow the set of axial wavenumbers is discrete, and the
modes form a basis for a general solution. The general solution may then be expressed
as a summation of modes,
∞
X
p 1 ( x, y, z, t) = A µ Pµ ( y, z) e i kµ x−iω t , (2.70)
µ=−∞
each mode having a certain amplitude A µ such that the series converges. For non-
uniform mean flow there is also continuous part of the spectrum, next to the discrete
set of eigenvalues. In this thesis we assume that this contribution is small, so that it
can be neglected [23, 33], and we will use (2.70) also for cases of non-uniform flow. We
will comment on this again at the end of Section 2.4. Since the problem is linear, the
modes can be determined separately. Once these modes are known, the amplitudes can
be found from the initial conditions at a certain axial position.
If the geometry is circularly cylindrical it is most natural to use cylindrical coordi-
nates ( x, r, θ ). Because of the periodicity in θ it is possible to express the solution as a
Fourier series in θ ,
∞
X ∞
X
p 1 ( x, y, z, t) = A mµ P mµ ( r ) e i k mµ x−iω t+i mθ , (2.71)
m=−∞ µ=−∞
where m is the circumferential (also: azimuthal) wavenumber. This is useful if the

medium and the boundary conditions are independent of θ , i.e. if the problem is axi-
symmetric. In this case the counter µ is referred to as the radial order.
§ 2.3 Wall boundary conditions 25
For hard walls a finite number of wavenumbers k µ is real-valued; the rest is complex-
valued. In the first case the wave is purely propagative, referred to as cut-on, while in
the second case the wave decays or increases exponentially, referred to as cut-off . For
frequencies below the cut-off frequency only the plane wave (for which the mode shape
is constant) exists—all higher order modes are cut-off.
2.3 Wall boundary conditions

In this section we will describe the boundary conditions that are used to model the
sound absorption in the wall. We start by defining the acoustic (normal) impedance,
which measures how much the motion of a fluid particle (or a surface) is impeded when
a pressure wave impinges on it, as
p̌ 1
Z (ω) = , (2.72)
·
v̌1 n
where the normal n is pointing into the wall. The impedance is a frequency-domain
quantity.
As discussed in Chapter 1, the acoustic field in an acoustically absorbing wall (for
example made of foam) and the acoustic field in the duct are coupled; a part of the
sound waves that travel inside the wall material will propagate parallel to the wall. As
a result, the impedance may vary for each position at the wall, so the wall can not be
described by just a single impedance value. These walls are referred to as non-locally
reacting. This is in contrast to locally reacting walls, which have an internal structure
that prohibits this parallel propagation (for example through a honeycomb structure),
and can consequently be described by a single impedance value Z = Z (ω) that depends
on ω only. It is also possible for non-locally reacting walls to find a single impedance
value that describes the entire surface if we consider only a single mode, since modal
solutions retain their shape as they propagate along the wall. Hence, for non-locally
reacting walls the impedance Z = Z (ω, k) depends on both ω and k. This can also be
interpreted as follows: for a non-locally reacting surface the impedance depends on
the angle of incidence of the sound wave, whereas for a locally reacting surface it is
angle-independent.
Due to viscosity, the mean flow velocity at the wall is zero—the flow is not slipping.
This gives rise to a boundary layer. Due to the very low viscosity of air this boundary
layer can be very thin (as is the case for the inlet duct of an aircraft jet engine). In the
limit of vanishing thickness the mean flow can be considered an inviscid slipping flow,
in which case the boundary layer can be modeled as a vortex sheet. This has implica-
tions for the impedance boundary condition. Ingard [52] proposed a boundary condition
for slipping flow, based on the continuity of particle displacement. This was later gen-
eralized by Myers [80] for a smooth but not flat surface. For this reason the commonly
used boundary condition bears their name; the Ingard-Myers boundary condition can
be formulated as
p̌ 1 p̌ 1
· ·
v̌1 n = (−iω + v0 ∇) −
−iω Z −iω Z
· ·
n ( n ∇ v 0 ). (2.73)
This condition is equivalent to continuity of particle displacement for the specific case
of parallel flow along a flat surface. We will first derive the boundary condition for
parallel slipping flow along a straight wall in the next section, as this is the geometry
u = u0 p+ +
1 , v1
S
δ
u=0 p− −
1 , v1
p̌− −
1 = Z v̌1
Figure 2.1: Schematic overview for the derivation of the Ingard boundary condition.
pertinent to this research. This boundary condition is valid for both locally ( Z = Z (ω))
and non-locally reacting ( Z = Z (ω, k)) walls when we interpret ∂ x as i k.
In Section 2.3.2 the propagation of sound through porous material will be discussed.
Porous materials can be characterized into three different skeleton types: rigid (e.g.
metal foam), limp (i.e. soft fibrous material) or elastic. Materials with an elastic skele-
ton are generally described by Biot theory, which governs the fluid density and the
displacement of the skeleton (i.e. four degrees of freedom). In this case both compres-
sion and shear waves may exist. However, to simplify the problem often an equivalent
fluid model [2, 76, 89] is used, in the same way as with rigid and limp materials. Only
the compression waves are considered in an equivalent fluid model, modeled through a
Helmholtz equation depending on two parameters: usually an effective density and an
effective sound speed, or equivalently, a characteristic impedance and a propagation
constant [110]. The various material properties like porosity, resistivity, flow resis-
tance, etc., are modeled through these two parameters. For this purpose many different
expressions exist, on both theoretical and empirical bases, such as the ones proposed
by Delaney and Bazley, Allard and Champoux, and Miki [3, 35, 71, 72].
2.3.1 Boundary condition for parallel slipping flow

In this section we give a derivation of the Ingard-Myers boundary condition for the
simple case of parallel slipping flow along a flat wall. A schematic overview of the
configuration is depicted in Figure 2.1. We use a Cartesian coordinate system ( x, y, z)
and denote the velocity components by u, v, w. For y > δ we have a uniform flow with
velocity u = u 0 , and for y < δ we have u = 0. A vortex sheet S is located at y = δ to
represent the boundary layer. This vortex sheet is perturbed by an incoming sound
wave. We use the superscripts + and − to denote variables above and below S . At the
wall at y = 0 we have the impedance boundary condition p̌− −
1 = Z v̌1 . A vortex sheet can
− +
not support a pressure difference, so we can use p 1 = p 1 . However, due to the fact that
the mean flow velocity is discontinuous (a result of the assumption that it is inviscid),
the perturbation velocity is discontinuous as well: v1− 6= v1+ . To model a slipping flow,
we wish to find an expression for p̌+ +
1 and v̌1 as δ → 0.
Any particle close to the surface S follows the streamlines of the interface, which
can be described by
¡ ¢
x = ξ( t), y = η( t) = δ + s 1 ξ( t), ζ( t), t , z = ζ( t), (2.74)
where s 1 is a small perturbation of the streamline due to the incoming sound wave.
Instead of looking at one particle that happens to be at position x = [ξ( t), η( t), ζ( t)] at
time t we can also consider a stream of particles for all x and t, so that we can take
partial derivatives to space and time. By taking the time derivative of η we then find,
at y = δ + s 1
∂s1 ∂s1 ∂s1
v1± = + u± + w± . (2.75)
∂t ∂x ∂z
Hence we have after linearization, above and below the streamline, at y = δ
µ ¶
∂ ∂ ∂
v1+ = + u0 s1 , v1− = s 1 . (2.76)
∂t ∂x ∂t
Eliminating s 1 yields µ ¶
∂ ∂ ∂
+ u0 v1− = v1+ . (2.77)
∂t ∂x ∂t
Across the surface S we have p+ − − −
1 = p 1 . Together with p̌ 1 = Z v̌1 we find for time-
−i t
harmonic signals ∼ e ω
µ ¶ +
+ ∂ p̌ 1
− iωv̌1 = −iω + u 0 . (2.78)
∂x Z
If s 1 ¿ δ ¿ 1 as δ → 0, i.e. if the perturbed streamline does not cross the liner surface
as we take the limit, then we can use (2.78) as the boundary condition at y = 0 [101].
Upon using the y component of (2.60b) to eliminate the velocity and assuming that Z
is constant in the x-direction we find for modes of the form P e−iω t+i kx the boundary
condition, in Cartesian or cylindrical coordinates
iρ 0 (ω − ku 0 )2 ∂P iρ 0 (ω − ku 0 )2 ∂P
P= , P= at the wall. (2.79)
ωZ ∂y ωZ ∂r
The Ingard-Myers boundary condition was shown by Brambley to be ill-posed [21].

It was shown in [33, 102] that this is the limit of a boundary layer–impedance wall in-
stability that exists for boundary layer thicknesses less than a critical thickness (which
depends on mean flow and wall properties). It was also shown that the critical thick-
nesses found for typical aeronautical applications are much smaller than the prevailing
boundary layer thicknesses. A regularized boundary condition that includes the effect
of a small but finite boundary layer thickness has been proposed, for which these in-
stabilities are absent. Because we will stick to a fixed and real-valued ω we will not
encounter this problem.
2.3.2 Sound propagation in porous material

In this section we summarize the basic equations for sound propagating through a
porous material; for more details see [51, 76]. The basic idea is to derive a wave equa-
tion for the porous material that has an equivalent form as for air. This is called an
equivalent fluid model.
We start with some definitions related to compressibility. For isentropic sound prop-
agation we develop ρ ( p, s) as a Taylor series around the steady state while keeping s
constant: µ ¶ ¯
¯ ∂ρ ¯¯
¯
ρ ( p)¯ = ρ ( p 0 ) + p 1 + O ( p21 ). (2.80)
p0 ∂ p s ¯ p0
When we define the adiabatic compressibility as

µ ¶
1 ∂ρ
βs := , (2.81)
ρ ∂p s
it follows, after neglecting the nonlinear terms, that we can write
ρ 1 = ρ 0 βs ( p 0 ) p 1 . (2.82)
If the sound perturbations are isothermal instead of isentropic we can replace βs with
the isothermal compressibility βT , which is defined as
µ ¶
1 ∂ρ
βT := . (2.83)
ρ ∂p T
Note that for an ideal gas with p = ρ RT we have βs = 1/(γ p) and βT = 1/ p.

Consider a porous material with a solid skeleton (e.g. metal foam) in which there
is no mean flow and a constant temperature. We also assume that the porous material
properties are uniform and isotropic (independent of orientation). Inside the porous
material the local velocity v is hard to define, since it is highly dependent of the local
orientation of the pores. We therefore use the averaged velocity u, which is defined
as the volume of fluid that crosses a unit area (which is perpendicular to the flow) per
unit time. Let us also define the porosity Ω as the fraction of the volume not occupied
by the solid material, usually very close to 1. The density inside the porous material
is increased slightly due to the presence of the solid; it becomes ρ /Ω. Correspondingly,
since the mass flux in air must be equivalent to the mass flux in the porous material,
the velocity u is slightly lower than v; we can use u = Ωv.
By multiplying the linearized mass conservation equation by Ω we find
∂
Ω
∂t
·
ρ 1 + ρ 0 ∇ u 1 = 0. (2.84)
We could use (2.82) to formulate a relation in terms of p 1 and u1 , however, the com-
pressibility β is frequency dependent. For low frequencies the presence of the porous
material causes the temperature to be constant—we need to use the isothermal com-
pressibility βT . For high frequencies there is not enough time for heat exchange, so we
need to use the adiabatic compressibility βs . For intermediate frequencies (typically of
the order of a few kHz) the process is neither isothermal nor adiabatic. Thus, we use a
compressibility β p that is frequency dependent, and we formulate the mass conserva-
tion law in the frequency domain. For time-harmonic solutions ∼ e−iω t it becomes
·
− iωβ p Ω p̌ 1 + ∇ ǔ1 = 0. (2.85)
In the momentum equation, a term σ u1 must be included to model the retardation

of the flow due to friction; the resistivity σ is the pressure drop required to force a unit
flow through the pores. Usually σ/ρ 0 c 0 is between 50 and 500 m−1 . Furthermore, the
effective mass of a fluid increases when it moves through constrictions, and also due to
the fact that some fibers may move with the flow. We therefore use the effective density
ρ p here, which is typically a factor of 1.5 to 5 times larger than the ambient density ρ 0 .
The linearized momentum equation inside the porous material thus reads
∂
ρp u1 + σ u1 + ∇ p 1 = 0. (2.86)
∂t
ρ p can be frequency dependent as well. In the frequency domain the momentum equa-
tion can be written as µ ¶
iσ
− iωρ p 1 + ǔ1 + ∇ p̌ 1 = 0. (2.87)
ρ pω
The linearized mass and momentum conservation equations (2.85) and (2.87) can
be combined into a dissipative Helmholtz equation
∇2 p̌ 1 + ω2 ρ p β p Ω p̌ 1 + iωβ p Ωσ p̌ 1 = 0. (2.88)
The wave speed can thus be defined as c2p := 1/(ρ p β p Ω). If we define an (again fre-
quency dependent) effective density as
µ ¶
iσ
ρ e := ρ p 1 + , (2.89)
ωρ p
we obtain
∇2 p̌ 1 + ω2 ρ e β p Ω p̌ 1 = 0. (2.90)
This motivates the definition of the effective wave speed as
1
c2e := , (2.91)
ρeβpΩ
which is frequency dependent as well.

In literature it is also very common to specify the bulk properties by a propaga-
tion constant and a characteristic impedance. For a plane wave satisfying (2.90) the
propagation constant µ p and characteristic impedance Z c can be computed as
ω
µp = , Zc = ρ e ce. (2.92)
ce
In that case the wave equation (2.90) and the momentum equation (2.87) have the form
∇2 p̌ 1 + µ2p p̌ 1 = 0, (2.93)
−iµ p Z c ǔ1 + ∇ p̌ 1 = 0. (2.94)
1
Note that for air we have ρ p = ρ 0 , Ω = 1, σ = 0, and β p ( p 0 ) = βs ( p 0 ) = γ p0
.
2.3.3 Bulk absorbing liners

We modeled the effect of slipping flow based on an impedance boundary condition at
the wall p̌− −
1 = Zv1 . The next step is to include in the boundary condition the effect of
the sound propagation inside the liner [110].
Consider the schematic overview of the geometry in Figure 2.2: we use superscripts
+ −
, and l to denote variables in the duct, boundary and liner regions. As before,
the boundary and duct regions are separated by a vortex sheet, and the liner and
boundary regions are separated by a facing sheet (for example a perforated plate [48]
covered with a wire mesh) that causes a pressure jump, so we have v+ 6= v− = v l and
p+ = p− 6= p l . The pressure jump across the facing sheet with impedance Z0 is related
to the velocity as
p̌− − p̌ l = Z0 v̌ l at r = d. (2.95)
hard wall r=b ¾

= d + dl
p 1l , v1l liner
u0 = 0
region
facing sheet r=d ¾
boundary
u0 = 0 p− −
1 , v1 δ layer
vortex sheet
(δ → 0)
¾ region
duct
u0 (r) p+ +
1 , v1
region
r=0
Figure 2.2: Schematic overview for derivation of boundary conditions, for bulk absorbing
and Helmholtz resonator array liners.
If we use
p̌ l
Z = Z0 + at r = d (2.96)
v̌ l
we find again p̌− −
1 = Z v̌1 .
Here we consider a circularly cylindrical geometry, so the wave propagation in the
porous material can be described in terms of modes of the form Q ( r ) e i kx+i mθ . It follows
from the wave equation ∇2 p̌ 1 + µ2p p̌ 1 = 0 that Q satisfies Bessel’s equation
µ ¶
d2 Q 1 dQ 2 2 m2
+ + µ p − k − Q=0 (2.97)
dr2 r dr r2
dQ
Taking into account the hard-wall boundary condition dr = 0 at r = b with b := d + d l ,
the solution to (2.97) is:
0
Q = A [ Jm (ζ r )Ym 0
(ζ b) − Ym (ζ r ) Jm (ζ b)], with ζ2 = µ2p − k2 . (2.98)
Consequently, for bulk absorbing liners we finally have a k-dependent impedance of

i µ p Z c N (ζ )
Z ( k) = Z0 + (2.99a)
ζ D (ζ)
with
0 0
N (ζ) = Jm (ζ d )Ym (ζ b) − Ym (ζ d ) Jm (ζ b), (2.99b)
0 0 0 0
D (ζ) = Jm (ζ d )Ym (ζ b) − Ym (ζ d ) Jm (ζ b), b = d + dl . (2.99c)
This expression for Z can be combined with the Ingard-Myers boundary condition
(2.79) to formulate the boundary condition for slipping flow along a bulk-absorbing
liner.
2.3.4 Honeycomb liner

A honeycomb liner can be considered a special case of the more general bulk absorbing
liner discussed above. It is a Helmholtz resonator array: an array of radially oriented
§ 2.4 Pridmore-Brown equation 31
cavities (tubes) covered with a facing sheet and having a cross-wise diameter that is
smaller than the wavelength. This prohibits the propagation of sound in the axial and
circumferential directions, so we can use m = k = 0. When the cavities are air-filled we
can also use µ p = ω/ c 0 , Z c = ρ 0 c 0 , and ζ2 = ω2 / c20 . It then follows that (2.97) reduces to
d2 Q 1 dQ ω2
+ + 2 Q = 0, (2.100)
dr2 r dr c0
with the solution, analogous to (2.99)
J0 (ω d / c 0 )Y00 (ω b/ c 0 ) − Y0 (ω d / c 0 ) J00 (ω b/ c 0 )
Z = Z0 + iρ 0 c 0 . (2.101)
J00 (ω d / c 0 )Y00 (ω b/ c 0 ) − Y00 (ω d / c 0 ) J00 (ω b/ c 0 )
d ωd
If it holds that dl ¿ 1 and c0 l = O(1), then it also holds that ωc0d À 1 and ωc0b À 1. By
using the fact that the Bessel Jm ( z) and Ym ( z) functions approximate cosine and sine
functions for fixed order m and large argument z (see [98], equation 10.7.8) we can find
the commonly used expression [101]
µ ¶
ω
Z = Z0 + iρ 0 c 0 cot dl . (2.102)
c0
2.4 Pridmore-Brown equation

2.4.1 Preform for arbitrary cross-section
In this section we first derive a ‘generalized Helmholtz equation’, which governs modes
(as discussed in Section 2.2) in a duct with arbitrary cross-section carrying parallel
mean flow [46, 108]. This is a more general version of the so-called Pridmore-Brown
equation [97] that includes the effect of temperature gradients [16, 56, 83], which will
be discussed in the next section.
We assume a mean flow with only a single non-zero velocity component in the axial
direction that varies only in the transverse direction. We use transverse coordinates
( y, z) as we first consider a cylindrical duct of arbitrary cross-sectional shape (i.e. not
necessarily circular). With constant mean pressure p 0 we then have v0 = u 0 ( y, z) e x ,
ρ 0 = ρ 0 ( y, z) and c 0 = c 0 ( y, z). With the definition of the convective derivative for paral-
lel flow
µ ¶
D ∂ ∂
:= + u0 , (2.103)
Dt ∂t ∂x
we can then use the following identities
·
∇ v0 = 0, ∇ p 0 = 0, ·
(v0 ∇)v0 = 0, ·
v0 ∇ρ 0 = 0,
∂ v1 D D ∂ v1 (2.104)
· ·
∇ (v1 ∇)v0 =
∂x
· ∇⊥ u0 , ∇ · D t v1 = D t ∇ · v1 + ∂x
· ∇⊥ u0 ,
where ∇⊥ is used to denote a transverse gradient operator in y and z.
It follows from the linearized mass (2.60a), momentum (2.60b) and energy (2.63)
equations that perturbations of a parallel mean flow are governed by
D
Dt
ρ 1 + ρ 0 ∇ v1 + v1 ∇ρ 0 = 0, · · (2.105a)
D 1
Dt ρ0
·
v1 + v1 ∇v0 + ∇ p 1 = 0, (2.105b)
D 1 D
Dt
ρ 1 + v1 ∇ρ 0 = 2
c0 D t
p1 . · (2.105c)
The last equation (2.105c) can be used to eliminate ρ 1 from the first equation (2.105a)
to find
1 D p1
ρ 0 c20 D t
+ ∇ v 1 = 0. ·
(2.106)
When we take the convective derivative of this equation and subtract from it the diver-
gence of the momentum equation (2.105b) we obtain (note that γ p 0 = ρ 0 c20 is constant)
µ ¶
1 D2 p 1 1 ∂ v1
ρ 0 c 0 D t2
2
−∇ · ρ0
∇ p1 − 2
∂x
·
∇⊥ u 0 = 0. (2.107)
To eliminate v1 we substitute the transverse components of the momentum equation

(2.105b) into the convective derivative of (2.107) to arrive at
D3 ∂ ³ ´ D ¡ ¢
Dt 3
p 1 + 2 c20
∂x
∇⊥ u 0 ∇⊥ p 1 −
Dt
·
∇ c20 ∇ p 1 = 0 · (2.108)
We are interested in solutions of the form p 1 ( x, y, z, t) = P ( y, z) e i kx−iω t (as we are

neglecting the continuous part of the spectrum, see the remarks at the end of the next
section) that satisfy a preform of the Pridmore-Brown equation
¡ ¡ ¢¢
·
iΩ3 P + 2i kc20 (∇⊥ u 0 ∇⊥ P ) + iΩ − k2 c20 P + ∇⊥ c20 ∇⊥ P = 0, · (2.109)
where Ω := ω − ku 0 . By noting that − k∇⊥ u 0 = ∇⊥ Ω this equation can be further sim-

plified to
µ 2 ¶ µ ¶
c0 k2 c20
∇⊥
Ω2
·
∇⊥ P + 1 − 2 P = 0.
Ω
(2.110)
If u 0 ≡ 0 this equation reduces to

¡ ¢ ¡ ¢
∇ · c20 ∇P + ω2 − k2 c20 P = 0, (2.111)
which is just equivalent to the Helmholtz equation if c 0 is constant.
2.4.2 Circularly cylindrical geometry

We can look for solutions of the form p 1 ( x, r, θ , t) = P ( r ) e i kx−iω t+i mθ if the duct cross-
section is circular and the temperature and flow speed depend only on the radial coor-
dinate r . By substituting P ( r ) e i mθ for P in (2.110) we find
" #0 " #
Ω2 c20 r 0 Ω2 m2
2
P + − k − 2 P = 0. (2.112)
c20 r Ω2 c20 r
§ 2.5 High-frequency approximations 33
This is the self-adjoint form of

· 0 ¸ · ¸
00 1 T0 2 ku00 0 (ω − ku 0 )2 2 m2
P + + + P + − k − 2 P = 0, (2.113)
r T0 ω − ku 0 γRT0 r
which is referred to as the Pridmore-Brown equation in cylindrical coordinates, here
written out explicitly in terms of the radial temperature and flow profiles.
The Pridmore-Brown equation together with the wall boundary condition (2.79)
(possibly combined with (2.99), (2.101) or (2.102)) and a regularity condition at r = 0
forms a boundary value problem (BVP). To motivate our choice of regularity condition,
we first note that (2.113) is a differential equation whose solutions can be found with
β( r ) γ( r )
the Frobenius method [115], as the equation can be formulated as P 00 + r P 0 + r2 P = 0,
where β( r ) and γ( r ) are assumed to be analytic. At least one solution has the form
P1 ( r ) = a 0 r λ1 + a 1 r λ1 +1 + a 2 r λ1 +2 + · · · , where λ1 = m is one of the two roots of the so-
called indicial equation, which is in our case λ2 − m2 = 0. With λ2 = − m the two roots
differ by an integer, so according to the Frobenius method the other solution is of the
form P2 ( r ) = cP1 ( r ) log( r ) + b 0 r −m + b 1 r −m+1 + b 2 r −m+2 +· · · . This second solution is sin-
gular at r = 0. To ensure that we find the non-singular solution, which behaves as r m
near r = 0, we choose the regularity condition
(
P 0 (0) = 0 if m 6= 1,
(2.114)
P (0) = 0 if m = 1.
We are interested in finding the eigensolutions of the BVP, also referred to as

(eigen)modes. Each mode is a pair of an eigenfunction P ( r ) and its corresponding
eigenvalue k. The eigenfunctions can be determined up to a multiplicative constant;
we assume that Z 6= 0 and hence P ( d ) 6= 0 so we choose the normalization condition
P ( d ) = 1. (2.115)
As mentioned in Section 2.2: the modal approach assumes that the modes form
a complete set, i.e. that all (acoustic) perturbation fields can be described in terms of
these modes, according to (2.71). This is in general not the case for sheared mean flow.
The coefficient of the first order term of the Pridmore-Brown equation is singular when
ω − ku 0 = 0; this happens when the phase speed of the perturbation wave, which is
ω/ k, is equal to the local mean flow velocity. The radial location where this happens
is referred to as the critical layer. By applying a proper forward and inverse Fourier
Transform in the axial direction, it follows that the total perturbation field may consist
of contributions not only from the discrete set of eigensolutions, but also from a contin-
uous part of the spectrum, which is due to the critical layer. However, it can be shown
that this last contribution is small in most cases of interest [23, 33]. In this thesis the
continuous part of the spectrum will be neglected.
2.5 High-frequency approximations

Since non-trivial analytical solutions of the Pridmore-Brown equation (2.113) are un-
known or at least rare it is useful to consider approximations, as these can provide
some analytical insight and can be used for comparison with numerical results. Simi-
lar to the approach described in [120] we can find some approximate solutions for high
frequency ω by using the WKB method.
2.5.1 WKB-type solutions for the radial modes

Let us first make (2.113) dimensionless based on a reference density
p ρ ∞ and temper-
ature T∞ , from which follows the reference sound speed c ∞ = γRT∞ . We use the
scaling
P T0 r u0 ωd
P̃ = , T˜0 = r̃ = , M= , ω̃ = , k̃ = kd, (2.116)
ρ ∞ c2∞ T∞ d c∞ c∞
to find a dimensionless form of the Pridmore-Brown equation (tildes omitted)

· 0 ¸ · ¸
1 T0 2 kM 0 (ω − kM )2 m2
P 00 + + + P0 + − k2 − 2 P = 0 on 0 < r É 1. (2.117)
r T0 ω − kM T0 r
We rewrite this equation as
P 00 + β( r, k)P 0 + γ( r, k)P = 0 (2.118)
and note that β = O(1) and γ = O(ω2 ) as ω → ∞.

