P. 1
692950

692950

|Views: 4|Likes:
Published by bahramymohsen

More info:

Published by: bahramymohsen on Mar 14, 2013
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

03/14/2013

pdf

text

original

Sections

  • Introduction
  • 1.1 Thermoacoustics
  • 1.2 History and applications
  • 1.2.1 Standing-wave or stack-based devices
  • 1.2.2 Traveling-wave or regenerator-based devices
  • 1.2.3 Special applications and phenomena
  • 1.3 Objectives
  • 1.4 Thesis outline
  • Theory and model
  • 2.1 General thermoacoustic theory
  • 2.1.1 Basic equations
  • 2.1.2 Linearization
  • 2.1.3 Velocity profile
  • 2.1.4 Thermoacoustic continuity equation
  • 2.1.5 Total energy flow
  • 2.2 Model
  • 2.2.1 Transfer matrix
  • 2.2.2 Energy equation and temperature
  • 2.2.3 Iteration
  • 2.3 Thermoacoustic devices
  • 2.3.1 Introduction
  • 2.3.2 Stack-based devices
  • 2.3.3 Regenerator-based devices
  • Electroacoustics
  • 3.1 Introduction
  • 3.2 Theory and model
  • 3.2.1 Speaker equations
  • 3.2.2 Acoustic impedance
  • 3.3 Experimental Setup
  • 3.4 Results
  • 3.4.1 Empty resonator
  • 3.4.2 Fixed speaker voltage
  • 3.4.3 Fixed frequency
  • 3.5 Discussion and conclusion
  • Acoustic measurements
  • 4.1 Introduction
  • 4.2 Experimental Set-up
  • 4.2.1 Microphones and speaker
  • 4.3 Multi-microphone method
  • 4.4 Acoustic energy losses
  • 4.4.1 Acoustic energy flow
  • 4.4.2 Fusco method
  • 4.4.3 Traveling-waves method
  • 4.4.4 Acoustic-energy-flow measurements
  • 4.4.5 Acoustic energy losses in a stack
  • 4.4.6 Minor losses
  • 4.5 Minor-loss correction
  • 4.6 Single-stack-position method
  • 4.7 Multiple-stack-positions method
  • 4.8 Discussion and conclusion
  • Temperature profile in a stack
  • 5.1 Introduction
  • 5.2 Experimental set-up
  • 5.3 Model
  • 5.3.1 Steady State model
  • 5.3.2 Time Dependent model
  • 5.4 Heat losses
  • 5.5 Temperature profile measurement
  • 5.5.1 Time dependent measurements
  • 5.5.2 Amplitude sweep
  • 5.5.3 Frequency sweep
  • 5.6 Discussion and conclusion
  • Flow visualization in and around a stack
  • 6.1 Introduction
  • 6.2 Experimental Set-up
  • 6.2.1 Perspex set-up
  • 6.2.2 Aluminium set-up
  • 6.2.3 Particle Image Velocimetry
  • 6.3 A typical measurement
  • 6.5 The influence of the plate-end shape
  • 6.6 Numerical simulations
  • 6.6.1 CFD model
  • 6.6.2 Subsequent cycles
  • 6.6.3 Minor losses
  • 6.7 Vortex street evolution
  • 6.8 Velocity profile in-between two plates
  • 6.8.1 Introduction
  • 6.8.2 A single measurement of the velocity profiles
  • 6.8.3 Fitting procedure
  • 6.8.5 Velocity fluctuations
  • 6.8.6 Streaming velocity
  • 6.9 Natural Convection
  • 6.9.1 Scale analysis on a vertical wall
  • 6.9.2 Measurements
  • 6.10 Streaming
  • 6.11 Discussion and conclusion
  • 7.1 Introduction
  • 7.2 Experimental set-up
  • 7.3 Results
  • 7.3.1 Instantaneous measurements
  • 7.3.2 Streaming in window 2
  • 7.3.3 Streaming in window 3
  • 7.4 Discussion and conclusion

High-Amplitude Thermoacoustic

Flow Interacting with Solid
Boundaries
Copyright c _2010 by P.C.H. Aben, Eindhoven, 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.
Printed by Print Service Technische Universiteit Eindhoven
Cover design by Paul Verspaget
A catalogue record is available from the Eindhoven University of Technology Library
Aben, Paul
High-Amplitude Thermoacoustic Flow Interacting with Solid Boundaries /
by Paul Aben. -
Eindhoven: Technische Universiteit Eindhoven, 2010.
Proefschrift. - ISBN 978-90-386-2399-3
NUR 928
This research was financially supported by the Technology Foundation (STW), grant
number ETTF.6668.
High-Amplitude Thermoacoustic
Flow Interacting with Solid
Boundaries
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 woensdag 8 december 2010 om 16.00 uur
door
Paul Cornelis Hubertus Aben
geboren te Sittard
Dit proefschrift is goedgekeurd door de promotor:
prof.dr. A.T.A.M. de Waele
Copromotor:
dr.ir. J.C.H. Zeegers
Nomenclature
List of symbols with a short description and the page on which it is introduced for the first time.
Symbol Unit Description Page
α kinetic-energy correction factor 67
α W/(m
2
K) heat transfer coefficient 88
β momentum correction factor 66
β K
−1
volume expansivity 131
γ ratio of specific heats 1
δ
ν
m viscous penetration depth 16
δ
κ
m thermal penetration depth 18
ζ m displacement 24
η efficiency 23
η
C
Carnot efficiency 24
θ rad phase angle 25
κ m
3
/s thermal diffusivity 18
λ m wavelength 4
µ Pas dynamic viscosity 86
ν m
3
/s kinematic viscosity 14
ξ m
−1
complex wave number 21
Π m wetted perimeter 13
ρ kg/m
3
density 14
σ

s
−1
viscous stress tensor 14
ψ porosity 54
ω rad/s angular frequency 15
ω
z
s
−1
vorticity 108
A m
2
cross-sectional area 13
A
c
m
2
cone area 36
b
µ
viscosity exponent 86
b
k
thermal conductivity exponent 86
B
l
N/A motor force factor 36
B
o
Boussinesq number 133
vi
c
p
J/(kgK) specific isobaric heat capacity 14
c
so
J/(kgK) solid heat capacity 86
C
c
contraction coefficient 68
C
+
, D
+
Pa amplitude of rightward traveling wave 55
C

, D

Pa amplitude of leftward traveling wave 55
COP coefficient of performance 24
COP
C
Carnot coefficient of performance 24
d m pore wall thickness 53
D m diameter 38
D m pore size 53
D
p
m plate separation 86
D
r
drive ratio 15
˙
E W total energy flow 19

˙
E
ml
W energy loss due to minor losses 66
f Hz frequency 41
f
0
Hz characteristic frequency 54
f
ν
viscous Rott function 17
f
κ
thermal Rott function 18
F N force 36
g m/s
2
acceleration due to gravity 131
h J/kg specific enthalpy 19
h
ν
, h
κ
channel-geometry dependant function 17
I A current 36
k W/(Km) thermal conductivity 14
k N/m spring constant 36
k
/
m
−1
real component of ξ 55
k
//
m
−1
imaginary component of ξ 55
K minor-loss coefficient 67
K
c
minor-loss coefficient for contraction 68
K
e
minor-loss coefficient for expansion 67
KC
D
Keulegan-Carpenter number based on D 113
KC
L
Keulegan-Carpenter number based on L 111
l
0
m half the plate thickness 16
L
el
Vs/A self-inductance 36
L m length 38
vii
˙ m kg/(m
2
s) mass streaming 130
M kg mass 36
N
u
Nusselt number 133
p Pa pressure 14
P
r
Prandtl number 18
˙
Q W cooling/heating power 23
R
s
J/(kgK) specific gas constant 14
R
e
Reynolds number 107
R

Reynolds number based on δ
ν
113
R
el
Ω electrical resistance 36
R
me
Ns/m friction constant 36
R
a
Rayleigh number 133
R
H
m hydraulic radius 13
R
r
reflection coefficient 57
s J/(kgK) entropy per unit mass 14
S
t
Strouhal number 107
S surface 65
t s time 14
T K or

C temperature 14
T transfer matrix 21
u m/s x-component of v 14
u
c
m/s coil velocity 37
U m
3
/s volume flow rate 15
v m/s velocity vector 14
v m/s y-component of v 14
[v
/
[ m/s root-mean-square value of v 129
V V voltage 36
V
b
m
3
back volume 36
V
ind
V induced voltage 36
w m/s z-component of v 14
˙
W W Work flow 23
˙
W W acoustic energy flow 60
˙
W
i j
W acoustic energy flow at position (x
i
+ x
j
)/2 62
W
p
m plate width 86
x m position along tube axis 14
y m 14
y
0
m half the plate distance 16
z m 14
Z impedance 36
Z
ac
N m
−5
s acoustical impedance 36
Z
el
V/A electrical impedance 37
(∇T)
crit
K/m critical temperature gradient 18
viii
List of sub- and superscripts and special operators.
Symbol Description Page
a
ν
viscous 17
a
κ
thermal 18
a
a
ambient 2
a
ac
acoustical 36
a
C
cold 2
a
el
electrical 36
a
H
hot 2
a
in
(power) into the device 24
a
L
left-hand side 21
a
me
mechanical 36
a
R
right-hand side 21
a
out
(power) out of the device 24
a
p
stack plate 86
a
so
solid 19
a
0
zeroth order (average over one period) 15
a
1
first harmonic 15
a
2
second harmonic 15
¸a¸ cross-sectional average 17
a

complex conjugate 19
˙ a per unit time 19
a
i j
matrix element of a with row i and column j 38
a
+
Moor-Penrose generalized inverse of K 57
a
H
Hermitian matrix of K 57
Contents
1 Introduction 1
1.1 Thermoacoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 History and applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Standing-wave or stack-based devices . . . . . . . . . . . . . . . . . 2
1.2.2 Traveling-wave or regenerator-based devices . . . . . . . . . . . . . 3
1.2.3 Special applications and phenomena . . . . . . . . . . . . . . . . . . 4
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Theory and model 13
2.1 General thermoacoustic theory . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1 Basic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.2 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.3 Velocity profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.4 Thermoacoustic continuity equation . . . . . . . . . . . . . . . . . . 17
2.1.5 Total energy flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.1 Transfer matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.2 Energy equation and temperature . . . . . . . . . . . . . . . . . . . 22
2.2.3 Iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Thermoacoustic devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.2 Stack-based devices . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.3 Regenerator-based devices . . . . . . . . . . . . . . . . . . . . . . . 30
3 Electroacoustics 35
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Theory and model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.1 Speaker equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.2 Acoustic impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.1 Empty resonator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.2 Fixed speaker voltage . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4.3 Fixed frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.5 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
x CONTENTS
4 Acoustic measurements 51
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2 Experimental Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2.1 Microphones and speaker . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3 Multi-microphone method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Acoustic energy losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4.1 Acoustic energy flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4.2 Fusco method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4.3 Traveling-waves method . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4.4 Acoustic-energy-flow measurements . . . . . . . . . . . . . . . . . . 62
4.4.5 Acoustic energy losses in a stack . . . . . . . . . . . . . . . . . . . . 63
4.4.6 Minor losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.5 Minor-loss correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 Single-stack-position method . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.7 Multiple-stack-positions method . . . . . . . . . . . . . . . . . . . . . . . . 75
4.8 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5 Temperature profile in a stack 81
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Experimental set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.3.1 Steady State model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.3.2 Time Dependent model . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4 Heat losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.5 Temperature profile measurement . . . . . . . . . . . . . . . . . . . . . . . 91
5.5.1 Time dependent measurements . . . . . . . . . . . . . . . . . . . . . 91
5.5.2 Amplitude sweep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.5.3 Frequency sweep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.6 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6 Flow visualization in and around a stack 99
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.2 Experimental Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.2.1 Perspex set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.2.2 Aluminium set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.2.3 Particle Image Velocimetry . . . . . . . . . . . . . . . . . . . . . . . 105
6.3 A typical measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.4 Different categories of vortex patterns and dimensionless numbers . . . . 108
6.5 The influence of the plate-end shape . . . . . . . . . . . . . . . . . . . . . . 113
6.6 Numerical simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.6.1 CFD model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.6.2 Subsequent cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.6.3 Minor losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.7 Vortex street evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.8 Velocity profile in-between two plates . . . . . . . . . . . . . . . . . . . . . 125
6.8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.8.2 A single measurement of the velocity profiles . . . . . . . . . . . . 125
CONTENTS xi
6.8.3 Fitting procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.8.4 First harmonic velocity and the transition to turbulence . . . . . . . 126
6.8.5 Velocity fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.8.6 Streaming velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.9 Natural Convection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.9.1 Scale analysis on a vertical wall . . . . . . . . . . . . . . . . . . . . . 130
6.9.2 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.10 Streaming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.11 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7 Flow measurements in co-axial regenerator-based devices 143
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.2 Experimental set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
7.3.1 Instantaneous measurements . . . . . . . . . . . . . . . . . . . . . . 145
7.3.2 Streaming in window 2 . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.3.3 Streaming in window 3 . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.4 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Chapter 1
Introduction
1.1 Thermoacoustics
Thermoacoustics concerns phenomena in which an interaction of acoustics with ther-
modynamics takes place. In an acoustic wave the gas parcels always undergo tempera-
ture variations, which is a consequence of their compression and expansion. When the
acoustic wave is adiabatic, the temperature variations to first order T
1
are given by
T
1
T
0
=
γ −1
γ
p
1
p
0
, (1.1)
in which T
0
is the average temperature, γ the ratio of specific heats, p
1
the pressure
variation, and p
0
the average pressure. In adiabatic sound waves these temperature
variations go unnoticed. However, when a solid is present near the acoustic wave, the
wave interacts with the solid and can cause a transfer of heat from one location in the
solid to another. This is called the thermoacoustic heat-pumping effect. This effect is the
driving mechanism in stack-based coolers and heat pumps. Vice versa, when a sound
wave interacts with a solid with a temperature gradient above a certain critical value,
the temperature gradient enhances the sound wave.
Lord Rayleigh was the first to give a qualitative description of thermoacoustic phe-
nomena. In his work ”The Theory of Sound” [1], published in 1887, he discussed the
ability to generate temperature differences using acoustic waves. The subject remained
largely untouched for over eighty years, until in 1969 Nicholas Rott began a series of
publications [2–7] that initiated a revival in thermoacoustic research. Rotts work forms
the theoretical basis of most of modern standing-wave thermoacoustic research. Later,
Rott’s publications have been reviewed by Swift [8]. Once the theory for parallel plates
was established, the thermoacoustic theory for different pore shapes (triangular, rectan-
gular, circular and pin arrays [9] instead of parallel plates) was developed by Arnott,
Bass, and Raspet [10]. Also other authors contributed to the understanding of ther-
moacoustics, including Wheatley and Atchley. Wheatley [11] introduced the thermoa-
coustic couple and compared his measurement results quantitatively with Rott’s theory.
Atchley worked on thermoacoustically generated temperature gradients in a thermoa-
coustic couple [12], on the analysis of thermoacoustic prime movers and their onset of
2 Introduction
self-oscillation [13–16], and together with Gaitan on the energy dissipation of higher
harmonics and how to avoid this by detuning the resonance tube [17].
In ’t panhuis [18] rederived the thermoacoustic theory in a very systematic way.
At the Los Alamos National laboratory a flexible code, called Design Environment for
Low-amplitude ThermoAcoustic Engines (DeltaE) [19], was developed for the numeri-
cal simulation of thermoacoustic devices. In 2007 a version with a better user interface
was developed, called DeltaEC [20]. This code has been used by many groups to de-
velop and test new devices.
1.2 History and applications
We distinguish heat-driven devices, called prime movers, and sound-driven devices,
called refrigerators or heat pumps. In a prime mover the absorbed heat at a higher tem-
perature T
H
is partially converted into work and the remaining energy is released at a
lower temperature T
C
< T
H
. A refrigerator or heat pump absorbs heat at a lower tem-
perature and requires input of work to release or pump heat to a higher temperature,
where the released heat is the sum of the absorbed heat and work. The difference be-
tween a heat pump and a refrigerator is that in a heat pump, the heat is transferred from
ambient temperature T
a
to a higher temperature T
H
> T
a
, whereas in a refrigerator heat
is transferred to ambient temperature T
a
from a lower temperature T
C
< T
a
.
1.2.1 Standing-wave or stack-based devices
Already in the 19th century, glass blowers noticed that when a hot glass bulb was at-
tached to a cool glass tube, it sometimes emitted sound, and Sondhauss [21] in 1850
quantitatively investigated the relation between the pitch of the sound and the dimen-
sions of the apparatus.
Standing-wave devices are called this way, since their operation is based on a stand-
ing wave that is created in a resonator tube. In case of a refrigerator one side of the
resonator is closed and at the other side an acoustic power source is located, e.g. a loud-
speaker, linear motor, or thermoacoustic prime mover. In case of a prime mover, one
end of the resonator is closed and at the other end an acoustic load is located, e.g. a
piston or a thermoacoustic refrigerator. For the operation of a standing-wave device
an imperfect heat transfer of the gas with a solid is essential. To optimize the gas-wall
interaction, a solid object, called a stack, is installed in the resonator. A stack consists
of pores whose size is of the same order of magnitude as the thermal penetration depth
in the gas. For a good performance the stack is located between a pressure node and a
pressure anti-node, since the thermoacoustic effect requires both pressure and displace-
ment oscillations. The waves in a thermoacoustic device are never pure standing waves,
since standing waves cannot transport energy. The phase angle between pressure and
velocity is close to 90