If γ = ω2 and β = 0 we have the exact solution P = A + e iωr + A − e−iωr (with some
constants A + and A − ), which is highly oscillatory for large ω. However, the mean flow
and temperature profiles (and consequently β and γ as well) are assumed to vary only
slowly compared to these oscillations. We can therefore assume a similar oscillatory
solution with a slowly varying amplitude and phase of the form
P = A ( r ) e i φ( r ) . (2.119)
Substituting this Ansatz into (2.117) leads to

¡ ¢ ¡ ¢ ¡ ¢
γ − (φ0 )2 A + i 2 A 0 φ0 + A (φ0 )0 + β A φ0 + A 00 + β A 0 = 0, (2.120)
| {z } | {z } | {z }
O (ω 2 ) O (ω ) O(1)
where the terms have been sorted based on order of magnitude. It then follows from
the terms of order ω2 and ω respectively that
p ω − kM
φ0 = ± γ, A=C p , (2.121)
γ1/4 rT0
for some constant C . The expression for the amplitude³ can

´ be found by integrating
d A 1 dγ d rT0
2 A + 2 γ + βd r = 0 and using the fact that β = dr log (ω−kM )2 . The final approximation
can now be expressed as
 
ω − kM ³ Zr p ´ ³ Zr p ´
P (r) ' p  A + exp i γ( s)d s + A − exp −i γ( s)d s  . (2.122)
γ1/4 rT0
Note that the exp(·) terms are oscillatory on the interval where γ( r ) > 0 and expo-
nentially small otherwise. Generally γ is negative near r = 0 when m 6= 0 (γ ≈ − m2 / r 2
yielding the P ∝ r m behavior near r = 0), with a zero at say r 1 . Depending on details
of the velocity and temperature profiles γ may also be negative near r = 1, with a zero
at say r 2 . The zeros r 1 and r 2 of γ are known as turning points [103]. If γ is indeed
negative near r = 1, in which case the solution is exponentially small near the wall,
§ 2.5 High-frequency approximations 35
then the effect of the boundary condition is negligible and the solution is practically
independent of the wall impedance. In that case the wave numbers k can be found
from a quantization condition [44] that requires that an integer number of radial semi-
wavelengths fits between the turning points r 1 and r 2 . On each side of the interval
( r 1 , r 2 ) an extra 14 π is required due to the matching of the oscillatory to the decaying
solution at the turning points [82]. If m 6= 0 (and γ has no more zeros than r 1 and r 2 )
this quantization condition is
Zr 2q
γ( r, k) d r = ( n − 12 )π, n = 1, 2, . . . (2.123)
r1
For m = 0 with only one turning point we have
Zr 2q
γ( r, k) d r = ( n − 14 )π, n = 1, 2, . . . (2.124)
0
2.5.2 Numerical results

Our numerical approach (which is based on the code COLNEW and will be described
in Chapter 3) can be validated for test cases with uniform mean flow and temper-
ature, as analytical solutions of the Pridmore-Brown equation are then available in
terms of Bessel functions. For non-trivial cases of non-uniform mean flow it is much
harder to find suitable test cases. One possible configuration is a strongly non-uniform
(parabolic) mean flow with upstream running modes of sufficiently high frequency. The
waves will refract to the part of the medium with the lowest (effective) sound speed
[108], which is in this case the duct center. The eigenfunction becomes exponentially
small near the wall, which may be challenging for numerical methods like shooting
[119]. However, the WKB solution of Section (2.5.1) is very applicable and will there-
fore be used here for making a comparison.
Consider the case where ω = 25, m = 5, Z = 2 − i, the mean flow temperature T0 = 1,
and the mean flow velocity is M ( r ) = 32 (1− 12 r 2 ). The first 6 upstream running eigenfunc-
tions, decaying exponentially towards the wall, are depicted in Figure 2.3. A compari-
son of the axial wavenumbers found by our numerical approach and the wavenumbers
determined by the quantization condition (2.123) shows an excellent agreement, see
Table 2.1. As there is little or no influence of the impedance wall the wave numbers
are practically real. Only for the higher order modes the damping of the wall becomes
little by little effective as the imaginary parts of the wave numbers become negative.
0 4 4
−1 2 2
Re(P)
Re(P)
Re(P)
−2 0 0
−3 −2 −2
−4 −4 −4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r r r
0 2 2
1 1
−1
Im(P)
Im(P)
Im(P)
0 0
−2
−1 −1
−3 −2 −2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r r r
(a) n = 1 (b) n = 2 (c) n = 3

4 4 4
2 2 2
Re(P)
Re(P)
Re(P)
0 0 0
−2 −2 −2
−4 −4 −4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r r r
2 1 1
1 0.5
0
Im(P)
Im(P)
Im(P)
0 0
−1
−1 −0.5
−2 −2 −1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r r r
(d) n = 4 (e) n = 5 (f) n = 6
Figure 2.3: Eigenfunctions of the first six upstream (left)-running modes in non-uniform
mean flow, refracted away from the duct wall.
n radial WKB numerical

1 -60.470050 -60.4392
2 -55.761477 -55.7281
3 -51.134220 -51.0980 - 0.0000i
4 -46.605341 -46.5659 - 0.0003i
5 -42.195806 -42.1422 - 0.0212i
6 -37.931062 -37.5622 - 0.3254i
Table 2.1: Modal wave numbers k of the first six modes, corresponding to figures 2.3, found
from the WKB quantization condition and by our numerical approach.
Chapter 3
Numerical approach
In this Chapter we describe our numerical approach for finding the eigensolutions of
the Pridmore-Brown equation for acoustic duct modes in parallel flow with tempera-
ture gradients.
One could ask: why would a semi-analytical approach based on modes be advan-
tageous over a fully numerical approach, based for instance on a direct discretization
of the Linearized Euler Equations (LEE) that govern the acoustic field? Computing
power is ever increasing; however, computational costs for complex geometries are still
prohibitively high for design calculations. This is especially the case for the high pre-
vailing frequencies in turbofans, with Helmholtz numbers up to 100 and circumferen-
tial wavenumbers up to 50. State of the art time-domain LEE computations for a 3D
exhaust analysis with a Helmholtz number of 28 take 40 hours on a cluster consisting
of 12 processors [12].
To reduce computational complexity, numerical codes used in aircraft industry are
therefore often based on the assumption that the mean flow and acoustic perturba-
tion field are irrotational. In this way a formulation based on a velocity potential can
be used [10], which amounts to one degree of freedom per grid point, where a full
discretization of the LEE would require four. However, the modeling of swirling or
parallel sheared flow, which is by definition not rotation-free, is then limited, although
some efforts have been made to compute sound propagation in sheared flow based on a
formulation in terms of the total fluid enthalpy (which is a scalar) [64]. The commercial
code ACTRAN/TM is an example of a finite element method based on the Helmholtz
equation for the acoustic velocity potential [43]. In [6] this code has been used to com-
pute sound propagation through a 3D bellmouth intake. The CPU time is about 2 hours
for a Helmholtz number of around 40. It is also noted in [6] that memory requirements
prevent computations for higher Helmholtz numbers.
Our approach is based on modes, which are the eigensolutions of a boundary value
problem (BVP) with a free parameter. We first describe in Section 3.1 how we solve this
BVP numerically. This enables the application of a path-following approach based on a
prediction-correction scheme, which will be described in Section 3.2. Finally, in Section
3.3 we describe how a good set of initial guesses for the duct modes can be obtained
with the aid of a root-finding method based on complex contour-integration.
37
38 Chapter 3. Numerical approach
3.1 Collocation to solve the BVP

The complete boundary value problem (BVP), which consists of the ordinary differen-
tial equation (ODE) (2.117) together with the wall boundary and regularity conditions
(2.79) and (2.114), made dimensionless by using the scaling of (2.116), is summarized
here as
P 00 + β( r, k)P 0 + γ( r, k)P = 0 on 0 < r É 1, (3.1a)
subject to
(
0 P 0 = 0 if m 6= 1
P + κ( r, k)P = 0 at r = 1, at r = 0, (3.1b)
P =0 if m = 1
with
−i (ω − kM )2
0
1 T0 2 kM 0 (ω − kM )2 m2
β= + + , γ= − k2 − 2 , κ= . (3.1c)
r T0 ω − kM T0 r ω ZT0
The impedance of a bulk reacting liner is made dimensionless by using the scaling
b dl
µ̃ p = µ p d, ζ̃ = ζ d, b̃ = = 1+ = 1 + d˜l ,
d d
which results in (tildes omitted)
iµ p Z c N (ζ)
Z ( k) = Z0 + , (3.2a)
ζ D (ζ )
with
0
N (ζ) = Jm (ζ)Ym 0
(ζ b) − Ym (ζ) Jm (ζ b ), ζ2 = µ2p − k2 ,
0 0 0 0
(3.2b)
D (ζ) = Jm (ζ)Ym (ζ b) − Ym (ζ) Jm (ζ b ), b = 1 + dl .
As mentioned before, this is a boundary value problem with a free parameter k,

or in other words: an eigenvalue problem. Even though the Pridmore-Brown equation
and the wall boundary condition are linear in P , the fact that we have a free (unknown)
parameter k makes the eigenvalue problem non-linear. Moreover, for non-locally react-
ing liners the impedance Z depends on k in a non-linear way according to (3.2).
As can be seen from (4.25), in principle it is possible to formulate the problem as
an (generalized) eigenvalue system of differential equations of the form LF = kK F in
terms of the vector F = [P,U, V ,W ]T of separate pressure and velocity components,
where L and K are differential operators in matrix form. The wall boundary condition
(2.78) can be expressed in this form as well. For reasons of computational efficiency we
use the problem formulation in terms of the Pridmore-Brown equation, since it is then
unnecessary to store the U , V and W components.
In case of a hard wall and a mean flow with uniform velocity and temperature,
the classical analytical solutions exist in the form of Bessel functions, and the axial
wavenumbers k are given through the easily found (because real) zeros of the deriva-
tive of the Bessel Jm function. For most other cases solutions have to be found entirely
numerically.
After its first publication [97] it took a long time before numerical solution of this
problem became tractable, due to the fact that the Pridmore-Brown equation is numer-
ically stiff for high frequencies and / or high values of k (i.e. higher order modes). One
§ 3.1 Collocation to solve the BVP 39
of the first works described a shooting approach based on a Runge-Kutta method to

find the two lowest order modes for a two-dimensional hard-walled duct with parallel
shear flow [78]. More recently an efficient way to numerically find (also higher order)
duct modes of an annular lined duct with parallel shear flow was proposed [119], which
is also based on the shooting method.
With the shooting method the problem is considered as an initial value problem;
the Pridmore-Brown equation is integrated several times for a fixed k from r = 0 to
r = 1, in [119] by using an implicit one-step scheme. The value of k for which the
boundary condition at r = 1 is satisfied is then found by using Newton’s method. Many
initial guesses of k are required if several eigensolutions are sought. Note that implicit
methods can be used in the shooting method without requiring the solution of a non-
linear system of equations, since the numerical integration of the ODE, which is linear
in P , is done while the parameter k is kept fixed.
Also global methods based on a finite differences or on spectral collocation have
been used, for example for the study of duct modes in swirling flow [28, 29, 47, 87, 113].
These methods discretize the previously mentioned problem formulation LF = kK F
on a given mesh. Together with equations that represent the boundary conditions
this leads to say N equations that form a generalized eigenvalue problem of the form
Ax = λBx, which can be solved by using standard methods.
While this approach yields N eigensolutions at once, only the smallest eigenvalue
(which corresponds to the value for the free parameter k) can be determined with the
accuracy of the discretization scheme, and the accuracy decreases for larger eigenval-
ues. (As a prototypical example, compare the analytically known eigenvalues of the
2
operator ddx2 with homogeneous Dirichlet boundary conditions to the eigenvalues of
the matrix resulting from a second order central differences discretization, [67, section
9.4.1].) Thus, to obtain a large number of sufficiently accurate eigensolutions a very
fine mesh (large N ) and a high order accurate discretization scheme are required, and
it is difficult to control the accuracy of specific eigenvalues other than the smallest one.
Another potential pitfall with global methods is that in some cases spurious (not
physically meaningful) modes are found due to difficulties related to the (ω − kM )−1
term, which can become singular [47]; however, these solutions can be ignored by rec-
ognizing that they change when the discretization is changed.
In order to control the accuracy of each of the required multiple eigensolutions we
will employ a path-following strategy (which will be described in Section 3.2) combined
with a separate BVP solver. This enables us to set the same error tolerance for each
individual eigensolution. Clearly, there is a host of numerical methods to solve BVPs,
like the above-mentioned shooting method [20, 119], finite elements, and collocation
methods [19]. The advantage of a shooting method is its simplicity, but it is known to
be potentially unstable. This is partly cured by employing multiple shooting. However,
for more complicated problems collocation was shown to perform best in practice.
We will specifically use a code called COLNEW [15], which is also described and
successfully applied to a number of practical problems in [9], and is freely available
from Netlib [49]. This code, which is a modification of the code COLSYS [7, 8], is ca-
pable of solving BVPs formulated as mixed-order systems of ODEs. It is based on
collocation at Gaussian points using a Runge-Kutta monomial basis representation, it
uses a damped Newton solver to solve the system of non-linear equations and it has
automatic meshing. The fact this code uses Gaussian points, which do not include the
boundary points of each mesh subinterval, prevents potential problems with the 1/ r
singularity in r = 0. Because of the availability of local error estimates COLNEW has

facilities to check the solution against user-prescribed tolerances, and to refine or re-
distribute the mesh points. This is one reason for the fact that the code is very efficient.
Finally, we mention that the computed solution can be evaluated at any point inside
the domain, not only at the mesh points.
We put the eigenvalue problem (3.1) in a form suitable for a standard BVP solver
like COLNEW by introducing the extra equation k0 = 0, adding a normalization con-
dition P (1) = 1, and splitting the equations in real and imaginary parts. We use sub-
scripts R and I to denote real and imaginary parts in order to formulate the problem
as
k0R = 0
k0I = 0
00 0 on 0 < r É 1, (3.3a)
PR = −βR P R + β I P I0 − γR P R + γ I P I
P I00 = −β I P R
0
− βR P I0 − γ I P R − γR P I
subject to
0
PR + κR P R − κ I P I = 0 PR = 1
0 and at r = 1,
PR + κ I P R + κR P I = 0 PI = 0
( (3.3b)
0
PR = P I0 = 0 if m 6= 1
at r = 0.
P R = P I = 0 if m = 1
Note that the coefficients βR , β I , γR , γ I , κR , κ I all depend on r , k R and k I , which shows

that the problem is non-linear. By introducing the vector z = [ k R , k I , P R , P R 0
, P I , P I0 ]T
the right-hand side of (3.3a) can be written in the form f ( r, z), and each of the boundary
conditions in the form g i (z( r )) = 0. In order to solve the non-linear system of equations
by using Newton’s method the Jacobian matrices of f and g are required, i.e. the terms
∂fi ∂g
∂z j
and ∂ z i . We compute these terms by using a second order central differences ap-
j
proximation.
3.2 Path-following based on linear extrapolation

As pointed out in the previous section, in order to find the relevant eigensolution it is
important to have a good initial guess. This is particularly important if we want to
make sure to find all (i.e. all physically relevant) modes. This is realized by a path-
following (or continuation) approach, where we start from an “easy” solution (for ex-
ample the analytically known solutions for a hard walled duct with uniform mean flow
velocity and temperature) and trace the solution when the relevant problem parame-
ters are varied until they reach their target values. Essentially, we embed the problem
in a formulation with a continuation parameter λ. Path-following can thus be seen as
an evolution problem with the problem evolving from a known solution for λ = 0 to the
target solution at λ = 1.
Apart from ensuring that we find all relevant eigensolutions there are more ad-
vantages of path-following: first, we can investigate the behavior of a solution as a
function of a parameter; and second, when we are interested in a series of λ-values of
the same continuation, for example for the case of an axially segmented liner (where
each segment has a different impedance), it is possible to save intermediate solutions
when they are encountered along the way, which makes this approach very efficient.
§ 3.2 Path-following based on linear extrapolation 41
numerical solutions to BVP k̃(λ2 ) = k pred

prediction
correction
k(λ1 ) prediction
k(λ2 ) = k corr
k(λ0 ) correction
k-curve that satisfies BVP ǫ = | k pred − k corr |
Figure 3.1: Prediction-correction scheme.
To compute the target solution with mean flow velocity M t ( r ), temperature T t ( r )

and impedance Z t starting from the analytical solution for uniform mean flow with
Mach number M0 , uniform temperature (T0 = 1) and a given impedance Z∞ we use
the embedding
M ( r ) = (1 − λ) M0 + λ M t ( r ),
T0 ( r ) = (1 − λ) + λT t ( r ), (3.4)
Z = (1 − λ) Z∞ + λ Z t .
(Other continuation parameterizations are also possible, but this was found to work
generally well.) The impedance, mean flow velocity and temperature are gradually
changed to their target values in parallel, or one after another.
Path-following has been used before for the case of uniform mean flow and temper-
ature [24], and a path in the impedance plane from hard wall to the target impedance
[42]. We will take into account, however, that not any path is suitable to find all
modes, as has been noted in [100]. It is known that surface modes exist for certain
impedance and mean flow values. Depending on the specific trajectory that is chosen
for the impedance variation, some of these surface modes may appear from or move
towards infinity as the impedance approaches a hard wall value. It was shown [100]
that the surface modes start from a finite value (and do not appear from infinity) if the
impedance is varied (in the right direction) along a specific trajectory in the complex
plane: starting from a nearly hard wall value we follow a straight vertical path (i.e.
with Re( Z t ) fixed) upwards to the target impedance. For our sign convention e−iω t and
mean flow in the positive x direction this ensures that (the axial wavenumbers of) any
possible surface waves start from a finite value and disappear to infinity.
Since each eigensolution is characterized by an eigenfunction and an axial wavenum-
ber k we can trace k(λ) as a curve in the complex plane (some examples are shown in
Figure 3.2 and more will be presented in Chapter 6). To determine the number of in-
termediate solutions and the corresponding values of λ we use a prediction-correction
scheme [4], see Figure 3.1. We use linear extrapolation of two previously computed k-
values for the prediction step. In general, the prediction will not satisfy the BVP. The
prediction is therefore corrected subsequently with the aid of COLNEW.
We trace the eigenvalue k = k(λ) and solution vector u = u(λ) for λ ∈ [0, 1] using
steps h j := λ j − λ j−1 . From two previous solutions k j−2 = k(λ j−2 ) and k j−1 = k(λ j−1 ) we
predict a value for k j by linear extrapolation
k j−1 − k j−2
k̃ j = k j−1 + h j . (3.5)
h j−1
The prediction of u is performed similarly. By substituting λ j−1 and λ j−2 in the Taylor
series expansion of k(λ) around λ = λ j we find expressions for k j−1 and k j−2 that can
be used with (3.5) to find
k̃ j = k j − 21 h j ( h j + h j−1 ) k00j + · · · . (3.6)
Thus the error ² j between the exact and predicted value is

¯ ¯ ¯ ¯
² j = ¯ k j − k̃ j ¯ = ¯ 1 h j ( h j + h j−1 ) k00 + · · · ¯ = ch2 ,
2 j j (3.7)
where k j is assumed to be sufficiently smooth (such that c ∼ k00j is not too large). Given
this j -th step error, we can compute the next step size h j+1 such that the error remains
around some tolerance ²tol = ch2j+1 . This leads to
v
u 2 s
h j+1 u ch j+1 ²tol
h j+1 = h j = h jt = h j . (3.8)
hj ch2j ²j
In some cases, a small change in λ results in a big change in the solution. It is even
possible to jump to the trajectory corresponding to another mode. This is especially
relevant when the minimum distance between the (a-priori unknown) trajectories of
two different modes is very small, in which case we wish to set a strict error tolerance.
On the other hand, if the trajectory of the mode under consideration is far away from
the trajectories of the other modes (as is the case of some surface modes) we can relax
²tol . This motivates the choice
²tol = ²̃tol ∆ k ref , (3.9)
where ∆ k ref depends on the mode considered and ²̃tol is a parameter, equal for all
modes, that has to be chosen.
One option is to base ∆ k ref on the distance between the axial wavenumbers corre-
sponding to a hard wall, uniform flow with Mach number M0 and uniform temperature.
In this case the axial wavenumbers k̂ of circumferential order m and radial order µ are
given by q
−ω M0 ± ω2 − α2mµ (1 − M02 )
k̂ mµ = , (3.10)
1 − M02
with αmµ the µ-th zero j 0mµ of the derivative of the Besselfunction Jm . These zeros can
be approximated by j 0mµ ' (µ + 21 m − 34 )π. Consequently, the difference ∆ k = | k̂ j+1 − k̂ j |
p
between two adjacent cut-off axial wavenumbers is approximately constant π/ 1 − M 2 ,
whereas for cut-on modes ∆ k depends on ω. Another option is to base ∆ k ref on the
distances of a set of a-priori known axial wavenumbers, if available. For example the
set of wavenumbers corresponding to a sound absorbing wall, uniform flow and uniform
temperature can be used. This set can be found by using the approach that will be
described in Section 3.3. Since now any possible surface waves are also known ²tol can
be relaxed on these modes.
§ 3.2 Path-following based on linear extrapolation 43
10 10
5 5
Im(k)
Im(k)
0 0
−5 −5
−10 −10
−20 −15 −10 −5 0 5 10 −20 −15 −10 −5 0 5 10

Re(k) Re(k)
(a) Collapsing trajectories. (b) Jumping from one trajectory to another.
10
5
Im(k)
−5
−10
−20 −15 −10 −5 0 5 10

Re(k)
(c) Trajectories for different embedding.
Figure 3.2: Trajectory of axial wavenumbers when the solution corresponding to uniform
flow and temperature (blue) is slowly varied to the solution for non-uniform flow and tem-
perature while keeping the duct wall hard.
In order to prevent continuing on the wrong trajectory after inadvertently jumping

to another mode we restart from the former solution with a halved step size when the
error is too large, i.e, when ² > ²max . We also add an upper limit h max to the step size.
We finally have:
" s #
∆ k ref ²̃tol
h j+1 = min h j , h max . (3.11)
²j
In practice there is a trade-off in choosing the parameters ²̃tol , ²̃max , h max and the
initial step size h 1 ; we would like to travel through the path quickly, while at the same
time maintaining a certain confidence that we do not jump to another mode.
Unfortunately, it is difficult to devise a method that prevents this jumping with
full confidence, since the trajectories of the modes may become arbitrarily close or they
might even collapse, depending on the specific choice of the problem embedding in the
continuation parameter λ. One such example is depicted in Figure 3.2. In this example
we are interested in the modes corresponding to non-uniform flow and non-uniform
temperature in a hard-walled duct. Figure 3.2a shows the trajectories when, starting
from the solution for a hard wall, uniform flow and uniform temperature (blue), we
follow the path to the target solution (red) while keeping the wall hard. It can be
seen that the trajectories of the lowest order right-running and left-running modes
collapse at the point where they change from cut-off to cut-on. For some choices of the
parameters ²̃tol , ²̃max , h max and h 1 this can result in jumping to the wrong mode, as
depicted in Figure 3.2b. In this case it is useful to choose a different embedding. With
2i
Z = 2− , 0 É λ É 1, (3.12)
4λ(1 − λ)
the impedance Z changes from Z = 2 − i∞ via Z = 2 − 2i back to Z = 2 − i∞ as λ changes

from 0 to 1. This specific path ensures that no surface waves are encountered (see
[100]). At the same time we change the flow and temperature from uniform to non-
uniform, as we did before. With this embedding the collapse of the k-trajectories is
prevented, as shown in Figure 3.2c.
3.3 Finding bulk-absorber modes by using contour

integration
For a duct carrying a uniform flow with uniform temperature the Pridmore-Brown
equation reduces to Bessel’s equation, and consequently the eigensolutions can be
found as the roots of a function involving Bessel functions. For a hard-walled duct
carrying uniform flow at uniform temperature the eigensolutions follow directly from
the roots of the derivative of the Bessel function via the dispersion relation. These roots
are easily found numerically, since they lie at approximately equidistant locations on
the real line and are consequently ordered.
For a soft-walled duct this is not the case; the complex-valued roots of a function
have to be found, which lie in the complex plane (consequently not ordered) at a-priori
unknown distances apart. Thus it is more difficult to ensure that all eigensolutions are
found as there is no way to systematically and efficiently scan the complex plane.
For locally reacting liners one option to find the modes is to start from the easily
found hard-wall modes and gradually vary the impedance to the desired value along a
certain trajectory. If this trajectory is chosen carefully, as discussed previously on page
41, then in principle a path-following strategy starting from the hard-wall modes can
be used to find all modes for a given impedance value.
For non-locally reacting liners, however, we have to resort to other means than
starting from the hard-wall modes. In this case a mode is determined through the
properties of both the duct and liner regions, i.e. the eigenfunction is defined on a
domain consisting of the duct and liner domains. The parts of the mode in the duct
and the annular liner regions are coupled as the two regions are coupled through the
facing sheet impedance. For a hard facing sheet the duct and liner regions are effec-
tively decoupled and separate duct and liner modes exist. In that case the eigenvalue
(i.e. axial wavenumber) of a liner mode is not necessarily also an eigenvalue of a duct
mode. Therefore, the modes that are related to the liner region can not be reached by
a path-following procedure starting from hard-walled duct modes. This motivates the
use of another approach that ensures that all modes, including bulk-absorber modes,
are found directly for given values of the liner properties: root-finding by contour inte-
gration, which is the topic of this section.
§ 3.3 Finding bulk-absorber modes by using contour integration 45
3.3.1 Eigenvalues as the roots of an analytic function

For uniform flow with Mach number M0 and uniform temperature T0 = 1 the Pridmore-
Brown equation (3.1a) reduces to Bessel’s equation
µ ¶
1 0
00 2 m2
P + P + α − 2 P = 0, (3.13)
r r
with the dispersion relation

α2 = (ω − kM0 )2 − k2 . (3.14)
The regular solution normalized such that P (1) = 1 is P ( r ) = Jm (α r )/ Jm (α). Together
with the wall boundary condition (3.1b) this leads to
(ω − kM0 )2 Jm (α)
Z+ 0 =0 (3.15)
iωα Jm (α)
for locally reacting liners, and
iµ p Z c N (ζ) (ω − kM0 )2 Jm (α)

Z0 + + 0 =0 (3.16)
ζ D (ζ) iωα Jm (α)
for bulk absorbers. The eigensolutions that we are after can be found by searching for
k’s that satisfy these equations.
The root-finding approach that will be described in the next section can only be ap-
plied to analytic functions. Hence, we first rewrite the eigenvalue equation of (3.16) in
the form of an analytic function. First note that the Bessel functions have the following
power series representations (see for example [98])
Jm ( z) = a 0 z m + a 1 z m+2 + a 2 z m+4 + · · · (3.17)

2 ³z´
Ym ( z) = b 0 z−m + b 1 z−m+2 + · · · + log Jm ( z). (3.18)
π 2
Thus, we can split the polynomial part from the logarithmic part by writing
µ ¶
2 ζ
Ym (ζ) = f (ζ) + log Jm (ζ), with f ( z) = b 0 z−m + b 1 z−m+2 + · · · , (3.19)
π 2
µ ¶
2 ζb 0
0
Ym (ζ b) = g(ζ) + log Jm (ζ b), with g( z) = c 0 z−m−1 + c 1 z−m+1 + · · · . (3.20)
π 2
It follows that the logarithmic terms in N (ζ) and D (ζ) vanish; we can write
N (ζ) = d 0 ζ−1 + d 1 ζ1 + d 2 ζ3 + · · · , (3.21)

−2 0 2
D (ζ) = e 0 ζ + e1ζ + e2ζ + · · · . (3.22)
The idea is to avoid difficulties related to the multi-valuedness of the square roots in α
and ζ by making sure that we construct a function that has a series representation in
even powers of α and ζ. This can be achieved through multiplying (3.16) by ζ2 D (ζ) and
α−m+1 Jm0
(α). This results in
· µ ¶ ¸
(ω − kM0 )2 Jm (α)
F ( k) = α−m+1 ζ2 D (ζ) Jm
0
(α) Z0 + 0
+ i µ p Z c ζ N (ζ ) J m (α) , (3.23)
iω α
of which all three terms have a series expansion of the form

¡ ¢¡ ¢
F ( k) = f 0 + f 1 α2 + f 2 α4 + · · · fˆ0 + fˆ1 ζ2 + fˆ2 ζ4 + · · · . (3.24)
As a result this function is analytic. Similarly, for locally reacting liners we multiply
(3.15) with α−m+1 Jm0
(α) to obtain
µ ¶
(ω − kM0 )2 Jm (α)
G ( k) = α−m+1 Jm
0
(α) Z0 + . (3.25)
iω α
Note that for m = 0 and Ztot → ∞ (i.e. a hard wall) the plane wave is a valid solution. In
that case (3.15) and (3.16) need to be multiplied by α−m+3 Jm0
(α) instead of α−m+1 Jm
0
(α)
in order to include the solution α = 0.
3.3.2 Finding roots using contour integration

The original method of Delves and Lyness [36] to compute the roots z i of an analytic
function f ( z) inside a contour Γ is based on the following.
For an analytic function f ( z) the number of roots N inside a contour Γ can be com-
puted by using (Cauchy’s) argument principle [1], which can be formulated as
Z 0
1 f ( z)
N= d z. (3.26)
2πi f ( z)
Γ
For the roots z i of f ( z) inside Γ the power sums S n can be computed as
Z N
1 f 0 ( z) X
Sn = zn d z = z ni , (3.27)
2πi f ( z) i
Γ
Note that in a quadrature implementation the function values f 0 ( z)/ f ( z) that are re-
quired to compute N can be reused for S n , n > 0. Moreover, since the integrand is
periodic if the contour Γ is smooth we can use a trapezoidal rule very efficiently, as will
be discussed in the next section.
These power sums S n , which are also sometimes referred to as Newton sums, can
be used to construct a polynomial that has the same roots as f ( z). Let us define this
polynomial as
N
Y N
X
p( z) = ( z − z i ) = ck zk . (3.28)
i =1 k=0
We use a monic polynomial, i.e. c N = 1. The remaining coefficients can then be com-
puted recursively by using Newton’s identities [55], which can be written as
1 NX −k
ck = S j c k+ j , k = N − 1, N − 2, . . . , 0. (3.29)
k − N j=1
Once the coefficients of the polynomial are known, the roots can be computed using a
standard polynomial root finding algorithm, like for example the following one. The
companion matrix to a monic polynomial of N -th degree is the N × N matrix
 
0 0 ··· 0 − c0
1 0 · · · 0 − c1 
 
 
A=  0 1 · · · 0 − c 2 . (3.30)
. . . 
 . . .. .. . 
. . . . . 
0 0 · · · 1 − c N −1
The roots of the polynomial are equivalent to the eigenvalues of the companion matrix.
In this formulation the computation of the power sums requires the evaluation of
the derivative f 0 ( z). In practical application f 0 ( z) may not be available or computation-
ally very expensive to compute. A slightly different approach [34] does not require the
derivative. Firstly, N can be computed by considering the total change of the argument
of f ( z) as z makes one round-trip along the contour Γ:
1
N= [∆ arg f ( z)]Γ . (3.31)
2π
Secondly, the integral of (3.27) can be rewritten as [53]

Z h i
n
Sn = − z n−1 log ( z − z0 )− N f ( z) d z + N z0n , (3.32)
2πi
Γ
where z0 is an arbitrary point inside Γ. Let us write the logarithmic term as log( g( z)) =
log(| g( z)|) + i θ with g( z) = ( z − z0 )− N f ( z). Note that, since g( z) has no zeros or poles, this
logarithmic term is single-valued along the contour Γ: θ returns to its original value
after traversing the complete contour. In a practical computer implementation, where
the available subroutines for computing a logarithm usually return its principal value
(for which the imaginary part is in (−π, π]), care must be taken that the imaginary part
of the integrand in (3.32) is continuous while traversing Γ.
3.3.3 Implementation details

The Bessel functions Jm and Ym grow exponentially for arguments with a large imag-
inary part. This provokes severe numerical cancelation errors in the computation of
(1) (2)
N (ζ) and D (ζ). Hankel functions of first and second kind H m and H m can be used
(1)
instead of the Bessel Ym function in order to prevent this. Using H m = Jm + iYm and
(2)
Hm = Jm − iYm we can rewrite N (ζ) in (3.2b) as
h i
(1) 0 (1) 0
N (ζ) = −i Jm (ζ) H m (ζ b) − H m (ζ) Jm (ζ b) , (3.33)
h i
(2) 0 (2) 0
N (ζ) = +i Jm (ζ) H m (ζ b) − H m (ζ) Jm (ζ b) . (3.34)
(1)
Hm decreases exponentially in the upper half of the complex plane, so the two terms in
N (ζ) are of order one, which avoids the cancelation problem. Similar reasoning holds
(2)
for H m , which decreases exponentially in the lower half of the complex plane. The
computation of D (ζ) can be done analogously. The various types of Bessel functions are
evaluated by using the subroutines developed by Amos [5].
We use circular contours centered at z = z0 with radius r 0 , which can be described
as
Γ : z ( t ) = z 0 + r 0 e 2πi t , t ∈ [0, 1], (3.35)
so (3.32) can be written as
Z1 h i
S n = −n ( z0 + r 0 e 2πi t )n−1 log ( r 0 e 2πi t )− N f ( z0 + r 0 e 2πi t ) r 0 e 2πi t d t + N z0n . (3.36)
0
A suitable quadrature rule must be chosen to evaluate the integral. We choose the
trapezoidal rule since it shows exponential convergence for periodic integrands. De-
pending on specific preliminary knowledge about the locations of the roots other con-
tours than circles might be more suitable, but they must always be smooth to ensure
exponential convergence of quadrature. The m-point trapezoidal rule approximation of
(3.36) can be written as
m µ ¶
n X i
T nm = − g , (3.37)
m i=1 m
where g( t) is the integrand in (3.36). An error estimate is [34]
n ϕmax ³ ³ϕ ´´
max
² = |T nm − S n | ' exp −π cot , (3.38)
N 2π 2
where ϕmax is the maximum jump in the argument of f ( z) for the given m-point dis-
cretization of the contour. When evaluating the logarithmic term in the integrand of
(3.36) it must be ensured that this term’s imaginary part is continuous, as mentioned
above. This amounts to adding or subtracting the required multiples of 2π to the com-
puted principle value of the logarithm.
In order to make the algorithm robust we implemented a number of checks. We
first start the algorithm with m function evaluations to compute N using (3.31), and
compute the maximum jump ϕmax of arg( f ( z)) along the contour. The computed N is
accepted if ϕmax < 43 π and m is not too low. If m is too low for this first test to be
accurate, we perform a second test [26]
¯ ¯
1 ¯ fk ¯
< ¯¯ ¯ < 6.1, k = 0, . . . , m − 1. (3.39)
6.1 f k+1 ¯
We also check whether the accuracy of the quadrature rule for n = N meets a given
tolerance: ² < tol. If one of the criteria is not met we double the number of function
evaluations by adding new function evaluations for the midpoints between the old grid
points.
Since the mapping from the sums S n to the zeros becomes ill-conditioned if the
number of zeros inside the contour is too large, we subdivide the circular contour into
9 smaller circular contours of radius r 1 = 5/12 r 0 . One of the smaller circles is centered
at z0 , the other circles are centered at
z1 = z0 + 0.76 r 0 e2πi j/8 , j = 1, . . . , 8. (3.40)
This subdivision is repeated recursively until the number of zeros inside each con-
tour N < Nmax , a predetermined number for which the polynomial can be reliably con-
structed from the sums S n . Based on our experience we choose Nmax around 5 to 10.
Note that the smaller circles are overlapping, so some zeros might be found multiple
times. To remove these multiple zeros we first group zeros that are located within a
distance d < grouptol from each other and subsequently compute the center z g of this
group. Suppose that the maximum distance from the zeros in the group to the center
is d max . We then apply the foregoing root-finding algorithm on a contour centered at
z g with radius r g = max(1.5 d max , quadtol) to find a new set of roots that replaces the
old roots in the group.
3.3.4 Numerical results

To illustrate the performance of this root-finding approach for bulk absorbing liners
we consider a test case with (all parameters dimensionless) ω = 10, m = 5, uniform
flow with Mach number M0 = 0.3 and uniform temperature T0 = 1. The liner has a
depth of d l = 0.33, and is filled with a porous material for which µ p /ω = 1.3 − 0.4i and
Z c = 1.3 − 0.3i, covered with a facing sheet having an impedance Z0 = 1 + 1i.
Figure 3.3 shows the 209 axial wavenumbers that are obtained. They are equiv-
alent to those obtained with the NLR code BAHAMAS, which uses a root-finding ap-
proach based on the Newton-Raphson method. We required 106688 function evalua-
tions, which is in our experience comparable to a Newton-based method with an excess
number of initial guesses. For this case three different groups of modes can be distin-
guished, namely: the cut-on modes, which are close to the real axis; the cut-off modes
with a real part around 3, which are connected to the duct; and the cut-off modes with
a real part smaller than approximately 3, which are connected to the liner. Compara-
tively more duct modes exist than liner modes.
Figure 3.4 illustrates the behavior of our root-finding strategy. Starting with the
blue circle centered at zero with radius 200, the area is subdivided into smaller circles.
The area is subdivided to a larger extent in those areas that contain the largest number
of zeros.
300
200
100
−100
−200
−300
−15 −10 −5 0 5 10 15
Figure 3.3: Axial wavenumbers for a bulk absorbing liner (209 modes are found using
106688 function evaluations).
250
200
150
100
50
−50
−100
−150
−200
−250
−300 −200 −100 0 100 200 300
Figure 3.4: Illustration of root-finding strategy; circles with many roots are subdivided
into smaller ones.
Chapter 4
Mode-matching for segmented

ducts
In the previous chapter we considered the numerical solution of the boundary value
problem that governs the modes of an axially invariant duct. In many applications,
however, the properties of the duct may suddenly change along the axial direction, for
example the cross-sectional area or the wall impedance. The incoming acoustic wave is
partly reflected and partly transmitted at the discontinuity. In an early work [32] the
scattering of plane waves is studied by applying the equations for conservation of mass,
momentum and energy to a small control volume around a sudden area discontinuity.
For higher frequencies also higher order modes must be taken into account; an in-
coming mode may scatter at a liner discontinuity into multiple outgoing modes in both
directions. In [61] the field is expressed as a summation of modes, and the unknown
modal amplitudes are found by matching the pressure and axial velocity at the axial lo-
cation of the liner discontinuity—this is the mode-matching method, which is the topic
of this chapter.
Mode-matching based on continuity of acoustic pressure and axial velocity has been
used more recently to study a turbofan inlet duct with a wall impedance discontinuity
due to a different liner depth [68]. At supersonic fan speeds the rotor-alone pressure
field is well cut-on. With the aid of mode-matching the liner could be designed such
that most of the acoustic energy of the fan tones is scattered into higher radial order
modes, which are better absorbed by the lining. A slightly different mode-matching
approach for non-axisymmetric configurations has been used to study the effect of cir-
cumferential liner splices [11]. Here, the finite element method was used to find the
eigensolution at a duct cross-section.
Besides mode-matching also the Wiener-Hopf technique has been used to investi-
gate problems with axial discontinuities [57, 74, 86]. However, this technique requires
a considerable amount of complex mathematical machinery and can only be applied to
very canonical geometries, which might explain the fact that its use is not widespread
in engineering contexts.
In this chapter we consider mode-matching for mean flow with non-uniform veloc-
ity and temperature. As will be discussed in Section 4.2, the system of mode-matching
equations results from taking inner products of the duct eigenfunctions with certain
test functions. These eigenfunctions are Bessel functions for a duct with a uniform
51
52 Chapter 4. Mode-matching for segmented ducts
medium. Since closed-form integrals are known for products of Bessel functions [121],
the choice of a suitable set of Bessel functions as test functions allows us to determine
the inner products analytically exactly. For non-uniform flow and temperature, how-
ever, the duct eigenfunctions are not Bessel functions anymore, and no exact integrals
of products of these eigenfunctions are known. For the classical mode-matching ap-
proach a set of Bessel functions may still be used as test functions, although the inner
products cannot (to our knowledge) be computed in closed form, so numerical quadra-
ture is needed. We here present a new approach [91] that uses the same Pridmore-
Brown modes as test functions, but instead of the standard inner product we use an
associated bilinear form1 that resembles an inner product. We will show that in this
way the integrals we need are available in closed form.
The outline of this chapter is as follows. We start in Section 4.1 by presenting
integrals of weighted combinations of products of eigenfunctions that can be evaluated
in closed form. These closed-form integrals, which will also be referred to as a bilinear
maps (BLM), form the basis of a new mode-matching approach which is described in
Section 4.2. First we describe classical mode-matching in Section 4.2.1 followed by
our new approach based on the bilinear map in Section 4.2.2. Mode-matching yields
a system of equations for the modal amplitudes of any two adjacent segments, hence,
to investigate the effect of more than two segments several of these systems have to
be combined. This will be done by using the scattering matrix formalism, which is be
described in Section 4.2.3. Finally, we compare the classical and BLM-based mode-
matching approaches based on some numerical results presented in Section 4.3.
4.1 Exact integrals of Pridmore-Brown eigenfunctions

In order to make the exposition as clear as possible, we start with the prototypical
example of the Helmholtz equation in Section 4.1.1). Next we consider the more general
case of the convected Helmholtz equation for parallel flow that govern modes in a duct
with an arbitrarily shaped cross-section in Section 4.1.2. We then continue in Section
4.1.3 to the circularly cylindrical case (governed by the Pridmore-Brown equation), the
results of which will be used in our new mode-matching approach.
4.1.1 Exact Integrals of Solutions of the Helmholtz Equation

Mode-matching is a particularly successful method for acoustic wave propagation in
segmented ducts of circular or rectangular cross section with a uniform medium. The
necessary integrals of the modal eigenfunctions (Bessel functions or (co)sine functions)
appear to be available in closed form, which greatly simplifies the numerical evalu-
ation. These closed-form integrals are a manifestation of a more general property of
solutions of the Helmholtz equation.
Suppose that we have for parameters α and β the solutions φ and ψ in a region
A ∈ R2 with boundary Γ (boundary conditions do not play a role for now) of
∇2 φ + α2 φ = 0, (4.1a)
2 2
∇ ψ + β ψ = 0. (4.1b)
1 Strictly speaking it is not an inner product. Although imprecise, we will refer to it here as inner product
because of the role it plays in the mode matching procedure.
§ 4.1 Exact integrals of Pridmore-Brown eigenfunctions 53
When we subtract φ times the second equation from ψ times the first and integrate
over A we obtain
Ï Ï Ï
¡ 2 ¢ ¡ ¢
(α2 − β2 ) φψdS = φ∇ ψ − ψ∇2 φ dS = ∇ φ∇ψ − ψ∇φ dS. ·
(4.2)
A A A
After using the divergence theorem this inner product of φ and ψ is given by an integral
along the boundary
Ï Z
1 ¡ ¢
φψdS = 2
α −β 2
φ∇ψ n − ψ∇φ n d`. · · (4.3)
A Γ
If α = β, this result can not be used. Suppose that we replace equation (4.1b) with the
far more general
∇2 χ + α2 χ = f , (4.4)
where f is an arbitrary (integrable) function. When we again cross-wise multiply and
subtract as before, we find
Ï Ï Z
¡ 2 ¢ ¡ ¢
φ f dS = φ∇ χ − χ∇2 φ dS = φ∇χ n − χ∇φ n d`. · (4.5) ·
A A Γ
This result was derived without specifying any boundary conditions on χ, so they
can be chosen arbitrarily as long as there exists a solution χ. This is guaranteed if (the
boundary conditions on χ are such, that) α is not an eigenvalue of the homogeneous
version of equation (4.4). This result is related to the Fredholm alternative for linear
·
operators [59]. Assume that A χ = B∇χ n on Γ and suppose that there exists a nonzero
solution w of ∇2 w + α2 w = 0 withÎ the same R boundary
¡ conditions.
¢ Then we obtain, for
arbitrary f , the contradiction A w f dS = Γ w∇χ n − χ∇w n d` = 0. · ·
The advantage of the result of (4.5) is its generality. We can substitute for f any
function we need, for example f = φ, the solution of (4.1a). The disadvantage is of
course that it requires the solution of the additional inhomogeneous equation (4.4). So
in practice we will use this result only if (4.3) breaks down. Note that the equation
∇2 χ + α2 χ = φ can be found by differentiating the homogeneous equation for φ to −α2 .
In the specific case of a circular disk of radius 1 and circularly symmetric solutions
φ = Jm (α r ) e i mθ and ψ = Jm (β r ) e−i mθ (we choose opposite signs of i mθ for non-trivial
results later) substituted in (4.3), we obtain the well-known [121] relation for Bessel
functions
Z1
1 £ 0 0
¤
Jm (α r ) Jm (β r ) r d r = β Jm (α) Jm (β) − α Jm (α) Jm (β) . (4.6)
α2 − β2
0
For the case when α = β one approach is to take the limit and use l’Hôpital’s rule, which
yields
Z1 µ ¶
1 m2 1 0
Jm (α r )2 r d r = 1 − 2 Jm (α)2 + Jm (α)2 . (4.7)
2 α 2
0
A more generic approach is the one discussed above. Suppose that we have χ( r, θ ) =
χ̂( r ) e−i mθ , regular in r = 0, where χ̂ is a solution of the inhomogeneous Bessel equation
µ ¶
1 m2
( r χ̂0 )0 + α2 − 2 χ̂ = Jm (α r ), (4.8)
r r
0
±
(for example χ̂( r ) = − rJm (α r ) 2α), then we have the equivalent result
Z1
Jm (α r )2 r d r = Jm (α)χ̂0 (1) − α Jm
0
(α)χ̂(1). (4.9)
0
Again, the boundary conditions on χ can be selected arbitrarily, except for the restric-
tion that A χ̂(1) + Bχ̂0 (1) 6= 0 if A Jm (α) + αBJm
0
(α) = 0.
The above manipulations (4.1)–(4.5) can be repeated for (2.111) (with u 0 ≡ 0) to
obtain weighted inner product integrals of the type
Ï
c20 P P̃ dS, (4.10)
A
but for the general case (2.110) this is not possible because Ω = Ω( k). Indeed, no closed
form expressions can be found for the standard inner products with eigenfunctions of
the Pridmore-Brown equation, which are required to set up the mode-matching equa-
tions, so it seems that we have to resort to numerical quadrature to compute the inte-
grals. With increasing radial order the eigenfunctions become more and more oscilla-
tory, resulting in increasingly more difficult numerical computations.
Moreover, note that the integrand is based on a numerical computation, since the
eigenfunctions of the Pridmore-Brown equation have to be determined numerically.
This requires that either the quadrature rule has its abscissas on the same grid points
as the BVP solver, or that the eigenfunctions can be interpolated on any location on or
around the grid points. All this is not the case if we replace the standard inner product
integrals with a dedicated integral associated to the prevailing equations, which we
call (since it is not really an inner product as mentioned before) a bilinear form or
bilinear map.
In the following we will construct two bilinear forms that are integrals of weighted
combinations of products of eigenfunctions. One for the general case of parallel mean
flow (applicable for example in distortion mode problems [111, 112]), much in the same
fashion as discussed above, and one for the particular case of circular ducts with ra-
dially symmetric mean flow, the Pridmore-Brown modes. This result was inspired by
[29], where a related integral was used to obtain a solvability condition for a multiple
scales solution of the disturbance field for a slowly varying duct with mean swirling
flow.
4.1.2 Exact Integrals of Parallel-Flow Modal Eigenfunctions

Analogous to (4.3) we want to construct an integral involving products of Pridmore-
Brown eigenfunctions and use the divergence theorem to evaluate its value through
the eigenfunction values on the boundary. Suppose that we have parallel mean flow in
the x-direction and modes of the form
[ρ 1 , p 1 , v1 ] = [R ( y, z), P ( y, z),U ( y, z) e x + V ( y, z) e y + W ( y, z) e z ] e i kx−iω t . (4.11)
U , V and W are the velocity components in the x, y and z direction of a Cartesian

coordinate system. For modal solutions of this form which are governed by (2.105) we
have
−iΩP + i ρ 0 c20 kU + ρ 0 c20 (Vy + Wz ) = 0, (4.12a)

¡ ¢
−iρ 0 ΩU + ρ 0 u 0 y V + u 0 z W + i kP = 0, (4.12b)
−iρ 0 ΩV + P y = 0, (4.12c)
−iρ 0 ΩW + P z = 0, (4.12d)
where Ω = ω − ku 0 , and R follows directly from the other amplitudes, for example with
2.105c. (Note that the system reduces to (2.110) if U , V and W are eliminated.) To-
gether with suitable boundary conditions this is an eigenvalue problem with eigen-
value k, but this will not be used here; k will be considered as a given constant.
When the individual equations in (4.12) are multiplied by suitable combinations
of other solutions of the same equations (say P̃ , Ũ , Ṽ and W̃ ) with constant k̃ and
corresponding auxiliary function Ω̃ = ω − k̃u 0 , and subsequently added together, we
obtain
¡ ¢ P̃
−iΩP + iρ 0 c20 kU + ρ 0 c20 Vy + ρ 0 c20 Wz
ρ 0 c20
¡ ¢ k̃ P̃
+ −iΩρ 0U + ρ 0 u 0 y V + ρ 0 u 0 z W + i kP
ρ 0 Ω̃
¡ ¢ ¡ ¢
− −iΩρ 0 V + P y Ṽ − −iΩρ 0 W + P z W̃ = 0. (4.13)
This choice of combinations of shape functions is clearly not self-evident; it was found
by first taking the products of the governing equations with arbitrary functions, and
then imposing the required conditions on these functions. The resulting equations
appeared to be equivalent to our original equations for P , U , V and W . For more
details see Appendix A. After reordering and splitting off a cross-wise divergence (4.13)
is equivalent to
Ã !
Ω k k̃ Ω k̃ − Ω̃ k ¡ ¢
−i − P̃P − i P̃U + iρ 0 Ω(Ṽ V + W̃W ) − V P̃ y − W P̃ z + Ṽy + W̃z P
ρ 0 c20 ρ 0 Ω̃ Ω̃
µ ¶ µ ¶
k̃ ¡ ¢ P̃V − Ṽ P P̃W − W̃ P
+ u 0 y Ṽ + u 0 z W̃ P + Ω̃ + Ω̃ = 0. (4.14)
Ω̃ Ω̃ y Ω̃ z
After using the defining equations (4.12) this becomes

Ã !
Ω k k̃ Ω k̃ − Ω̃ k
−i − P̃P − i P̃U + iρ 0 Ω(Ṽ V + W̃W ) − iΩ̃ρ 0 (Ṽ V + W̃W )
ρ 0 c20 ρ 0 Ω̃ Ω̃
Ã ! µ ¶ µ ¶
Ω̃ k̃2 P̃V − Ṽ P P̃W − W̃ P
+i − P̃P + Ω̃ + Ω̃ = 0. (4.15)
ρ 0 c20 ρ 0 Ω̃ Ω̃ y Ω̃ z
Recombining and dividing by Ω̃ yields

"Ã ! #
1 u0 k̃ ω
− i( k − k̃) + P̃P + P̃U − ρ 0 u 0 (Ṽ V + W̃W )
Ω̃ ρ 0 c20 ρ 0 Ω̃ Ω̃
µ ¶ µ ¶
∂ P̃V − Ṽ P ∂ P̃W − W̃ P
= + . (4.16)
∂y Ω̃ ∂z Ω̃
We can use the defining equations for U , V and W in case we want to write the left hand
side in terms of P only. When we integrate (4.16) over a cross section A with bound-
ary Γ and use the divergence theorem we obtain an integral over A of parallel-flow
shape functions, in particular (with suitable boundary conditions and eigenvalues k)
parallel-flow eigenfunctions, expressed
£ as an¤ integral along boundary Γ. We introduce
the vector of shape functions F = P,U, V ,W and denote this integral by
Ï "Ã ! #
1 u0 k̃ ω
〈〈F, F̃ 〉〉 = + P̃P + P̃U − ρ 0 u 0 (Ṽ V + W̃W ) dS
Ω̃ ρ 0 c20 ρ 0 Ω̃ Ω̃
A
Z
i P̃ (V n y + W n z ) − (Ṽ n y + W̃ n z )P
= d` , (4.17)
k − k̃ Ω̃
Γ
where n y and n z denote the y and z components of the outward normal vector on Γ and
k 6= k̃. If a symmetric form is preferred, we can replace 〈〈F, F̃ 〉〉 by 21 〈〈F, F̃ 〉〉 + 12 〈〈F̃, F 〉〉
and the right-hand side correspondingly.
If u 0 ≡ 0 this reduces to a regular integral inner product (with a weight function
1 2
∝ ρ− 0 ∝ c0 )
Ï Z
k + k̃ P̃P 1 P̃ (P y n y + P z n z ) − (P̃ y n y + P̃ z n z )P
〈〈F, F̃ 〉〉 = dS = d`, (4.18)
ω2 ρ0 ( k − k̃)ω2 ρ0
A Γ
a result very similar to (4.3), which could have been obtained directly from (2.111).
For a slipping mean flow along an impedance wall at Γ we apply Ingard-Myers
conditions V n y + W n z = ΩP /ω Z and Ṽ n y + W̃ n z = Ω̃P̃ /ω Z̃ and obtain
Z µ ¶
i P̃P Ω Ω̃
〈〈F, F̃ 〉〉 = − d` . (4.19)
k − k̃ Ω̃ω Z Z̃
Γ
The integrals vanish for hard walls, i.e. when Z = Z̃ = ∞. For a no-slip mean flow
with u 0 = 0 along Γ and impedance boundary conditions P = Z (V n y + W n z ) and P̃ =
Z̃ (Ṽ n y + W̃ n z ) we obtain
Z µ ¶
i P̃P 1 1
〈〈F, F̃ 〉〉 = − d`. (4.20)
k − k̃ ω Z Z̃
Γ
Interestingly, the integral vanishes for different modes corresponding to the same Z .
Although this surface integral resembles a non-standard inner product between
vectors F and F̃ , it is not an inner product because it lacks positive-definiteness for
〈〈F, F 〉〉. We therefore refer to it as a bilinear map, although occasionally it may be
referred to as an inner product since it plays the same role as an inner product in the
classical mode matching procedure.
The above result is evidently not valid for k = k̃. In practice, the limit of k = k̃
corresponds with F = F̃ when we deal with modal eigenfunctions which all satisfy the
same boundary condition, so we will consider that situation here.
We start with the associated inhomogeneous system of (4.12) with solution [P̂, Û, V̂ , Ŵ ],
with the same k as in the original system, and a solution vector [P,U, V ,W ] satisfying
(4.12). This inhomogeneous system is
−iΩP̂ + iρ 0 c20 kÛ + ρ 0 c20 (V̂y + Ŵz ) = i( u 0 P + ρ 0 c20U ), (4.21a)

−iΩρ 0Û + ρ 0 u 0 y V̂ + ρ 0 u 0 z Ŵ + i k P̂ = i(ρ 0 u 0U + P ), (4.21b)
−iΩρ 0 V̂ + P̂ y = iρ 0 u 0 V , (4.21c)
−iΩρ 0 Ŵ + P̂ z = iρ 0 u 0 W, (4.21d)
with boundary conditions such that k is not an eigenvalue of the left-hand side in
order to guarantee the existence of a solution [P̂, Û, V̂ , Ŵ ]. (Note that this system can
be found by differentiating (4.12) with respect to k.) This system is equivalent to an
inhomogeneous variant of (2.110)
µ ¶ µ ¶ µ 2 ¶
c20 k2 c20 c20 c0 u0
∇⊥ · Ω2
∇⊥ P̂ + 1 −
Ω2
P̂ = 2
Ω3
k ω P − 2∇⊥
Ω3
∇⊥·P . (4.22)
We multiply left- and right-hand sides of (4.21) with P /ρ 0 c20 , kP /ρ 0 Ω, −V and −W

respectively, add and do exactly the same manipulations as before. We find that the
factor k − k̃ vanishes and obtain the final result
Ï "Ã ! #
1 u0 k 2 ω 2 2
〈〈F, F 〉〉 = + P + U P − ρ 0 u 0 (V + W ) dS
Ω ρ 0 c20 ρ 0 Ω Ω
A
Z
P̂ (V n y + W n z ) − (V̂ n y + Ŵ n z )P
=i d`. (4.23)
Ω
Γ
If u 0 ≡ 0 this reduces to
Ï Z
2k P2 1 P̂ (P y n y + P z n z ) − (P̂ y n y + P̂ z n z )P
〈〈F, F 〉〉 = dS = 2 d` . (4.24)
ω2 ρ0 ω ρ0
A Γ
4.1.3 Exact Integrals of Radial Pridmore-Brown Modes

A special application of the above results will be for a circularly symmetric mean
flow u 0 ( r ), c 0 ( r ), ρ 0 ( r ) in a circular duct of radius d and cross section A (an annu-
lar duct would require only little changes) with polar coordinate system ( x, r, θ ) and
v1 = u 1 e x + v1 e r + w1 eθ . In this case the solution can be written as a sum over cir-
cumferential Fourier modes, so we can assume modal shape solutions of the form
F ( r ) e i mθ = [P ( r ),U ( r ), V ( r ),W ( r )] e i mθ satisfying
³ 1 im ´
−iΩP + iρ 0 c20 kU + ρ 0 c20 V 0 + V + W = 0, (4.25a)
r r
−iρ 0 ΩU + ρ 0 u00 V + i kP = 0, (4.25b)
0
−iρ 0 ΩV + P = 0, (4.25c)
im
−iρ 0 ΩW + P = 0, (4.25d)
r
where Ω = ω − ku 0 and the 0 denotes an r -derivative. As before we will assume k to
be just a constant, but with suitable boundary conditions this system is an eigenvalue
problem with eigenvalue k.
Due to the symmetry it is no restriction to assume another solution of (4.25) with

k̃ 6= k of the form F̃ ( r ) e−i mθ = [P̃ ( r ), Ũ ( r ), Ṽ ( r ), −W̃ ( r )] e−i mθ , such that the surface inte-
gral in (4.17) divided by 2π simplifies to
Zd "Ã ! # · ¸
r u0 k̃ ω id P̃V − Ṽ P
〈F, F̃ 〉 = + P P̃ + U P̃ − ρ u
0 0 (V Ṽ + W W̃ ) d r = ,
Ω̃ ρ 0 c20 ρ 0 Ω̃ Ω̃ k − k̃ Ω̃ r=d
0
(4.26)
where we assumed that the solutions are regular in r = 0.
If u 0 ≡ 0 we obtain
Zd · ¸
k + k̃ r d P̃P 0 − P̃ 0 P
〈F, F̃ 〉 = P P̃ d r = . (4.27)
ω2 ρ0 ( k − k̃)ω ρ0 r=d
0
With slipping flow and impedance walls along r = d we apply Ingard-Myers boundary
conditions V = ΩP /ω Z with Ω = ω − ku 0 ( d ) for both solutions and obtain
µ ¶
i d P̃P Ω Ω̃
〈F, F̃ 〉 = − , (4.28)
( k − k̃)Ω̃ω Z Z̃
which vanishes if Z = Z̃ = ∞. With no-slip flow u 0 ( d ) = 0 this simplifies to
µ ¶
i d P̃P 1 1
〈F, F̃ 〉 = − . (4.29)
( k − k̃)ω Z Z̃
If k and k̃ are different eigenvalues from the same impedance condition Z̃ = Z , then all
integrals vanish in this case.
To find the degenerate case of k̃ = k and F̃ = F we consider with constant k the
solution F e−i mθ = [P,U, V , −W ] e−i mθ of (4.25), and the associated solution F̂ e−i mθ =
[P̂, Û, V̂ , −Ŵ ] e−i mθ of the inhomogeneous system (with the same k)
µ ¶
1 im
−iΩP̂ + iρ 0 c20 kÛ + ρ 0 c20 V̂ 0 + V̂ + Ŵ = i( u 0 P + ρ 0 c20U ), (4.30a)
r r
−iΩρ 0Û + ρ 0 u00 V̂ + i k P̂ = i(ρ 0 u 0U + P ), (4.30b)
0
−iΩρ 0 V̂ + P̂ = iρ 0 u 0 V , (4.30c)
im
−iΩρ 0 Ŵ +
P̂ = iρ 0 u 0 W. (4.30d)
r
(Note that this system can also be found by differentiating (4.25) with respect to k.)
In actual practice the system (4.30) will be reduced to the following inhomogeneous
Pridmore-Brown equation in P̂
Ã ! Ã !
Ω2 ³ rc20 0 ´0 Ω2 2 m2 ω u00 0 u0 Ω
P̂ + 2 − k − 2 P̂ = 2 2 P − 2 + k P, (4.31)
rc20 Ω2 c0 r Ω c20
which may be solved by almost the same numerical routine as is used for the Pridmore-
Brown equation itself. Note, however, that it is not an eigenvalue problem in this case
since k is fixed, which reduces the required computational effort. The surface integral
(divided by 2π) of (4.23) now simplifies to
Zd "Ã ! # · ¸
r u0 k 2 ω 2 2 P̂V − V̂ P
〈F, F 〉 = + P + U P − ρ u
0 0 (V + W ) d r = i d ,
Ω ρ 0 c20 ρ 0 Ω Ω Ω r=d
0
(4.32)
§ 4.2 Mode-matching at an interface 59
a+
l −1
a+
l
b+
l
a+
l +1
a+
l +2
a−
l −1
a−
l b−
l +1
a−
l +1
a−
l +2
xl −1 xl xl +1
Figure 4.1: Mode-matching at several interfaces.
where we assumed that the solutions are regular in r = 0. As before, it should be noted
that the inhomogeneous equation (4.31) has no solutions if the problem for P is an
eigenvalue problem with homogeneous boundary conditions, and the same conditions
are applied to P̂ . Finally, if u 0 ≡ 0 we obtain
Zd · ¸
2k r 2 d P̂P 0 − P̂ 0 P
〈F, F 〉 = P dr = 2 . (4.33)
ω2 ρ0 ω ρ0 r=d
0
4.2 Mode-matching at an interface