, but never exactly 90

.
In the Sondhauss tube the thermoacoustic effect occurs in a single pore. In standing-
wave engines, however, the process occurs in a stack, which consists of many pores in
parallel, all of which contribute to the acoustic power generation. Such a stack was not
added to a Sondhauss tube until the 1960s. This important development allowed fill-
ing a large-diameter tube with relatively small pores, creating a large volume of strong
1.2 History and applications 3
thermoacoustic power production, while leaving the rest of the resonator open and rela-
tively low in dissipation. Heat exchangers spanning the ends of the stack are needed for
efficient delivery and extraction of the large amounts of heat needed by a stack. Early
use of such heat exchangers was described by Feldman and Carter [22] and by Wheatley
et al. [23].
The first stranding-wave refrigerator was developed by Hofler [24] in 1986. He used
a loudspeaker to drive a closed resonator tube with a stack positioned near the loud-
speaker. At the other part of the tube a volume was attached to simulate an open end-
ing. The tube is effectively a quarter-wave resonator. At both sides of the stack a heat
exchanger is located, one at low temperature and the other at ambient temperature. Ti-
jani et al. [25, 26] optimized this standing-wave refrigerator, using the back volume of
the loudspeaker as a gas-spring system to adept the effective resonance frequency of the
loudspeaker to its load. They managed to cool down to -70

C. They also demonstrated
the advantage of using gas mixtures, in order to get a low Prandtl number.
One application of stack-based devices is an ice-cream freezer that is developed by
Garret [27] from the Pennsylvania State University in 2004 in cooperation with Ben &
Jerrys Homemade. It is 25 cm in diameter, 48 cm tall, is filled with 10 bar helium, and
is driven by a moving-magnet linear motor at around 100 Hz. The cooling capacity is
119 W at -24.6