In the following we consider a circularly cylindrical duct that is divided in N axial
segments, and assume that the wall properties are constant within each segment. We
furthermore assume that the acoustic field in each segment can be expressed as a
summation of eigenmodes of the Pridmore-Brown equation according to (2.71), so we
ignore the contribution of the critical layer, as discussed at the end of Section 2.4.
We also note that the issue of possible instabilities due to the interaction of shear
layer and impedance wall (as for example discussed in [23, 102]) will not be addressed
here. Ill-posedness problems associated with a vanishing boundary layer only occur
in time domain calculations, while the detection of possible unstable modes requires a
causality analysis that we have not undertaken in the present context.
4.2.1 Construction of Matrix Equations (Classical Mode-Matching)

In this section we describe the classical mode-matching (CMM) approach based on
continuity of pressure and axial velocity. The total field for a given circumferential
wavenumber m in each segment is a superposition of all right and left-running modes:
∞ ³
X ´
i k+ ( x− xl −1 ) i k− ( x− x l )
p l ( x, r ) = a+l,µ P l,+µ ( r ) e l,µ + a− −
l,µ P l,µ ( r )
e l,µ , xl −1 É x É xl , (4.34)
µ=1
In a numerical implementation this infinite series has to be truncated; the finite num-
ber of modes µl to represent the field of the l -th segment depends on the type of liner.
We use µmax modes to represent the field for segments with a locally reacting liner. For
bulk-reacting liners we use µl = µmax + µadd
l
modes, where the number of extra modes
µadd
l
depends on the depth of the liner (it can be different for different segments). In
the following, we consider the general case of an interface between two segments with
a bulk reacting liner (for locally reacting liners µadd

l
= 0). At the interface at x = xl we
have for the pressure in segment l
µl ³
X ´
p l (r) = b+ + − −
l,µ P l,µ ( r ) + a l,µ P l,µ ( r ) , (4.35a)
µ=1
µl ³
X ´
q l (r) = b+ + − −
l,µ Q l,µ ( r ) + a l,µ Q l,µ ( r ) , (4.35b)
µ=1
where q and Q are used for the pressure field in the bulk region.
Consider the hard-wall uniform flow eigenfunctions Ψν ( r ) = Jm (αν r ) where αν are
the hard-wall radial wavenumbers, which satisfy Ψν0 ( d ) = 0. These functions form a
complete L 2 -basis and are at least locally, for high orders, similar in behavior as the
Pridmore-Brown modes. Therefore they are suitable to serve as test functions when
we set up the matrix system for the modal vectors.
Inside the duct region we impose continuity of pressure and axial velocity (approx-
imated due to the truncation) at an interface at xl and subsequently project onto the
set of test functions Ψν , ν = 1, . . . , νmax , which yields
µl
X µX
l +1
b+ + − −
l,µ (P l,µ , Ψν ) + a l,µ (P l,µ , Ψν ) = a+l +1,µ (P l++1,µ , Ψν ) + b− −
l +1,µ (P l +1,µ , Ψν ), (4.36a)
µ=1 µ=1
µl
X µX
l +1
b+ + − −
l,µ (U l,µ , Ψν ) + a l,µ (U l,µ , Ψν ) = a+l +1,µ (Ul++1,µ , Ψν ) + b− −
l +1,µ (U l +1,µ , Ψν ), (4.36b)
µ=1 µ=1
where we use the standard inner product

Zd
( f , g) = f ( r ) g( r ) r d r. (4.37)
0
The axial velocity amplitudes follow from the pressure amplitudes via the relation
k u00
U= P− P 0, (4.38)
ρ0Ω ρ 0 Ω2
which follows from (4.25b) and (4.25c).
For the annular liner region we use (similarly to the duct region) hard-walled duct
eigenfunctions as test functions, of the form (2.98). Let
0 0
Φl,ν ( r ) = Jm (η l,ν r )Ym (η l,ν ( d + d l )) − Ym (η l,ν r ) Jm (η l,ν ( d + d l )), (4.39)
where η l,ν are the hard wall radial wave numbers, which satisfy Φ0l,ν ( d ) = Φ0l,ν ( d + d l ) = 0
(note the l -dependency due to the fact that the depth d l of the bulk absorber can be
different for each segment). Furthermore, for the axial velocity in the bulk region it
follows from (2.94) that
k
U bulk = Q. (4.40)
µp Zc
Consequently, imposing vanishing axial velocity at at the left side of the liner wall at
x = xl and subsequently projecting onto the set of hard-wall bulk eigenfunctions Φl,ν ,
ν = 1, . . . , νadd
l
yields
µl
X ¡ + + ¢ ¡ − − ¢
b+ −
l,µ k l,µ Q l,µ , Φ l,ν + a l,µ k l,µ Q l,µ , Φ l,ν = 0. (4.41)
µ=1
We again use the standard inner product (4.37) here, except the domain of integration
is now the bulk region d É r É d + d l .
All matching conditions for the duct and bulk regions together yield the following
system of equations:
 +   + 
A A− B B−
· ¸ · ¸
C+ C−  b+  + D−  a+
  l = D  l +1 . (4.42)
E+ E−  a−  0 0  b−
l l +1
+ −
0 0 F F
The matrix entries corresponding to the duct region are inner products of Pridmore-
Brown eigenfunctions and Bessel functions; for the matrices that correspond to the
pressure equations we have
Zd Zd
A±
νµ = (P l,±µ , Ψν ) = P l,±µ ( r )Ψν ( r ) r d r, B±
νµ = (P l±+1,µ , Ψν ) = P l±+1,µ ( r )Ψν ( r ) r d r.
0 0
(4.43)
The matrix entries of C± and D± corresponding to the axial velocity equations are
computed analogously. The matrix entries E± corresponding to the liner region can be
computed in closed form by using the well-known [121] relation
Z
r £ ¤
Cm (αr )C˜m (βr )r dr = 2 2
βCm (α r )C˜m
0
(β r ) − αC˜m (β r )Cm
0
(α r ) , (4.44)
α −β
which holds
± for two different solutions Cm (α r ) and C˜m (β r ) of the Bessel equation. The
term d (µ p Z c ) cancels out so we find (superscripts ± and subscripts l omitted)
1
E νµ = k µ Q µ0 ( d )Φν ( d ). (4.45)
ζ2µ − η2ν
From the boundary condition P = ZV bulk at r = d (see Section 2.3.3) and the momentum
equation for the liner region (2.94) we find
iµ p Z c
Q µ0 ( d ) = i µ p Z c N (ζ µ )
P µ ( d ). (4.46)
Z0 + ζ µ D (ζ µ )
We can choose the normalization of Φ as Φ( d ) = 1 so we finally have

1 iµ p Z c
E νµ = k µ Pµ ( d ). (4.47)
ζ2µ − η2ν Z0 +
i µ p Z c N (ζ µ )
ζ µ D (ζ µ )
The matrix entries of F± corresponding to the liner segment to the right side of the
interface can be computed analogously.
We consider the amplitudes of the modes propagating towards the interface as the
the unknowns. Rearranging leads to
µl +1 µl µl µl +1
 +   
νmax B −A− νmax A+ −B−
 D+ −  · a+ ¸
 C+ −  · b+ ¸
νmax
 −C  l +1 νmax
 −D  l
 0 −  =  E+ . (4.48)
νadd
l
−E al−
νadd
l
0  b−
l +1
νadd
l +1
F+ 0 νadd
l +1
0 −F −
where the dimensions of the submatrices have been included. Thus, we have 2νmax +
νadd
l
+ νadd
l +1
equations and µl + µl +1 = 2µmax + µadd
l
+ µadd
l +1
unknowns.
The solution to the matching problem at a liner discontinuity is not unique. This
is due to the fact that at the discontinuity the boundary condition is not defined. This
problem is very similar to the problem of determining the electromagnetic field around
a sharp edge [18, 40, 70]. The solution can be made unique by imposing the so-called
edge condition: the requirement that the stored energy in the field around the edge is
finite. For the case of a duct with uniform cross-section and a locally reacting liner it
can be shown that this edge condition is automatically satisfied if the number of modes
on both sides of the matching interface is equal. Ducts with non-locally reacting liners
that have a discontinuity in liner depth can be viewed as ducts with a discontinuity
in cross-sectional area. For these cases the ratio of the total radii on both sides of the
interface must be chosen equal to the ratio of the number of modes in order to satisfy
the edge condition [66, 73, 88, 110]. For non-locally reacting liners with a discontinuity
in depth µadd (and νadd ) could be chosen based on a similar criterion.
Finally we note that in [11] the matching conditions are formulated as a varia-
tional statement based on the time-harmonic linearized continuity and momentum
equations for parallel flow and uniform temperature. For zero mean flow these condi-
tions are equivalent to continuity of pressure and axial velocity at the interface. For
non-zero mean flow, however, these conditions include an extra term: an integral over
the boundary of the cross-section, which is due to the Ingard-Myers boundary condition
for slipping flow.
4.2.2 Matching Conditions Based on the Bilinear Map

In this section we set up a system of equations which has the same structure as (4.42)
for the classical mode-matching approach, but now we base the matching conditions on
the bilinear map that was described in Section 4.1.
Let us define the vector f whose components are the acoustic pressure and the
velocity components as
£ ¤
f l ( x, r ) = p l ( x, r ), u l ( x, r ), vl ( x, r ), wl ( x, r ) . (4.49)
The total field for a given circumferential wavenumber m in each segment is a super-
position of all modes
∞ ³
X ´
i k+ ( x− xl −1 ) i k− ( x− x l )
f l ( x, r ) = a+l,µ F +
l,µ ( r )
e l,µ + a− −
l,µ F l,µ ( r )
e l,µ , xl −1 É x É xl , (4.50)
µ=1
where F again denotes the vector of perturbation amplitudes. At the interface at x = xl

we have
µl ³
X ´
f l (r) = b+ F +
l,µ l,µ ( r ) + a −
F −
l,µ l,µ ( r ) . (4.51)
µ=1
Inside the duct we impose continuity of p( x, r ), u( x, r ), v( x, r ) and w( x, r ) by applying

the bilinear form to f l = f l +1 with the solution of the associated problem Ψν , which
results in
µl
X µX
l +1
b+ + − −
l,µ 〈F l,µ , Ψν 〉 + a l,µ 〈F l,µ , Ψν 〉 = a+l +1,µ 〈F + − −
l +1,µ , Ψν 〉 + b l +1,µ 〈F l +1,µ , Ψν 〉, (4.52)
µ=1 µ=1
for ν = −νmax , . . . , −1, 1, . . . , νmax . When we split the range of ν into left (ν < 0) and right
(ν > 0) running parts we again obtain in matrix format
· + ¸· ¸ · + ¸· ¸
A A− b+l = B B− a+ l +1 . (4.53)
C+ C− a− l
D+ D− b − l +1
In order to prevent unnecessary computation of an extra set of eigenfunctions we can

choose as test functions Ψν the eigensolutions of, say, segment n. In that case the
matrix entries can be computed as
{ A, C }±
νµ = 〈F l µ , F nν 〉, (4.54a)
{B, D }±
νµ = 〈F l +1,µ , F nν 〉, (4.54b)
where
{ A, B}+ : µ > 0, ν > 0, {C, D }+ : µ > 0, ν < 0,

{ A, B}− : µ < 0, ν > 0, {C, D }− : µ < 0, ν < 0.
The values of the bilinear forms in (4.54) can be computed as follows. Suppose that
we know the set of eigensolutions for two segments l and n, with possibly different liner
properties. When Z l 6= Z n then the sets of axial wavenumbers have in general (except
for a rare coincidence) no values in common, so it holds that k l µ 6= k nν . Consequently
we can use (4.28) to compute
· ¸¯
i dP l µ P nν Ω l µ Ω nν ¯
〈 F l µ , F nν 〉 = − ¯ , for Z l 6= Z n . (4.55)
Ω n ν ω( k l µ − k n ν ) Z l µ Z n ν ¯ r = d
For the case when Z := Z l = Z n we can have µ 6= ν, which means that k l µ 6= k nν , so we
can compute
¯
i dP l µ P nν u 0 ¯
〈 F l µ , F nν 〉 = − ¯ , for Z l = Z n , µ 6= ν, (4.56)
Ω nν ω Z ¯ r = d
which is identically zero for non-slipping flow ( u 0 ( d ) = 0) or a hard wall ( Z → ∞). When
µ = ν we have k µ := k l µ = k nν and Pµ := P l µ = P nν . We require the solution P̂µ of the
inhomogeneous Pridmore-Brown equation (4.31) to compute
· ¸
d 0 u0 0 0
〈 F l µ , F nν 〉 = P̂ P
µ µ − P P µ − P̂ P
µ µ , for Z l = Z n , µ = ν. (4.57)
ρ 0 Ω2µ Ωµ µ r=d
Note that this implies that A+ and C− are diagonal matrices, and A− and C+ zero
matrices for non-slipping flow or hard wall.
The use of this new inner product for mode-matching thus may require the com-
putation of an extra set of Pridmore-Brown modes to be used as test functions, or the
solution of an inhomogeneous Pridmore-Brown equation. Any of these solutions have
to be computed only once, regardless of the number of segments, whereas for the clas-
sical approach we need to compute all inner products at each interface. Furthermore,
in some occasions the off-diagonal inner products are zero, which simplifies the cal-
culations even more. Since the computational work to determine the extra set of (in
some cases inhomogeneous) Pridmore-Brown modes are, to our experience, at worst
comparable to the numerical quadrature for a single interface while the inherent nu-
merical integration errors are avoided, we conclude that our new approach is both more
accurate and cheaper than the classical mode matching methods.
a+
1 a+
2 a+
3 a+
l−1 a+
l a+
l+1
a−
1 a−
2 a−
3 a−
l−1 a−
l a−
l+1
x0 = 0 x1 x2 x3 xl−2 xl−1 xl xl+1 xN = L

S1 S2 S3 Sl−2 Sl−1 Sl Sl+1
| {z }
S̄1 = S1 segment segment segment
| {z } l−1 l l+1
S̄2 = S̄1 ∗ S2 = S1 ∗ S2
interface interface
| {z }
l−1 l
S̄3 = S̄2 ∗ S3 = S1 ∗ S2 ∗ S3
| {z }
S̄l = S̄l−1 ∗ Sl
Figure 4.2: Schematic representation of the S-matrix algorithm.
4.2.3 Scattering Matrix Formalism

The effects of multiple reflections and transmissions have to be combined when liner
discontinuities exist at more than one axial location. A naive coupling of the duct
sections via the transmission and reflection matrices is possible, but this process is
unstable in case of a large numbers of sections due to the exponentially decaying and
increasing cut-off modes that are involved. In other words: the matrix that describes
the combined effect of all individual segments is ill-conditioned.
An alternative approach would be an iterative one [61], where the propagation of a
mode is only considered in the direction in which it decays. With this approach, starting
from the amplitudes of the incident modes, the modal amplitudes in the intermediate
sections are updated as more and more reflections and transmissions at different inter-
faces are taken into account at each new iteration, until the change in the amplitudes
is below a certain threshold. However, this procedure may not converge for geometries
with a large number of segments. We therefore use the scattering matrix formalism
(see for example [65, 96]), which has no convergence issues and is numerically stable.
We want to express the modal amplitude vectors of the outgoing waves in terms of
the amplitude vectors of the incident waves. We therefore write
M− 1
1 M2
z }| {z }| {
  −1  
B+ −A− A+ −B− µl µl +1
· ¸ −  · b+ ¸ · 11 12 ¸· ¸
a+  +
l +1 = D
−  +
−C  C −D  l µl +1 Ŝ Ŝ b+l
= 21 , (4.58)
al−  0 −E−  E+ 0  b− l +1 µ l Ŝ Ŝ22 b−
l +1
+ −
F 0 0 −F
where we introduced the interface scattering matrix Ŝ, including the sizes of the sub-
matrices. Next, we want to combine the effect of scattering at the interface and propa-
gation through the segment. We therefore introduce the segment scattering matrix Sl
according to
· ¸ · 11 ¸· ¸· ¸ · 11 ¸· ¸
a+
l +1 =
Ŝl Ŝ12
l
X+l
0 a+l
Sl S12
l
a+l
= 21 . (4.59)
a−l Ŝ21
l
Ŝ22
l
0 X−
l +1
a−
l +1
Sl S22
l
a−
l +1
where the propagation is accounted for by the diagonal matrices
i k+ h −i k− h
(X+
l )µµ =
e l,µ l , (X−
l +1 )µµ =
e l +1,µ l +1 , h l = xl − xl −1 . (4.60)
Note that the exponentials are always decaying.

The effect of all layers up to layer l can be combined in the cumulative scattering
matrix S̄l (see also Figure 4.2):
· ¸ · 11 ¸· ¸
a+
l +1 =
S̄l S̄12
l
a+1 . (4.61)
a1−
S̄21
l
S̄22
l
−
al +1
The cumulative scattering matrix of a certain set of segments can be computed by using
the Redheffer star product, which is defined as
· 11 ¸ · 11 ¸
A A12 B B12
∗ 21
A21 A22 B B22
· ¸
B11 (I − A12 B21 )−1 A11 B12 + B11 A12 (I − B21 A12 )−1 B22
= 21 . (4.62)
A + A22 B21 (I − A12 B21 )−1 A11 A22 (I − B21 A12 )−1 B22
By using this definition S̄l can be computed as
S̄l = S̄l −1 ∗ Sl = S1 ∗ · · · ∗ Sl . (4.63)
The Redheffer star product can be constructed as as follows. We would like to construct
the cumulative scattering matrix S̄l in order to compute the effect of segment l . Let us
assume that we can describe the effect of all segments up to segment l − 1 via
· ¸ · 11 ¸· ¸
a+
l =
S̄l −1 S̄12
l −1
a+
1 .
21 (4.64)
a−
1 S̄ l −1
S̄22
l −1
a−
l
Furthermore, we have at interface l

· + ¸ · 11 ¸· ¸
al +1 Sl S12
l
a+l
= 21 . (4.65)
a−l
Sl S22
l
−
al +1
Substitute the second row of (4.65) into the first row of (4.64) to obtain
11 + 12 − 11 + 12
£ 21 + 22 −
¤
a+
l = S̄ l −1 a1 + S̄ l −1 a l = S̄ l −1 a1 + S̄ l −1 S l a l + S l a l +1 . (4.66)
Collecting the terms gives

£ ¤ +
I − S̄12 21 11 + 12 22 −
l −1 S l a l = S̄ l −1 a1 + S̄ l −1 S l a l +1 , (4.67)
so
£ 12
¤
21 −1 11 +
£ ¤
21 −1 12
a+
l = I − S̄ l −1 S l S̄l −1 a1 + I − S̄12
l −1 S l S̄l −1 S22 −
l a l +1 . (4.68)
Substituting this into the first row of (4.65) yields

£ ¤ n £ ¤ o
11 12 21 −1 11 11 12 21 −1 12 22 12
a+
l +1 = S I − S̄ S
l −1 l S̄ a
l −1 1
+
+ S I − S̄ S
l −1 l S̄ S
l −1 l + S a−
l +1 . (4.69)
|l {z } |
l
{z
l
}
S̄11
l S̄12
l
Analogously, we can substitute the first row of (4.64) into the second row of (4.65) to
obtain
£ ¤ −
I − S21 12 21 11 + 22 −
l S̄ l −1 a l = S l S̄ l −1 a1 + S l a l +1 , (4.70)
from which follows
£ ¤
21 12 −1 21 11 +
£ ¤
12 −1 22 −
a−
l = I − S l S̄ l −1 Sl S̄l −1 a1 + I − S21
l S̄ l −1 Sl al +1 . (4.71)
Substituting this into the second row of (4.64) yields

n £ ¤ o £ ¤
21 22 21 12 −1 21 11 22 21 12 −1 22 −
a−1 = S̄ l −1 + S̄ l −1 I − S l S̄ l −1 S l S̄ +
l −1 a1 + S̄ l −1 I − S l S̄ l −1 Sl al +1 . (4.72)
| {z } | {z }
S̄1l S̄12
l
These equations can still be used when the number of modes is different for each
segment (i.e. when the four blocks of the scattering matrices are not square). To see
this, consider the total number of amplitudes for the three segments numbered 1, l and
l + 1, which is 2µ1 + 2µl + 2µl +1 . Of these amplitudes the µ1 + µl +1 amplitudes of the
incident waves in segments 1 and l + 1 are known. Hence, the other amplitudes can be
determined by using the µ1 + 2µl + µl +1 equations of (4.64) and (4.65). Also note that
the dimensions of the terms between square brackets in (4.69) and (4.72) are µl × µl , so
only the solution of square systems is required.
Consequently, if all of the segment scattering matrices and the incoming ampli-
tudes a+ −
1 and a N of the outer segments are known, then the outgoing amplitudes and
hence the total field in the outer segments can be computed. To compute the field in-
side the entire duct the amplitudes in intermediate segments are required as well. We
compute these by using (4.66) and the second row of (4.65). Note that we did not invert
the propagation matrices Xl , which would have caused growing exponentials (which
might provoke numerical problems).
4.3 Numerical results

In order to compare the results of the classical (CMM) and the bilinear-map-based
(BLM) mode-matching approaches we consider the test cases which are listed in Table
4.1. The 1m long duct with radius 0.15m is split into two segments at x = 0.5m (except
for configuration IV); the left segment has a hard wall, and the right hand side segment
has a locally reacting impedance wall. The incident field consists of one right-running
mode, either µ = 1 or 2 for the pertinent configurations2 . For the BLM-based results
the modes of the left (hard-wall) segment are used as test functions.
The results in this section are made dimensionless by scaling on the duct radius d , a
reference density ρ ∞ and a reference ptemperature T∞ . This implies that velocities are
scaled on reference sound speed c ∞ = γRT∞ and time on d / c ∞ . The non-dimensional
mean flow axial velocity is denoted by M (the Mach number), i.e. u 0 = c ∞ M .
Configuration II has a slipping flow, so it is necessary to use the Ingard-Myers
boundary condition here; this is in contrast to configurations I and III, which have
non-slipping flow. The mean flow profile for configuration II has the same mass flow
2 For non-uniform flow and/or temperature the numbering of the modes is not unambiguous; we number
them according their similarity in eigenfunction shape and axial wavenumber value to the uniform flow
modes.
§ 4.3 Numerical results 67
Configuration I II III IV
Helmholtz & azi. ω = 13.86, m = 5 ω = 8.86, m = 5 ω = 15, m = 5 ω = 15, m = 5
2
Temperature T =1 T =1 T = 2 log(2)(1 − r2 ) T =1
2
Mean flow M = 0.5(1 − r 2 ) M = 0.3 · 34 (1 − r2 ) M = 0.3 · tanh(10(1 − r )) M = 0.3 · tanh(10(1 − r ))
Soft-wall impedance Z = 1 − 1i Z = 1 + 1i Z = 1 − 1i Z = 1 − 3 i, . . . , 1 + 3 i,
N = 20
Incident rad. mode nr. µ=1 µ=1 µ=2 µ=1
Table 4.1: Overview of test configurations.
0.15
1
0.1 0.5
r (m)
0
0.05 −0.5
−1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x (m)
(a) Classical mode-matching.

0.15
1
0.1 0.5
r (m)
0
0.05 −0.5
−1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x (m)
(b) Bilinear map-based mode-matching.
Figure 4.3: Real part of pressure field for configuration I.
as a uniform flow with Mach number 0.3. The non-uniform temperature profile of
configuration III has the same mass flow as a flow having uniform mean temperature
(T = 1). The flow profile of configuration III is more representative of a uniform flow
with a thin boundary layer. Configuration IV is a duct with N = 20 segments, where
the impedances of the segments have an imaginary part that varies linearly between
−3 and 3. For all configurations we use µmax = 50 modes to represent the field and an
equal number of test functions.
To compute the solutions of the boundary value problems we use the approach that
is described in Chapter 3. It is important to note that it is not necessary to use path-
following for the inhomogeneous problem, since the axial wave number k µ is given and
not part of the solution, and the inhomogeneous problem (with homogeneous Dirichlet
boundary condition at r = d ) has a unique solution. Consequently, it is much cheaper
to solve the inhomogeneous problem than the original eigenvalue problem.
Figure 4.3 shows the acoustic pressure field for both the classical (CMM) and the
bilinear map (BLM) based mode-matching approaches for configuration I. Figures 4.4,
4.5 and 4.6 compare the pressure, axial and radial velocity for both mode-matching
methods at several radial locations. The figures show that the results of the two ap-
proaches are in very good agreement for all configuration I–III.
1.5
0.5
Re(P) (dimless)
−0.5
Re(P) (BLM), r=0.035m
Re(P) (CMM), r=0.035m
Re(P) (BLM), r=0.075m
−1 Re(P) (CMM), r=0.075m
Re(P) (BLM), r=0.15m
Re(P) (CMM), r=0.15m
−1.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)
1
0.8
0.6
0.4
Re(U) (dimless)
0.2
−0.2
−0.4 Re(U) (BLM), r=0.035m

Re(U) (CMM), r=0.035m
−0.6 Re(U) (BLM), r=0.075m
Re(U) (CMM), r=0.075m
−0.8 Re(U) (BLM), r=0.15m
Re(U) (CMM), r=0.15m
−1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)
0.5
0.4
0.3
0.2
Re(V) (dimless)
0.1
−0.1
Re(V) (BLM), r=0.035m
Re(V) (CMM), r=0.035m
−0.2 Re(V) (BLM), r=0.075m
Re(V) (CMM), r=0.075m
−0.3 Re(V) (BLM), r=0.15m
Re(V) (CMM), r=0.15m
−0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)
Figure 4.4: Comparison of classical (CMM) and bilinear-map-based (BLM) mode-matching

approaches for configuration I. Pressure, axial and radial velocity at radial locations: r =
{0.035, 0.075, 0.15} m.
1.5
0.5
Re(P) (dimless)
−0.5
Re(P) (BLM), r=0.035m
Re(P) (CMM), r=0.035m
Re(P) (BLM), r=0.075m
−1 Re(P) (CMM), r=0.075m
−1.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)
0.8
0.6
0.4
0.2
Re(U) (dimless)
−0.2
Re(U) (BLM), r=0.035m
−0.4 Re(U) (CMM), r=0.035m
Re(U) (BLM), r=0.075m
Re(U) (CMM), r=0.075m
−0.6 Re(U) (BLM), r=0.15m
−0.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)
0.3
0.2
0.1
0
Re(V) (dimless)
−0.1
−0.2
Re(V) (BLM), r=0.035m
−0.3 Re(V) (CMM), r=0.035m
Re(V) (BLM), r=0.075m
Re(V) (CMM), r=0.075m
−0.4 Re(V) (BLM), r=0.15m
−0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)

approaches for configuration II. Pressure, axial and radial velocity at radial locations: r =
{0.035, 0.075, 0.15} m.
1.5
0.5
Re(P) (dimless)
−0.5
Re(P) (BLM), r=0.035m
Re(P) (CMM), r=0.035m
Re(P) (BLM), r=0.075m
−1 Re(P) (CMM), r=0.075m
−1.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)
0.8
0.6
0.4
0.2
Re(U) (dimless)
−0.2
Re(U) (BLM), r=0.035m
−0.4 Re(U) (CMM), r=0.035m
Re(U) (BLM), r=0.075m
Re(U) (CMM), r=0.075m
−0.6 Re(U) (BLM), r=0.15m
−0.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)
0.8
0.6
0.4
Re(V) (dimless)
0.2
Re(V) (BLM), r=0.035m

−0.2
Re(V) (CMM), r=0.035m
Re(V) (BLM), r=0.075m
−0.4 Re(V) (CMM), r=0.075m
Re(V) (BLM), r=0.15m
−0.6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x(m)

approaches for configuration III. Pressure, axial and radial velocity at radial locations: r =
{0.035, 0.075, 0.15} m.
−1.5
test1 (BLM)
test1 (CMM)
−2
log10 of normalized energy balance