C and the coefficient of performance is 19% of Carnot. Although this
refrigerator is more expensive than conventional refrigerators, no chemicals or gases
are used that can be harmful to the environment. Another advantage of thermoacous-
tic devices, including this refrigerator, is that they have only a few moving parts, and
even more important, no moving parts in the cold, which makes them very reliable
and relatively simple and flexible to build. This makes them also interesting for space
applications [28].
When pumping natural gas through pipes to the surface, sensors are located far
below the ground. To power these sensors a reliable electrical power supply is required.
One way to provide this power is to create a side branch to the main flow, which was
studied by Slaton [29–32]. At the edges of the side branch with the main pipe, vortices
will be shed. When the vortex-shedding frequency corresponds with the resonance
frequency of the side branch, a standing wave is created. By installing a stack in the
side branch the acoustic power can be used to generate a temperature difference, that
can produce electric power with thermoelectric elements.
1.2.2 Traveling-wave or regenerator-based devices
In Stirling engines and so-called traveling-wave engines, the conversion of heat to acous-
tic power occurs in the regenerator, which smoothly spans the temperature difference
between the hot heat exchanger and the ambient heat exchanger and contains small
pores through which the gas oscillates. A regenerator differs from a stack by a much
smaller pore size.
The pores in a regenerator are small enough that the gas in them is in excellent local
thermal contact with their walls. Asolid matrix such as a pile of fine-mesh metal screens
or spherical particles is often used. Proper design causes the gas in the channels to move
toward the hot heat exchanger while the pressure is high, and toward the ambient heat
exchanger while the pressure is low. The time phasing described above is that of a
traveling acoustic wave, which carries acoustic power from ambient to hot. In contrast
4 Introduction
to standing-wave engines, acoustic power must be injected into the ambient end of a
regenerator in order to create more acoustic power; the regenerator is an amplifier of
acoustic power. Yazaki et al. [8] demonstrated a traveling-wave engine very similar
to that first conceived by Ceperley [33–35], with the path length around the toroidal
waveguide nearly equal to 2λ. At about the same time, De Blok [36] and the Los Alamos
group [37, 38] invented a traveling-wave engine with the heat exchangers imbedded in
a lumped-acoustic impedance torus much shorter than the wavelength λ.
Backhaus and Swift [37–39], at the Los Alamos National Laboratory, have developed
a heat-driven thermoacoustic refrigerator to liquify natural gas. It is a combination of a
thermoacoustic engine that drives a orifice pulse tube. The engine is a regenator based
on toroidal geometry.
In the process industry large quantities of waste heat are released to the environment
that cannot be reused, mostly because the temperature level is too low. In order to
reuse part of this waste heat, a heat pump is necessary that can provide a temperature
lift to the required temperature levels. A thermoacoustic device can accomplish this
goal. The energy research center of the Netherlands (ECN) [40, 41] develops traveling-
wave devices to upgrade waste heat to either process heat or to generate cooling. Their
thermoacoustic devices are a combination of an engine and a heat pump. The waste
heat is used to create a temperature gradient and is converted into acoustic power. The
acoustic power is used to drive a heat pump that upgrades part of the waste heat to a
required temperature level.
The traditional Stirling engine has high efficiency, but it has moving parts. The
thermoacoustic-Stirling hybrid engine has reasonably high efficiency and very high re-
liability, but the toroidal topology needed is responsible for high fabrication costs. Fi-
nally, the stack-based standing-wave thermoacoustic engine is reliable and costs little to
fabricate, but its efficiency is only about 2/3 that of a regenerator-based system.
1.2.3 Special applications and phenomena
A Rijke tube is similar to the Sondhauss tube, in the sense that it turns heat into sound,
by creating a self-amplifying standing wave. Rijke [42] discovered in 1859 a way of
using heat to sustain a sound in a cylindrical tube, open at both ends. The tube was
oriented vertically and a hot wire gauze was located at the lower half of the tube. The
flow of air past the gauze is a combination of two motions. There is a uniform upwards
motion of the air due to a convection current resulting from the gauze heating up the
air. Superimposed on this is the motion due to the sound wave. Rijke oscillations only
occur in a pipe having both ends open, since the upward motion, due to natural convec-
tion past the hot gauze is an essential part of the operation of the Rijke tube. Feldman
produced a literature overview of both the Rijke [43] and the Sondhauss [44] thermoa-
coustic phenomena.
A completely different use of thermoacoustics is the separation of gas mixtures by
thermoacoustic waves, which was discovered by Swift and Spoor [45–47]. The super-
position of nonzero time-averaged mole flux on a thermoacoustic wave in a binary gas
mixture in a tube produces continuous mixture separation.
A cascade thermoacoustic engine, developed by Gardner and Swift [48], is a device
in which one standing-wave engine and two traveling-wave engines are cascaded in
series. Most of the acoustic power is produced in the efficient traveling-wave stages.
1.3 Objectives 5
The straight-line series configuration is easy to build and allows no Gedeon streaming.
The engine delivers up to 2 kWof acoustic power, with an efficiency (the ratio of acoustic
power to heater power) of up to 20%.
Luo et al. [49, 50] have been working on a novel cascade thermoacoustic prime
mover. And currently Hu et al. [51–53] are still working on cascade thermoacoustic
engines. They use much higher frequencies than Gardner and Swift did.
Bauwens et al. have worked on numerical simulations of the flow and heat transfer
in a stack [54, 55] and on a transient theory [56].
1.3 Objectives
The thermoacoustic theory, as developed by Rott and Swift [57], is a linear theory. How-
ever, in many situations the amplitudes are so high that the use of the linear theory is
not justified [38, 58–61]. Therefore we decided to study the effects that appear at high
amplitudes. The nonlinear effects that we are interested in, are:
• vortex shedding at the end of a parallel-plate stack,
• dissipation at the ends of a stack due to the sudden change in cross section,
• transition to turbulence in-between plates,
• streaming.
In order to study these effects we have built a set-up to do detailed measurements of
pressure, stack temperature, and local velocity fields. Since nonlinear effects that we
want to study occur at high amplitudes (≈ 3% [62–67]), we need to build a set-up that
can produce such high amplitudes. We want to emphasize that the goal is not to build
a device that is optimized for efficiency, but rather to do measurements to gain a better
understanding of different phenomena that are occurring in thermoacoustic devices.
The work is a dual PhDproject, in which Peter in ’t panhuis focuses on the theoretical
and mathematical aspects and we focus more on the practical and experimental part.
An additional goal of this work is to help several Dutch companies, including ECN,
Aster Thermoacoustics, and Shell, that work on thermoacoustics and also partially
funded this project. By getting more insight in thermoacoustics we hope that they will
be able to improve their thermoacoustic devices.
1.4 Thesis outline
In this thesis we will first derive a well-known set of three differential equations that
give a mathematical description of thermoacoustics for low amplitudes. These equa-
tions can be applied to many different geometries, including parallel-plate stacks, pores,
even various cross-section shapes, and resonator tubes. The equations include viscous
and thermal dissipation, and thermal conduction. Losses due to entrance effects in a
stack, changes in cross section, and streaming are not included. Since the time depen-
dency is assumed to be harmonic, higher harmonics are also not included. The set of
equations can only be solved analytically for special cases. For all the other cases we
6 Introduction
wrote a code to solve the equations numerically. The thermoacoustic theory, the code,
and the different types of thermoacoustic devices are discussed in chapter 2.
In chapter 3 we present the experimental set-up that is used, with a focus on the
electroacoustics, i.e. the acoustic output of the loudspeaker with respect to the electrical
input. By including the speaker equations in our code we can calculate the pressure
and velocity in the resonator as a function of the speaker voltage and frequency. We
have measured the current through the loudspeaker, its excursion, and also the pres-
sure at different positions in the resonator, using microphones. The measurements are
compared with the model.
In chapter 4 we describe the use of a multi-microphone method to determine the
transfer matrix of various stacks. From the transfer matrix elements the Rott functions
can be determined as functions of the frequency. The Rott functions are fitted to an
analytical solution with the pore size and porosity as fitting parameters. Furthermore,
the multi-microphone method is used to determine energy fluxes at both ends of a stack.
This way the dissipated energy in the stack is measured and compared with our model.
The temperature profile in a parallel-plate is registered as a function of time, using
32 thermometers that are inserted in the center plate of the stack. Since the stack is not
isolated from the environment, the stack temperature is not uniform over the cross sec-
tion. The radial temperature gradients are included in the model, which is discussed in
chapter 5. Temperature measurements in time are compared with our time-dependent
model.
In chapter 6 we visualize the flow using Particle Imaging Velocimetry (PIV). A large
part of the chapter discusses the vortex shedding that occurs behind the plate ends.
The flow-visualization measurements are compared with a numerical simulation. In
the center of the stack in-between two plates the velocity is measured. The velocity
profile is compared with an analytical solution. At sufficiently high Reynolds numbers
the profiles are expected to deviate from this analytical solution, due to turbulence. We
will measure the transition of laminar to turbulent. The PIV technique is also used to
measure jet streaming and natural convection.
Regenerator-based devices, consisting of a coaxial loop, are designed by Aster Ther-
moacoustics and ECN. It is valuable to know how an oscillatory flow behaves in such a
coaxial loop, which is studied using PIV in chapter 7. Especially, the flow at the sharp
corners is of interest.
We conclude this thesis with the most important conclusions and recommendations.
Bibliography
[1] L. Rayleigh, Theory of Sound, Vol. II. New York: Dover, 1945.
[2] N. Rott, “Damped and thermally driven acoustic oscillations in wide and narrow
tubes,” Journal of Applied Mathematics and Physics, vol. 20, pp. 230–243, 1969.
[3] N. Rott, “Thermally driven acoustic oscillations. part II: Stability limit for helium,”
Journal of Applied Mathematics and Physics, vol. 24, pp. 54–72, 1973.
[4] N. Rott, “Thermally driven acoustic oscillations. part III: Second-order heat flux,”
Journal of Applied Mathematics and Physics, vol. 26, pp. 43–49, 1975.
[5] N. Rott and G. Zouzoulas, “Thermally driven acoustic oscillations. part IV: Tubes
with variable cross-section,” Journal of Applied Mathematics and Physics, vol. 27,
pp. 197–224, 1976.
[6] N. Rott, “Thermoacoustics,” Advances in Applied Mechanics, vol. 20, pp. 135–175,
1980.
[7] N. Rott, “The influence of heat conduction on acoustic streaming,” Journal of Ap-
plied Mathematics and Physics, vol. 25, pp. 417–421, 1974.
[8] G. Swift, “Thermoacoustic engines,” Journal of the Acoustical Society of America,
vol. 84, pp. 1146–1180, 1988.
[9] G. W. Swift and R. M. Keolian, “Thermoacoustics in pin-array stacks,” Journal of the
Acoustical Society of America, vol. 94, no. 2, pp. 941–943, 1993.
[10] W. Arnott, H. E. Bass, and R. Raspet, “General formulation of thermoacoustics for
stacks having arbitrarily shaped pore cross sections,” Journal of the Acoustical Society
of America, vol. 90, no. 6, pp. 3228–3237, 1991.
[11] J. Wheatley, T. Hofler, G. Swift, and A. Migliori, “An intrinsically irreversible ther-
moacoustic heat engine,” Journal of the Acoustical Society of America, vol. 74, pp. 153–
170, 1983.
[12] A. Atchley, T. Hofler, M. Muzzerall, M. Kite, and C. Ao, “Acoustically generated
temperature gradients in short plates,” Journal of the Acoustical Society of America,
vol. 88, no. 1, pp. 251–263, 1990.
8 BIBLIOGRAPHY
[13] A. Atchley, “Analysis of the initial buildup of oscillations in a thermoacoustic
prime mover,” Journal of the Acoustical Society of America, vol. 95, no. 3, pp. 1661–
1664, 1994.
[14] A. Atchley, “Standing wave analysis of a thermoacoustic prime mover below onset
of self-oscillation,” Journal of the Acoustical Society of America, vol. 92, no. 5, pp. 2907–
2914, 1992.
[15] A. Atchley, H. Bass, T. Hofler, and H.-T. Lin, “Study of a thermoacoustic prime
mover below onset of self-oscillation,” Journal of the Acoustical Society of America,
vol. 91, no. 2, pp. 734–743, 1992.
[16] A. Atchley and F.-M. Kuo, “Stability curves for a thermoacoustic prime mover,”
Journal of the Acoustical Society of America, vol. 95, no. 3, pp. 1401–1404, 1994.
[17] D. Gaitan and A. Atchley, “Finite amplitude standing waves in harmonic and an-
harmonic tubes,” Journal of the Acoustical Society of America, vol. 93, no. 5, pp. 2489–
2495, 1993.
[18] P. H. M. W. in ’t Panhuis, S. W. Rienstra, J. Molenaar, and J. J. M. Slot, “Weakly non-
linear thermoacoustics for stacks with slowly varying pore cross-sections,” Journal
of Fluid Mechanics, vol. 618, pp. 41–70, 2009.
[19] B. Ward and G. Swift, “Design environment for low-amplitude thermoacoustic en-
gines,” Journal of the Acoustical Society of America, vol. 95, pp. 3671–3672, 1996.
[20] W. Ward, J. Clark, and G. Swift, Design Environment for Low-amplitude
Thermoacoustic Energy Conversion. Los Alamos National Laboratory,
http://www.lanl.gov/thermoacoustics/DeltaEC.html, 2008.
[21] C. Sondhauss, “Ueber die schallschwingungen der luft in erhitzten glasr¨ ohren un
in gedeckten pfeifen von ungleicher weite,” Annalen der Physik, vol. 79, p. 1, 1850.
[22] K. T. Feldman and R. L. Carter, “A study of heat driven pressure oscillations in a
gas,” Journal of Heat Transfer, vol. 92, pp. 536–541, 1970.
[23] J. C. Wheatley, T. Hofler, G. W. Swift, and A. Migliori, “Understanding some sim-
ple phenomena in thermoacoustics with applications to acoustical heat engines,”
American Journal of Physics, vol. 53, pp. 147–162, 1985.
[24] T. J. Hofler, Thermoacoustic refrigerator design and performance. Phd thesis, Physics
Dept., University of California, San Diego, 1986.
[25] M. E. H. Tijani, J. C. H. Zeegers, and A. T. A. M. De Waele, “A gas-spring sys-
tem for optimizing loudspeakers in thermoacoustic refrigerators,” Journal of Ap-
plied Physics, vol. 92, no. 4, p. 2159, 2002.
[26] M. Tijani, Loudspeaker-driven Thermoacoustic Refrigeration. Eindhoven: Universiteits-
drukkerij Eindhoven, 2001.
[27] M. E. Poese, R. W. M. Smith, S. L. Garrett, R. van Gerwen, and P. Gosselin,
“Thermoacoustic refrigeration for ice cream sales,” in Proceedings of the 6th Gustav
Lorentzen Natural Working Fluids Conference, 2004.
BIBLIOGRAPHY 9
[28] S. L. Garrett, J. A. Adeff, and T. J. Hofler, “Thermoacoustic refrigeration for space
application,” Journal of Thermophysics and Heat Transfer, vol. 7, no. 4, pp. 595–599,
1993.
[29] W. Slaton and J. Zeegers, “An aeroacoustically driven thermoacoustic heat pump,”
Journal of the Acoustical Society of America, vol. 117, no. 6, pp. 3628–3635, 2005.
[30] W. Slaton and J. Zeegers, “Acoustic power measurements of a damped aeroacous-
tically driven resonator,” Journal of the Acoustical Society of America, vol. 118, no. 1,
pp. 83–91, 2005.
[31] W. Slaton and J. Zeegers, “Thermoelectric power generation in a thermoacoustic
refrigerator,” Applied Acoustics, vol. 67, no. 5, pp. 450–460, 2006.
[32] J. Zeegers and W. Slaton, “Aeroacoustic thermoacoustics,” in AIP Conference Pro-
ceedings, vol. 838, pp. 416–423, 2006.
[33] P. H. Ceperley, “Apistonless stirling enginethe traveling wave heat engine,” Journal
of the Acoustical Society of America, vol. 66, pp. 1508–1513, 1979.
[34] P. H. Ceperley, “Gresonant travelling wave heat engine,” US Patent 4, pp. 355,517,
1982.
[35] P. H. Ceperley, “Gain and efficiency of a short traveling wave heat engine,” Journal
of the Acoustical Society of America, vol. 77, pp. 1239–1244, 1985.
[36] C. M. de Blok, “Thermoacoustic system,” Dutch Patent. International Application
Number PCT/NL98/00515, 1998.
[37] S. Backhaus and G. W. Swift, “A thermoacoustic-stirling heat engine,” Nature,
vol. 399, pp. 335–338, 1999.
[38] S. Backhaus and G. W. Swift, “A thermoacoustic-stirling heat engine: Detailed
study,” Journal of the Acoustical Society of America, vol. 107, pp. 3148–3166, 2000.
[39] B. Arman, J. Wollan, G. Swift, and S. Backhaus, “Thermoacoustic natural gas liq-
uefiers and recent developments,” in Proceedings of the 2003 International Conference
on Cryogenics and Refrigeration, pp. 123–127, 2003.
[40] L. A. Nijeholt, T. J.A., and S. M.E.H., Spoelstra, “Simulation of a traveling-wave
thermoacoustic engine using computational fluid dynamics,” Journal of the Acous-
tical Society of America, vol. 118, no. 4, pp. 2265–2270, 2005.
[41] M. Tijani and S. Spoelstra, “Study of a coaxial thermoacoustic-stirling cooler,” Cryo-
genics, vol. 48, no. 1-2, pp. 77–82, 2008.
[42] P. L. Rijke, “On the vibration of the air in a tube open at both ends,” Philosophical
Magazine, vol. 17, pp. 419–422, 1859.
[43] K. T. Feldman, “Reviewof the literature on rijke thermoacoustic phenomena,” Jour-
nal of Sound and Vibration, vol. 7, pp. 83–89, 1968.
10 BIBLIOGRAPHY
[44] K. T. Feldman, “Review of the literature on sondhauss thermoacoustic phenom-
ena,” Journal of Sound and Vibration, vol. 7, pp. 71–82, 1968.
[45] P. Spoor and G. Swift, “Mode-locking of acoustic resonators and its application to
vibration cancellation in acoustic heat engines,” Journal of the Acoustical Society of
America, vol. 106, no. 3 I, pp. 1353–1362, 1999.
[46] P. Spoor and G. Swift, “Thermoacoustic separation of a he-ar mixture,” Physical
Review Letters, vol. 85, no. 8, pp. 1646–1649, 2000.
[47] G. Swift and P. Spoor, “Thermal diffusion and mixture separation in the acous-
tic boundary layer,” Journal of the Acoustical Society of America, vol. 106, no. 4 I,
pp. 1794–1800, 1999.
[48] D. Gardner and G. Swift, “A cascade thermoacoustic engine,” Journal of the Acous-
tical Society of America, vol. 114, no. 4 I, pp. 1905–1919, 2003.
[49] H. Ling, E. Luo, W. Dai, and X. Li, “Numerical simulation on a novel cascade ther-
moacoustic prime mover,” in Proceedings of the Twentieth International Cryogenic En-
gineering Conference, ICEC 20, pp. 333–336, 2005.
[50] E. Luo, E. Dai, and H. Ling, “Basic operation principle of a novel cascade thermoa-
coustic prime mover,” in Proceedings of the Twentieth International Cryogenic Engi-
neering Conference, ICEC 20, pp. 317–320, 2005.
[51] Z. Hu, Q. Li, X. Xie, G. Zhou, and Q. Li, “Design and experiment on a mini cascade
thermoacoustic engine,” Ultrasonics, vol. 44, no. SUPPL., pp. e1515–e1517, 2006.
[52] Z. Hu, Q. Li, Q. Li, and Z. Li, “A high frequency cascade thermoacoustic engine,”
Cryogenics, vol. 46, no. 11, pp. 771–777, 2006.
[53] Z. Hu, Q. Li, Z. Li, and Q. Li, “Orthogonal experimental study on high frequency
cascade thermoacoustic engine,” Energy Conversion and Management, vol. 49, no. 5,
pp. 1211–1217, 2008.
[54] L. Bauwens, “Oscillating flow of a heat-conducting fluid in a narrow tube,” Journal
of Fluid Mechanics, vol. 324, pp. 135–161, 1996.
[55] O. Hireche, C. Weisman, D. Baltean-Carls, P. L. Qu´ er´ e, M.-X. Franc¸ois, and
L. Bauwens, “Numerical model of a thermoacoustic engine,” Comptes Rendus
Mcanique, vol. 338, no. 1, pp. 18 – 23, 2010.
[56] L. Bauwens, “Thermoacoustics: Transient regimes and singular temperature pro-
files,” Physics of Fluids, vol. 10, no. 4, pp. 807–818, 1998.
[57] G. W. Swift, “Thermoacoustic engines,” Journal of the Acoustical Society of America,
vol. 84, no. 4, pp. 1145–1180, 1988.
[58] G. Swift, A Unifying Perspective for Some Engines and Refrigerators. Melville: Acous-
tical Society of America, 2002.
[59] G. Swift, “Thermo acoustic engines and refrigerators,” Physics Today, vol. 48, no. 7,
pp. 22–28, 1995.
BIBLIOGRAPHY 11
[60] M. E. Poese and S. L. Garrett, “Performance measurements on a thermoacoustic
refrigerator driven at high amplitudes,” Journal of the Acoustical Society of America,
vol. 107, no. 5, pp. 2480–2486, 2000.
[61] H. S. Roh, R. Raspet, and H. E. Bass, “Parallel capillary-tube-based extension of
thermoacoustic theory for random porous media,” Journal of the Acoustical Society
of America, vol. 121, no. 3, pp. 1413–1422, 2007.
[62] A. Berson and P. Blanc-Benon, “Nonperiodicity of the flow within the gap of a ther-
moacoustic couple at high amplitudes,” Journal of the Acoustical Society of America,
vol. 122, no. 4, pp. EL122–EL127, 2007.
[63] A. Berson, P. Blanc-Benon, and G. Comte-Bellot, “A strategy to eliminate all non-
linear effects in constant-voltage hot-wire anemometry,” Review of Scientific Instru-
ments, vol. 80, no. 4, 2009.
[64] A. Berson, G. Poignand, P. Blanc-Benon, and G. Comte-Bellot, “Capture of instan-
taneous temperature in oscillating flows: Use of constant-voltage anemometry to
correct the thermal lag of cold wires operated by constant-current anemometry,”
Review of Scientific Instruments, vol. 81, no. 1, 2010.
[65] D. Marx and P. Blanc-Benon, “Numerical calculation of the temperature difference
between the extremities of a thermoacoustic stack plate,” Cryogenics, vol. 45, no. 3,
pp. 163–172, 2005.
[66] D. Marx and P. Blanc-Benon, “Computation of the mean velocity field above a stack
plate in a thermoacoustic refrigerator,” Comptes Rendus M´ ecanique, vol. 332, no. 11,
pp. 867–874, 2004.
[67] D. Marx and P. Blanc-Benon, “Numerical simulation of stack-heat exchangers cou-
pling in a thermoacoustic refrigerator,” AIAA Journal, vol. 42, no. 7, pp. 1338–1347,
2004.
12 BIBLIOGRAPHY
Chapter 2
Theory and model
2.1 General thermoacoustic theory
We derive three basic thermoacoustic differential equations. Two of them relate the
pressure and volume-flow rate for a given temperature gradient and geometry. To-
gether they form Rott’s wave equations. The third equation is the energy equation,
from which a temperature profile can be determined for a given pressure, volume-flow
rate, and geometry. The theory was developed in the seventies by Rott et al. [1–6] and
was later reviewed by Swift [7], for the application of parallel plates. The theory is gen-
eral in the sense that it applies to a whole range of different geometries. The size of
the flow channels can vary from smaller than the thermal penetration depth, in case
of regenerators, to a few times the thermal penetration depth, in case of a stack, to big
resonator tubes. The hydraulic radius R
H
, defined as the ratio of the cross-sectional
area of the flow A to the wetted perimeter Π, is a measure of the flow channel size. The
shape of the geometry can be parallel plates, circular cylinders, rectangular cylinders, or
any other cylindrical regular shape. The generalization to arbitrary pore geometry was
developed by Arnott et al. [8] using the assumption that the temperature at the bound-
aries is constant. In ’t panhuis et al. [9] repeated this generalization, without using this
assumption, by taking into account the heat equation in the solid, and also allowed for
gradual changes of the hydraulic radius of the pore.
In the derivation of the three basic thermoacoustic equations we make the following
major assumptions:
• The relevant parameters, pressure, density, velocity, and temperature can be lin-
earized, i.e. can be written as the sum of a mean value and a first-order variation.
The assumption that second-order effects are ignored yields to a major limitation
of this ’linear’ theory to the amplitudes. For high-pressure amplitudes (higher
than 10% of the mean pressure, according to literature) the second-order terms
cannot be neglected.
• The acoustic wave has a single, fixed frequency and is propagating in only one
direction that is along the tube axis. This implies that the hydraulic radius is re-
quired to be smaller than the wavelength, to avoid cut-off modes.
• The system is in steady state, i.e. at the end of a cycle the system returns to the
14 Theory and model
same state as in the beginning of the cycle.
• The local time-average mean velocity is equal to zero.
In addition to these major assumptions, some additional assumptions that are com-
mon in fluid dynamics and thermodynamics, like perfect heat contact between the solid
and gas, no-slip condition at wall, the gas is assumed to be a Newtonian fluid and a
perfect gas, the only relevant viscous force to be shear viscosity, and the kinetic energy
to be negligible in the energy equation.
2.1.1 Basic equations
Here we will derive the three basic equations that follow from conservation of mass,
momentum, and energy respectively. The second law of thermodynamics and the ideal
gas law are two additional basic equations. From the conservation of mass the well-
known continuity equation follows
∂ρ
∂t
+∇ (ρv) = 0, (2.1)
where ρ is the density, t the time, and v the velocity. From the conservation of mo-
mentum and approximating the viscous term by ν∇
2
v, the well-known Navier-Stokes
equation
∂v
∂t
+ (v ∇)v = −
∇p
ρ
+ν∇
2
v, (2.2)
can be derived. Here p is the pressure and ν the kinematic viscosity. By combining the
first law of thermodynamics with the Eqs. 2.1 and 2.2, Landau and Lifshitz [10] derived
a general equation of heat transfer for fluids
ρT
_
∂s
∂t
+v ∇s
_
= ∇ k∇T + (σ

∇) v, (2.3)
where s represents the entropy per unit mass, k the thermal conductivity, T the temper-
ature, and σ

the viscous stress tensor. Using ρTds = −dp +ρc
p
dT (for ideal gases),
with c
p
the specific isobaric heat capacity, we can rewrite the heat equation to
ρc
p
_
∂T
∂t
+v ∇T
_

_
∂p
∂t
+v ∇p
_
= ∇ k∇T + (σ

∇) v. (2.4)
The ideal gas law yields
p = ρR
s
T, (2.5)
where R
s
is the specific gas constant.
2.1.2 Linearization
The coordinate system for a flow channel with arbitrary cross section is shown in figure
2.1. The velocity in Cartesian coordinates v = (u, v, w), with u the x-component of v.
2.1 General thermoacoustic theory 15
x
z
y
gas area
solid
Figure 2.1: A geometry consisting of a homogeneous flow channel with an arbitrarily shaped
cross-section. The flow-channel length (in x direction) is much larger than its perimeter.
The basic equations can be simplified by linearizing the relevant variables:
p(x, t) = p
0
+Re¦p
1
(x)e
iωt
¦, (2.6a)
ρ(x, y, z, t) = ρ
0
(x) +Re¦ρ
1
(x, y, z)e
iωt
¦, (2.6b)
T(x, y, z, t) = T
0
(x) +Re¦T
1
(x, y, z)e
iωt
¦, (2.6c)
u(x, y, z, t) = Re¦u
1
(x, y, z)e
iωt
¦, (2.6d)
U(x, t) = Re¦U
1
(x)e
iωt
¦, (2.6e)
where ω is the angular frequency and U
1
is the volume flow rate, i.e. the sectional
integrated velocity over a surface ⊥x. The indexes denote the order. In case of the
pressure, for instance, p
0
is the zeroth-order term, which is the time mean value, and
p
1
is the first-order term, also called acoustic velocity. The drive ratio, D
r
, is defined as
the ratio of [p
1
[ and p
0
. Note that p
1
is not a function of y and z and that p
0
is constant,
as was shown with small-parameter asymptotics by In ’t panhuis [11]. He also showed
that T
0
and ρ
0
are only x dependent. We will now substitute the linearized variables
into the three basic equations and will then gather all first-order terms.
Substituting the linearized density and velocity into Eq. (2.1) yields
iωρ
1
+

∂x

0
u
1
) +ρ
0
∂v
∂y

0
∂w
∂z
= 0. (2.7)
Using a similar concept for the x-component of the momentum equation results in
iωρ
0
u
1
= −
dp
1
dx