−2.5
−3
−3.5
−4
−4.5
5 10 15 20 25 30 35 40 45 50
mumax

approaches for configuration I. Energy balance (normalized on flux through inlet plane)
versus µmax .
The validity of the numerical results can be assessed by checking whether they
satisfy the balance of energy. For that purpose we use Myers’ Energy Corollary (see
Section 2.1.5), which holds exactly for the disturbance field. Figure 4.7 shows the sum
of the acoustic fluxes through the wall, the inlet and the outlet plane, and the volume
integral over the source term. We use a relative numerical accuracy of 10−6 for the
boundary value problem solver, and use Simpson’s rule for the numerical quadrature
on a grid of 151 by 1001 grid points to compute the energy terms. Thus, this sum (which
is normalized on the flux through the inlet plane), which ideally should be zero, is not
expected to be bigger than 10−6 . Figure 4.7 shows that the energy balance is satisfied
better as the number of modes µmax increases, which is to be expected. Furthermore,
the BLM-based mode-matching method performs even better than the classical one
for this configuration I. Incidentally, the energy integral being so accurately satisfied
confirms the assumption of a negligible continuous spectrum contribution.
To verify the regular behavior of the solution at the liner discontinuities along the
wall (where the edge condition has to be satisfied, which is that any volume surround-
ing the edge carries finite energy, as discussed on page 62) we check the uniform con-
vergence of the modal series via the convergence rate of the found modal amplitudes
A n . If we assume that A n = O( n p ) for n → ∞ such that log | A n | = p log n + O (1), then p n
defined as
log | A n |
p n := (4.73)
log n
is expected to approach p with p n = p + O(1/ log( n)). Anticipating a local behavior of
the modal functions that is asymptotically similar to a Fourier series, a convergence
rate p < −1 will be sufficient for uniform convergence. As shown in Figure 4.8, p ' −2
for the considered configurations I–III, so we see that the condition is indeed satisfied,
in particular for both mode matching methods in the same way.
Moreover, there is another interesting observation possible from these plots. The
behavior of p n from the classical mode-matching method is not as smooth as from the
−1
−1.2
−1.4
−1.6
−1.8
pn
−2
−2.2
−2.4
−2.6
BLM
−2.8
CMM
−3
0 10 20 30 40 50
n
−1
−1.2
−1.4
BLM
−1.6
CMM
−1.8
pn
−2
−2.2
−2.4
−2.6
−2.8
−3
0 10 20 30 40 50
n
−1
−1.2
−1.4
−1.6
−1.8
pn
−2
−2.2
−2.4
−2.6
BLM
−2.8
CMM
−3
0 10 20 30 40 50
n
Figure 4.8: Convergence rate of amplitudes for classical and bilinear-map-based mode-
matching, for configurations I, II and III.
0.15 1
0.5
0.1
r (m)
0
0.05
−0.5
0 −1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x (m)
Figure 4.9: Real part of pressure field for configuration IV.
BLM-based mode-matching as n becomes larger. Apparently the amplitudes from the

classical method are more inaccurate for large n. This is a confirmation of the fact that
the inner products based on closed-form expressions are not prone to the quadrature
errors for large n, for which the eigenfunctions are most oscillatory and quadrature is
consequently most difficult.
Finally, to illustrate that a high number of segments poses no problem for mode-
matching with the scattering matrix formalism we included Figure 4.9, which depicts
the pressure field for a configuration with N = 20 segments. From our experience an
iterative procedure often does not converge for more than 10 segments. The artifacts
near r = 0 are due to the fact that at that location the field is almost zero, so very close
to the level curve at zero.
Chapter 5
Asymptotic solutions for

slowly varying impedance
In the previous chapters we considered configurations whose properties are (at least
locally, in case of a segmented liner) invariant in the axial direction; the shape of the
cross-section, the mean flow and the wall impedance were independent of the axial
position. This permitted us to describe the sound field in terms of duct modes. Modes
in a strict sense cease to exist if one of the problem parameters varies with the axial
coordinate.
Fortunately, if this variation is slow it is possible to find a small parameter that may
be used for solutions of multiple scales or WKB type (a special case of multiple scales
analysis). This approach is based on slowly varying modes, for which it is assumed
that the fast scale of the wave motion and the slow scale of the axially varying ‘wave
envelope’ (mode shape and wave number) are independent. (Note that in Section 2.5
we also employed the WKB method, but in cross-wise direction.)
One specific example is the case of sound propagation through flow ducts with a
slowly varying cross-section [103, 104]. The usefulness of the approximation was shown
by a numerical comparison for a geometry that is very typical of the shape of an aero-
engine duct [107]. Moreover, the reflection of modes at turning points was studied,
where a mode changes from cut-on to cut-off due to a decrease in cross-sectional area.
WKB solutions for slowly varying cross-section have also been used to match the flow
perturbations of the source region (in and near the region of rotor and stator blades,
computed from fully numerical simulations) to the acoustic region in the inlet and
bypass duct [92, 118].
The multiple-scales method can in principle be used for the variation of any of the
problem parameters [22, 28, 29, 63, 93], but with the mean flow it is difficult since we
do not have an analytic expression available for a slowly varying ducted shear flow.
We will consider here a slowly varying impedance. As mentioned in Chapter 1 and
depicted in Figure 1.2, the sound absorbing liner of an APU exhaust duct has an axially
decreasing depth, which results in an axially varying wall impedance.
In Section 5.1 we present a solution1 in terms of slowly varying modes for a duct
with an axially varying impedance and a mean flow with non-uniform velocity and
1 The work described in this chapter is the result of a collaboration with W. Lazeroms, who derived most
of the analytical results of Section 5.1, which can also be found in [63].
75
76 Chapter 5. Asymptotic solutions for slowly varying impedance
temperature [63, 90]. The cross-wise mode shape of a slowly varying mode is given
by the Pridmore-Brown equation, which is therefore solved at each axial location with
the numerical approach described in Chapter 3. In Section 5.2 some numerical results
obtained with the WKB approach will be compared to some mode-matching results.
5.1 WKB solutions

In this section we use the Linearized Euler Equations for parallel flow in the formula-
tion of (2.105a), (2.105b) and (2.64), in which the last term vanishes for parallel flow.
Together with the linearized ideal gas law (2.61) these equations are made dimension-
less in the same way as the Pridmore-Brown equation in Sections 2.5 and 3.1, i.e. based
on the duct radius d , a referencepdensity ρ ∞ and temperature T∞ , from which follows
the reference sound speed c ∞ = γRT∞ . This amounts to the scaling
u0 ρ0 T0 p0 ωd
M= , R= , Θ= , Π= , ω̃ = .
c∞ ρ∞ T∞ ρ ∞ c2∞ c∞
(5.1)
v1 ρ1 T1 p1 x
ṽ1 = , ρ̃ 1 = , T̃1 = , p̃ 1 = , x̃ = ,
c∞ ρ∞ T∞ ρ ∞ c2∞ d
which implies that we can use R = 1/Θ and Π = 1/γ. In the following we will suppress
the tildes and use dimensionless quantities unless stated otherwise. Up to this point
we denoted the mean flow and perturbations by subscripts 0 and 1 , but these will be
used later to denote the orders in an asymptotic expansion, and we will now write the
total field consisting of the steady parallel mean flow plus perturbations of the form
e−iω t+i mθ as
£ ¤ £ ¤ n£ ¤ o
p, u, v, w, ρ , T ( x, r, θ , t) = Π, M, 0, 0, R, Θ ( r ) + Re P , U , V , W , D, T ( x, r ) e i mθ−iω t .
(5.2)
The governing equations (2.105a), (2.105b), (2.64) and (2.61) then become
µ ¶ µ ¶
∂ 1 ∂( rR V ) im ∂U
−iω + M D+ +R W+ = 0, (5.3a)
∂x r ∂r r ∂x
µ ¶
∂ ∂ M 1 ∂P
−iω + M U +V + = 0, (5.3b)
∂x ∂r R ∂x
µ ¶
∂ 1 ∂P
−iω + M V+ = 0, (5.3c)
∂x R ∂r
µ ¶
∂ im
−iω + M W+ P = 0, (5.3d)
∂x rR
µ ¶ µ ¶
∂ ∂Θ 1 ∂ im ∂U
−iω + M T+ V + (γ − 1)Θ (rV ) + W+ = 0, (5.3e)
∂x ∂r r ∂r r ∂x
γP = R T + D Θ. (5.3f)
Note that for modes of the form P ( x, r ) = P ( r ) e i kx we obtain a system of equations

for the modal amplitudes depending only on r . This system can be reduced to the
Pridmore-Brown equation. The Ingard-Myers boundary condition (2.78) in dimension-
less form is in the present notation
µ ¶
∂ P
− iωV = −iω + M at r = 1. (5.3g)
∂x Z
§ 5.1 WKB solutions 77
If the impedance varies in the axial direction, the problem is not axially invariant
anymore, so strictly speaking it is not possible to describe the field in terms of modes
of the form P ( r ) e i kx . However, if we assume that the inherent length scale L of typical
variations of Z ( x̄/L) (where x̄ = xd and x are the dimensional and dimensionless axial
coordinates) is large compared to the duct radius d , i.e.
µ ¶ µ ¶
x̄ d d
Z =Z x = Z (ε x) = Z ( X ), ε= ¿ 1, X := ε x, (5.4)
L L L
we can use this small parameter ε to construct so-called slowly varying modes by a
variant of the WKB method. Assuming that the axial wavenumbers are typically equal
or larger than O(1), we approximate the acoustic field by modes of which the amplitude,
mode shape and axial wavenumber vary only slowly in the axial direction; we look for
modal solutions of the form
 X 
Z
£ ¤ £ ¤ i
P , U , V , W , D, T (r, X ; ε) = P,U, V ,W, D, T (r, X ; ε) exp  µ(η)dη , (5.5)
ε
0
where µ( X ) is the axial wavenumber depending on the slow coordinate X . Substituting

this WKB Ansatz (5.5) into (5.3) yields
µ ¶
1 ∂
∂D im ∂U
− iΛD + ε M ( rRV ) + R
+ W +ε + iµU = 0, (5.6a)
∂ X r ∂r r ∂X
µ ¶
∂U d M 1 ∂P
− iΛU + ε M + V+ ε + iµP = 0, (5.6b)
∂X dr R ∂X
∂V 1 ∂P
− iΛV + ε M + = 0, (5.6c)
∂ X R ∂r
∂W i m
− iΛW + ε M + P = 0, (5.6d)
∂ X rR
· ¸
∂T dΘ 1 ∂ im ∂U
− iΛT + ε M + V + (γ − 1)Θ rV
( ) + W + ε + i µU = 0, (5.6e)
∂X dr r ∂r r ∂X
γP = RT + ΘD, (5.6f)
with Λ := ω − µ M . The corresponding boundary condition obtained from (5.3g) is
µ ¶
iΛ ∂ P
− i ωV = − P + εM at r = 1. (5.6g)
Z ∂X Z
Now we expand the amplitude functions in ε as
£ ¤ £ ¤
P,U, V ,W, D, T ( r, X ; ε) = P0 ,U0 , V0 ,W0 , D 0 , T0 ( r, X )
£ ¤
+ ε P1 ,U1 , V1 ,W1 , D 1 , T1 ( r, X ) + O (ε2 ). (5.7)
The leading order equations obtained after substitution of this expansion in (5.6) are
µ ¶
1 ∂ im
− iΛD 0 + ( rRV0 ) + R W0 + iµU0 = 0, (5.8a)
r ∂r r
dM iµ
− iΛU0 + V0 + P0 = 0, (5.8b)
dr R
1 ∂P0
− iΛV0 + = 0, (5.8c)
R ∂r
im
− iΛW0 + P0 = 0, (5.8d)
rR · ¸
dΘ 1 ∂ im
− iΛT0 + V0 + (γ − 1)Θ ( rV0 ) + W0 + iµU0 = 0, (5.8e)
dr r ∂r r
γP0 = RT0 + ΘD 0 , (5.8f)
with the boundary condition
iΛ
− iωV0 = − P0 at r = 1. (5.8g)
Z
The system of equations that arises for modal solutions of the form P ( x, r ) = P ( r ) e i kx
has the same structure as (5.8), except that in (5.8) the modal amplitudes and the axial
wavenumber depend on X , which serves as a parameter. It follows that P0 satisfies the
Pridmore-Brown equation (2.117), i.e.
· µ ¶ µ 2 ¶¸
∂2 1 2 dΛ 1 dΘ ∂ Λ 2 m2
L P 0 := + − + + −µ − 2 P0 = 0, (5.9a)
∂r2 r Λ dr Θ dr ∂r Θ r
together with the Ingard-Myers boundary condition (3.1b), i.e.
∂P0 iΛ2 R
− P0 = 0 at r = 1. (5.9b)
∂r ωZ
For every fixed axial location X this problem (together with the regularity condition at
r = 0) is exactly the same as the constant impedance problem described by (3.1). The
general leading order solution will be of the form
P0 ( r, X ) = N ( X )ψmn ( r, X ), (5.10)
where ψmn is again an eigenfunction (for clarity considered normalized), this time
parametrically depending on X , and N ( X ) is a slowly varying amplitude function that
is still unknown. As before, solving the Pridmore-Brown equation to find the eigen-
function ψmn ( X ) of radial order n also yields the axial wavenumber µmn ( X ) for any
given frequency ω and circumferential wavenumber m.
To determine N ( X ) the first order equations obtained after substituting the ex-
pansion (5.7) in (5.6) are needed. N ( X ) can be determined from a condition that the
asymptotic expansion remains uniform in X and the first order problem is solvable, as
we will see later—we do not have to solve this first order system explicitly. These first
§ 5.1 WKB solutions 79
order equations are

µ ¶ µ ¶
1 ∂ im ∂D 0 ∂U0
− iΛD 1 + ( rRV1 ) + R W1 + iµU1 = − M +R , (5.11a)
r ∂r r ∂X ∂X
µ ¶
dM iµ ∂U0 1 ∂P0
− iΛU1 + V1 + P1 = − M + , (5.11b)
dr R ∂X R ∂X
1 ∂P1 ∂V0
− iΛV1 + = −M , (5.11c)
R ∂r ∂X
im ∂W0
− iΛW1 + P1 = − M , (5.11d)
rR ∂X
µ ¶ µ ¶
dΘ 1 ∂ im ∂T0 ∂U0
− iΛT1 + V1 + (γ − 1)Θ ( rV1 ) + W1 + iµU1 = − M + (γ − 1)Θ ,
dr r ∂r r ∂X ∂X
(5.11e)
γP1 = RT1 + ΘD 1 , (5.11f)

µ ¶
iΛ ∂ P0
− iωV1 + P1 = M at r = 1. (5.11g)
Z ∂X Z
In a similar way as before we can find a single equation for P1 with a right-hand side
that only contains P0 . The result is
· µ ¶ ¸ ·µ ¶ ¸
2i M ∂ ∂ Λ ∂P0 M ∂ Λ µ
L P 1 = F := − ln −i + P2 , (5.12a)
Λ ∂ X ∂r M ∂r P0 ∂ X Θ M 0

Ã 2
!
∂P1 iΛ2 R MRZ ∂ ΛP0
− P1 = − at r = 1. (5.12b)
∂r ωZ ωP0 ∂ X Z 2
Instead of solving for P1 , which would lead to yet another undetermined amplitude
function analogous to N ( X ) in (5.10), we will derive a solvability condition [82, chapter
15] for (5.12) that only contains leading order terms. If this condition holds, then the
solution of the form (5.5) exists, and the leading and first order systems (5.9) and (5.12)
are consistent with a regular asymptotic expansion. This solvability condition can be
found as follows.
We start by multiplying (5.12a) by r ΘP0 /Λ2 , (5.9a) by r ΘP1 /Λ2 , subtracting the two
equations and integrating the result. This yields
Z1 Z1
Θ Θ
(P0 LP1 − P1 LP0 ) r d r = P0 F r d r. (5.13)
Λ2 Λ2
0 0
¡ ¢
Because the operator r Θ/Λ2 L is the self-adjoint form of L (compare with (2.112)), i.e.
· ¸ µ ¶
rΘ ∂ r Θ ∂P r Θ Λ2 2 m2
L P = + − µ − P, (5.14)
Λ2 ∂ r Λ2 ∂ r Λ2 Θ r2
the left-hand side of (5.13) can be reduced to an expression containing only boundary
terms, i.e.
Z1 Z1 · µ ¶¸
Θ ∂ rΘ ∂P1 ∂P0
(P0 LP1 − P1 LP0 ) r d r = P 0 − P 1 dr
Λ2 ∂ r Λ2 ∂r ∂r
0 0
· µ ¶¸ µ ¶¯
rΘ ∂P1 ∂P0 r=1 Θ ∂P1 ∂P0 ¯¯
= P0 − P1 = 2 P0 − P1
Λ2 ∂r ∂r r =0 Λ ∂r ∂ r ¯r=1
Ã !
2 ¯
MZ ∂ ΛP0 ¯¯
=− , (5.15)
ωΛ2 ∂ X Z 2 ¯ r =1
where we used the boundary conditions (5.9b) and (5.12b) and the relation R Θ = 1 in
the last step. Using this result together with (5.12a) in (5.13) then leads to the following
solvability condition for the first order problem:
Z1 ½ · ¶µ ¸ ·µ ¶ ¸¾ Ã !
2 ¯
2MΘ ∂ ∂ Λ ∂P0 MΘ ∂ Λ µ 2 MZ ∂ ΛP0 ¯¯
i P0 ln + 2 + P rdr = .
Λ3 ∂X ∂r M ∂r Λ ∂X Θ M 0 ωΛ2 ∂ X Z 2 ¯r=1
0
(5.16)
The next step is to substitute the general solution of P0 shown in (5.10) into (5.16).
After working out and rearranging terms we arrive at a first order equation for the
amplitude function N
d
g( X ) [ N ( X )]2 = − f ( X ) [ N ( X )]2 , (5.17a)
dX
with (in principle known) functions
Z1 ½ · µ ¶ ¸ ¾ µ ¶¯
2ψ ∂ ∂ Λ ∂ψ 1 ∂ h¡ ¢ 2i MZ ∂ Λψ2 ¯¯
f ( X ) := i M Θ ln + M Λ + µ Θ ψ r d r − ,
Λ3 ∂ X ∂r M ∂r Λ2 ∂ X ωΛ2 ∂ X Z 2 ¯r=1
0
Z1 ½ ¶ µ ¾ ¯
ψ Λ ∂ψ M Λ + µΘ 2
∂ M ¯
2¯
g( X ) := i M Θ ln + ψ rdr − ψ . (5.17b)
Λ3 ∂r M ∂r Λ2 ωΛ Z ¯r=1
0
This equation has the solution

 
ZX
2 f (η) 
N = N02 exp − dη , (5.18)
g(η)
0
where N0 is a normalization constant that can be determined from the source, e.g. at
the beginning
± of the duct. (Note that any normalization of ψ drops out in the term
f ( X ) g( X ) .) ±
±In principle (5.10) can now be computed, provided that the derivatives ∂ψ ∂ X and
dµ d X are known. Since µ( X ) and ψ( r, X ) are only known from the solution of the
Pridmore-Brown equation, which are solved numerically (unless flow and temperature
are uniform), the X -derivatives have to be computed numerically as well.
The results can be simplified for some special cases. For uniform mean flow and
arbitrary temperature we have
µ ¶
f (X ) 1 ∂
= a( X ) − h( X ) , (5.19a)
g ( X ) a( X ) ∂ X
§ 5.2 Numerical comparison between WKB and mode-matching 81
with
Z1 · ¸ · ¸
¡ ¢ iMΛ 2 i MΛ 2 dZ
a ( X ) := M Λ + µΘ ψ2 r d r + ψ , h ( X ) := ψ , (5.19b)
ωZ r =1 ωZ2 dX r =1
0
such that (5.18) reduces to

 
ZX
N02 h(η) 
N 2(X ) = exp − dη . (5.20)
a( X ) a(η)
For zero mean flow and arbitrary temperature this can be reduced even further, to
 −1
Z1
2
N ( X ) = N02 µ 2
Θψ r d r  . (5.21)
0
± ±
Note that for these special cases it is not required to compute ∂ψ ∂ X and dµ d X .
The result in (5.18) may be compared with the one given by [28], where the duct
varies in diameter and the mean flow includes swirl, but the impedance is taken con-
stant. With the APU geometry in mind of a constant duct but varying impedance,
we could obtain a simpler result, that is relatively easily applicable when numerical
solutions of the Pridmore-Brown equation are available.
Finally we note that the small parameter ε only acts as a bookkeeping parameter;
since (5.17a) does not depend explicitly on ε and both sides of the equality sign contain
only single X -derivatives we can replace X by x.
5.2 Numerical comparison between WKB and mode-

matching
In this section we show some numerical results for test cases motivated by a realistic
APU exhaust duct, having a length of 1m and a radius d of 0.15 m. We choose the
reference temperature T∞ = 700 K and density ρ ∞ = 0.5 kg/m3 , corresponding to a
reference sound speed of c ∞ = 549 m/s. The frequency ω, wavenumber m, and the
details of the velocity and temperature of the mean flow are given in the captions
of the respective figures. We consider cases where Im( Z ( x)) varies linearly with fixed
Re( Z ) = 1.5, and cases where the liner is modeled as a Helmholtz resonator (see Section
2.3.4).
In the following we will use x̄ = xd and x to denote the dimensional and dimension-
less axial coordinates respectively, as we did in (5.4).
To assess the applicability of the WKB method and estimate the truncation error
in the WKB expansion (which is O(ε), see (5.7)) we need to estimate ε. Since by as-
sumption Z 0 ( X )/ Z ( X ) = O(1) this can be done by noting that if Z varies typically over a
length scale L we have
d
Z ( x̄/L) d Z 0 ( X )
d d x̄ = = O (ε). (5.22)
Z ( x̄/L) L Z(X )
We plot contour lines of the pressure field, where the source at x̄ = 0 m consists of a
single n-th order radial mode (usually, n = 1). For cases of uniform mean flow velocity
and temperature we compare the (axial) WKB results with mode-matching2 results
(see Chapter 4). The eigenfunctions are normalized according to
Z1
|ψ|2 r d r = 1, ψ0 (1) ∈ R+ , (5.23)
0
i.e. the phase is chosen such that ψ0 (1) is real-valued and positive.
For the WKB results we first compute the eigenfunctions and axial wavenumbers
for 200 equally spaced axial positions along the duct with the numerical approach de-
scribed in Chapter 3. The amplitudes N ( X ) are computed by using (5.18) or the special
cases (5.20) and (5.21). The integrals over η are computed with a trapezoidal rule,
which has equally spaced grid-points; this enables us to use all computed eigenfunc-
tions (for all axial positions) for the field plot. The integrals over r in (5.17b), (5.19b)
and (5.21) are computed by using the QUADPACK [95] code (which is based on an
adaptive Gauss-Kronrod rule). We motivate this choice by noting that the eigenfunc-
tion (and hence the integrand of the radial integral) can be oscillatory, while the inte-
grands of the axial integrals are only slowly varying. For non-uniform mean flow we
need to compute f ( X ). Working out f ( X ) in (5.17b) yields
Z1 ½ · ¸ · ¸¾
2ωψΘ ∂ M ∂2 ψ ∂µ ∂ψ ψ 2 ∂µ ∂ψ
f (X ) = i − Λ +M + 2 (Θ − M ) ψ + 2( M Λ + µΘ) rdr
Λ5 ∂ r ∂r∂ X ∂ X ∂r Λ ∂X ∂X
0
· µ ¶ ¸
Mψ ∂ψ ∂µ ∂Z
− 2Λ Z − ZM + 2 Λ ψ . (5.24)
ωΛ2 Z 2 ∂X ∂X ∂X r =1
± ±
The X -derivatives ∂ψ ∂ X and dµ d X (which are not available analytically, as men-
tioned before) are computed by using a second order finite differences approximation.
To test the WKB approach we first let Z ( x̄) vary linearly from 1.5 − i to 1.5 + i, which
corresponds to an estimated ε = 0.2. Figure 5.1 shows the piecewise impedance for the
segments that are used in the mode-matching approach. From Figure 5.2a and Figure
5.2b it is clear that the difference between constant impedance and axially varying
impedance is significant, so the x̄-dependency of Z has to be taken into account. Figure
5.2b shows that the WKB and the mode-matching results agree reasonably well for
this not very small choice of ε. From Figure 5.2c, with the same parameter values and
mean flow mass flux as before but a parabolic mean flow profile, we conclude that the
effect of a non-uniform mean flow should not be neglected. The present difference is
explained by the fact that downstream running sound waves are refracted towards the
lined wall [108], which results in more damping.
In the next configuration we let Z ( x̄) vary linearly from 1.5 − 2i to 1.5 + 2i along
the same interval, which corresponds to a higher estimated ε of 0.4. The truncation
error in the WKB approximation is now larger and we may expect a bigger difference
between the WKB and the mode-matching solutions. This is indeed the case, as is
shown in Figure 5.3. Near x̄ = 0.7 m some signs of intermodal scattering are visible,
which is explicitly not taken into account by the WKB method.
To evaluate the applicability of the WKB method for a realistic APU exhaust duct
geometry we now consider a locally reacting liner with a cell depth d̄ l ( x̄) that is axially
2 In [90] (on which this chapter is based) we referred to the NLR-developed mode-matching code
BAHAMAS, which handles uniform flow and temperature, since a mode-matching code that handles non-
uniform flow and temperature was not available at the time of writing.
0.5
Im(Z/(ρ0 c0))
0
−0.5
WKB
−1 modematch
0 0.2 0.4 0.6 0.8 1

x (m)
Figure 5.1: Z varies linearly from 1.5 − i to 1.5 + i, ε ≈ 0.2.
constant impedance
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(a) Z = 1.5 − i, uniform mean flow velocity M0 = 0.3.

mode−matching
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
WKB
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(b) Same as Figure 5.2a, except Z varies linearly from 1.5 − i to 1.5 + i so ε ≈ 0.2.
WKB
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(c) Same as Figure 5.2b, except mean flow velocity M(r) = M0 43 (1 − 12 r 2 ) with M0 = 0.3.
Figure 5.2: ω = 10, m = 2, n = 1

mode−matching
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
WKB
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 5.3: ω = 10, m = 2, n = 1, uniform mean flow velocity M0 = 0.3, Z varies linearly
from 1.5 − 2i to 1.5 + 2i so ε ≈ 0.4.
1
0
0
Im(Z/(ρ0 c0))
Im(Z/(ρ0 c0))
−5
WKB
−1
modematch
−2 −10
WKB
−3 modematch
−15
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m)
Figure 5.4: Z modeled as Helmholtz resonator with liner depth that varies linearly from
7 cm to 1 cm. Left: ω = 6 (ε ≈ 0.3), right: ω = 10 (resonance occurs).
varying from 7 cm to 1 cm along the duct. The liner is modeled as a Helmholtz resonator
array (see Section 2.3.4), which has an impedance given by
µ ¶
ω̄
Z̄ = Z̄0 + iρ ∞ c ∞ cot d̄ l ( x̄) , (5.25)
c∞
where the overbars are used to denote dimensional quantities, and Z̄0 is the face sheet
impedance
Z̄0 = R̄ 0 − iω̄χ̄. (5.26)
We choose a facing sheet resistance of R̄ 0 = 400 kgm−2 s−1 , and a mass reactance of
χ̄ = 0.001 kg/m2 . We remark that this is only a model and the reference sound speed
c ∞ may be different from the sound speed at the wall for non-uniform temperatures.
Figure 5.4 shows the imaginary part of the impedance as a function of x̄ for two differ-
ent frequencies. Note that for ω = 10 resonance occurs; close to x̄ = 0.4 m the impedance
becomes very large so the liner behaves as a hard wall.
Figure 5.5 show the acoustic field for ω = 6. On the interval considered there is no
location where the Helmholtz resonator is in resonance; Z ( x̄) is slowly varying with an
estimated ε = 0.3. The WKB and mode-matching results show rather good agreement,
about what can be expected from this value of ε.
However, for ω = 10 we do have resonance near x̄ = 0.4 m so the assumption of a
slowly varying Z ( x̄) is, at least near this point, not valid anymore. This can indeed be
mode−matching
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
WKB
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 5.5: ω = 6, m = 2, n = 1, uniform mean flow velocity M0 = 0.3, Z modeled as

Helmholtz resonator with liner depth that varies linearly from 7 cm to 1 cm.
mode−matching
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
WKB
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 5.6: ω = 10, m = 2, n = 1, uniform mean flow velocity M0 = 0.3, Z modeled as

Helmholtz resonator with liner depth that varies linearly from 7 cm to 1 cm.
identified from Figure 5.6, where the region of resonance seems to excite the second
radial mode, an effect which cannot be described by (straight-forward) application of
the WKB method [92].
In a realistic APU exhaust duct cool air is let in near x̄ = 0 m along the wall (see
Figure 1.4). This produces a strong radial temperature gradient. We modeled this
by the tanh-type profile that is depicted in Figure 5.7. The effect of this temperature
gradient is that it creates effectively two concentric ducts, each with its own propaga-
tion properties. These duct fields are not completely independent of each other. Sound
waves from the center region (with the highest sound speed) will refract (by a form of
Snell’s law) to the colder annular region. However, sound waves in the annular region
refracts only if the angle between duct axis and their propagation direction is not too
small. Otherwise the annular region will act as a duct on its own.
This is illustrated in Figure 5.8, where the fields are plotted for the first two right-
running radial modes. In case of the first mode the field is virtually only existent in
the colder outer region. The field of the second mode exists in both, but such that the
sound waves refract from inside to outside.
1.6
1.4
1.2
0.8
0.6
0.4
0.2
0 0.2 0.4 0.6 0.8 1
¡ 5£ ¢¤
Figure 5.7: Temperature profile Θ(r) = 14 + 8 1 + tanh 50( 43 − r) .
WKB
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
WKB
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 5.8: ω = 10, m = 2, n = 1 and n = 2, uniform mean flow velocity M0 = 0.3, strong
temperature gradient (see Figure 5.7), Z(x) varies linearly from 1.5 − i to 1.5 + i.
Chapter 6
Application to APU exhaust

duct
In this chapter we present numerical results based on the theory that was described
in the preceding chapters. The parameters have been chosen with the application of
an APU exhaust duct in mind. We aim to illustrate the practical applicability of the
methods developed above and to present results and insights that facilitate design
studies.
For easier interpretation, all parameters and variables in this chapter are made
dimensionless by scaling on the duct radius d , a reference density ρ ∞ and a refer-
ence temperature
p T∞ . This implies that velocities are scaled on reference sound speed
c ∞ = γRT∞ , time on d / c ∞ and impedance on ρ ∞ c ∞ (note that for non-uniform tem-
perature the local sound speed at the wall might be different from c ∞ ). The dimension-
less frequency, also referred to as Helmholtz number, is computed as ω = 2π f d / c ∞ . The
dimensionless mean flow axial velocity is denoted by M (the Mach number) and the di-
mensionless mean temperature is by T , in other words, u 0 = c ∞ M ( r ) and T0 = T∞ T ( r ).
Unless mentioned otherwise, we choose the duct length L = 1 m, radius d = 0.15 m,
the reference density ρ ∞ = 0.5 kg/m3 , and reference temperature T∞ = 700 K, and the
gas to be a perfect gas (see Section 2.1.1) with γ = 1.4. This amounts to a reference
sound speed c ∞ = 531.7 m/s and pressure p ∞ = ρ ∞ RT∞ = 100800 Pa. We choose the
kinematic viscosity ν = 6.8 · 10−5 m2 /s.
As mentioned in Section 1.3, we assume that the duct end is reflection free. We
justify this by noting that for high dimensionless frequencies, and for modes not too
close to cut-off, the reflections are very small. In addition, when the duct is treated,
the amplitude of the sound field is reduced, and the influence of the end reflection may
be even smaller. Finally, we note that including the duct-end reflection and radiation
could obfuscate the propagation effects that we wish to investigate.
6.1 Effect of non-uniform mean flow