0
ν
_

2
u
1
∂y
2
+

2
u
1
∂z
2
_
. (2.8)
A linearization of the heat equation, Eq. 2.4, yields
ρ
0
c
p
_
iωT
1
+ u
1
dT
0
dx
_
−iωp
1
= k
_

2
T
1
∂y
2
+

2
T
1
∂z
2
_
, (2.9)
where we assume that k is temperature independent.
16 Theory and model
2.1.3 Velocity profile
First we will derive the velocity profile for the special case of a parallel-plate channel
and then for the general case of a flow channel with an arbitrary cross section.
Parallel-plate channel:
x
z
y
2y0
l0
gas
solid
center of plate
center of plate
Figure 2.2: Parallel-plate geometry. The center of the gas channel is at y = 0 and the boundaries
are at y = ±y
0
. At the centers of the plates a periodic boundary condition in the y-direction is
applied. The plate thickness is 2l
0
.
We will solve the momentum equation, Eq. 2.8, first for a parallel-plate geometry
(figure 2.2), and then for an arbitrary geometry. For parallel plates, separated by 2y
0
with y = 0 in the center, the following boundary conditions apply:
u
1
(x, −y
0
) = 0, (2.10a)
u
1
(x, y
0
) = 0. (2.10b)
Since the problem is two-dimensional, it is independent of the z-position: u
1
= u
1
(x, y).
We assume to be sufficiently far away, i.e. more than twice the displacement length,
from the plate ends. Solving the differential equation, Eq. 2.8, leads to
u
1
=
i
ωρ
0
dp
1
dx
+ C
1
cosh[(1 + i)y/δ
ν
] + C
2
sinh[(1 + i)y/δ
ν
], (2.11)
where C
1
and C
2
are complex constants and δ
ν
the viscous penetration depth, defined
as
δ
ν
=
_

ω
. (2.12)
From the boundary conditions, Eqs. 2.10a and 2.10b, it follows that
C
1
=
i
ωρ
0
dp
1
dx
1
cosh[(1 + i)y
0

ν
]
, (2.13a)
C
2
= 0. (2.13b)
This makes the solution for parallel plates
u
1
=
i
ωρ
0
_
1 −
cosh[(1 + i)y/δ
ν
]
cosh[(1 + i)y
0

ν
]
_
dp
1
dx
. (2.14)
2.1 General thermoacoustic theory 17
General solution for a flow channel:
Other channel geometries (figure 2.1) lead to different solutions for u
1
, but all of them
can be written as a general solution
u
1
=
i
ωρ
0
[1 −h
ν
(y, z)]
dp
1
dx
, (2.15)
with h
ν
(y, z) a complex function, depending on the channel geometry, the frequency,
and weakly on x due to the temperature dependence of the viscosity. The volume flow
rate is the cross-sectional integrated velocity
U
1
=
__
A
u
1
dA =
iA
ωρ
0
[1 − f
ν
]
dp
1
dx
, (2.16)
where A is the cross-sectional area and f
ν
is the spatial average of h
ν
and is called the
viscous Rott function
f
ν
=
1
A
__
A
h
ν
(y, z)dA. (2.17)
Eq. 2.16 is an important equation, as it is one of the three equations that is used by our
model that will be discussed in section 2.2. For parallel plates
h
ν
=
cosh[(1 + i)y/δ
ν
]
cosh[(1 + i)y
0

ν
]
, (2.18a)
f
ν
=
tanh[(1 + i)y
0

ν
]
(1 + i)y
0

ν
. (2.18b)
The h
ν
and f
ν
functions of other geometries, cylindrical, triangular, rectangular, and pin
arrays, are derived by Arnott et al. [8].
2.1.4 Thermoacoustic continuity equation
Here we will derive a second equation, the thermoacoustic continuity equation, for a
general pore geometry. T
1
is derived from the linearized heat equation and is used to
determine ρ
1
, using the ideal gas law. ρ
1
is substituted into the linearized continuity
equation, which is integrated over the cross section.
By using the divergence theorem and that v = 0 at the pore boundaries, it can be
shown that
__
A
∂v/∂ydA =
__
A
∂w/∂zdA = 0. This equation can be used to integrate Eq.
2.7 over the cross section, resulting in
iAω¸ρ
1
¸ +
d
dx

0
U
1
) = 0, (2.19)
where ¸¸ denotes the cross-sectional average. Solving the heat equation (Eq. 2.9) in the
gas, with the boundary condition that the temperature at the solid is constant (T
1
= 0),
results in
T
1
=
(1 −h
κ
)p
1
ρ
0
c
p

1
iωA
dT
0
dx
(1 −h
κ
) −P
r
(1 −h
ν
)
(1 − f
ν
)(1 −P
r
)
U
1
, (2.20)
18 Theory and model
where P
r
= c
p
µ/k is the Prandtl number and h
κ
and f
κ
are channel-geometry dependant
functions that are very similar to h
ν
and f
ν
, with the only difference that δ
ν
is replaced
with δ
κ
. For instance, for parallel plates
h
κ
=
cosh[(1 + i)y/δ
κ
]
cosh[(1 + i)y
0

κ
]
, (2.21a)
f
κ
=
tanh[(1 + i)y
0

κ
]
(1 + i)y
0

κ
. (2.21b)
The thermal penetration depth δ
κ
is given by
δ
κ
=
_

ω
, (2.22)
with κ = k/ρ
0
c
p
the thermal diffusivity. The viscous and thermal penetration depths
are related by the Prandtl number
P
r
=
δ
2
ν
δ
2
κ
. (2.23)
If we take the cross-sectional average of Eq. 2.20
¸T
1
¸ =
(1 − f
κ
)p
1
ρ
0
c
p

1
iωA
dT
0
dx
(1 − f
κ
) −P
r
(1 − f
ν
)
(1 − f
ν
)(1 −P
r
)
U
1
. (2.24)
For flow channels with a hydraulic radius much larger than both the thermal and
viscous penetration depth (R
H
¸δ
ν
, R
H
¸δ
κ
) the Rott functions approach zero f
ν

0, f
κ
→0. In case of a parallel-plate stack this can be easily seen by taking y
0
→∞. For
gas parcels sufficiently far away from the flow-channel wall h
ν
→ 0, h
κ
→ 0, Eq. 2.20
can be simplified to
T
1
=
p
1
ρ
0
c
p

1
iωA
dT
0
dx
U
1
. (2.25)
From this equation it is easily seen that T
1
=0 if,
dT
0
dx
=
iωA
ρ
0
c
p
p
1
U
1
. (2.26)
This equation can be fulfilled in case of a standing wave with a critical temperature
gradient
(∇T)
crit
=
ωA[p
1
[
ρ
0
c
p
[U
1
[
. (2.27)
If [
dT
0
dx
[ = (∇T)
crit
, the gas parcels are at the same temperature as the plate during the
whole cycle, and consequently no heat is exchanged between the plates and the gas.
This would be the case for an inviscid thermoacoustic couple in steady state. Inviscid
standing-wave engines have [
dT
0
dx
[ > (∇T)
crit
and inviscid standing-wave refrigerators
or heat pumps have [
dT
0
dx
[ < (∇T)
crit
.
2.1 General thermoacoustic theory 19
To solve Eq. 2.24 we need boundary conditions for T
1
at the gas channel walls. For
parallel plates Swift [7] takes into account the heat equation in both the gas and the
solid. For flow channels with arbitrary cross sections this is more difficult. Arnott et
al. [8] assume that the temperature on the boundary of the flow channel is constant.
When the thermal capacity and conductivity of the solid is much larger than that of
the gas, this is a reasonable assumption. In ’t panhuis et al. [9] showed how to solve
differential equation for arbitrary pores without using this assumption. Just like Swift
did for parallel plates, they take into account the heat equation in both the gas and the
solid.
Using Eq. 2.5 in combination with Eq. 2.24, substituting it in Eq. 2.19, and using Eq.
2.16 results in
iAω(
ρ
0
p
0
p
1

ρ
0
T
0
¸T
1
¸) + U
1
d
dx
(
p
0
RT
0
) +ρ
0
dU
1
dx
= 0. (2.28)
Substituting Eq. 2.24 for ¸T
1
¸ leads to
iAω(
ρ
0
p
0
p
1

(1 − f
κ
)p
1
c
p
T
0
) +
dT
0
dx
(1 − f
κ
) −P
r
(1 − f
ν
)
(1 − f
ν
)(1 −P
r
)
ρ
0
U
1
T
0

ρ
0
U
1
T
0
dT
0
dx

0
dU
1
dx
= 0.
(2.29)
Using c
p
=
γ
γ−1
p
0
ρ
0
T
0
, combining terms, and dividing by ρ
0
leads to the final differential
equation
dU
1
dx
= −
iωA[1 + (γ −1) f
κ
]
γp
0
p
1
+
f
κ
− f
ν
(1 − f
ν
)(1 −P
r
)
dT
0
dx
U
1
T
0
. (2.30)
2.1.5 Total energy flow
The total energy flow
˙
E is the sum of the enthalpy flow and heat flow:
˙
E =
1
2
ρ
0
__
A
Re[h
1
u

1
]dA −(Ak + A
so
k
so
)
dT
0
dx
, (2.31)
with h the specific enthalpy, A
so
the solid area, k and k
so
, the thermal conduction coeffi-
cient of the gas and the solid, and

denoting the complex conjugate. For ideal gases
h
1
= c
p
T
1
. (2.32)
A substitution of this into Eq. 2.31 leads to
˙
E =
1
2
ρ
0
c
p
__
A
Re[T
1
u

1
]dA −(Ak + A
so
k
so
)
dT
0
dx
. (2.33)
Substituting Eq. 2.20 and u

1
=
1−h

ν
1−f

ν
U

1
A
into this equation and performing the integra-
tion results in
20 Theory and model
˙
E =
1
2
Re
_
p
1
U

1
_
1 −
f
κ
− f

ν
(1 + P
r
)(1 − f

ν
)
__
+
ρ
0
c
p
[U
1
[
2
2Aω(1 −P
2
r
)[1 − f
ν
[
2
Im( f
κ
+ P
r
f

ν
)
dT
0
dx
−(Ak + A
so
k
so
)
dT
0
dx
. (2.34)
This is the third important equation that is used in the model.
2.2 Model
In section 2.1 three 1-D differential equations were derived (Eq. 2.16, 2.30, and 2.34),
with three variables (U
1
(x), p
1
(x), and T
0
(x)), and one unknown constant (
˙
E). When the
boundary conditions are known, the system can be solved. In general, the set of three
differential equations cannot be solved analytically. For this reason a numerical code is
developed in Matlab. A number of thermoacoustic numerical codes exist, among which
the famous DeltaE, which was developed by Swift [12]. We have chosen to write our
own code, in collaboration with Wei Dai from the Chinese Academy of Sciences, for the
following reasons:
Flexibility: Having a code of our own provides flexibility. Measured data can be used
as input for the code.
Better insight: We want to gain more insight in the equations and algorithms behind
the code. This way we also learn about the limitations of the code.
Independence: Not being dependent on others.
Easy to use: In Matlab it is easy to write scripts for making good plots and connect
them to the model.
2.2.1 Transfer matrix
A stack with a fixed temperature gradient can be described by the first two equations,
Eqs. 2.16 and 2.30,
dp
1
dx
+ R
2
U
1
= 0, (2.35a)
dU
1
dx
+ R
1
p
1
−R
3
U
1
= 0. (2.35b)
Note that p
1
and U
1
are explicit functions of x, whereas the parameters
R
1
= iωA(1 + (γ −1) f
κ
)/γp
0
, (2.36a)
R
2
= iωρ
0
/(1 − f
ν
)A, (2.36b)
R
3
=
( f
κ
− f
ν
)dT
0
/dx
(1 − f
ν
)(1 −P
r
)T
0
, (2.36c)
2.2 Model 21
depend only implicitly on x by the T dependency. A substitution of (2.35a) into (2.35b)
leads to the ’wave’ equation of Rott:
d
2
p
1
dx
2
−R
3
dp
1
dx
−R
1
R
2
p
1
= 0. (2.37)
For constant R
1
, R
2
, and R
3
this differential equation has as general solution
p
1
= (C
1
cosh(R
4
x) + C
2
sinh(R
4
x)) exp(R
3
x/2), (2.38a)
U
1
= −
_
2R
4
C
2
+ R
3
C
1
2R
2
cosh(R
4
x) +
2R
4
C
1
+ R
3
C
2
2R
2
sinh(R
4
x)
_
exp(R
3
x/2),
(2.38b)
where R
4
=
_
R
2
3
/4 + R
1
R
2
and C
1
and C
2
are complex constants. The values of C
1
and
C
2
depend on the boundary conditions:
p
1
= p
L
, U
1
= U
L
, for x = x
L
= 0, (2.39)
where p
L
and U
L
are the pressure and volume flow rate at the left side of the compo-
nent. For C
1
= p
L
and C
2
= −
R
3
p
L
+2R
2
U
L
2R
4
both boundary conditions are fulfilled. The
pressure and volume flow rate at the right-hand side of the component, at x = x
R
, p
R
and U
R
, are linear combinations of p
L
and U
L
and can be written as
_
p
R
U
R
_
= T
_
p
L
U
L
_
. (2.40)
Here T is a transfer matrix given by
T = exp(R
3
x
R
/2)
_
_
cosh(R
4
x
R
) −
R
3
sinh(R
4
x
R
)
2R
4

R
2
R
4
sinh(R
4
x
R
)

R
1
R
4
sinh(R
4
x
R
) cosh(R
4
x
R
) +
R
3
sinh(R
4
x
R
)
2R
4
_
_
.
(2.41)
When R
3
x
R
¸1 the transfer matrix simplifies to
T =
_
_
cosh(
_
R
1
R
2
x
R
) −
_
R
2
R
1
sinh(
_
R
1
R
2
x
R
)

_
R
1
R
2
sinh(
_
R
1
R
2
x
R
) cosh(
_
R
1
R
2
x
R
)
_
_
. (2.42)
Since in this case cosh(R
4
x
R
) = cos(
ω
c
_
1+(γ−1) f
k
1−f
ν
x
R
), the pressure waves (using 2.38a)
can be written as
p
1
(x) = C
+
e
−iξx
+ C

e
iξx
, (2.43)
with ξ the complex wave number given by
ξ =
ω
c
¸
1 + (γ −1) f
k
1 − f
ν
. (2.44)
22 Theory and model
2.2.2 Energy equation and temperature
When p
1
, U
1
, and two boundary conditions for the temperature are known, T
0
(x) can
be determined from the energy equation. This equation is solved using a discretization,
as is shown in figure 2.3. The geometry is equidistant divided in n sections with size ∆x.
In the section i the temperature is T
0,i
, the pressure is p
1,i
, and so on. When considering
the energy flow equation at the dashed contour, it follows that
˙
E(x
L
) =
˙
Q
i
+
˙
E(x
R
), (2.45)
where
˙
Q
i
is the heat flow from the environment.
˙
E(x
L
) and
˙
E(x
R
) are computed from
Eq. 2.30. When ∆x is sufficiently small, we can use the approximations dT
0
/dx[
x
L
=
(T
0,i
−T
0,i−1
)/∆x and dT
0
/dx[
x
R
= (T
0,i+1
−T
0,i
)/∆x. We also use the approximations
T
0,x
L
= (T
0,i−1
+ T
0,i
)/2 and T
0,x
R
= (T
0,i
+ T
0,i+1
)/2 for the temperature, and the
pressure and volume flow rate at positions x
L
and x
R
are approximated similarly. Four
T0,i-1 T0,i T0,i+1
xL xR
T0,i+2 T0,i-2
∆x
Q
i
Figure 2.3: An overview of the discretization for the x-dependence that is used for the energy
flow equation.
˙
Q
i
is the heat exchange with the environment and T
0,i
is the discretized tempera-
ture.
different boundary conditions are possible: the energy flow at the left hand side of a
component; the energy flow at the right hand side; the temperature at the left hand
side; and the temperature at the right hand side. Out of these four different possible
conditions, two are required to solve the energy flowequation. Eq. 2.45 for 2 ≤ i ≤ n−1
in addition to the two boundary conditions, give us n equations for n unknown T
0
’s.
2.2.3 Iteration
We have a method to determine p
1
and U
1
when T
0
is known and a method to determine
T
0
when p
1
and U
1
are known. But in practice none of them are known. For this reason
an iteration technique is used (figure 2.4).
step 1 We make a first guess of the temperature T
0
(x)
step 2 We calculate p
1
(x) and U
1
(x) using the transfer matrix (Eq. 2.41)
2.3 Thermoacoustic devices 23
step 3 From the calculated p
1
(x) and U
1
(x) we compute T
0
(x) as explained in section
2.2.2
step 4 We compare the computed T
0
(x) with the profile from the previous calculation
step. If the difference is sufficiently small, the iteration process is completed, oth-
erwise we repeat steps 2 and 3.
3. calculate new T0(x)
2. calculate p1(x), U1(x)
1. initial guess of T0(x)
4. is
|T0
new
-T0
old
|
< ?
ε
no
yes
end
Figure 2.4: The schematic of the iteration algorithm.
2.3 Thermoacoustic devices
2.3.1 Introduction
Two types of thermoacoustic devices that make use of the thermoacoustic effect can be
distinguished:
Engines Atemperature difference, T
H
−T
C
, is used to generate power
˙
W
out
, as is shown
in figure 2.5(a).
Refrigerators or heat pumps Power
˙
W
in
is used to extract heat from one location and
release heat at another location at a higher temperature, as is shown in figure
2.5(b). In case of a refrigerator the objective is to create cooling power
˙
Q
C
by
keeping T
H
at the environmental temperature and in case of a heat pump the
objective is to generate heating power
˙
Q
H
by keeping T
C
at the environmental
temperature.
From the first law of thermodynamics it follows that in steady state
˙
W
in,out
=
˙
Q
H

˙
Q
C
.
For engines the efficiency is defined as η =
˙
W
out
/
˙
Q
H
. The maximum efficiency, which
can only be reached in an ideal device, is called Carnot efficiency η
C
, and η
C
= (T
H

24 Theory and model
T
C
)/T
H
. For refrigerators the coefficient of performance is defined as COP =
˙
Q
C
/
˙
W
in
and the Carnot coefficient of performance is defined as COP
C
= T
C
/(T
H
−T
C
).
TH
TC
engine Wout
QC
QH
(a) engine
TH
TC
refrigerator
or heat pump
Win
QC
QH
(b) refrigerator or heat
pump
Figure 2.5: Schematic drawing of the working principle of two different types of thermodynamic
devices. The squares represent thermal reservoirs. The two top reservoirs are at temperature T
H
and the two bottom reservoirs at T
C
. The arrows show the energy-flow directions.
a) An engine converts the heating power
˙
Q
H
at temperature T
H
partially into power
˙
W
out
and
transfer the waste heat
˙
Q
L
to the cold reservoir.
b) The power
˙
W
in
is used to generate a cooling power
˙
Q
L
in case of a refrigerator and a heating
power
˙
Q
H
in case of a heat pump.
The devices can also be categorized in a different way: stack-based devices (section
2.3.2) and regenerator-based devices (2.3.3). We define a stack as a geometry for which
the hydraulic radius of the pores is of similar size as the thermal penetration depth,
R
H
· δ
κ
, and a regenerator as a geometry for which the hydraulic radius of the pores
is much smaller than the thermal penetration depth, R
H
¸ δ
κ
. In literature these cat-
egories are often referred to as standing-wave and traveling-wave devices. Since, in
practice, the waves are never purely standing nor purely traveling, this nomenclature
can lead to confusion.
2.3.2 Stack-based devices
We consider a standing wave in a straight resonator, closed at both ends, at the first
resonance frequency (figure 2.6). The pressure and velocity are out of phase. If we
neglect the interaction of the wave with the wall, the wave is a perfect standing wave.
When we follow a gas parcel during one cycle, we can draw a pressure-displacement,
p −ζ, plot of it (figure 2.7(a)). The displacement ζ is the position of one gas parcel
during one cycle with respect to its average position x
0
. We write the pressure as
p = p
0
+[p
1
[ cos(ωt +θ
p
), (2.46)
and the displacement as
ζ = [ζ
1
[ cos(ωt +θ
ζ
), (2.47)
2.3 Thermoacoustic devices 25
ωt = 0
ωt = π
p-p0
x
ωt = π/2; 3π/2
(a) pressure
ωt = 3π/2
ωt = π/2
u
x
ωt = 0; π
(b) velocity
Figure 2.6: Acoustic pressure and velocity as functions of the position in a straight resonator at
the first resonance frequency, at four different phase angles ωt.
with θ a phase angle. A substitution from 2.47 into 2.46 results in two equations for the
pressure as a function of the displacement
p = p
0
+[p
1
[ cos
_
arccos(
ζ
ζ
1
) ±(θ
p
−θ
ζ
)
_
, −1 ≤
ζ

1
[
≤ 1. (2.48)
The two equations in (2.48) together describe an ellipse. If we normalize the pressure
p
/
= (p − p
0
)/[p
1
[ and displacement ζ
/
= ζ/[ζ
1
[, the ellipse axis is at an angle of 45

with the ζ-axis for [θ
p
−θ
ζ
[ < π/2 and -45

for π/2 < [θ
p
−θ
ζ
[ < π. In case of

p
−θ
ζ
[ = π/2 the ellipse is a circle and in case of [θ
p
−θ
ζ
[ = π or [θ
p
−θ
ζ
[ = 0 it is a
straight line. The eccentricity of the ellipse yields e = cos(θ
p
−θ
ζ
).
p
0
p0
(a) p-ζ
p
x0
(b) p-ζ at various positions
Figure 2.7: Pressure-displacement plots at different positions, x, in the resonator.
In figure 2.7(b) the p −ζ plots of parcels at different positions are shown for a stand-
ing wave. Since the pressure and displacement are in phase, i.e. θ
p
= θ
ζ
, it follows
that
p = p
0
+
[p
1
[

1
[
ζ, (2.49)
26 Theory and model
which is a straight line with steepness [p
1
[/[ζ
1
[ in the pressure-displacement plot. The
T −ζ plots have a similar shape as the p −ζ plots. For this reason the enthalpy flow
(Eq. 2.31) is zero in a standing wave.
If the pressure and displacement are out of phase, i.e. [θ
p
−θ
ζ
[ = π/2, if follows
that
(
p − p
0
[p
1
[
)
2
+ (
ζ

1
[
)
2
= 1, (2.50)
which is the description of a circle in a normalized pressure-displacement plot.
For the thermoacoustic effect to take place, we need an interaction of the sound wave
with a solid boundary. This is accomplished by installing a stack in the resonator, as is
shown in figure 2.8.
Even in an ideal engine, performing at Carnot efficiency, a nonzero power
˙
W =
(T
H
/T
C
−1)
˙
Q
C
is required. This power is delivered by a driver, which is a loudspeaker
in our configuration (figure 2.8a), delivering a power
˙
W
in
. The power is converted to an
acoustic energy flow
˙
E
in
in the resonator. As a consequence a traveling-wave compo-
nent is added to the sound wave. The cold part of the stack is connected to a reservoir
at a temperature T
C
using a heat exchanger and the hot part is connected to a reservoir
at temperature T
H
. The sound wave transfers an energy flow
˙
Q
C
from the cold to the
hot reservoir.
For the engine (figure 2.8b), the temperature gradient is used to generate an acoustic
power
˙
W
out
. A heating power
˙
Q
H
from the hot reservoir is required to sustain this
temperature gradient.
˙
Q
H
is partially converted to acoustic power and flows partially
to the cold (environmental) reservoir. The acoustic power
˙
W
out
can be used e.g. to drive
a piston.
Now we will follow a gas parcel that is oscillating near a stack plate, at a distance

κ
. The displacement and pressure as functions of time during one cycle are shown
in figure 2.9. In a standing wave the pressure is out of phase with the velocity and in
phase with the displacement. We will explain the working of a stack-based device from
a Lagrange point of view. In figure 2.9 four moments during a cycle are indicated by A,
B, C, and D. In figure 2.10 the thermodynamic processes occurring in-between the four
succeeding moments in time are schematically shown. In the left column a gas parcel in
a refrigerator and in the right column a parcel in an engine is shown. In both devices a
positive temperature gradient is present in the solid. For a refrigerator or a heat pump
the temperature gradient is lower than the critical temperature gradient, which is given
by Eq. 2.27, and in the engine the temperature gradient is lower than (∇T)
crit
. The gas
parcel is continuously moving and exchanging heat with the solid, but for simplicity we
neglect the parcel movement in steps B→C and D→A, and neglect the heat exchange in
steps A→B and C→D.
We will first discuss the four steps in the refrigerator or heat pump (left column):
• Step A→B: The parcel moves to the left while the pressure decreases, which causes
the parcel to expand and cool down.
• Step B→C: The parcel is at its leftmost position and the right half of the resonator
is at its minimum pressure. The temperature of the solid at the same x-position is
higher than that of the parcel, resulting in heat flow δq from the solid to the gas
parcel.
2.3 Thermoacoustic devices 27
TH
TC
Ein
QC
QH
E
hot heat
exchanger
cold heat
exchanger
stack speaker
Win
(a) Stack-based refrigerator/heat pump
TH
TC
Eout
QC
QH
E
hot heat
exchanger
cold heat
exchanger
stack
piston
Wout
(b) Stack-based engine
Figure 2.8: A schematic overview of a loudspeaker-driven refrigerator or heat pump (a) and a
thermoacoustic engine driving a piston.
A B C D
p-p0
t
p-p0
Figure 2.9: The displacement and pressure as functions of time during one cycle.
28 Theory and model
• Step C→D: The parcel moves to the right while the pressure increases, which
causes the parcel to compress and heat up.
• Step D→A: The parcel is at its rightmost position and the right half of the resonator
is at its maximum pressure. The temperature of the solid at the same x-position is
lower than that of the parcel, resulting in heat flow δq from the gas parcel to the
solid.
Now we will give a stepwise description of the processes in an engine (right col-
umn):
• Step A→B: The parcel moves to the left while the pressure decreases, which causes
the parcel to expand and cool down.
• Step B→C: The parcel is at its leftmost position and the left half of the resonator
is at its maximum pressure. The temperature of the solid at the same x-position is
lower than that of the parcel, resulting in heat flowδq from the parcel to the solid.
• Step C→D: The parcel moves to the right while the pressure increases, which
causes the parcel to compress and heat up.
• Step D→A: The parcel is at its rightmost position and the right half of the resonator
is at its minimum pressure. The temperature of the solid at the same x-position
is lower than that of the parcel, resulting in heat flow δq from the solid to the gas
parcel.
2.3 Thermoacoustic devices 29
column 1: column 2:
refrigerator or heat pump engine
A→B: movement, pressure decrease A→B: movement, pressure decrease
δq δq
B→C: heat absorption, expansion B→C: heat rejection, compression
C→D: movement, pressure increase C→D: movement, pressure increase
δq δq
D→A: heat rejection, compression D→A: heat absorption, expansion
Figure 2.10: Moving with a gas parcel during one cycle in a refrigerator (column 1) and an
engine (column 2).
30 Theory and model
2.3.3 Regenerator-based devices
Aregenerator-based engine is shown in figure 2.11. The right open-end of the T-junction
is connected to a large resonator. The size of the loop is much smaller than the wave-
length. The regenerator acts like an amplifier in the ideal case. The volume flow rate at
the hot side of an ideal regenerator is amplified by a factor T
H
/T
C
in comparison with
the cold side. Since the pores in the regenerator are relatively small (R
H
¸ δ
κ
), the
flow through the regenerator could lead to a significant viscous dissipation. To reduce
this effect the loop geometry is designed in such a way that the volume flow rate U
1
is
relatively small at the position of the regenerator. In many practical regenerator-based
devices [p
1
/U
1
ρ
0
c[ is in-between 10 and 20 at the position of the regenerator [13]. Since
only the traveling-wave component of the sound wave have a useful effect on the per-
formance of a regenerator-based device, in practice p
1
and U
1
are in phase at the location
of the regenerator. To fulfill these requirements a change in cross-section is included in
the feedback loop (inertance to compliance).
The heating power
˙
Q
H
is converted to an acoustic energy flow
˙
E
out
, which is split
up in the T-junction. One part of
˙
E
out
flows into the resonance tube, where it is used to
drive a piston, producing power
˙
W
out
. The other part of
˙
E
out
flows into the inertance,
through the compliance, and is rejected as wasted heat
˙
Q
C
into the heat sink T
C
. In
an ideal regenerator the enthalpy flow is zero and
˙
E in the regenerator only consists of
conduction.
QC
QH
E
hot heat
exchanger
cold heat
exchanger
regenerator
TH
TC
piston
intertance
compliance
T-junction
environmental
heat exchanger,
flow straightener
thermal
buffer tube
resonance tube
Ein
Eout
Wout
Figure 2.11: A schematic drawing of a regenerator-based engine.
Different geometries can be used for regenerator-based engines. In chapter 6 a coax-
ial design, developed by De Blok [14], is discussed.
2.3 Thermoacoustic devices 31
T
x0
cold heat
exchanger
hot heat
exchanger
regenerator
thermal
buffer tube
compliance
Figure 2.12: The temperature T of gas parcels as a function of their displacement, during a
cycle, at different positions x
0
. In the compliance the mean temperature T
0
as function of x
0
is constant, in the regenerator it increases linearly, and in the thermal buffer tube it decreases
slowly.
Regenerator-based devices are often referred to as traveling-wave devices since the
pressure and velocity are ideally in phase in the loop. As a consequence the waves
transfer energy and lead to an enthalpy flow. In figure 2.12 the T-ζ plots in and nearby
the regenerator are shown. In the compliance T and ζ are out of phase, resulting in
ellipses in the T-ζ plots. Since the parcels move towards the regenerator with higher
temperature and enthalpy than when they move away from the regenerator, this leads
to a positive enthalpy flow in the compliance.
In an ideal regenerator the gas parcels have the same temperature as the solid. In the
T-ζ plot this results in straight lines with steepness equal to ∂T/∂x of the solid. Since
the gas parcels have the same temperature when moving from the hot to the cold side
as from the cold to the hot side, the enthalpy flow in an ideal regenerator is zero. The
difference in the energy flow between compliance and regenerator is wasted to the cold
reservoir at temperature T
C
in the cold heat exchanger. In the thermal buffer tube T and
ζ are out of phase, just as in the compliance, which corresponds to positive enthalpy
flow
˙
W
out
, that is taken from the hot reservoir. The heating power
˙
Q
H
is larger than the
cooling power
˙
Q
C
and the difference is converted to power to drive a piston at the end
of the resonator tube.
In a regenerator-based device the enthalpy flow is zero inside the regenerator and
nonzero in all other parts of the device. Compare this to a stack-based device, in which
the enthalpy flow out of the stack is close to zero, whereas the enthalpy flow into the
stack is significant. For both devices the difference in enthalpy flow between into and
out of the stack or regenerator leads to heat flows in and out of the heat exchangers.
The heat transfer between solid and gas is imperfect in a stack, inevitably leading to an
entropy production. In an ideal regenerator the heat transfer between solid and gas is
adiabatic. By avoiding the entropy production due to heat transfer, regenerator-based
devices potentially have a higher efficiency.
32 Theory and model
Bibliography
[1] N. Rott, “Damped and thermally driven acoustic oscillations in wide and narrow
tubes,” Journal of Applied Mathematics and Physics, vol. 20, pp. 230–243, 1969.
[2] N. Rott, “Thermally driven acoustic oscillations. part II: Stability limit for helium,”
Journal of Applied Mathematics and Physics, vol. 24, pp. 54–72, 1973.
[3] N. Rott, “Thermally driven acoustic oscillations. part III: Second-order heat flux,”
Journal of Applied Mathematics and Physics, vol. 26, pp. 43–49, 1975.
[4] N. Rott and G. Zouzoulas, “Thermally driven acoustic oscillations. part IV: Tubes
with variable cross-section,” Journal of Applied Mathematics and Physics, vol. 27,
pp. 197–224, 1976.
[5] N. Rott, “Thermoacoustics,” Advances in Applied Mechanics, vol. 20, pp. 135–175,
1980.
[6] N. Rott, “The influence of heat conduction on acoustic streaming,” Journal of Ap-
plied Mathematics and Physics, vol. 25, pp. 417–421, 1974.
[7] G. Swift, “Thermoacoustic engines,” Journal of the Acoustical Society of America,
vol. 84, pp. 1146–1180, 1988.
[8] W. P. Arnott, H. E. Bass, and R. Raspet, “General formulation of thermoacoustics
for stacks having arbitrarily shaped pore cross sections,” Journal of the Acoustical
Society of America, vol. 90, no. 6, pp. 3228–3237, 1991.
[9] P. H. M. W. in ’t panhuis, S. W. Rienstra, J. Molenaar, and J. J. M. Slot, “Weakly non-
linear thermoacoustics for stacks with slowly varying pore cross-sections,” Journal
of Fluid Mechanics, vol. 618, pp. 41–70, 2009.
[10] L. Landau and E. Lifshitz, Fluid Mechanics. Pergamon, 1982.
[11] P. in ’t panhuis, Mathematical aspects of Thermoacoustics. PhD thesis, Einhoven Uni-
versity of Technology, 2009.
[12] B. Ward and G. Swift, “Design environment for low-amplitude thermoacoustic en-
gines,” Journal of the Acoustical Society of America, vol. 95, pp. 3671–3672, 1996.
[13] S. Backhaus and G. W. Swift, “A thermoacoustic-stirling heat engine: Detailed
study,” Journal of the Acoustical Society of America, vol. 107, pp. 3148–3166, 2000.
[14] de Blok and van Rijt, “(wo/1999/020957) thermo-acoustic system,” 1999.
34 BIBLIOGRAPHY
Chapter 3
Electroacoustics
3.1 Introduction
Here we will study the coupling of a loudspeaker to an acoustic set-up. First we will
model the loudspeaker behavior by coupling two electric variables, the voltage over
and current through the coil, to two mechanical variables, the coil movement and the
Lorentz force, which are coupled to two acoustic variables, the pressure and the volume
flow rate. The electroacoustical model in addition to the thermoacoustic model (chapter
3.2) allows us to model a complete set-up.
To study nonlinearities we had to build an experimental set-up in which high drive
ratios (up to 20%) could be reached. We choose the, in thermoacoustics commonly used,
set-up of a loudspeaker connected to a resonator. In the development of this set-up the
model was used to predict the attainable drive ratios. The set-up was to be used for
acoustic measurement as well as flow visualization measurements and had to fulfill the
following requirements:
• sufficiently high drive ratios, up to 20%,
• suitable for acoustic measurement,
• suitable for flow visualization measurements,
• flexible in frequency and amplitude,
• fit into a lab of 4.2 m by 3.5 m.
The experimental set-up is described and explained in section 3.3. Once the set-up was
completed we validated our model (section 3.4). Besides empty resonators also res-
onators with parallel-plate stacks are studied.
In stack-based thermoacoustics heat-pumps or refrigerators are often driven by
moving-coil loudspeakers. Therefore it is important to have a complete model of a
driver coupled to a geometry. Bailliet et al. [1] used linear speaker equations to cal-
culate the speaker amplitudes (current, displacement, and force) for known frequency,
input voltage and acoustic impedance of the geometry coupled to the speaker. Tijani et
al. [2] introduced a concept of using an adjustable back volume to match the mechani-
cal resonance frequency with the acoustical resonance frequency. Marx et al. [3] took a
36 Electroacoustics
closer look at the agreement between the speaker model and experimental results, but
only for an empty resonator. We performed a similar study, but with a stack enclosed in
the resonator, to test a model based on transfer matrices.
3.2 Theory and model
3.2.1 Speaker equations
We consider a loudspeaker, driven with a constant angular frequency ω at a voltage V.
The behavior of the loudspeaker depends on the acoustical impedance of the system it
is driving. The speaker is connected to a thermoacoustic device with a known acoustic
impedance. The acoustic impedance is defined as the ratio of pressure to volume flow
rate at the speaker membrane position [4]
Z
ac
=
p
1
U
1
. (3.1)
To reduce sound waves leaking to the environment, the loudspeaker is enclosed in a
back enclosure.
By using Newton’s second law and adding all the forces acting on the loudspeaker
coil we find the mechanical equation describing the motion of the coil
M
me
d
2
ζ
dt
2
= F −R
me