First we investigate the influence of the flow profile on the axial wavenumbers. We
illustrate the effect of a widening boundary layer by introducing a (simple) flow profile,
87
88 Chapter 6. Application to APU exhaust duct
0.8
0.35
0.7
0.3
0.6
0.1
0.25
0.2
0.5
0.3
0.2 0.4
M(r)
M(r)
0.4 0.5
0.6
0.15 0.7
0.3 100
0.8
50
0.9
20
0.2 0.1
1
10
5
0.1 3 0.05
2
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.8 0.82 0.84 0.86 0.88 0.9 0.92 0.94 0.96 0.98 1
r (dimless) r (dimless)
2
(a) M(r) = M0 b+ b
2 (1 − r ) with M0 = 0.4 for var-
(b) Blasius-like profile of (6.2) with M0 = 0.3 at
ious b values. several values of x = ξ d.
Figure 6.1: Mean flow profiles with varying boundary layer thickness.
only depending on the parameter b
( b + 2)
M ( r ) = M0 (1 − r b ), b ∈ (2, 3, 5, 10, 20, 50, 100), (6.1)
b
see Figure 6.1a. This profile has the same mass flux for all b, equal to that of a flow
with uniform temperature and Mach number M0 .
Figure 6.2a shows the trajectories of the ten axial wavenumbers that are closest to
the real axis for dimensionless frequency ω = 10, circumferential wavenumber m = 5
and locally reacting impedance Z = 1 + 1i. It is seen that the influence of the mean
flow profile is relatively minor, except for the wavenumber ending at k = −11.3 − 3.6i.
For the highest b-values, when the mean flow profile is closest to uniform, this mode
behaves as a surface wave, which is transformed into a mode which does not exhibit
this surface wave behavior as the boundary layer is widened. This is also illustrated in
Figure 6.2b, which depicts the modulus of the eigenfunctions. This figure shows that
the eigenfunction of the third left-running mode (µ = −3) changes its character, from
surface wave to regular mode, whereas the other modes do not.
Furthermore, it can be seen that the right-running mode starting at k = 6.1 + 1.4i,
which is close to cut-on, becomes more cut-off as the boundary layer is widened. The
reverse effect can be seen for the left-running mode starting at k = −11.4 − 0.8i, which
becomes more cut-on. In the direction of the flow the wave is refracted towards the
wall and hence damped more strongly. In the upstream direction the wave is refracted
away from the wall, which almost annihilates the damping (see also Section 2.5). This
refraction effect increases when the flow becomes less uniform.
Next we consider a more physically inspired profile, which is an approximation of
the Blasius solution for the boundary layer development of a flow parallel to a semi-
infinite flat plate, see Appendix B. We choose the mean flow profile M ( r ; x) at a given
(dimensionless) axial location ξ = x/ d with respect to where the flow is still uniform
§ 6.1 Effect of non-uniform mean flow 89
15
10
5
Im(k)
−5
−10
−15
−25 −20 −15 −10 −5 0 5 10 15 20

Re(k)
(a) Trajectory of axial wavenumbers, which move from from blue to red pass-
ing the b-values b = 100, 50, 20, 10, 5, 3 (indicated by dots).
µ=−1, k=−13.8 −0.1i µ=−2, k= −7.3 −8.6i
3 3
2 2
1 1
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
µ=−3, k=−11.3 −3.6i µ=−4, k= −7.1−14.6i
4 3
3
2
2
1
1
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
µ=−5, k= −7.2−19.5i µ=−6, k= −7.3−24.1i
4 4
100
3 3
50
2 2 20
10
1 1 5
0 0 3
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(b) Magnitude of normalized eigenfunctions of the first six left running modes
as a function of the radial coordinate, for different values of b.
Figure 6.2: Influence of boundary layer thickness for mean flow profile (6.1).
−30
−35
−40
−45
Im(k)
−50
−55
−60
−65
−5 0 5 10 15 20 25 30 35
Re(k)
Figure 6.3: Trajectory of axial wavenumbers for the mean flow profile of (6.2) with x = 1 m,
Z = 1.5 + 0.5i.
(corresponding to the plate edge for the Blasius solution) as

±
h ³ í1 b(r;ξ) ±p ±p
b ( r ;ξ )
M ( r ; x) = M0 tanh a( r ; ξ) ,
a ( r ; ξ) = B 0 r ξ, b ( r ; ξ ) = a 0 + B 1 r ξ
q ± q ±
with ξ = x/ d, B0 = λ U0 d ν, B1 = a 1 U0 d ν, U0 = M0 c ∞ ,
a 0 = 1.4853, a 1 = 0.0555, λ = 0.33206.
(6.2)
where U0 is the dimensional mean flow velocity far away from the wall. See Figure
6.1b for the boundary layer development for axial locations up to x = 1 m.
Figure 6.3 shows the axial wavenumber trajectories for a test case with ω = 3, m = 1,
and a Blasius-like boundary layer with M0 = 0.3 and x = 1 m. The impedance of the
locally reacting liner has been chosen such that two acoustic and two hydrodynamic
surface waves exists (see [100], note the different sign convention). The trajectories,
which move from blue to red, consist of two paths designated by the dots; first the
impedance is varied from hard-wall to Z = 1.5+0.5i, after which the mean flow profile is
varied from uniform flow to the boundary layer profile. As the boundary layer thickness
is increased the hydrodynamic surface wave ceases to exist, which is demonstrated by
the fact that the corresponding axial wavenumber disappears to infinity in the lower
half of the complex k-plane.
To illustrate the effect of a developing boundary layer on the axial wavenumbers we
plotted in Figure 6.4 the k-trajectories for a test case with ω = 10, m = 0 and Z = 1 + 1i,
as the mean flow profile is gradually changed from the Blasius-like profile correspond-
ing to x = 10 cm to the profile corresponding to x = 1 m. It can be seen that the influence
§ 6.2 Effect of non-uniform temperature 91
20
15
10
5
Im(k)
−5
−10
−15
−20
−25 −20 −15 −10 −5 0 5 10 15 20

Re(k)
Figure 6.4: Trajectory of axial wavenumbers for the mean flow profile of (6.2) with varying
x = 0.1, . . . , 1 m, Z = 1 + 1i.
of the mean flow profile is in this case very small, which is to be expected in view of the
very thin boundary layer (99% of the centerline Mach number is reached at a distance
from the wall of approximately 1.0–3.5% of the duct radius). The surface wave in the
third quadrant is influenced most strongly.
6.2 Effect of non-uniform temperature

To investigate the influence of a temperature gradient we consider a (dimensionless)
temperature profile of the form
1
T (r) = 1 + tanh ( q( r 0 − r )) , (6.3)
2
where the parameter q is used to set the magnitude of the gradient, and the parame-
ter r 0 to set its radial location. We take a dimensionless impedance close to hard wall
( Z = 100 + 1i), in order to avoid the special embedding of (3.12) for non-uniform flow and
temperature combined with a hard wall as described in Section 3.2. The circumferen-
tial wavenumber m = 0. With zero mean flow, the left and right axial wavenumbers are
identical except for their sign, so we plot only the right-running modes.
For a hard-walled duct with uniform flow and temperature we have the dispersion
relation
(ω − kM )2
α2 = − k2 . (6.4)
T
2 2
1.5 1.5
T(r)
T(r)
1 1
0.5 0.5
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r (dimless) r (dimless)
(a) Temperature profile of (6.3) with q = 50 and (b) Temperature profile of (6.3) with r 0 = 0.8
r 0 = {0.65, 0.8, 0.95}. and q = {100, 22.3, 5}.
Figure 6.5: Temperature profiles.
The radial wavenumber α follows from the boundary conditions; the axial wavenumber
is then p p
−ω M ± T ω2 − α2 (T − M 2 )
k= . (6.5)
T − M2
For cases where M 2 ¿ T (which is reasonable for an APU since T is typically close to 1
and M to 0.3) it follows that
s
ω2
k≈± − α2 for M 2 ¿ T. (6.6)
T
This means that the cut-on modes move towards the origin as T increases. For the cut-
off modes (for which holds α2 > ω2 /T ) that are close to cut-on the wavenumbers move
away from the origin, and for the modes that are strongly cut-off (α2 À ω2 /T ) there is
little influence of the temperature on k.
Figure 6.6a shows the trajectories of the axial wavenumbers when the temperature
gradually varies from a uniform T = 1 to the non-uniform profile of (6.3) with r 0 =
0.95 and q = 30. The dimensionless frequency ω = 10. This figure illustrates that
for this case, with the location of the gradient close to the wall, the behavior of the
wavenumbers can be understood like above, in other words, for the axial wavenumbers
the situation is very similar to uniform temperature.
Looking at the eigenfunctions depicted in Figure 6.6b it can be seen that the first
mode (µ = 1) is strongly influenced by the non-uniform temperature. As waves refract
towards the pregion with the lowest wave speed (the region with the lowest temperature,
since c20 = γRT0 ) the energy of each mode is mostly concentrated in the region near
the wall. This is also visualized in Figure 6.6c, where the time-averaged perturbation
(‘acoustic’) energy flux vector field is plotted on top of the pressure field.
It is also clear from Figure 6.6b that the influence of the temperature gradient
decreases for the higher order modes. Let us assume that a mode is made by two
waves bouncing back and forth between the duct wall and the center of the duct, these
waves having a propagation direction along the vector with axial component k and
radial component ±α. (This is a strongly simplified description, which is nevertheless
25
20
15
Im(k)
10
0
−10 −5 0 5 10 15 20
Re(k)
(a) Trajectory of axial wavenumbers, which move from from blue to red as
the temperature varies from T = 1 to T(r).
µ=1, k= 8.7 +0.0i µ=2, k= 7.4 +0.0i
2.5 4
2 3
1.5 2
1 1
0.5 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
µ=3, k= 4.4 +0.0i µ=4, k= 0.0 +6.1i
6 6
∞
4 4 0.95
2 2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(b) Magnitude of normalized eigenfunctions of the first six right running

modes for T = 1 (denoted by ∞) and T(r).
(c) Pressure field and time-averaged acoustic intensity vectors for the first mode.
Figure 6.6: Influence of the non-uniform temperature profile T(r) of (6.3) with q = 30 and
r 0 = 0.95 (see Figure 6.5a), ω = 10, m = 0 and Z = 100 + 1i.
0.2
0.18
20
0.16
0.14
0.12
15
Im(k)
0.1
0.08
Im(k)
0.06
10
0.04
0.02
0
5 0 5 10 15 20 25
Re(k)
0 5 10 15 20 25
Re(k)
Figure 6.7: Trajectory of axial wavenumbers, which move from from blue to red as the
temperature varies from T = 1 to T(r) of (6.3) with q = 50 and r 0 = 0.8, with ω = 20, m = 0
and Z = 100 + 1i. The inset is a vertical zoom of the area close to the real axis.
suitable for our purposes; a more precise description of the ray structure of circular
duct modes can be found in [27].) The higher radial orders, for which α becomes larger,
make a larger angle with the duct axis. Consequently, the wavefront of these modes
becomes more perpendicular to, and hence is less influenced by, the radial temperature
gradient.
For the next test cases we increase the frequency to ω = 20. We first consider a
temperature profile with q = 50 and r 0 = 0.8. Figure 6.7 shows that only the first order
mode moves away from the origin, contrary to what would be expected in view of the
average temperature increase. Note also that modes can change from cut-on to cut-off
or vice versa.
To investigate the effect of the location r 0 of the temperature gradient we now con-
sider a temperature profile with q = 50 and r 0 = {0.65, 0.8, 0.95}. Figure 6.8 illustrates
that the steep temperature gradient can produce a tunneling effect. Considering the
case for which r 0 = 0.95 we see that only the first radial-order mode is fully trapped
inside the low temperature region (it is zero elsewhere). For r 0 = 0.8 the first two
radial-order modes are trapped, and for r 0 = 0.65 the first three modes.
Finally we investigate the influence of the magnitude of the temperature gradient
by considering a profile with r 0 = 0.8 and q = {100, 22.3, 5}. The tunneling effect is
clearly the most pronounced for the steepest temperature gradients. However, it is
still present for the smaller temperature gradients as well. As before, for higher radial
orders the influence of the temperature profile decreases.
µ=1, k= 28.0 +0.0i µ=2, k= 25.1 +0.0i

6 4
3
4
2
2
1
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
µ=3, k= 19.1 +0.0i µ=4, k= 15.9 +0.0i
6 6
4 4
2 2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
µ=5, k= 14.4 +0.0i µ=6, k= 11.7 +0.0i
8 8
∞
6 6
0.95
4 4 0.8
0.65
2 2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Figure 6.8: Magnitude of normalized eigenfunctions of the first six right running modes as
a function of the radial coordinate, for T = 1 or the temperature profile of (6.3) with q = 50
and r 0 = {0.65, 0.8, 0.95}, with ω = 20, m = 0 and Z = 100 + 1i.
µ=1, k= 22.9 +0.1i µ=2, k= 17.4 +0.0i

4 4
2 2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
µ=3, k= 16.0 +0.0i µ=4, k= 14.6 +0.0i
5 10
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
µ=5, k= 12.0 +0.0i µ=6, k= 7.3 +0.0i
10 10
5 5
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
µ=7, k= 0.0 +7.6i µ=8, k= 0.0+13.8i
10 10
T=1
5 5 100.00
22.36
0 0 5.00
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Figure 6.9: Magnitude of normalized eigenfunctions of the first six right-running modes as
a function of the radial coordinate, for T = 1 or the temperature profile of (6.3) with q = 50
and r 0 = {0.65, 0.8, 0.95}, with ω = 20, m = 0 and Z = 100 + 1i.
§ 6.3 Role of source in transmission loss 97
6.3 Role of source in transmission loss

The transmission loss always depends on the (assumed) distribution of the modal
source amplitudes. Very often this source is not (exactly) known. A common approach
is then to create a source field that consists of all cut-on modes and a few cut-off modes,
where the individual modes have equal modal energy (often referred to as ‘equiparti-
tion of energy’) and random phases, in other words, they are treated as being incoher-
ent. In this section we describe how the expected value of the transmission loss can be
computed directly by considering the source modes separately.
Each source mode produces a different acoustic field in the entire duct. The com-
bined acoustic field due to all source modes is the sum of these fields. When the source
modes have different azimuthal orders, the power of the combined field is equal to the
sum of the powers of the separate fields, since the azimuthal eigenfunctions e i mθ are
orthogonal. However, with fixed m the fields couple radially (irrespective of any possi-
ble orthogonality of the radial eigenfunctions), as we will show next. Consequently, in
that case the power of the combined field is not equal to the sum of the powers of the
separate fields. We will show this in the following.
Suppose that we have a certain set of source mode amplitudes A µ at the duct en-
trance. If we keep only the j -th source mode and set all others to zero this gives rise to
a certain field at the duct end. Suppose also that the j -th source mode has an arbitrary
phase shift φ j . Let us denote the mean density by R , the mean flow in the axial direc-
tion by M , disturbance pressure, density, axial velocity, temperature and entropy fields
R 2π
respectively by p, ρ , u, τ and s. The time average over one period A = 2ωπ 0 ω A ( t)d t
of the product of two time-harmonic perturbations of the form of q̃ = q̂ e−iω t can be
computed as q̃ 1 q̃ 2 = 21 Re( q̂ 1 q̂∗2 ), where (·)∗ is used to denote the complex conjugate.
Consequently, the power flux (see Section 2.1.5) through a circular duct cross section
due to a single source mode j can be defined as
Zd ½· ¸· ¸∗ ¾
M
P j := π Re uj + ρj p j + MRu j + MR τ j s∗j r d r. (6.7)
R
0
The power of the combined field due to two source modes 1 and 2 with phase shifts e iφ1
and e iφ2 can then be computed as
Zd·³ ´ M³ ´¸·³ ´ ³ ´¸∗

i φ1 i φ2 i φ1 i φ2 i φ1 i φ2 i φ1 i φ2
P12 = π Re u1 e +u2 e + ρ1 e +ρ 2 e p1 e + p2 e + MR u 1 e +u2 e rdr
R
0
Zd ³ ´³ ´∗
+ π Re MR τ1 e iφ1 +τ2 e iφ2 s 1 e iφ1 + s 2 e iφ2 r d r
0
 
 Zd ½· ¸· ¸∗ ¾ 
i(φ2 −φ1 ) M
= P1 + P2 + π Re e u 2 + ρ 2 p 1 + MRu 1 + MR τ2 s 1 r d r
 R 
0
 
 Zd ½· ¸· ¸∗ ¾ 
M
+ π Re e i(φ1 −φ2 ) u 1 + ρ 1 p 2 + MRu 2 + MR τ1 s 2 r d r . (6.8)
 R 
0
More generally, the power of a combined field due to multiple source modes con-
sists of the sum of the powers of the individual fields plus their interaction terms, so
generally X
Ptot 6= P j . (6.9)
j
One option to obtain a representative transmission loss while taking into account the
random phases of the source modes would be to compute the power of the combined
field for multiple realizations of a set of random input phases, and then average the
results. This is computationally expensive for many realizations. However, by taking
the expected value of (6.8) the cross-terms vanish, since the phases are random. More
generally we have X
E {Ptot } = P j . (6.10)
j
Consequently, we can compute the expected value of the transmission loss directly
based on the output powers due to separate source modes.
Therefore, in the following we compute the transmission loss as
ÃP in !
m, j P m j
TL := 10 log10 P out
, (6.11)
m, j P m j
where the sum is over all cut-on modes for a given (ω, m) combination and add two extra
cut-off modes. The extra cut-off modes are included to capture a possible near field
that might be scattered into propagating modes after the first segment. We compute
the power flux through the duct entrance and exit due to the separate incident modes
by means of numerical quadrature. In order to produce easily reproducible results and
adhere at the same time to the ‘equipartition of energy’ assumption mentioned earlier,
we set the amplitudes of the individual source modes to one. This approximates
R1 equal
energy per mode, since the eigenfunctions are normalized according to 0 P 2 r d r = 1,
and the power through a hard-walled duct cross section carrying uniform flow is similar
P Re(kµ )
to µ ω | A µ |2 . We could have included a factor ω, but this term cancels in (6.11).
6.4 Liner with varying depth

As discussed in the introductory chapter, due to the geometrical constraint of the coni-
cally shaped tail of an aircraft, the APU exhaust duct typically has a liner depth which
varies in axial direction. Furthermore, strong radial temperature gradients may exist
due to the inflow of cold air from the sides. In this section we investigate the trans-
mission loss for some representative test cases based on this knowledge. We consider
a case (I) with uniform flow and temperature, a case (II) with non-uniform flow and
constant temperature, and a case (III) with both non-uniform temperature and flow.
For these test cases we consider frequencies ranging from ω = 1 to 20 (which corre-
sponds to dimensional frequencies of approximately 550–11000 Hz). The duct of length
1 m is divided into 12 segments; the first and last segments are hard-walled and have
a width of 1 cm. The remaining 10 segments have an equal width and a locally react-
ing liner with a depth d l which varies from 7 to 1 cm in the positive x-direction (i.e.
downstream). The liner is modeled as a Helmholtz resonator having a dimensional
impedance µ µ ¶¶
2π f d l
Z = ρ ∞ c ∞ Z0 + i cot , (6.12)
c∞
§ 6.4 Liner with varying depth 99
where f is the (dimensional) frequency and Z0 the dimensionless face sheet impedance,
here Z0 = 1.5. The imaginary part of the dimensionless impedance is plotted for all
frequencies in Figure 6.10.
The non-uniform mean flow profile that we use here is in dimensionless form
¡ ¢
M ( r ) = 0.3 tanh 20(1 − r ) . (6.13)
For the case with non-uniform temperature we use the profile (again in dimensionless
form) £ ¡ ¢¤
T ( r ) = N 1 + 12 tanh 30(0.8 − r ) , N = 1.10594, (6.14)
where the normalization constant N is computed such that this profile has the equiva-
lent mass flux as a constant temperature (T = 1) profile.
For the mode-matching the field is expressed as a sum of µmax = 15 modes in both
directions. Bilinear map based (BLM) mode-matching is used; classical mode-matching
gave identical results, which are not included here.
The transmission loss is computed based on all modes that are cut-on in the hard-
walled segments at a given frequency plus two extra cut-off modes. Figures 6.12, 6.13
and 6.14 show the number of cut-on modes for all (ω, m) combinations for case I, II and
III respectively (for m > ω cut-on modes are not to be expected, so they have not been
tested).
Figure 6.15a depicts the transmission loss versus the frequency for case I. For very
low frequencies the transmission loss is very small; this can be explained by the fact
that the lower radial order modes propagate mostly in the axial direction and hence do
not interact with the wall. It can be seen from Figure 6.10 that for ω = 7 and ω = 14 a
nearly hard-wall segment (a liner resonance) is close to the source plane. This matches
with the local minima at these frequencies in the transmission loss curve.
When the transmission loss is split out for individual m-values up to 5, as depicted
in Figure 6.15b, we see that the lowest m-values are least attenuated, and hence are of
the most interest for design calculations. The figure also shows that the transmission
loss decays far stronger with frequency if m is high. This trend can be explained on a
simplified ray-acoustics picture (for details of the ray structure of duct modes, see [27,
99]). The wave vector of length ω has axial and radial components k and α. For the
angle θ between the wave vector and the duct axis we have θ = arcsin(α/ω). Further-
more, the radial eigenvalue of the first radial-order mode αm1 = j 0m1 ≈ m, where j 0m1 is
the first zero of the derivative of the Bessel Jm function. Hence, θ increases with m
at fixed frequency, and consequently the mode is damped more strongly. Increasing ω
while keeping m fixed has the opposite effect.
The non-uniform flow results of case II, depicted in Figures 6.16a and 6.16b, are
very similar to the uniform flow results of case I. There is a slight increase of the trans-
mission loss, which can be explained by the fact that the acoustic field that propagates
downstream is refracted towards the wall, and hence damped more strongly.
Finally, we consider case III for which both the flow and the temperature are non-
uniform. Figure 6.17 shows again that the attenuation is relatively small for low fre-
quencies. Overall the transmission loss is of the order of 10 dB higher compared to
the uniform temperature cases. It appears that the refraction of the acoustic energy
towards the lower temperature region near the wall is very beneficial for the transmis-
sion loss. The local minima are now not visible anymore. As before, only the lowest
m-values are of most interest as they are the least attenuated.
f = 564.1369 Hz f = 1128.2737 Hz f = 1692.4106 Hz

5 5 5
0 0 0
−5 −5 −5
0 0.5 1 0 0.5 1 0 0.5 1
f = 2256.5475 Hz f = 2820.6843 Hz f = 3384.8212 Hz
5 5 5
0 0 0
−5 −5 −5
0 0.5 1 0 0.5 1 0 0.5 1
f = 3948.9581 Hz f = 4513.0949 Hz f = 5077.2318 Hz
5 5 5
0 0 0
−5 −5 −5
0 0.5 1 0 0.5 1 0 0.5 1
f = 5641.3687 Hz f = 6769.6424 Hz f = 7897.9161 Hz
5 5 5
0 0 0
−5 −5 −5
0 0.5 1 0 0.5 1 0 0.5 1
f = 9026.1899 Hz f = 10154.4636 Hz f = 11282.7373 Hz
5 5 5
0 0 0
−5 −5 −5
0 0.5 1 0 0.5 1 0 0.5 1
Figure 6.10: Imaginary part of dimensionless impedance versus axial location (m); the
liner depth d varies from 7 to 1 cm for all frequencies.
0.4 2
0.3 1.5
M(r)
T(r)
0.2 1
0.1 0.5
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r r
Figure 6.11: Non-uniform mean flow and temperature profiles of (6.13) and (6.14).
1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 2 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
5 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6 2 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 3 2 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
8 3 2 2 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
ω
9 3 3 2 2 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0
10 4 3 3 2 2 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0
12 4 4 3 3 2 2 2 1 1 1 1 0 0 0 0 0 0 0 0 0 0
14 5 4 4 3 3 3 2 2 2 1 1 1 0 0 0 0 0 0 0 0 0
16 6 5 5 4 4 3 3 3 2 2 1 1 1 1 0 0 0 0 0 0 0
18 6 6 5 5 4 4 4 3 3 2 2 2 1 1 1 1 0 0 0 0 0
20 7 6 6 5 5 5 4 4 3 3 3 2 2 2 1 1 1 1 0 0 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
m
Figure 6.12: Number of cut-on modes for all (ω, m) combinations, uniform flow and uniform
temperature (case I).
1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 2 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
5 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6 2 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 3 2 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
8 3 2 2 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
ω
9 3 3 2 2 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0
10 4 3 3 2 2 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0
12 4 4 3 3 2 2 2 1 1 1 1 0 0 0 0 0 0 0 0 0 0
14 5 4 4 4 3 3 2 2 2 1 1 1 1 0 0 0 0 0 0 0 0
16 6 5 5 4 4 3 3 3 2 2 2 1 1 1 1 0 0 0 0 0 0
18 6 6 5 5 4 4 4 3 3 2 2 2 2 1 1 1 1 0 0 0 0
20 7 6 6 5 5 5 4 4 3 3 3 2 2 2 1 1 1 1 1 0 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
m
Figure 6.13: Number of cut-on modes for all (ω, m) combinations, non-uniform flow and
uniform temperature (case II).
1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
5 2 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6 2 2 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 2 2 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0
8 3 2 2 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0
ω
9 3 2 2 2 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0
10 3 3 2 2 2 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0
12 4 3 3 2 2 2 2 1 1 1 1 1 1 0 0 0 0 0 0 0 0
14 4 4 3 3 3 2 2 2 1 1 1 1 1 1 1 0 0 0 0 0 0
16 5 4 4 3 3 3 2 2 2 2 2 1 1 1 1 1 1 0 0 0 0
18 5 5 5 4 4 3 3 3 2 2 2 2 2 1 1 1 1 1 1 0 0
20 6 6 5 5 4 4 3 3 3 2 2 2 2 2 2 2 1 1 1 1 1
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
m
Figure 6.14: Number of cut-on modes for all (ω, m) combinations, non-uniform temperature
and non-uniform flow (case III).
25
20
Transmission Loss (dB)
15
10
0
0 2 4 6 8 10 12 14 16 18 20
ω
(a) Transmission loss based on cut-on modes for all m’s.
100
0
90 1
2
3
80 4
5
70
60
50
40
30
20
10
0
0 2 4 6 8 10 12 14 16 18 20
ω
(b) Transmission loss based on cut-on modes for separate m = 0, . . . , 5.
Figure 6.15: Transmission loss versus Helmholtz number ω for case with uniform flow and
uniform temperature (case I).
25
20
15
10
0
0 2 4 6 8 10 12 14 16 18 20
ω
120
0
1
2
100 3
4
5
80
60
40
20
0
0 2 4 6 8 10 12 14 16 18 20
ω
Figure 6.16: Transmission loss versus Helmholtz number ω for case with non-uniform flow
and uniform temperature (case II).
40
35
30
25
20
15
10
0
0 2 4 6 8 10 12 14 16 18 20
ω
250
0
1
2
3
200 4
5
150
100
50
0
0 2 4 6 8 10 12 14 16 18 20
ω
Figure 6.17: Transmission loss versus Helmholtz number ω for case with both non-uniform
flow and temperature (case III).
Chapter 7
Concluding remarks
In this concluding chapter we formulate some open questions which could serve as the
starting point of future work.
In Section 2.5 we presented an approximate high-frequency solution of the Pridmore-
Brown equation with non-uniform flow in the form of a WKB approximation. As
pointed out, analytical solutions of this equation are rare, which is unfortunate as
they are very useful for comparison with numerical computations and can provide ana-
lytical insight. It is suggested that next to this WKB approximation a larger collection
of asymptotic results for various flow and temperature profiles will be derived. One ex-
ample of an ‘exact’ analytical solution that could be considered is the case of cross-wise
waves in parallel flow [45]; the frequency could be chosen such that k = 0 is a solu-
tion (hence the mode is x-independent), in which case the Pridmore-Brown equation
reduces to Bessel’s equation, and the other components follow directly from a system
like (4.25). Furthermore, next to the exact solution for a linear flow velocity profile,
also the analytical solution for an exponential boundary layer [25] could be considered.
In Section 3.1 we discussed the use of the collocation code COLNEW that is used
to numerically solve the boundary value problem consisting of the Pridmore-Brown
equation subject to the pertinent boundary conditions. Even though this code is very
efficient, for high order modes the computations become increasingly expensive due
to the highly oscillatory nature of the eigenfunctions. Specialized methods for highly
oscillatory ODEs exist, like Magnus methods [54, 122], which have been successfully
applied to Bessel’s and Airy’s equations. It might be interesting to investigate whether
the Pridmore-Brown equation is amenable to such methods.
In order to describe the acoustic field as a sum of eigenmodes, and use this in a
mode-matching approach, it is very important to find all relevant modes with certainty.
For simple cases of an absorbing wall combined with uniform flow and temperature
this can be achieved with the aid of a contour-integration based root-finding method,
as discussed in Section 3.3. The axial wavenumbers (eigenvalues) are the roots of an
analytic function that follows from substituting the solutions of the governing Bessel’s
equation into the wall boundary condition, hence this function is available in closed
form. For non-uniform flow and temperature we used a path-following approach, which
certainly improves certainty about finding all the modes, but it is never completely fail-
safe: the paths of two modes may come close to each other, which might result in the
path-following algorithm jumping from one mode to another, or the path may traverse
an unanticipated singular situation for a certain value of the continuation parameter.
107
108 Chapter 7. Concluding remarks
This raises the question whether a contour-integration based root-finding approach

could also be applied for the non-uniform flow case. In that case the roots are sought of
a function of the axial wavenumber that follows from substituting a numerical solution
of the Pridmore-Brown equation for a certain fixed k into the wall boundary condition.
At least two difficulties will need to be addressed: firstly, this function of k (the dis-
persion relation) is not analytic in the whole complex plane due to the continuous part
of the spectrum, so the integration contour needs to be deformed. Secondly, each func-
tion evaluation consists of the solution of an initial value problem (IVP), instead of a
Bessel-function evaluation. Even though the solution of this IVP is less expensive than
the solution of the BVP, it might render the whole procedure very costly.
In Chapter 4 a new mode-matching procedure for non-uniform flow was presented,
based on closed-form expressions for integrals of weighted combinations of products of
Pridmore-Brown modes. We noted that these integrals, which serve the same purpose
as a standard inner product in the classical mode-matching method, are in fact not
really inner products, so we referred to them as bilinear maps. By their construction
these bilinear maps can be considered as the natural generalizations of the Bessel
function product integrals. Although we did not encounter it in practice, the fact that
the bilinear map is not an inner product might lead to problems in the mode-matching
procedure (e.g. ill-conditioned scattering matrices, since the Pridmore-Brown modes
might not be independent enough as they can not be proven to be orthogonal). It would
therefore be an important step theoretically if a real inner product of Pridmore-Brown
modes that can be evaluated in closed-form could be constructed. We also mention here
that the bilinear maps might be useful to formulate a solvability condition for slowly
varying solutions of WKB-type.
As mentioned in Chapter 4 we checked whether our numerical mode-matching so-
lutions satisfied the edge condition (finite energy around the liner discontinuity) by
investigating the convergence of the found modal amplitudes. This was for locally re-
acting liners. We also mentioned that for non-locally reacting liners the situation can
be compared to that of a thin iris. For a thin iris it was shown [73] that the ratio of
the number of modes in each of the adjacent segments has to be chosen proportionally
to the ratio of the radii in order to satisfy the edge condition. For non-locally reacting
liners it is expected that the number of extra modes νadd has to be chosen according to
a similar criterion; however, this is not completely understood and therefore requires
further research.
Finally, we remark that it is of great practical interest to the aircraft industry to
develop models for cases with developing (i.e. axially varying) flow. Concerning the
possible application of mode-matching for such cases we note that the bilinear map in
the current form can not be used to match segments with different flow profiles, since
its construction was based on the fact that the governing ODE (including its coeffi-
cients, as for example the mean flow profile) is the same in both segments. Moreover,
it is unclear how the mean flow should be modeled; first of all, an analytical expression
(that satisfies the continuity and momentum equations) for a developing flow is re-
quired, and also the discontinuities at the interfaces between segments (i.e. the jumps
of the mean flow velocity along the streamlines) have to be handled appropriately. Ad-
ditionally, it is unclear how discontinuities in the mean flow should be accounted for in
the matching conditions; at any rate straightaway application of continuity of acoustic
pressure and velocity will be ill-advised.
Appendix A
Construction of bilinear map
For modal solutions of the form
[ρ 1 , p 1 , v1 ] = [R ( y, z), P ( y, z),U ( y, z) e x + V ( y, z) e y + W ( y, z) e z ] e i kx−iω t (A.1)
which are governed by (2.105) we have
−iΩP + i ρ 0 c20 kU + ρ 0 c20 (Vy + Wz ) = 0, (A.2a)