dt
−[k
me
+ k
b
]ζ − A
2
c
Z
ac

dt
, (3.2)
where M
me
is the moving mass, ζ the coil displacement, F the Lorentz force, R
me
the
friction constant, k
me
the spring constant of the speaker, k
b
the gas spring constant of
the back enclosure, A
c
the cone area, and Z
ac
the acoustical impedance of the geometry
connected to the speaker. The Lorentz force is given by
F = B
l
I, (3.3)
with B
l
the motor force factor and I the current through the coil. The spring constant of
the back enclosure is given by
k
b
=
γp
0
A
2
c
V
b
, (3.4)
where V
b
is the average volume of the back enclosure.
For determining an electric equation the speaker is modeled as a resistance and coil
in series, and the total voltage is a superposition of the speaker voltage V and the in-
duced voltage V
ind
V + V
ind
= R
el
I + L
el
dI
dt
, (3.5)
with R
el
the DC resistance and L
el
the self-inductance of the coil. The induced voltage
due to the coil movement yields
V
ind
= −B
l

dt
, (3.6)
3.2 Theory and model 37
with u
c
= dζ/dx the coil velocity.
We linearize the speaker variables analogously to Eqs. 2.6
V(t) = Re¦V
1
e
iωt
¦, (3.7a)
I(t) = Re¦I
1
e
iωt
¦, (3.7b)
F(t) = Re¦F
1
e
iωt
¦, (3.7c)
ζ(t) = Re¦ζ
1
e
iωt
¦, (3.7d)
u
c
(t) = Re¦u
c1
e
iωt
¦. (3.7e)
We define the mechanical impedance as the Lorentz force to velocity ratio
Z
me
=
F
1
u
1
. (3.8)
The mechanical impedance can be found by substituting the linearized variables from
Eq. 3.7
Z
me
= R
me
+ i (M
me
ω−[k
me
+ k
b
]/ω) + A
2
c
Z
ac
. (3.9)
The total effective electrical impedance is defined as the voltage to current ratio
Z
el
=
V
1
I
1
. (3.10)
A substitution of Eqs. 3.7 into Eq. 3.5 results in
Z
el
= R
el
+ iωL
el
+ B
l
u
c1
I
1
. (3.11)
Using Eqs. 3.8 and 3.3 the last term of 3.11 can be rewritten
Z
el
= R
el
+ iωL
el
+
B
2
l
Z
me
. (3.12)
For a given voltage using Eq. 3.10 the current can be calculated. Next, using Eq.
3.3, the Lorentz force can be determined and by using Eq. 3.8 the coil velocity can be
computed. Multiplying the coil velocity with the cone area results in the volume flow
rate and Eq. 3.1 gives us the pressure.
3.2.2 Acoustic impedance
If all parameters are known the speaker equations can be solved. The only unknown
parameter is Z
ac
, which depends on the geometry that the speaker is coupled to. This
geometry can consist of multiple components varying in diameter, length, shape and
temperature distribution. To determine the behavior of the system we need to deter-
mine the acoustic impedance of the complete geometry. Our method to do this is to
calculate the transfer matrix of each individual component and then by a matrix multi-
plication determining the total transfer matrix.
38 Electroacoustics
The transfer matrix T relates the p −u vectors at both sides of a component (x
L
and
x
R
) and is given by Eq. 2.41. In this chapter the influence of the temperature gradient
on the acoustic equations is neglected. In this case the simplified transfer matrix of Eq.
2.42 can be used. The total transfer matrix of a system of n components is
T
total
= T
1
T
2
. . . T
n
. (3.13)
If the right end of the geometry is closed (U
R
= 0), the pressure and volume flow rate
have to fulfill the condition
T
21
p
L
+ T
22
U
L
= 0, (3.14)
which results in the impedance at the speaker side of the geometry
Z
ac
=
p
1
U
1
¸
¸
¸
¸
x=0
= −T
total,22
/T
total,21
. (3.15)
3.3 Experimental Setup
The set-up consists of a moving-coil loudspeaker
1
enclosed in a cylindrical back volume,
V
b
, as is shown in figure 3.1. The speaker parameters are presented in table 3.1. The back
volume is connected to a horn by a flexible rubber membrane to reduced the influence
of mechanical speaker vibrations on the resonator. The inner diameter of the speaker
side of the horn D
0
= 0.356 m, the inner diameter at the resonator-tube side D
end
= 60
mm, and the length L
horn
= 1.01 m. The resonator tube has an inner diameter, D
tube
, of
60 mm and a length of 1.707 m.
Table 3.1: Speaker parameters.
Resistance R
el
2.95 Ω
Self inductance @ 1 kHz L
el
1.06 mH
Motor force factor B
l
15.39 Tm
Moving mass M
me
248.15 g
Spring constant k
me
6.535 N/mm
Friction constant R
me
5.73 Ns/m
Resonance frequency f
res
25.2 Hz
Cone area A
c
0.078 m
2
Maximum excursion ζ
max
20.32 mm
Back volume V
b
60.3 L
At different locations in the tube a parallel-plate stack can be placed. On both sides
of the stack three microphones are mounted with their measurement surface at the inner
tube wall. These piezoresistive pressure transducers
2
are connected to an Endevco DC-
differential voltage amplifier
3
. The microphone positions, stack position and end-plate
1
JBL W15GTi 15
//
woofer
2
Endevco 8510B-2
3
Endevco 136 DC
3.3 Experimental Setup 39
V
I
0
Lhorn
xend
Vback
p1 p2 p4 p5 p6
x2 x1 x6 x5 x3 x4 xL xR
Dend
D0
p3
Figure 3.1: A schematic drawing of the experimental set-up (not to scale). A loudspeaker,
enclosed in a back enclosure, is connected to a conical horn. Inside the resonator tube a parallel-
plate stack is located. Six microphones, p
1
. . . p
6
, are mounted in the tube wall. The voltage over
and current through the speaker coil are registered. The conical horn starts with diameter D
0
and ends with the tube diameter D
end
.
position are presented in table 3.2.
Table 3.2: Microphone and stack-end positions.
x
1
1.210 m
x
2
1.410 m
x
3
1.860 m
x
L
2.049 m
x
R
2.251 m
x
4
2.367 m
x
5
2.517 m
x
6
2.667 m
x
end
2.717 m
The horn diameter varies as function of the axial position according to four different
curves:
D
0
, 0 ≤ x < a
0
(3.16)
a
1
cosh([x −a
0
]/a
2
) + a
3
, a
0
≤ x < 0.14 m (3.17)
tan(18
o
)(a
4
+ a
0
−x) + a
5
, 0.14 m ≤ x < 0.29 m (3.18)
D
end
cosh((L
horn
−x)/a
6
), 0.29 m ≤ x ≤ L
horn
(3.19)
40 Electroacoustics
with D
0
= 0.356 m, a
0
= 0.050 m, a
1
= -0.0080 m, a
2
= 0.04487 m, a
3
= 0.364 m, a
4
=
0.240 m, a
5
= 0.236 m, a
6
= 0.351 m, and L
horn
= 1.01 m. Both D(x) and dD/dx have no
discontinuities, in order to avoid boundary-layer separation.
The displacement of the speaker membrane is measured using a laser triangulation
sensor
4
. The current through the speaker is recorded as well as the voltage, using a
voltage divider. Using the temperature dependance of the DC-resistance of the coil, we
can estimate the coil temperature. The coil resistance is determined by applying a small
DC current through the coil.
A sinusoid with a certain amplitude and frequency is generated by a function gen-
erator and, after amplification, used as the input for the loudspeaker. All measured
variables (six microphones, voltage, current, displacement) are recorded using a data-
acquisition card at a frequency of 200 kHz. The amplitudes and phase angles (in relation
to a reference signal) of all signals are determined using a digital lock-in function, with
the function-generator signal as a reference signal.
Inside the resonator tube a stack of parallel plates made of perspex is placed. The
plate distance is 1.0 mm, the plate thickness is 1.0 mm, and the stack length x
R
− x
L
=
206.0 mm.
3.4 Results
3.4.1 Empty resonator
As the acoustics department of Philips have experience with loudspeakers, we have
cooperated in the modeling of the speaker and resonator. Together with Ouweltjes from
Philips [5] we have calculated the pressures and velocities as function of the position in
the resonator and as function of the frequency.
We use Eq. 2.42 to determine the transfer matrix of the resonator tube. For the f
ν
and f
κ
functions we use the boundary layer approximation. The transfer matrix of the
horn is more difficult to calculate, since its diameter is not constant. For this reason
we divided the horn into 200 sections with length L
horn
/200. The sections are modeled
as tubes with constant diameters. The diameters of the sections are chosen in such a
way that the horn shape is approximated. By multiplying the transfer matrices of the
individual sections, the transfer matrix of the whole horn is determined.
This method of discretization of the horn diameter is tested for an exponential cone,
for which the transfer matrix is known analytically and works very well (within 0.1%
accuracy).
By multiplying all individual transfer matrices, we calculate the transfer matrix of
the whole resonator T
total
which can be used to determine the acoustical impedance Z
ac
from Eq. 3.15. By coupling the acoustic impedance of the geometry to the electroacous-
tical model, we can calculate the pressure and velocity amplitudes at every position in
the resonator for a specific frequency range (10 to 400 Hz in this case). We use a color
plot to show both the frequency and position dependency in one figure 3.2. In figure
3.2a the pressure amplitude [p
1
[ is shown and in figure 3.2b the velocity amplitude. The
colors are a measure of the amplitude, as is shown by the color bars to the right of the
plots.
4
LMI LDSc90/45
3.4 Results 41
x [m]
f
[
H
z
]


0 1 2
10
1
10
2
p
1
[
k
P
a
]
10
10
10
10
-3
-5
-1
1
(a) pressure amplitude [p
1
[
x [m]
f
[
H
z
]


0 1 2
10
1
10
2
u
1
[
m
/
s
]
10
10
10
10
-3
-5
-1
1
(b) velocity amplitude [u
1
[
Figure 3.2: Numerical calculation of the pressure amplitude (a) and velocity amplitude (b)
as function of the frequency f , logarithmically on the vertical axis, and of the position in the
resonator x, linearly on the horizontal axis. The colors are a measure of the amplitude on a
logarithmical scale, as is shown by the color bars to the right of the plots. The white dashed
vertical lines represent the position x = L
horn
.
42 Electroacoustics
The first two resonance frequencies can be easily recognized in both figure 3.2a and
3.2b by the horizontal lines of high intensity (45 Hz and 133 Hz). The exponential horn
is located in-between x = 0 m and x = 1.01 m. At the resonance frequencies the nodes
and anti-nodes in both the pressure and velocity can be clearly observed. From these
plots we can also easily observe for which positions and frequencies we can reach the
highest velocities and the highest pressures.
3.4.2 Fixed speaker voltage
First we consider a resonator with a parallel-plate stack enclosed in it with a plate thick-
ness of 1 mm and a plate distance of 3 mm. In figures 3.3 and 3.4 the measured data are
compared with the model. In figures 3.3(a)-3.3(d) the amplitudes of the speaker voltage,
speaker current, voltage-to-current ratio, speaker excursion and in figures 3.4 the pres-
sure amplitudes at four positions are shown as functions of the frequency. The dashed
plots represent the measured data and the solid curves the model.
0 50 100 150 200
0
2
4
6
8
f [Hz]
|
V
1
|

[
V
]
(a) [V
1
[
0 50 100 150 200
0
0.5
1
1.5
2
2.5
f [Hz]
|
I
1
|

[
A
]


measurement
model
(b) [I
1
[
0 50 100 150 200
0
10
20
30
40
50
f [Hz]
|
V
1
/
I
1
|