¡ ¢
−iρ 0 ΩU + ρ 0 u 0 y V + u 0 z W + i kP = 0, (A.2b)
−iρ 0 ΩV + P y = 0, (A.2c)
−iρ 0 ΩW + P z = 0, (A.2d)
where Ω = ω − ku 0 , and R follows directly from the other amplitudes, for example with
2.105c. When the individual equations in (A.2) are multiplied respectively by ψ, ξ, φ
and χ and added together it follows that
(−iΩρ 0 ξ + i kρ 0 c20 ψ)U + ρ 0 (−iΩφ + u 0 y ξ)V + ρ 0 (−iΩχ + u 0 z ξ)W − i(Ωψ − kξ)P

+ ρ 0 c20 Vy ψ + ρ 0 c20 Wz ψ + P y φ + P z χ = 0. (A.3)
A cross-wise divergence can be split off, i.e.
ρ 0 (−iΩξ + i kc20 ψ)U + ρ 0 (−iΩφ + u 0 y ξ − c20 ψ y )V

+ ρ 0 (−iΩχ + u 0 z ξ − c20 ψ z )W + (−iΩψ + i kξ − φ y − χ z )P
+ (ρ 0 c20 V ψ + P φ) y + (ρ 0 c20 W ψ + P χ) z = 0. (A.4)
Suppose that ψ, ξ, φ, χ satisfy the following ‘associated’ system

¡ ¢
ρ 0 −iΩ̃ξ + i k̃c20 ψ = 0, (A.5a)
¡ ¢
ρ 0 −iΩ̃φ + u 0 y ξ − c20 ψ y = 0, (A.5b)
¡ ¢
ρ 0 −iΩ̃χ + u 0 z ξ − c20 ψ z = 0, (A.5c)
−iΩ̃ψ + i k̃ξ − φ y − χ z = 0, (A.5d)
109
110 Appendix A. Construction of bilinear map
with k̃ and Ω̃ = ω − k̃u 0 . After multiplying these equations respectively by U , V , W and

P and subtracting them from (A.4) we have
(ρ 0 c20 V ψ + P φ) y + (ρ 0 c20 W ψ + P χ) z
£ ¤
= −i( k − k̃) ( u 0 ρ 0U + P )ξ + (ρ 0 c20U + u 0 P )ψ + ρ 0 u 0 (φV + χW ) . (A.6)
Note that (A.5) is equivalent to (A.2) after the transformation
k ≡ k̃, ρ 0 c20 Ω̃ψ ≡ P̃, Ω̃φ ≡ −Ṽ , Ω̃χ ≡ −W̃, ρ 0 Ω̃2 ξ ≡ k̃ P̃. (A.7)
This explains why in Section 4.1.2 we multiply the individual equations of (4.12) with
P /ρ 0 c20 , kP /ρ 0 Ω, −V and −W respectively to obtain the bilinear map.
Appendix B
Approximate Blasius solution
In this chapter we discuss the classical solution found by Blasius of the problem of
the development of a boundary layer along a flat plate, see for example [60]. In the
following we use a Cartesian coordinate system ( x, y) with velocity components u and
v. Consider a uniform flow U0 parallel to a semi-infinite flat plate oriented along the x
direction ( y = 0, x ∈ [0, ∞)), where y is the coordinate orthogonal to the plate. Assume
that we have a Reynolds number Re x , a boundary layer thickness δ( x) and a scaled y
coordinate η as
U0 x x y
Re x = , δ( x ) = p , η= , (B.1)
ν Re x δ ( x)
where ν is the kinematic viscosity. It can be shown that the velocity u( y) can be written
as
u = U0 f 0 (η), (B.2)
where the shape of boundary layer (equivalent at each x) follows from the solution f of
the following boundary value problem
1 00
f 000 + f f = 0,
2 (B.3)
f (0) = f 0 (0) = 0, f 0 (∞) = 1.
It appears that f 0 is well approximated by

h ³ í1/a(η)
f 0 (η) ≈ tanh (λη)a(η) , a(η) = a 0 + a 1 η,
(B.4)
a 0 = 1.4853, a 1 = 0.0555, λ = f 00 (0) ≈ 0.33206.
This is evidenced by Figure B.1, where we compare the numerical solution of (B.3)
with the approximation (B.4). The numerical solution of (B.3) is found with the Matlab
routine ode45 (which is based on the explicit Runge-Kutta (4,5) formula of Dormand
and Prince), where we used a relative error tolerance of 10−13 . It can be seen that the
difference with the approximation (B.4) is of the order 10−4 .
We now switch to cylindrical coordinates and make everything dimensionless. Based
on the approximation of B.4 we choose the mean flow profile M ( r ) at a given axial loca-
111
112 Appendix B. Approximate Blasius solution
−4
x 10
1 2
0.8 1
difference
f’(η)
0.6
0
0.4 ode45
−1
0.2 tanh
0 −2
0 2 4 6 8 10 0 2 4 6 8 10
η η
(a) Blasius boundary layer profile. (b) Difference.
Figure B.1: Numerical solution of (B.3) compared to the approximation (B.4).
tion ξ (non-dimensional) away from the plate edge as

±
h ³ í1 b(r;ξ) ±p ±p
b ( r ;ξ )
M ( r ; x) = M0 tanh a( r ; ξ) , a ( r ; ξ) = B 0 r ξ, b ( r ; ξ ) = a 0 + B 1 r ξ
q ± q ±
with ξ = x/ d, B0 = λ U0 d ν, B1 = a 1 U0 d ν, U0 = M0 c ∞ ,
(B.5)
where d , U0 , c ∞ and ν are the duct radius, mean flow velocity, reference sound speed
and kinematic viscosity (all dimensional).
Bibliography
[1] M. J. Ablowitz and A. S. Fokas. Complex Variables: Introduction and Applica-

tions. 2nd ed. Cambridge University Press, 2003.
[2] J. Allard and N. Atalla. Propagation of Sound in Porous Media: Modelling Sound
Absorbing Materials. 2nd ed. John Wiley & Sons, 2009.
[3] J. Allard and Y. Champoux. “New Empirical Equations for Sound Propagation
in Rigid Frame Fibrous Materials”. J. Acoust. Soc. Am. 91 (1992), pp. 3346–
3353.
[4] E. Allgower and K. Georg. Introduction to Numerical Continuation Methods.
SIAM, 2003.
[5] D. E. Amos. “Algorithm 644: A Portable Package for Bessel Functions of a Com-
plex Argument and Nonnegative Order”. ACM Trans. Math. Softw. 12.3 (Sept.
1986), pp. 265–273.
[6] B. van Antwerpen, R. Leneveu, S. Caro, and P. Ferrante. “New advances in the
use of ACTRAN/TM for nacelle simulations”. 14th AIAA/CEAS Aeroacoustics
Conference, Vancouver, British Columbia, Canada. AIAA 2008-2827. May 5–7,
2008.
[7] U. Ascher, J. Christiansen, and R. D. Russell. “Collocation Software for Boundary-
Value ODEs”. ACM Trans. Math. Softw. 7.2 (June 1981), pp. 209–222.
[8] U. Ascher, J. Christiansen, and R. Russel. “Collocation software for Boundary-
Value ODEs”. ACM Transactions on Mathematical Software 7.2 (June 1981),
pp. 209–222.
[9] U. M. Ascher, R. M. M. Mattheij, and R. D. Russell. Numerical solution of bound-
ary value problems for ordinary differential equations. SIAM, 1995.
[10] R. J. Astley and W. Eversman. “A finite element method for transmission in non-
uniform ducts without flow: Comparison with the method of weighted residu-
als”. Journal of Sound and Vibration 57.3 (1978), pp. 367–388.
[11] R. Astley, V. Hii, and G. Gabard. “A computational mode-matching aproach for
propagation in three-dimensional ducts with flow”. 12th AIAA/CEAS Aeroa-
coustics Conference, Cambridge(MA), USA. AIAA 2006-2528. May 2006.
[12] R. J. Astley, R. Sugimoto, and P. Mustafi. “Computational aero-acoustics for
fan duct propagation and radiation. Current status and application to turbofan
liner optimisation”. Journal of Sound and Vibration 330.16 (2011), pp. 3832–
3845.
113
114 Bibliography
[13] O. V. Atassi. “Computing the sound power in non-uniform flow”. Journal of

Sound and Vibration 266.1 (2003), pp. 75–92.
[14] O. V. Atassi. “Propagation and stability of vorticity–entropy waves in a non-
uniform flow”. Journal of Fluid Mechanics 575.1 (2007), pp. 149–176.
[15] G. Bader and U. Ascher. “A New Basis Implementation for a Mixed Order
Boundary Value ODE Solver”. SIAM J. Sci. and Stat. Comput. 8.4 (July 1987),
pp. 483–500.
[16] A. Bihhadi and Y. Gervais. “Analysis of the distribution and attenuation of
acoustic energy flux in lined duct containing inhomogeneous medium by the
finite difference method”. Acta Acustica united with Acustica 83.1 (1997), pp. 1–
12.
[17] D. T. Blackstock. Fundamentals of physical acoustics. John Wiley & Sons, Inc.,
2000.
[18] C. J. Bouwkamp. “A note on singularities occuring at sharp edges in electro-
magnetic diffraction theory”. Physica XII 7 (Oct. 1946), pp. 467–474.
[19] G. Boyer, E. Piot, and J.-P. Brazier. “Theoretical investigation of hydrodynamic
surface mode in a lined duct with sheared flow and comparison with experi-
ment”. Journal of Sound and Vibration 330 (2011), pp. 1793–1809.
[20] E. J. Brambley. “A well-posed boundary condition for acoustic liners in straight
ducts with flow”. AIAA Journal 49 (2011), pp. 1272–1282.
[21] E. J. Brambley. “Fundamental problems with the model of uniform flow over
acoustic linings”. Journal of Sound and Vibration 322.4 (2009), pp. 1026–1037.
[22] E. J. Brambley and N. Peake. “Propagation of unsteady disturbances in a slowly
varying duct with mean swirling flow”. J. Fluid Mech. 596 (2008), pp. 387–412.
[23] E. J. Brambley, M. Darau, and S. W. Rienstra. “The critical layer in linear-shear
boundary layers over acoustic linings”. Journal of Fluid Mechanics 710 (2012),
pp. 545–568.
[24] C. J. Brooks and A. McAlpine. “Sound transmission in ducts with sheared mean
flow”. 13th AIAA/CEAS Aeroacoustics Conference, Rome, Italy. AIAA 2007-3545.
May 20–22, 2007.
[25] L. M. B. C. Campos and P. G. T. A. Serrão. “On the acoustics of an exponential
boundary layer”. Philosophical Transactions of the Royal Society of London.
Series A: Mathematical, Physical and Engineering Sciences 356.1746 (1998),
pp. 2335–2378.
[26] M. Carpentier and A. Dos Santos. “Solution of equations involving analytic
functions”. Journal of Computational Physics 45.2 (1982), pp. 210 –220.
[27] C. J. Chapman. “Sound radiation from a cylindrical duct. Part 1. Ray structure
of the duct modes and of the external field”. Journal of Fluid Mechanics 281
(1994), pp. 293–311.
[28] A. J. Cooper. “Effect of mean entropy on unsteady disturbance propagation in a
slowly varying duct with mean swirling flow”. Journal of Sound and Vibration
291 (2006), pp. 779–801.
[29] A. J. Cooper and N. Peake. “Propagation of unsteady disturbances in a slowly
varying duct with mean swirling flow”. J. Fluid Mech. 445 (2001), pp. 207–234.
Bibliography 115
[30] D. G. Crighton. “Asymptotics—an indispensable complement to thought, com-

putation and experiment in applied mathematical modeling”. Proceedings of the
7th European Conference on Mathematics in Industry. 1994, pp. 3–19.
[31] D. G. Crighton. “Basic principles of aerodynamic noise generation”. Progress in
Aerospace Sciences 16.1 (1975), pp. 31–96.
[32] A. Cummings. “Sound transmission at sudden area expansions in circular ducts,
with superimposed mean flow”. Journal of Sound and Vibration 38.1 (1975),
pp. 149–155.
[33] M. Darau. “Acoustic liner – mean flow interaction”. PhD thesis. Eindhoven Uni-
versity of Technology, 2012.
[34] B. Davies. “Locating the zeros of an analytic function”. Journal of Computa-
tional Physics 66.1 (1986), pp. 36 –49.
[35] M. E. Delany and E. N. Bazley. “Acoustical properties of fibrous absorbent ma-
terials”. Applied Acoustics 3 (1970), pp. 105–116.
[36] L. Delves and J. Lyness. “A numerical method for locating the zeros of an ana-
lytic function”. Math. Comp. 21 (1966), pp. 543–560.
[37] S. Dequand. “Duct Aeroacoustics: from Technological Application to the Flute”.
PhD thesis. Technische Universiteit Eindhoven, 2000.
[38] Directive 2003/10/EC of the European Parliament and of the Council. Minimum
health and safety requirements regarding the exposure of workers to the risks
arising from physical agents (noise). June 2003.
[39] A. P. Dowling and J. E. Ffowcs Williams. Sound and sources of sound. Ellis
Horwood Limited, 1983.
[40] D. S. Jones. The Theory of Electromagnetism. Pergamon Press, 1964, pp. 566–
569.
[41] W. Eversman. “Aeroacoustics of Flight Vehicles”. Ed. by H. H. Hubbard. Vol. 2:
Noise Control. Acoustical Society of America, 1995. Chap. 13: Theoretical mod-
els for duct acoustic propagation and radiation.
[42] W. Eversman. “Computation of axial and transverse wave numbers for uniform
two-dimensional ducts with flow using a numerical integration scheme”. Jour-
nal of Sound Vibration 41 (July 1975), pp. 252–255.
[43] Free Field Technologies SA. Actran 13 User’s Guide. Vol. 1 – Installation, Oper-
ations, Theory and Utilities. Aug. 2012.
[44] N. Fröman and P. Fröman. JWKB approximation: contributions to the theory.
Amsterdam: North-Holland, 1965.
[45] G. Gabard. “A comparison of impedance boundary conditions for flow acoustics”.
Journal of Sound and Vibration 332.4 (2013), pp. 714 –724.
[46] M. E. Goldstein. Aeroacoustics. New York: McGraw-Hill International Book Co.,
1976.
[47] V. V. Golubev and H. M. Atassi. “Acoustic-vorticity waves in swirling flows”.
Journal of Sound and Vibration 209.2 (1998), pp. 203–222.
[48] A. W. Guess. “Calculation of perforated plate liner parameters from specified
acoustic resistance and reactance”. Journal of Sound and Vibration 40.1 (1975).
116 Bibliography
[49] http://www.netlib.org/ode/colnew.f.
[50] ICAO—International Civil Aviation Organization. Annex 16—Environmental
Protection. 5th. Vol. I—Aircraft Noise. Attachment C.—Guidelines for noise cer-
tification of installed auxiliary power units (APU) and associated aircraft sys-
tems during ground operation. July 2008.
[51] K. U. Ingard. “Locally and nonlocally reacting flexible porous layers—a compar-
ison of acoustical properties”. Transactions of the American Society of Mechani-
cal Engineers. 1980.
[52] K. Ingard. “Influence of fluid motion past a plane boundary on sound reflection,
absorption, and transmission”. J. Acoust. Soc. Am. 31 (1959), pp. 1035–1036.
[53] N. Ioakimidis and E. Anastasselou. “A modification of the Delves-Lyness method
for locating the zeros of analytic functions”. Journal of Computational Physics
59.3 (1985), pp. 490 –492.
[54] A. Iserles, H. Z. Munthe-Kaas, S. P. Nørsett, and A. Zanna. “Lie-group meth-
ods”. Acta Numerica 2000 9 (2000), pp. 215–365.
[55] D. Kalman. “A matrix proof of Newton’s identities”. Mathematics Magazine
(2000), pp. 313–315.
[56] A. Kapur and P. Mungur. “On the propagation of sound in a rectangular duct
with gradients of mean flow and temperature in both transverse directions”.
Journal of Sound and Vibration 23.3 (1972), pp. 401 –404.
[57] W. Koch and W. Möhring. “Eigensolutions for liners in uniform mean flow ducts”.
AIAA Journal 21.2 (1983), pp. 200–213.
[58] L. S. G. Kovásznay. “Turbulence in Supersonic Flow”. Journal of the Aeronauti-
cal Sciences (Institute of the Aeronautical Sciences) 20.10 (1953), pp. 657–674.
[59] E. Kreyszig. Introductory Functional Analysis with Applications. Wiley Classics
Library. John Wiley & Sons, 1989.
[60] P. K. Kundu and I. M. Cohen. Fluid Mechanics. 3rd ed. Elsevier Academic Press,
2004.
[61] D. L. Lansing and W. E. Zorumski. “Effects of wall admittance changes on duct
transmission and radiation of sound”. Journal of Sound and Vibration 27.1
(1973), pp. 85–100.
[62] M. Lavieille, D. Brown, and F. Vieuille. “Numerical Modeling and Experimen-
tal Validation of the Acoustic Efficiency of Treated Ducts on an Aircraft Auxil-
iary Power System”. 17th AIAA/CEAS Aeroacoustics Conference, Portland(OR),
USA. AIAA 2011-2810. June 2011.
[63] W. Lazeroms. “Sound propagation in slowly varying lined ducts with tempera-
ture gradients”. MSc thesis. Eindhoven University of Technology, Dec. 2010.
[64] C. Legendre, G. Lielens, and J.-P. Coyette. “Sound Propagation in a Sheared
Flow Based on Fluctuating Total Enthalpy as Generalized Acoustic Variable”.
Proceedings of the Internoise 2012/ASME NCAD meeting. IN12-838. American
Society of Mechanical Engineers. Aug. 19–22, 2012, pp. 279–284.
[65] L. Li. “Formulation and comparison of two recursive matrix algorithms for mod-
eling layered diffraction gratings”. J. Opt. Soc. Am. A 13.5 (May 1996), pp. 1024–
1035.
Bibliography 117
[66] P. H. Masterman and P. J. B. Clarricoats. “Computer field-matching solution of

waveguide transverse discontinuities”. Proceedings of the Institution of Electri-
cal Engineers. Vol. 118. 1. 1971, pp. 51–63.
[67] R. M. M. Mattheij, S. W. Rienstra, and J. H. M. ten Thije Boonkkamp. Partial
differential equations. Modeling, analysis, computation. SIAM Monographs on
Mathematical Modeling and Computation. Society for Industrial and Applied
Mathematics (SIAM), 2005.
[68] A. McAlpine, R. J. Astley, V. J. T. Hii, N. J. Baker, and A. J. Kempton. “Acoustic
scattering by an axially-segmented turbofan inlet duct liner at supersonic fan
speeds”. Journal of Sound and Vibration 294 (2006).
[69] A. McAlpine, M. Wright, H. Batard, and S. Thezelais. “Finite/Boundary Ele-
ment Assessment of a Turbofan Spliced Intake Liner at Supersonic Fan Operat-
ing Conditions”. 9th AIAA/CEAS Aeroacoustics Conference and Exhibit. AIAA
2003-3305. May 12–14, 2003.
[70] J. Meixner. “The behavior of electromagnetic fields at edges”. IEEE Transac-
tions on Antennas and Propagation 20.4 (1972), pp. 442–446.
[71] Y. Miki. “Acoustical properties of porous materials. Modifications of Delany-
Bazley models”. Journal of the Acoustical Society of Japan 11.1 (1990), pp. 19–
24.
[72] Y. Miki. “Acoustical properties of porous materials. Generalizations of emperi-
cal models”. Journal of the Acoustical Society of Japan 11.1 (1990), pp. 25–28.
[73] R. Mittra, T. Itoh, and S. Li. “Analytical and numerical studies of the rela-
tive convergence phenomenon arising in the solution of an integral equation by
the moment method”. IEEE Transactions on Microwave Theory and Techniques
20.2 (1972), pp. 96–104.
[74] R. Mittra and S.-W. Lee. Analytical techniques in the theory of guided waves.
Macmillan New York, 1971.
[75] C. L. Morfey. “Acoustic energy in non-uniform flows”. Journal of Sound and
Vibration 14.2 (1971), pp. 159 –170.
[76] P. M. Morse and K. U. Ingard. Theoretical acoustics. International series in pure
and applied physics. McGraw Hill, 1968.
[77] R. E. Motsinger and R. E. Kraft. “Aeroacoustics of Flight Vehicles”. Ed. by H. H.
Hubbard. Vol. 2: Noise Control. Acoustical Society of America, 1995. Chap. 14:
Design and performance of duct acoustic treatment.
[78] P. Mungur and G. M. L. Gladwell. “Acoustic wave propagation in a sheared fluid
contained in a duct”. Journal of Sound and Vibration 9.1 (1969), pp. 28–48.
[79] M. K. Myers. “An exact energy corollary for homentropic flow”. Journal of Sound
and Vibration 109.2 (1986), pp. 277 –284.
[80] M. K. Myers. “On the acoustic boundary condition in the presence of flow”. J.
Sound Vib. 71 (1980), pp. 429–434.
[81] M. K. Myers. “Transport of energy by disturbances in arbitrary steady flows”.
Journal of Fluid Mechanics 226 (1991), pp. 383–400.
[82] A. H. Nayfeh. Introduction to Perturbation Techniques. John Wiley and Sons,
Inc., 1993.
118 Bibliography
[83] A. H. Nayfeh and J. Sun. “Effect of transverse velocity and temperature gra-
dients on sound attenuation in two-dimensional ducts”. Journal of Sound and
vibration 34.4 (1974), pp. 505–517.
[84] A. H. Nayfeh, J. E. Kaiser, and D. P. Telionis. “Acoustics of Aircraf Engine-Duct
Systems”. AIAA Journal 13.2 (Feb. 1975), pp. 130–153.
[85] E. Nielsen. Airport noise and emission regulations for Copenhagen Airport. Boe-
ing. URL: http://www.boeing.com/commercial/noise/copenhagen.html#
apu Nov. 1, 2013.
[86] R. J. Nijboer and P. Sijtsma. “Sound diffraction by the splitter of a turbofan
engine”. 6th International Congress on Sound and Vibration, Lyngby, Denmark.
Also published as NLR report NLR-TP-99133. July 5–8, 1999, pp. 415–422.
[87] R. Nijboer. “Eigenvalues and eigenfunctions of ducted swirling flows”. 7th AIAA/
CEAS Aeroacoustics Conference and Exhibit, Maastricht, The Netherlands. AIAA
2001-2178. May 28–30, 2001.
[88] B. E. Nilsson. “The propagation of sound in cylindrical ducts with mean flow
and bulkreacting lining”. Also published as 4 seperate parts in: IMA Journal of
Applied Mathematics, 1980-1981. PhD thesis. University of Gothenburg, Swe-
den, 1980.
[89] T. Nodé-Langlois, P. Sijtsma, S. Moal, and F. Vieuille. “Modelling of Non-Locally
Reacting Acoustic Treatments for Aircraft Ramp Noise Reduction”. 16th AIAA/
CEAS Aeroacoustics Conference, Stockholm, Sweden. AIAA 2010-3769. June
2010.
[90] M. Oppeneer, W. M. Lazeroms, S. W. Rienstra, P. Sijtsma, and R. M. Mattheij.
“Acoustic modes in a duct with slowly varying impedance and non-uniform
mean flow and temperature”. 17th AIAA/CEAS Aeroacoustics Conference, Port-
land(OR), USA. AIAA 2011-2871. June 2011.
[91] M. Oppeneer, S. W. Rienstra, and P. Sijtsma. “Efficient Mode-Matching Based
on Closed Form Integrals of Pridmore-Brown Modes”. 19th AIAA/CEAS Aeroa-
coustics Conference, Berlin, Germany. AIAA 2013-2172. May 2013.
[92] N. C. Ovenden and S. W. Rienstra. “Mode-matching strategies in slowly varying
engine ducts”. AIAA journal 42.9 (2004), pp. 1832–1840.
[93] N. Peake and A. J. Cooper. “Acoustic propagation in ducts with slowly varying
elliptic cross-section”. J. Sound Vib. 243 (2001), pp. 381–401.
[94] A. D. Pierce. Acoustics. An Introduction to Its Physical Principles and Applica-
tions. Acoustical Society of America, 1989.
[95] R. Piessens, E. De Doncker-Kapenga, and C. Überhuber. QUADPACK: a sub-
routine package for automatic integration. Springer, 1983. URL: http://www.
netlib.org/quadpack/.
[96] M. Pisarenco, J. Maubach, I. Setija, and R. Mattheij. “Modified S-matrix algo-
rithm for the aperiodic Fourier modal method in contrast-field formulation”. J.
Opt. Soc. Am. A 28.7 (July 2011), pp. 1364–1371.
[97] D. Pridmore-Brown. “Sound propagation in a fluid flowing through an attenu-
ating duct”. J. Fluid Mech. 4 (1958), pp. 393–406.
Bibliography 119
[98] NIST Digital Library of Mathematical Functions. http://dlmf.nist.gov/, Release

1.0.6 of 2013-05-06. Online companion to NIST Handbook of Mathematical
Functions. URL: http://dlmf.nist.gov/.
[99] E. J. Rice, M. F. Heidmann, and T. G. Sofrin. “Modal propagation angles in a
cylindrical duct with flow and their relation to sound radiation”. 17th AIAA
Aerospace Sciences Meeting, New Orleans. AIAA 79-0183. Jan. 15–17, 1979.
[100] S. W. Rienstra. “A classification of duct modes based on surface waves”. Wave
Motion 37.2 (2003), pp. 119–135.
[101] S. W. Rienstra. “Impedance Models in Time Domain including the Extended
Helmholtz Resonator Model”. 12th AIAA/CEAS Aeroacoustics Conference. AIAA
2006-2686. Cambridge MA, 2006.
[102] S. W. Rienstra and M. Darau. “Boundary Layer Thickness Effects of the Hy-
drodynamic Instability along an Impedance Wall”. J. Fluid Mech. 671 (2011),
pp. 559–573.
[103] S. W. Rienstra. “Sound propagation in slowly varying lined flow ducts of arbi-
trary cross-section”. Journal of Fluid Mechanics 495 (2003), pp. 157–173.
[104] S. W. Rienstra. “Sound transmission in slowly varying circular and annular
ducts with flow”. Journal of Fluid Mechanics 380 (1999), pp. 279–296.
[105] S. W. Rienstra. “Acoustic radiation from a semi-infinite annular duct in a uni-
form subsonic mean flow”. Journal of Sound and Vibration 94.2 (1984), pp. 267–
288.
[106] S. W. Rienstra. “Contributions to the theory of sound propagation in ducts with
bulk-reacting lining”. J. Acoust. Soc. Am. 77.5 (May 1985), pp. 1681 –1685.
[107] S. W. Rienstra and W. Eversman. “A numerical comparison between multiple-
scales and finite-element solution for sound propagation in lined flow ducts”.
Journal of Fluid Mechanics 437 (2001), pp. 367–384.
[108] S. W. Rienstra and A. Hirschberg. An Introduction to Acoustics. Aug. 26, 2013.
URL : http://www.win.tue.nl/~sjoerdr/papers/boek.pdf.
[109] A. Shavit and C. Gutfinger. Thermodynamics. From concepts to applications.