[
V
/
A
]
(c) [V
1
/I
1
[
0 50 100 150 200
0
0.5
1
1.5
f [Hz]
|
ζ
1
|

[
m
m
]
(d) excursion [ζ
1
[
Figure 3.3: Amplitudes of (a) voltage, (b) current, (c) voltage-to-current ratio, and (d) cone
excursion as functions of the frequency. Measurements are plotted dashed and the model solid.
3.4 Results 43
0 50 100 150 200
0
0.5
1
1.5
2
f [Hz]
|
p
1
,
1
|

[
k
P
a
]


measurement
model
(a) pressure p
1,1
0 50 100 150 200
0
0.5
1
1.5
2
2.5
3
f [Hz]
|
p
1
,
2
|

[
k
P
a
]
(b) pressure p
1,2
0 50 100 150 200
0
0.5
1
1.5
2
2.5
f [Hz]
|
p
1
,
3
|

[
k
P
a
]
(c) pressure p
1,3
0 50 100 150 200
0
1
2
3
4
5
f [Hz]
|
p
1
,
4
|

[
k
P
a
]
(d) pressure p
1,4
Figure 3.4: The pressure amplitudes [p
1
[ measured with microphones 1 to 4 at positions x
1
to
x
4
as functions of the frequency. Measurements are plotted dashed and the model solid.
44 Electroacoustics
In the frequency range 13 to 24 Hz, the set-up vibrated heavily. Therefore no mea-
surements were made in this frequency range. The amplifier driving the speaker is un-
able to maintain a constant V
1
of 6.8 V for low frequencies (figure 3.3(a)). The measured
voltage is used as an input parameter for the model. The measured [V
1
/I
1
[ (figure 3.3(c))
is in good agreement with the model, especially for relatively low frequencies (smaller
than 120 Hz). At relatively high frequencies (higher than 150 Hz) the calculated value
is 20% lower than the measured one. The measured displacement corresponds well
with the model, except in the frequency range 25 to 75 Hz. The discrepancy is probably
caused by the vibrations of the set-up. The measured pressures agree very well with the
model, with the exception that the measured peak values are lower than the computed
ones. The deviations at the peak values are probably caused by nonlinear effects. We
will look into the measurement at the resonance frequencies in more detail in the next
subsection.
3.4.3 Fixed frequency
The first resonance frequency is 44 Hz. We performed an amplitude sweep from 1 to
52 V at 44 Hz (figure 3.5). For the fixed-frequency measurements a different stack, with
plate thickness and plate distance both 2 mm, is used.
According to the theory (section 3.2.1), all variables are expected to be linear with the
speaker voltage. The speaker current (figure 3.5a) is linear with the speaker voltage, but
the other variables (displacement and pressures, figure 3.5b-d) are strongly nonlinear,
which is in agreement with Marx et al. [3]. The nonlinearity can be caused by various
effects:
1. nonlinearities in the electric equation of the speaker, e.g. the heating of the coil,
leading to a higher electric resistance.
2. nonlinearities in the mechanical equation of the speaker, e.g. an increase of the
friction coefficient as the cone excursion increases.
3. nonlinearities of the sound wave at high drive ratios effecting the acoustical
impedance.
4. small changes in the resonance frequency, due to the heating up of the gas.
Since the current is very linear with the speaker voltage, nonlinearities in the electric
equation can not be the cause of the nonlinearity of the pressure and cone excursion
with the speaker voltage. Also from measurement of the DC resistance, it was found
not to increase significantly. Therefore we can conclude that the effect of the coil heating
can be neglected.
Small changes in the resonance frequency can lead to large changes in the measured
pressures, when measuring at a peak that is very steep. When the resonance frequency
changes only a little bit (1 Hz for instance) the system is not measuring at its resonance
frequency, resulting in a much lower pressure amplitude. When we increase the speaker
voltage, more power goes into the system, resulting in a higher gas temperature, leading
to a higher resonance frequency. While the speaker frequency remains at 44 Hz and
the resonance frequency grows above 44 Hz, the difference between measurement and
calculation increases. In theory this effect could be an explanation of the discrepancy
3.4 Results 45
0 20 40 60
0
2
4
6
8
10
|V
1
| [V]
|
I
1
|

[
A
]


measurement
model
(a) current [I
1
[
0 20 40 60
0
2
4
6
8
|V
1
| [V]
|
ζ
1
|

[
m
m
]
(b) excursion [ζ
1
[
0 20 40 60
0
0.05
0.1
0.15
0.2
|V
1
| [V]
|
p
2
|
/
p
0
(c) pressure [p
1,2
[
0 20 40 60
0
0.1
0.2
0.3
0.4
|V
1
| [V]
|
p
6
|
/
p
0
(d) pressure [p
1,6
[
Figure 3.5: The current, membrane excursion and pressures as functions of the speaker voltage.
The plate thickness and plate distance of the stack are both 2 mm. Measurements are plotted
dashed and the model solid.
46 Electroacoustics
between measurements and theory at high amplitudes in figure 3.5. To verify whether
the resonance frequency truly changes at higher amplitudes we have compared two
different measurements of the pressure amplitudes as function of the frequency at two
different but fixed speaker voltages. We do not see any changes in resonance frequency
of the two measurements, thus we conclude that this effect can not be the explanation
of the discrepancy in figure 3.5.
Two options that can explain the discrepancy remain. Additional measurements at
100 Hz are performed to determine which of these two nonlinear mechanisms is domi-
nating. In figure 3.6a the pressure is shown as functions of the speaker voltage for 100
Hz. Since the resonator is not in resonance at this frequency, the pressure amplitudes
are significantly lower than at 44 Hz. We expect the nonlinear acoustic effects to reduce.
The measured pressure is still nonlinear with the speaker voltage, so we can conclude
that the nonlinear acoustic effects are not dominant here. The nonlinearities of the mea-
sured pressures to the speaker voltage are caused by nonlinearities in the mechanical
equation of the speaker.
Now we will look at the relative deviation of the measured pressure to the calcu-
lated pressure (according to our linear model). The relative deviations of the pressures
are plotted as functions of the modeled cone excursion (figure 3.6b) and modeled pres-
sure (figure 3.6c), at 44 Hz and at 100 Hz. These plots show that the relative deviations
are linear with ζ
1
(figure 3.6b) and that both plots have the same steepness. The steep-
ness of the two plots in figure 3.6c are very different. From this we conclude that the
nonlinearities are probably caused by speaker nonlinearities due to high cone excur-
sions. To strengthen this conclusion, we have plotted the relative pressure deviations
for three different microphones for both frequencies as functions of the cone excursion
in figure 3.6d. The steepness of the six plots are consistent.
3.5 Discussion and conclusion
We have succeeded in building a set-up that can reach drive ratios of more than 20%,
even in the presence of a stack in the resonator. The decoupling of speaker and resonator
with an elastic membrane works well for frequencies above 30 Hz. A recommendation
for future studies is to install the membrane in-between the horn end and the resonator
tube, instead of between the speaker and the beginning of the horn. This reduces the
forces between the two decoupled parts, as the cross-sectional area is smaller at this
position.
The results of the electro-acoustical model are in good agreement with the measure-
ments. The peak frequencies are predicted within an accuracy of 1 Hz. The cone ex-
cursion shows the highest deviation (20%) at 24 Hz, due to mechanical vibrations of
the set-up. The calculated pressures, peaks excluded, are within 5% accuracy of the
measurements. At the first resonance frequency the discrepancy between pressure cal-
culations and measurements grows with the amplitude up to 50%, at a drive ratio of
20%. The discrepancy between theory and measurement may have several causes:
• Nonlinearities within the loud-speaker, due to a large cone excursion.
• Acoustic nonlinearities within the resonator.
• Small changes in the resonance frequency. Since the peaks are very narrow, small
3.5 Discussion and conclusion 47
0 20 40 60
0
0.005
0.01
0.015
0.02
0.025
|V
1
| [V]
|
p
6
|
/
p
0


measurement
model
(a) [p
1,6
[ versus [V
1
[ at 100 Hz
0 2 4 6 8
−0.4
−0.3
−0.2
−0.1
0

1
|
|


p
1
,
6
|
/
|
p
1
,
6
|


44 Hz
100 Hz
(b) [∆p
1,6
[ versus [ζ
1
[
0 0.1 0.2 0.3 0.4
−0.4
−0.3
−0.2
−0.1
0
|p
1,6
|/p
0
|


p
1
,
6
|
/
|
p
1
,
6
|


44 Hz
100 Hz
(c) [∆p
1,6
[ versus [p
1,6
/p
0
[
0 2 4 6 8
−0.4
−0.3
−0.2
−0.1
0
0.1

1
|
|


p
1
|
/
|
p
1
|


mic 1, 44 Hz
mic 1, 100 Hz
mic 2, 44 Hz
mic 2, 100 Hz
mic 3, 44 Hz
mic 3, 100 Hz
(d) p
1,6
, p
4,6
, and p
6,6
Figure 3.6:
(a) The pressure at the end of the resonator p
6
as a function of the speaker voltage at 100 Hz.
Measurements are plotted dashed and the model solid.
(b) The relative pressure deviations as functions of the cone excursion, for two different frequen-
cies (dashed at 44 Hz and solid at 100 Hz).
(c) The relative pressure deviations as functions of the drive ratio, for two different frequencies
(dashed at 44 Hz and solid at 100 Hz).
(d) The relative pressure deviations for microphones 1 (dashed), 4 (dash-dotted), and 6 (solid),
measured at two different frequencies (grey at 44 Hz and black at 100 Hz).
48 Electroacoustics
deviations of the resonance frequency, for instance due to the heating up of the
resonator gas, can lead to large changes in the pressure amplitudes.
We have explained that the nonlinearities within the loud-speaker, due to a large cone
excursion, are the most sensible explanation for the discrepancy.
The good agreement of the calculation with the measurements, for low amplitudes,
for a wide frequency range (5 to 200 Hz) and a wide amplitude range, does not only
validate the electro-acoustical model and the acoustical model of the resonator and the
stack, but also gives confidence on our measurement techniques.
Bibliography
[1] H. Bailliet, P. Lotton, M. Bruneau, and V. Gusev, “Coupling between electrody-
namic loudspeakers and thermoacoustic cavities,” Acta Acustica united with Acustica,
vol. 86, no. 2, pp. 363–373, 2000.
[2] M. E. H. Tijani, J. C. H. Zeegers, and A. T. A. M. De Waele, “A gas-spring system for
optimizing loudspeakers in thermoacoustic refrigerators,” Journal of Applied Physics,
vol. 92, no. 4, p. 2159, 2002.
[3] D. Marx, X. Mao, and A. J. Jaworski, “Acoustic coupling between the loudspeaker
and the resonator in a standing-wave thermoacoustic device,” Applied Acoustics,
vol. 67, no. 5, pp. 402–419, 2006.
[4] L. E. Kinsler, A. R. Frey, A. B. Coppens, and J. V. Sanders, Fundamentals of acoustics,
vol. 4th edition. Wiley, 1999.
[5] R. Aarts, J. Nieuwendijk, and O. Ouweltjes, “Efficient resonant loudspeakers with
large form-factor design freedom,” Journal of the Audio Engineering Society, vol. 54,
no. 10, pp. 940–953, 2006.
50 BIBLIOGRAPHY
Chapter 4
Acoustic measurements
4.1 Introduction
Seybert and Ross [1] proposed a new method for impedance tube measurements of
sound absorbtion using the pressure at two positions, called the two-microphone
method, which was much faster than the conventional standing-wave-ratio method.
Chung and Blaser [2–4] further developed this method and later Chu [5, 6] also in-
cluded the tube-attenuation effect, allowing the microphones to be placed farther away
from the sample. Bod´ en and
˚
Abom [7] showed by numerical simulations that the two-
microphone method has its lowest sensitivity when the two microphones are separated
by a quarter wavelength. Chu [8] has shown experimentally that for accurate measure-
ments one of the microphones has to be located close to a pressure minimum and the
choice of the other microphone is not as critical as long as the separation is not close
to half a wavelength. Fujimori et al., Pope, Chu [9], Jones and Parrot [10] describe a
multiple-microphone method: a least square method on the pressure measurements at
more than three positions. This way it is possible to cover a much wider frequency
range. Jang and Ih [11] theoretically and experimentally studied the influence of the
microphone positions on the accuracy of the multiple-microphone method. They have
shown that the equidistant positioning of sensors yields the smallest error within the ef-
fective frequency range. In addition they showed that the measurement accuracy can be
increased and the frequency range can be widened by increasing the number of equidis-
tant sensors. The transfer-matrix approach is a well-known method for characterizing
porous materials [11]. Also in thermoacoustics the transfer-matrix approach has proven
to be useful. Penelet et al. [12] used this approach to make an analytical model of the
acoustic field for an arbitrary temperature profile. This model was used to study the
influence of the temperature profile on the thermoacoustic amplification in an annular
thermoacoustic prime mover.
We use the multi-microphone method to determine the transfer matrix elements of
a thermoacoustic sample. From the transfer matrix elements the viscous and thermal
Rott functions can be determined. The theoretical Rott functions of pores of various
shapes are well-known [13]. Not much literature is available about experimental mea-
surements of the Rott functions of small pores. Wilen measured the thermoviscous Rott
functions of a single pore [14, 15] and Petculescu and Wilen also studied the influence of
52 Acoustic measurements
a temperature gradient [16] and of high amplitudes [17] on the Rott functions of a sin-
gle pore. In this paper instead of a single pore we study practical thermoacoustic stacks.
We present a method to determine the acoustic properties of a stack in a resonator only
with the use of microphones.
In chapter 2 the linear theory of thermoacoustics is derived, which resulted in three
1-D differential equations (Eq. 2.16, 2.30, and 2.34). In this chapter we focus on the
first two equations, the momentum equation and the continuity equation. For a given
temperature gradient dT
0
/dx the two differential equations can be solved for p
1
and
U
1
. The p
1
and U
1
at the right side of a domain can be related to the p
1
and U
1
at the
left side, by a transfer matrix. This is shown in subsection 2.2.1 and the transfer matrix
is given by Eq. 2.41.
If we assume that
dT
0
/dx
T
0
/L
¸1, the influence of the temperature gradient on the trans-
fer matrix can be ignored. The transfer matrix without the presence of a temperature
gradient is given by Eq. 2.42. Later we will verify this assumption using our numerical
model.
We want to determine the transfer matrix of a stack experimentally. A multi-micro-
phone method (section 4.3) is used to measure the leftward and rightward traveling-
wave amplitudes at both sides of the stack. From the amplitudes the p
1
and U
1
at
both sides of the stack can be calculated. This gives us two equations for the transfer
matrix elements. Since the transfer matrix has four elements, two additional equations
are required. Two different methods are explored to do this: the single-stack-position
method and the multiple-stack-positions method.
For the single-stack-position method (section 4.6) we use the symmetry of the ge-
ometry to obtain two additional equations for the transfer matrix. This method is only
valid for a symmetrical geometry in which the influence of the temperature gradient is
negligible.
For the multiple-stack-positions method (section 4.7) the two additional equations
are obtained by using different p
1
-U
1
vectors at one side of the stack. This is accom-
plished by putting the stack at different positions in the acoustic wave. Two different
stack positions yield four equations for the transfer matrix. When more than two dif-
ferent positions are used, the system is overdetermined, and a least squares method is
used to determine the transfer matrix elements. Since no symmetry equations are used,
this method can in principle also be used for stacks with a nonzero temperature gra-
dient. This is possible only when the temperature gradient is the same at the different
locations.
Once the transfer matrix elements are determined, the viscous and thermal Rott
functions, f
ν
and f
κ
, can be calculated using Eq. 2.42. The objective of the experi-
ments is to measure the Rott function of different stack geometries as a function of the
frequency. We hope to verify the linear model and also to study under which conditions
(below which drive ratio) the results are in agreement with the linear theory, as well as
where differences occur with the linear model.
In addition to the transfer matrix, we also study the acoustic-energy dissipation of a
stack by using only microphones. We will show the importance of minor losses.
4.2 Experimental Set-up 53
4.2 Experimental Set-up
To determine different properties of stacks and to see wether the transfer matrix method
holds for large amplitudes, an experimental setup (fig. 4.1) is built.
Stack Damper Speaker
Elastic
connection
Mic. 1 Mic. 2 Mic. 3 Mic. 4 Mic. 5 Mic. 6
DC Amplifier DC Amplifier
Lresonator
Dresonator
Closed back volume
Figure 4.1: Schematic drawing of the multi-microphone setup.
The whole construction is placed on a 5 m long aluminium rail table
1
. The resonator
tube is a long aluminiumcylindrical tube filled with air at ambient pressure and at room
temperature. The left end of the tube is connected to a linear horn with a loudspeaker
and a back volume. The right end of the tube is closed by an end plate and a damper
made of foam. The tube consists of separate parts that can be connected together. This
allows us to change stacks, change the stack position, and change the tube length. To re-
duce the vibration from the speaker to the tube, an elastic connection is placed between
the resonator tube and the linear horn. This connection consists of two different parts
that can slide over each other. Between them, a rubber ring is located, so no gas can
escape out of the system. This way, the vibrations generated in the speaker housing can
not transfer to the resonator tube wall.
For our measurement different ceramic stacks with square pores are used. Aschematic
drawing is shown in figure 4.2 The different parameters of the stack are the pore size D
d
D
L
D
Figure 4.2: Schematic drawing of a ceramic stack. The image on the right is a close-up of the
front of the stack. The pore size D and the pore wall thickness d vary for the different stacks.
and the pore wall thickness d. Another parameter which is related to the pore diameter
1
Item
54 Acoustic measurements
and pore wall thickness is the porosity. It is given by:
ψ =
D
2
(d + D)
2
. (4.1)
In this experiment five different ceramic stacks are used. The diameter of each stack is
38.1 mm. The stack properties can be found in table 4.1.
Table 4.1: Stack properties: the pore size D; the pore wall thickness d; the porosity; and the stack
length L.
D [mm] d [mm] porosity L [cm]
Ceramic 1 0.36 0.76 0.68 2.6
Ceramic 2 0.36 0.76 0.68 5.0
Ceramic 3 0.62 0.50 0.85 16.0
Ceramic 4 1.30 0.20 0.75 2.6
Ceramic 5 0.98 0.76 0.86 2.6
4.2.1 Microphones and speaker
In our setup the signal coming from a function generator, is amplified using a 21200
watt linear amplifier
2
. This amplified signal is sent to the loudspeaker
3
. The speaker
properties are shown in table 4.2.
To measure the acoustics in the resonator, six microphones
4
are used. These micro-
phones generate an electrical signal which is sent to an amplifier
5
.
Table 4.2: Speaker properties.
Diameter 25.4 cm
f
s
26 Hz
x
max
(one way) 3.8 cm
RMS power handling 750 W
Peak power handling 1500 W
The Rott functions show interesting behavior when the viscous penetration depth is
of the same size as the hydraulic diameter of the pores. The characteristic frequency f
0
at which this happens depends on the hydraulic radius. We normalize the frequency by
2
Dynacord L2400 21200 watt linear amplifier
3
10
//
Eclipse SW9102
4
Endevco 8510B-2
5
3-channel Endevco DC amplifier, model 136 DC
4.3 Multi-microphone method 55
dividing by the characteristic frequency, which is defined as
f
0
=
ν
πR
2
H
. (4.2)
A normalized frequency of unity ( f /f
0
= 1) corresponds to δ
ν
= R
H
.
Another step, for which Matlab is used, is calculating the values of f
ν
and f
κ
. From
these calculations, a fit is made. From this fit, the values of the hydraulic radius of the
pores R
H
and the porosity are determined.
4.3 Multi-microphone method
The complex pressure in a tube is the sum of two traveling waves:
p
1
(x) = (C
+
e
−iξ(x−x
L
)
+ C