Pearson Education Limited, 1995.
[110] P. Sijtsma and H. van der Wal. “Modelling a Spiralling Type of Non-Locally
Reacting Liner”. 9th AIAA/CEAS Aeroacoustics Conference and Exhibit, Hilton
Head, South Carolina. AIAA 2003-3308. 2003.
[111] T. Sofrin and D. Cicon. “Ducted fan noise propagation in non-uniform flow. II-
Wave equation solution”. AIAA 11th Aeroacoustics Conference, Palo Alto (CA),
USA. AIAA 87-2702. Oct. 1987.
[112] T. Sofrin and D. Cicon. “Ducted fan noise propagation in non-uniform flow. I-
Test background and simplified model”. AIAA 11th Aeroacoustics Conference,
Palo Alto (CA), USA. AIAA 87-2701. Oct. 1987.
[113] C. K. W. Tam and L. Auriault. “The wave modes in ducted swirling flow”. Jour-
nal of Fluid Mechanics 371 (1998), pp. 1–20.
[114] C. K. W. Tam and J. C. Webb. “Dispersion-relation-preserving finite difference
schemes for computational acoustics”. Journal of computational physics 107.2
(1993), pp. 262–281.
120 Bibliography
[115] G. Teschl. Ordinary Differential Equations and Dynamical Systems. American

Mathematical Society, 2012. URL: http://www.mat.univie.ac.at/~gerald/
ftp/book-ode/.
[116] The Jet Engine. 5th ed. Rolls-Royce plc, 1996.
[117] J. M. Tyler and T. G. Sofrin. “Axial flow compressor noise studies”. Transactions
of the Society of Automotive Engineers 70 (1962), pp. 309–332.
[118] G. G. Vilenski. “Mode matching in engine ducts with vortical flows”. 12th AIAA/
CEAS Aeroacoustics Conference, Boston, MA, USA. AIAA 2006-2584. May 8–10,
2006.
[119] G. G. Vilenski and S. W. Rienstra. “Numerical study of acoustic modes in ducted
shear flow”. Journal of Sound and Vibration 307 (2007), pp. 610–626.
[120] G. G. Vilenski and S. W. Rienstra. “On hydrodynamic and acoustic modes in
a ducted shear flow with wall lining”. Journal of Fluid Mechanics 583 (2007),
pp. 45–70.
[121] G. N. Watson. A Treatise on the Theory of Bessel Functions. Cambridge Univer-
sity Press, 1995.
[122] J. Wensch, M. Däne, W. Hergert, and A. Ernst. “The solution of stationary ODE
problems in quantum mechanics by Magnus methods with stepsize control”.
Computer physics communications 160.2 (2004), pp. 129–139.
Index
absolute temperature, 12 at constant pressure, 12

APU exhaust duct, 1, 6 at constant volume, 12
argument principle, 46 ratio of heat capacities, 12
Auxiliary Power Unit (APU), 1, 123 heat flux vector, 13
Helmholtz number, 87
bilinear map, 9, 52, 54, 56 Helmholtz resonator array, 4, 30
boundary layer, 3, 111 homentropic flow, 15, 18
Blasius solution, 111
bulk absorber, 4 ideal gas, 13
impedance, 4, 25
COLNEW, 8, 39 Ingard-Myers boundary condition, 25
companion matrix, 46 internal energy, 11
compressibility
adiabatic, 28 Linearized Euler Equations, 21, 37
isothermal, 28 liner, 4
continuation, see path-following locally reacting, 4, 25
convective derivative, 14 non-locally reacting, 4, 25
critical layer, 33
cut-off frequency, 25 Mach number, 3
cut-off mode, 3, 25 mode order, see also order, radial
cut-on mode, 3, 25 mode shape, see also eigenfunction, 24
mode, slowly varying, 77
deviatoric stress tensor, 13 mode-matching method, 7, 8, 51, 124
duct acoustics, 1, 2, 123 multiple scales, 75
duct mode, 2, 23 Myers’ Energy Corollary, 22, 71
Eckert number, 17 Navier-Stokes equations, 14

edge condition, 62 Newton’s identities, 46
eigenfunction, 2 noise source, 9, 97
eigenvalue, see also duct mode
enthalpy, 12 order, radial, 24
entropy, 12
equation of state, 13 Péclet number, 17
equivalent fluid model, 26, 27 parallel flow, 18
Euler equations, 15 path-following, 8, 40
perfect gas, 13
face sheet, 4 porous material, 4
fundamental thermodynamic relation, 12 power sums, 46
Prandtl number, 17
heat, 11 prediction-correction scheme, 8, 41
heat capacity, 12 Pridmore-Brown equation, 3, 24, 33
121
122 Index
quantization condition, 35
ramp noise, 1, 123

Redheffer star product, 65
refraction of sound, 3, 9, 92
Reynolds number, 16
root-finding, 8, 44
scattering matrix formalism, 7, 9, 64

shear flow, 3
shooting method, 39
solvability condition, 79, 80
Sound Pressure Level (SPL), 19
sound speed, 12
specific gas constant, 13
surface wave, 8, 90
temperature gradient, 3
total energy, 13
transmission loss, 7, 97
turbofan, 3
uniform flow, 3
wavenumber
axial, 24
azimuthal, 24
circumferential, 24
WKB method, 33, 75, 77
work, 11
Summary
Sound propagation in lined ducts with parallel flow

Over the past decades aeroacoustic research for lined flow ducts (duct acoustics) was
primarily aimed at reducing the noise levels in inlet and exhaust ducts of the main
turbofan engines of aircraft. Recently the so-called ramp noise—to which workers are
exposed while the aircraft is parked on the ground—has been given more attention, as
regulations have become more stringent.
A large contribution of the ramp noise comes from the auxiliary power unit (APU),
which is a turbine engine in the tail of the aircraft that produces power while the
main engines are switched off. The APU exhaust duct, which is typically straight and
circular-cylindrical, carries a non-uniform flow with strong temperature gradients due
to the inflow of cold air from the sides. To reduce the noise coming out of the APU
exhaust duct, its walls are generally treated with an acoustically absorbing lining,
which typically is a locally reacting liner in the form of a honeycomb structure, or a non-
locally reacting liner in the form of a series of hollow annular segments, each covered
by a porous facing sheet. The liner depth typically varies along the axial direction due
to the conical geometry of the aircraft fuselage tail.
To find an optimum for the trade-off between minimal weight and noise emissions
there is a need for more insight in the main noise source propagation mechanisms and
accurate design tools. This motivates the main goal of this work, which is to provide
semi-analytical solutions for the propagation of sound in lined flow ducts with parallel
flow and strong thermal gradients, and axially varying lining. We formulate the sound
propagation problem in terms of acoustic duct modes. These modes are solutions of the
Pridmore-Brown equation; together with suitable boundary conditions this equation
forms a boundary value problem that describes small (acoustic) perturbations of a par-
allel mean flow with transverse temperature gradients. We present some asymptotic
solutions of the Pridmore-Brown equation, which are used to validate our numerical
approach.
For non-uniform flow and temperature we solve the Pridmore-Brown equation
numerically in a robust and efficient manner with the aid of the COLNEW code, which
solves boundary value problems in ordinary differential equations by collocation. To
ensure that we find all relevant solutions we use a path-following (or continuation)
approach, where we start from an easy solution (e.g. uniform flow and temperature)
and trace the solution when the relevant problem parameters are varied to the val-
ues of interest. Our path-following approach is refined by using a prediction-correction
scheme, where the prediction is found by linear extrapolation of the previous solutions.
The correction step is then an updated solution by COLNEW, with the prediction as
123
124 Summary
the starting value.

For uniform flow and temperature an exact dispersion relation is available (in terms
of Bessel functions), both for locally and non-locally reacting boundary conditions. This
enables us to find the eigenvalues (modal axial wavenumbers) as the roots of an ana-
lytic function. To ensure that all relevant eigenvalues are found numerically we employ
the root-finding method of Davies (a derivative-free adaptation of the method of Delves
and Lyness), which is based on complex contour integration. This method guarantees
that all roots inside a given area of the complex plane are found, whereas a Newton
method only converges to all of the roots if it is started from sufficiently close initial
guesses.
A suitable approach to compute sound propagation for a geometry that consists
of several annular segments with different liner properties is mode-matching. Here,
the acoustic field is expressed as a summation of modes, and the unknown modal am-
plitudes are found by imposing continuity of the acoustic pressure and axial velocity
fields at the axial location of the liner discontinuity. The matching is based on the
projection (by inner products) of the continuity conditions onto a suitable set of test
functions. The scattering matrix formalism is used to compute the combined propaga-
tion through multiple interfaces. This approach is numerically stable because it only
considers the propagation of cut-off modes in the direction of exponential decay.
For uniform flow the necessary inner products can be found in closed form if Bessel
functions are used as test functions (the classical approach). For non-uniform flow how-
ever, numerical quadrature is required to compute the inner products for the classical
approach. In this dissertation a new bilinear form is presented, which is an integral of
a weighted combination of products of Pridmore-Brown modes that can be evaluated
in closed form. As this bilinear form resembles an inner product of Pridmore-Brown
functions, it can be used instead of the standard inner product to form the basis of a
new mode-matching procedure. Since this new approach avoids the inherently inac-
curate numerical quadrature of oscillating functions we conclude that it is both more
accurate and cheaper than the classical approach, and demonstrate this by numerical
experiments.
The availability of the numerical Pridmore-Brown eigenfunctions enables us to con-
sider asymptotic solutions in the form of slowly varying modes of WKB type for the case
of an impedance that is slowly varying in the axial direction due to varying liner depth.
We compare these asymptotic results with mode-matching results and illustrate that
for most APU exhaust duct configurations, for which the impedance is not slowly vary-
ing due to liner resonances, the asymptotic results are not applicable.
Finally, the developed methods are applied to illustrate some refraction effects due
to non-uniform mean flow and temperature, and we describe how this influences the
sound attenuation. We also discuss an approach to model the noise source for trans-
mission loss calculations, when the exact modal amplitude distribution for the noise
source is not known (which is generally the case in practice). Based on this approach
we consider the transmission loss of three test cases that are typical of a realistic APU
exhaust duct geometry. We conclude that especially the temperature non-uniformity is
very beneficial for the attenuation, and that modes of the lowest circumferential orders
are most relevant for design calculations since these are the least attenuated.
Samenvatting
Geluidsvoortplanting door akoestisch beklede kanalen

met parallelstroming
Gedurende de afgelopen decennia was aero-akoestisch onderzoek voor stromingskana-
len met wandbekleding (kanaalakoestiek) voornamelijk gericht op geluidsreductie in
de in- en uitlaatkanalen van de hoofdmotoren (turbofans) van vliegtuigen. Sinds enige
tijd heeft ook het zogenaamde ramp-lawaai—waaraan grondpersoneel wordt blootge-
steld terwijl het vliegtuig aan de grond staat—meer aandacht gekregen, aangezien de
regelgeving strenger is geworden.
Een belangrijke bijdrage aan het ramp-lawaai wordt geleverd door de Auxilliary
Power Unit (APU), een turbinemotor in de staart van een vliegtuig die vermogen le-
vert zolang de hoofdmotoren uitgeschakeld zijn. Het uitlaatkanaal van de APU, dat
doorgaans recht en cirkelcylindrisch is, voert een niet-uniforme stroming met sterke
temperatuurgradiënten die veroorzaakt worden door het binnenstromen van koude
lucht via de zijkanten. Om het lawaai uit het APU-uitlaatkanaal te verminderen,
wordt de wand over het algemeen bekleed met geluidsabsorberend materiaal: door-
gaans een lokaal-reagerende bekleding in de vorm van een honingraatstructuur, of
een niet lokaal-reagerende bekleding in de vorm van een reeks holle, annulaire seg-
menten, ieder bedekt met een poreuze afdekplaat. De diepte van de wandbekleding
varieert meestal in de lengterichting vanwege de conische vorm van de staart van de
vliegtuigromp.
Om een goede afweging te kunnen maken tussen minimaal gewicht en zo min moge-
lijk lawaai bestaat de behoefte aan meer begrip van de belangrijkste geluidsvoortplan-
tingsmechanismen en aan nauwkeurige ontwerpgereedschappen. Dit motiveert het
hoofddoel van dit onderzoek, namelijk het beschikbaar maken van semi-analytische
oplossingen voor geluidsvoortplanting in akoestisch beklede kanalen met parallelstro-
ming en sterke temperatuurgradiënten, en axiaal variërende wandeigenschappen. Wij
formuleren het geluidsvoortplantingsprobleem in termen van akoestische kanaalmo-
des. Deze modes zijn oplossingen van de Pridmore-Brown-vergelijking; samen met ge-
schikte randvoorwaarden vormt deze vergelijking een randwaardeprobleem dat kleine
(akoestische) verstoringen van een parallelstroming met transversale temperatuurgra-
diënten beschrijft. We presenteren enkele asymptotische oplossingen van de Pridmore-
Brown-vergelijking, die gebruikt worden om onze numerieke aanpak te valideren.
Voor niet-uniforme stroming en temperatuur bepalen we numerieke oplossingen
van de Pridmore-Brown-vergelijking op een robuuste en efficiënte wijze door gebruik-
making van de COLNEW computercode, die randwaardeproblemen in termen van ge-
125
126 Samenvatting
wone differentiaalvergelijkingen oplost met behulp van collocatie. Om te garanderen

dat alle relevante oplossingen gevonden worden, gebruiken we een padvolgende (of
continuerings-) aanpak, waarbij we starten met een gemakkelijke oplossing (bijvoor-
beeld uniforme stroming en temperatuur) en vervolgens deze oplossing volgen ter-
wijl de relevante parameters worden gevarieerd tot de gewenste waarden bereikt zijn.
Onze padvolgingsaanpak is verfijnd door gebruikmaking van een predictie-correctie-
regeling, waarbij de predictie wordt bepaald door lineaire extrapolatie van de voor-
gaande oplossingen. De correctiestap bestaat vervolgens uit het aanpassen van de
oplossing door middel van COLNEW, met de predictie als startwaarde.
Voor uniforme stroming en temperatuur is een exacte dispersierelatie beschikbaar
(door gebruik te maken van Besselfuncties), zowel voor lokaal- als voor niet lokaal-
reagerende randvoorwaarden. Dit maakt het mogelijk om de akoestische modes te
vinden door hun axiale golfgetallen op te lossen als de nulpunten van een analytische
functie. Om te garanderen dat alle relevante oplossingen worden gevonden, passen
we voor het zoeken van nulpunten de methode van Davies toe, die gebaseerd is op
contourintegratie in het complexe vlak (een aanpassing van de methode van Delves
and Lyness zodanig dat afgeleiden niet benodigd zijn). Deze methode garandeert dat
alle nulpunten binnen een bepaald gebied van het complexe vlak worden gevonden,
dus dat er geen afhankelijkheid is van de beginschattingen, zoals bij de methode van
Newton.
Een geschikte aanpak om geluidsvoortplanting te berekenen voor een configuratie
die bestaat uit meerdere annulaire segmenten met verschillende wandeigenschappen
is mode-matching. Hierbij wordt het akoestische veld beschreven als een sommatie van
modes en worden de onbekende modale amplitudes gevonden door het opleggen van de
continuïteit van akoestische druk en axiaalsnelheid op de plaats waar de wandeigen-
schappen discontinu zijn. Het koppelen van de akoestische velden gebeurt op basis van
de projectie van de continuïteitsvoorwaarden op een geschikte basis van testfuncties.
Het scattering matrix-formalisme wordt toegepast om de geluidsvoortplanting door een
reeks van opeenvolgende segmenten te berekenen. Deze methode is numeriek stabiel
aangezien de voortplanting van niet-propagerende modes alleen wordt beschouwd in
de richting van exponentiële afname.
Voor uniforme stroming kunnen de benodigde inproducten worden gevonden als
analytische uitdrukkingen indien Besselfuncties worden gebruikt als testfuncties (de
klassieke aanpak). Echter, in het geval van niet-uniforme stroming is numerieke inte-
gratie vereist om de inproducten te berekenen bij toepassing van de klassieke aanpak.
In dit proefschrift wordt een nieuwe bilineaire afbeelding voorgesteld in de vorm van
een integraal van een gewogen combinatie van producten van Pridmore-Brown modes
waarvan de waarde kan worden berekend met behulp van een analytische uitdrukking.
Aangezien deze bilineaire afbeelding lijkt op een inproduct van Pridmore-Brown modes
kan deze worden toegepast in plaats van het standaard integraal-inproduct om hier-
mee de basis te vormen van een nieuwe mode-matching-procedure. Omdat met deze
nieuwe aanpak de inherent onnauwkeurige numerieke integratie van oscillatorische
functies vermeden wordt, is deze aanpak nauwkeuriger numeriek goedkoper dan de
klassieke aanpak, wat aangetoond wordt met numerieke experimenten.
De beschikbaarheid van de numerieke Pridmore-Brown modes stelt ons in staat
om asymptotische oplossingen in de vorm van langzaam variërende modes van het
WKB-type te beschouwen, voor het geval dat de impedantie langzaam varieert in de
lengterichting vanwege de variërende wanddikte. We vergelijken deze asymptotische
Samenvatting 127
benaderingen met de resultaten van mode-matching en laten zien dat voor de meeste
APU-uitlaatkanalen, waarvoor geldt dat de impedantie niet langzaam varieert als ge-
volg van resonanties in de wand, deze asymptotische benaderingen niet toepasbaar
zijn.
Ten slotte worden de ontwikkelde methodes toegepast om enkele afbuigingseffecten
te illustreren die het gevolg zijn van niet-uniforme stroming en temperatuur, en bespre-
ken we hoe dit de geluidsdemping beïnvloedt. Daarnaast presenteren we een aanpak
om bij transmissieverliesberekeningen het brongeluid te modelleren als de exacte mo-
dale verdeling hiervan niet bekend is (wat meestal het geval is in de praktijk). Op
basis van deze aanpak beschouwen we het transmissieverlies voor drie testgevallen
die representatief zijn voor een realistische configuratie van een APU uitlaatkanaal.
We concluderen dat in het bijzonder de niet-uniformiteit van de temperatuur een erg
gunstige invloed heeft op de demping, en dat modes met een laag omtreksgolfgetal
het meest relevant zijn voor ontwerpberekeningen aangezien deze het minst worden
gedempt.
Curriculum Vitae
Martien Oppeneer was born on the 28th of May 1982 in Axel, The Netherlands. Af-
ter finishing his pre-university education (VWO) in 2000 at the Zeldenrustcollege in
Terneuzen he began his studies in Electrical Engineering at the TU Eindhoven. As
part of his Master’s education he did internships at the Netherlands Organisation for
Applied Scientific Research (TNO) on the topic of radio wave propagation of cellular
networks, and at the Electromagnetics department of the TU Eindhoven on the topic
of thin-wire antennas.
In 2008 he conducted his Master’s thesis research under the supervision of prof.dr.
A.C. Cangellaris at the University of Illinois at Urbana-Champaign (UIUC) where he
worked on numerical models for multi-terminal resistance extraction of the power dis-
tribution network of microchips, using algebraic multigrid based on element interpola-
tion (AMGe) and stochastic collocation techniques. He graduated on this topic within
the Elecromagnetics (EM) group of prof.dr. A.G. Tijhuis at the Department of Electrical
Engineering of the TU Eindhoven in 2010. During his studies he has been an active
member of the student organization VGSEi, of which he was president from October
2003 to October 2004.
From September 2009 he was employed as a PhD student at the Dutch National
Aerospace Laboratory (Nationaal Lucht- en Ruimtevaartlaboratorium—NLR), where
he worked on a PhD project in cooperation with the Center for Analysis, Scientific
Computing and Applications (CASA) under the supervision of prof.dr. R.M.M. Mattheij,
dr. S.W. Rienstra, and dr. P. Sijtsma (NLR); the results are presented in this disserta-
tion. As a part of this project he did a short internship in 2012 at the Acoustics and
Environment Department of Airbus in Toulouse, France.
Since March 2014 he is employed as a Scientific Software Engineer at VORtech
Computing in Delft.
129
Dankwoord /
Acknowledgements
Binnen de antropologie wordt wel de term rite de passage gebruikt om een overgangs-
ritueel aan te duiden dat de verandering van iemands sociale status markeert, zoals
bijvoorbeeld een huwelijk. Een rite de passage gaat vaak gepaard met beproevingen
en veel symboliek. De promotieplechtigheid kan wellicht ook als een voorbeeld hiervan
worden gezien, een soort coming of age in wetenschappelijke zin. Het volbrengen en
afronden van het promotietraject dat zijn weerslag gevonden heeft in dit proefschrift is
voor mij op sommige momenten werkelijk een beproeving geweest. Hoewel een doctors-
titel een bewijs is van wetenschappelijke zelfstandigheid, is de hulp en het getoonde
vertrouwen van anderen op sommige momenten voor mij onontbeerlijk geweest, zowel
op inhoudelijk als op persoonlijk vlak.
Voor dit project ben ik in dienst geweest bij de afdeling AVHA van het Nationaal
Lucht- en Ruimtevaartlaboratorium (NLR) en heb ik een aanzienlijk deel van mijn
tijd doorgebracht bij het Centre for Analysis, Scientific computing and Applications
(CASA) van de faculteit Wiskunde en Informatica van de TU Eindhoven. Allereerst
dank ik het NLR voor het financieel mogelijk maken van dit project en CASA voor
het bieden van een interessante multi-disciplinaire wetenschappelijke omgeving. Ik
heb het waardevol gevonden om deel uit te maken van zowel de meer bedrijfsmatige
omgeving van het NLR, waar een sterke link is met de experimentele wereld, als de
wetenschappelijke en meer theoretische omgeving van de TU.
Ik dank mijn eerste promotor Bob Mattheij voor het geschonken vertrouwen vanaf
het begin, de vele inhoudelijke discussies gedurende het project en het op vriendelijke
en tegelijkertijd resolute wijze bewaken van de voortgang tijdens de afrondende fase.
Ook dank ik Pieter Sijtsma en Sjoerd Rienstra, de geestelijke vaders van dit project.
Ik dank Pieter, niet alleen voor het feit dat ik met ‘zijn’ BAHAMAS een uitstekend
beginpunt voor mijn eigen onderzoek had en voor het trekken van mijn project binnen
het NLR, maar ook voor het laten zien hoe je geavanceerde wiskunde in kunt zetten
voor concrete technische toepassingen. Ik ben ook zeer veel dank verschuldigd aan
Sjoerd; voor zijn fundamentele inzichten en diepgaande kennis op het gebied van aero-
akoestiek, voor de inspirerende discussies en creatieve ideeën, voor het demonstreren
van het belang van goed modelleren, en voor het regelmatig (soms wat uitgebreid) uit-
wijden over een onderwerp, al dan niet wetenschappelijk van aard. Ik dank Barry
Koren voor zijn bereidwilligheid om als tweede promotor te fungeren en voor zijn be-
trokkenheid gedurende de afrondende fase van het project. I would also like to thank
the professors Schröder, Hoeijmakers and Slot for carefully evaluating my thesis, and
professor Aarts for being the chairman during the defense ceremony.
131
132 Dankwoord / Acknowledgments
Ik heb mij binnen de afdeling AVHA van het NLR goed thuis gevoeld; ik wil mijn
huidige chef Joost Hakkaart en zijn voorganger Christophe Hermans hartelijk danken
voor de ondersteuning en prettige samenwerking. Ook mijn overige (oud-)afdelings-
genoten uit de Noordoostpolder en de andere collega’s bij het NLR dank ik hartelijk
voor de goede sfeer. De therapeutische boswandelingen hebben daaraan zeker bijge-
dragen.
Also at the CASA department I very much enjoyed the friendly and open atmosp-
here. I would like to thank the current and former PhD students, postdocs and staff
members for that. In het bijzonder dank ik Werner Lazeroms voor de prettige samen-
werking v.w.b. de WKB-oplossingen die beschreven zijn in Hoofdstuk 5. At CASA an
atmosphere is fostered in which it is easy to cross the gaps between people from very
different backgrounds, both scientifically as well as culturally, which is very valuable.
This PhD project was motivated by the practical problem of noise coming from an
APU exhaust duct, a problem that is studied extensively at Airbus. For this reason I
spent some time in Toulouse during the fall of 2012. This has been a great learning
experience, and also in other respects very enjoyable due to the friendly people, the
nice climate and the good food. Je remercie Maud Lavieille pour l’avoir rendu possible.
Ten slotte wil ik mijn familie en schoonfamilie en vrienden bedanken voor hun
niet aflatende interesse in mijn voortgang, naast de morele en praktische ondersteu-
ning, met name in periodes van grote drukte en stress. Geertje verdient een speciale
vermelding; hartelijk dank voor het ontwerpen van de omslag van dit proefschrift. Pa-
ranimfen Joep en broer Daan, bedankt dat jullie bereid zijn om mijn ‘bruidjonkers’ te
zijn tijdens mijn huwelijk met de wetenschap.
Marjan en Karlijn, jullie geven mijn leven zoveel meer kleur, bedankt daarvoor.
Inderdaad Marjan, het was niet makkelijk om allebei ongeveer gelijktijdig een proef-
schrift af te ronden, van baan te wisselen en de zorg voor onze prachtige dochter Karlijn
te delen, terwijl je zwanger bent van onze zoon, die het daglicht zal zien vlak voor of na
het verschijnen van dit proefschrift. Ik dank je voor je ondersteuning, zorg en liefde.
Marjan, ik draag dit proefschrift op aan jou.

Sound Propagation in Lined Ducts With Parallel Flow: WWW - Tue.nl/taverne

Uploaded by

Copyright:

Available Formats

Sound Propagation in Lined Ducts With Parallel Flow: WWW - Tue.nl/taverne

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sound Propagation in Lined Ducts With Parallel Flow: WWW - Tue.nl/taverne

Uploaded by

Copyright:

Available Formats

Sound propagation in lined ducts with parallel flow

Citation for published version (APA):

Document status and date:

Please check the document version of this publication:

Take down policy

Download date: 17. janv.. 2024

A catalogue record is available from the Eindhoven University of Technology Library.

Copyright © 2014 by M. Oppeneer, Amsterdam, The Netherlands.

The work described in this thesis has

ter verkrijging van de graad van doctor aan de

voorzitter: prof.dr. E.H.L. Aarts

2 Basic equations and model 11

4 Mode-matching for segmented ducts 51

5 Asymptotic solutions for slowly varying impedance 75

6 Application to APU exhaust duct 87

7 Concluding remarks 107

A Construction of bilinear map 109

B Approximate Blasius solution 111

Curriculum Vitae 129

Dankwoord / Acknowledgments 131

M Mach number (dim. less axial mean velocity)

1.1 Motivation: ramp noise

1.2 Context: aircraft duct acoustics

Figure 1.3: Honeycomb Helmholtz resonator array liner. Copied with

an APU exhaust duct, as we will discuss in the next section.

drainage slots are non-locally reacting.

1.3 Problem description

cool air inlet

Figure 1.4: Schematic overview of a typical APU exhaust duct geometry.

1.4 Objectives and main results

Path-following to find Pridmore-Brown eigensolutions

Our numerical approach to find Pridmore-Brown eigensolutions for non-uniform flow

Root finding by contour integration to find initial guesses

Mode-matching based on closed-form integrals

in Chapter 4 a new mode-matching method, which consists of replacing the inner-

Application to APU exhaust duct

Basic equations and model

2.1 Governing equations

This is the fundamental thermodynamic relation for a closed single-component sys-

Complying with the common notation in thermodynamics literature, the subscript s

where R is the specific

can be used, from which it follows that d h = d( e + pv) = d e +d(RT ) so c p dT = c v dT +RdT

s = c v log p − c p log ρ + ( s init − c v log p init + c p log ρ init ), (2.14)

2.1.2 Conservation laws for a Newtonian fluid

In acoustics we are interested in small unsteady perturbations of a steady mean

2.1.3 Mean flow equations

v( x, t) = v0 ( x) + v1 ( x, t), p = p 0 ( x) + p 1 ( x, t), etc.. (2.25)

where the Reynolds number is defined as

A measure of the viscous term is

This formulation can be found by substituting T d s = c p dT − d p/ρ (the fundamental

where ∆T is a characteristic temperature difference. The energy equation then be-

To summarize, the steady mean flow satisfies

Moreover, from (2.21) we have as a different form of the energy equation

We also have the ideal gas law

This research concerns sound propagation through a straight circularly symmetric

• uniform flow with v0 = u 0 e x ;

• no mean flow, i.e. v0 = 0.

• constant mean temperature T0 ;

• a mean temperature that varies only in the radial coordinate: T0 = T0 ( r )