e
iξ(x−x
L
)
), (0 ≤ x ≤ x
L
), (4.3a)
where C
+
and C

are the amplitudes of the rightward and leftward traveling waves
respectively, x
L
is the reflection position andξ is the complex wave number. It is conve-
nient to chose x
L
equal to zero.
The complex wave number was derived in chapter 2 (Eq. 2.44)
ξ =
ω
c
¸
1 + (γ −1) f
k
1 − f
ν
. (4.4)
For cylindrical resonator ducts, with radius R = 2R
H
, the f
ν
and f
κ
functions are given
by [13]
f
ν
=
2J
1
[(i −1)R/δ
ν
]
J
0
[(i −1)R/δ
ν
](i −1)R/δ
ν
, (4.5a)
f
κ
=
2J
1
[(i −1)R/δ
κ
]
J
0
[(i −1)R/δ
κ
](i −1)R/δ
κ
, (4.5b)
where J
0
and J
1
are Bessel functions of the first kind. If R
H
¸δ
ν
the boundary approx-
imation can be used, yielding [13]
f
ν
=
(1 −i)δ
ν
2R
H
(4.6a)
f
κ
=
(1 −i)δ
κ
2R
H
. (4.6b)
The complex wave number can be written as the sum of a real and imaginary compo-
nent:
ξ = k
/
−ik
//
. (4.7)
To determine the complex amplitudes of both traveling waves, C
+
and C

, it is nec-
essary to measure the pressure at two or more different locations. For now we will as-
56 Acoustic measurements
sume that two microphones are used (two-microphone method), but later we will show
that it is advantageous to use more than two (multi-microphone method). By using a
lock-in function the complex amplitudes of the microphone pressures p
x1
and p
x2
are
determined. The complex sound pressure amplitudes in terms of traveling waves are:
p
x1
= C
+
e
−iξx
1
+ C

e
iξx
1
, (4.8a)
p
x2
= C
+
e
−iξx
2
+ C

e
iξx
2
, (4.8b)
where x
1
and x
2
are the microphones positions (x
1
< x
2
< 0). Using equations (4.8) the
amplitudes of the traveling waves can be determined:
C
+
=
p
x2
e
iξx
1
−p
x1
e
iξx
2
2i sinξ(x
1
−x
2
)
, (4.9a)
C

=
p
x1
e
−iξx
2
−p
x2
e
−iξx
1
2i sinξ(x
1
−x
2
)
, (4.9b)

x
x
2
x
n
x
3
x
1
C
+
C
-
Figure 4.3: A schematic drawing of the multi-microphone method. To the left of the sample n
microphones (n ≥ 2) are located at distances x
1
, x
2
, . . . , x
n
from the left side of the sample.
When using the two-microphone method to determine the left and rightward trav-
eling waves, the distance between the two microphones is very critical. We will focus
on the region to the left of the sample as is shown in figure 4.3. The two microphone
method fails if
sinξ(x
1
−x
2
) = 0. (4.10)
And even when sinξ(x
1
−x
2
) is not equal but close to zero the method is very sensitive
to errors. When the measurements are done at a fixed frequency the situation of (4.10)
can be easily avoided by choosing the microphone positions wisely, but for a whole
frequency range this is a real problem. A way to overcome this problem is to use more
than two microphones: the multiple microphone method. When n microphones are
used, at positions x
1
, x
2
, . . . , x
n
measuring pressures p
1
, p
2
, . . . , p
n
, the traveling wave
components C
+
and C

are described by the following overdetermined (for n > 2)
4.3 Multi-microphone method 57
linear matrix equation
Kx = p, (4.11)
with
K =
_
_
_
_
_
exp(−iξx
1
) exp(iξx
1
)
exp(−iξx
2
) exp(iξx
2
)
.
.
.
.
.
.
exp(−iξx
n
) exp(iξx
n
)
_
_
_
_
_
, x =
_
C
+
C

_
, p =
_
_
_
_
_
p
1
p
2
.
.
.
p
n
_
_
_
_
_
. (4.12)
One can compute the optimal solution for x using the least-squares method. The best
approximate solution for x (|p−Kx| is minimal) is obtained by using the Moor-Penrose
generalized inverse K
+
[18]
x = K
+
p = (K
H
K)
−1
K
H
p, (4.13)
where K
H
denotes the Hermitian matrix of K. Using the matrix definition of K in Eq.
4.12
K
H
K =
_

n
j=1
exp(2k
//
x
j
) ∑
n
j=1
exp(2ik
/
x
j
)

n
j=1
exp(2k
//
x
j
) ∑
n
j=1
exp(2ik
/
x
j
)
_
. (4.14)
By applying Cramer’s rule to Eq. 4.13 and substituting Eq. 4.14 we find for x:
x =
1
det(K
H
K)
_
_

n
j=1
p
j
e


x
j

n
l=1
e
−2k
//
x
l
−∑
n
j=1
p
j
e
−iξ

x
j

n
l=1
e
2ik
/
x
l

n
j=1
p
j
e
−iξ

x
j

n
l=1
e
2k
//
x
l
−∑
n
j=1
p
j
e


x
j

n
l=1
e
−2ik
/
x
l
_
_
, (4.15)
where (

) denotes the complex conjugate. The denominator in 4.15 is equal to zero
when det(K
H
K) = 0. As
det(K
H
K) =
_
n

j=1
e
2iξx
j
__
n

l=1
e
−2iξx
l
_
−n
2
, (4.16)
it follows for n = 2 (the two microphone method) that det(K
H
K) = 0 if cos 2ξ(x
1

x
2
) = 1.
To verify our experimental set-up and to determine the accuracy of the multi-micro-
phone method we measure the reflection coefficient R
r
of a solid end plate. The reflec-
tion coefficient is defined as
R
r
=
C

C
+
, (4.17)
and should be equal to unity in our set-up. In figure 4.4 the reflection coefficient, mea-
sured by the two-microphone method, is shown as a function of the frequency. The two
microphones are 1 m apart from each other. Better results could be achieved by putting
the two microphones closer to each other, but we deliberately choose to put them at a
large distance to demonstrate the peaks in the reflection coefficient that occur when Eq.
4.10 is satisfied.
The reflection coefficient measured by the multi-microphone method with five mi-
58 Acoustic measurements
10
2
10
3
0.8
0.85
0.9
0.95
1
1.05
1.1
1.15
frequency [Hz]
|
R
|
Figure 4.4: The reflection coefficient as a function of the frequency. The vertical dashed lines are
drawn at all frequencies that are multitudes of 171.5 Hz.
crophones is shown in figure 4.5. In figure 4.5(a) the reflection coefficient is plotted
linearly and in 4.5(b) the difference with a unity reflection coefficient is plotted logarith-
mically. The microphones are at distances to the resonator end: 1.600 m, 1.300 m, 1.000
m, 0.940 m, 0.600 m, and 0.300 m. When comparing figure 4.4 with figure 4.5(a) the
peaks have disappeared. The multi-microphone method is working well at almost the
whole frequency range. The only big peaks are located at 6.3 10
2
Hz and in between
1.0 10
3
Hz and 1.1 10
3
Hz. These peaks excluded, the error in the reflection coefficient
is smaller than 0.03.
4.3 Multi-microphone method 59
10
2
10
3
0.8
0.85
0.9
0.95
1
1.05
1.1
1.15
frequency [Hz]
|
R
|
(a) linear scale
10
2
10
3
10
−3
10
−2
10
−1
10
0
|
|
R
|

1
|
frequency [Hz]
(b) logarithmic scale
Figure 4.5: The reflection coefficient measured by the multi-microphone method as a function of
the frequency. In subfigure (a) it is plotted linearly and in (b) logarithmically.
60 Acoustic measurements
4.4 Acoustic energy losses
4.4.1 Acoustic energy flow
We want to study the acoustic energy ∆
˙
W that is dissipated in a stack. The acoustic
energy loss is determined by measuring the acoustic energy flows at both sides of the
stack. The acoustic energy flow
˙
W is defined as
˙
W =
ω

_
p(t)U(t)dt =
1
2
Re¦p
1
U

1
¦. (4.18)
The acoustic energy flow in the resonator is determined by microphone measurements.
Two methods will be presented: the Fusco method and the traveling-waves method.
These methods are based on measuring the pressure at two or more different positions.
A more direct approach is to measure both the velocity and the pressure at a single po-
sition. Among others, this approach was used by Bailliet et al. [19]. They measured the
acoustic power flow in a thermoacoustic resonator by means of laser Doppler anemom-
etry together with microphonic measurement and found their results in good agreement
with analytical calculations.
4.4.2 Fusco method
This method is called after Fusco, who introduced a technique that determines acoustic
energy flow in a wide duct from measurements of pressure based on the boundary
layer approximation [20]. He achieved an accuracy of 5% with this method, even at
high amplitudes where the acoustic flow is turbulent. Biwa et al. [21] validated this
technique experimentally also for ducts with a radius smaller than the viscous boundary
layer thickness.
In the Fusco method the relation between the velocity and the pressure gradient Eq.
2.8 is used to determine the volume flow rate. Since this method is used in a resonator
with R
H
¸δ
ν
, we can neglect the viscous term in Eq. 2.8, resulting in
iωρ
0
u
1
= −
dp
1
dx
. (4.19)
Two microphones, separated by ∆x, are used, as is shown in figure 4.6. If the separation
between the microphones is sufficiently small (∆x ¸ λ), the pressure gradient can be
approximated by
dp
1
dx
=
p
1B
− p
1A
∆x
, (4.20)
where p
1A
and p
1B
are the pressures measured with microphones A and B respectively.
The pressure midway between the microphones can be approximated as the average of
the two pressures
p
1
=
p
1A
+ p
1B
2
. (4.21)
4.4 Acoustic energy losses 61
A
xA xB xQ
∆x
A B
p1A p1B
WQ
Figure 4.6: A schematic drawing of the Fusco method for energy flows.
By substituting Eqs. 4.20-4.21 into Eq. 4.18 and using u
1
= U
1
/A it follows that
˙
W =
iA
4ωρ
0
∆x
Im¦(p
1A
+ p
1B
)(p
1A
− p
1B
)

¦ = −
iA
4ωρ
0
∆x
Im¦p
1A
p

1B
− p

1A
p
1B
¦.
(4.22)
Using z −z

= 2Im¦z¦ yields
˙
W =
iA
2ωρ
0
∆x
Im¦p
1A
p

1B
¦ =
iA
2ωρ
0
∆x
[p
1A
[[p
1B
[ sin(θ), (4.23)
with θ the phase angle by which p
1A
leads p
1B
. Fusco et al. [20] extended Eq. 4.23 to
an equation which is also valid if the requirement ∆x ¸ λ is not fulfilled, by including
attenuation in the laminar boundary-layer approximation
˙
W =
A

0
c
0
sin(ω∆x/c
0
)

_
Im¦p
1A
p

1B
¦
_
1 −
δ
ν
4R
H
_
1 −
γ −1
_
P
r
+
_
1 +
γ −1
_
P
r
_
ω∆x
c
0
cot
_
ω∆x
c
0
_
__
+
δ
ν
8R
h
([p
1A
[
2
−[p
1B
[
2
)
_
1 −
γ −1
_
P
r
+
_
1 +
γ −1
_
P
r
_
ω∆x
c
0
csc
_
ω∆x
c
0
_
__
.
(4.24)
4.4.3 Traveling-waves method
The complex pressure in a tube is the sum of two traveling waves:
p
1
(x) = (C
+
e
−ik(x−x
Q
)
+ C

e
ik(x−x
Q
)
), (4.25)
62 Acoustic measurements
with x
Q
the position at which we want to determine
˙
W and the complex wave number
k given by Eq. 2.44. The traveling-wave amplitudes C
+
and C

can be determined
experimentally with either a two-microphone method or a multi-microphone method
as is explained in section 4.3. Using Eq. 4.25 we can determine pressure at position
x = x
Q
and using Eq. 2.16 we can also determine the volume flow rate
p
1Q
= C
+
+ C

, (4.26a)
U
1Q
=
Ak(1 − f
ν
)
ρ
0
ω
(C
+
−C

), (4.26b)
where f
ν
is given by Eq. 4.6a. Substituting Eqs. 4.26 into Eq. 4.18 yields
˙
W
Q
=
A

0
ω
Re¦(C
+
+ C

)(C
+
−C

)

k

(1 − f

ν
)¦. (4.27)
If the imaginary part of k(1 − f
ν
) can be neglected in Eq. 4.27, it can be simplified to
˙
W
Q
=
A

0
c
0
([C
+
[
2
−[C

[
2
). (4.28)
This equation is intuitively understandable, as
A
2ρc
0
[C
+
[
2
is the energy flow of the right-
ward traveling wave C
+
and
A

0
c
0
[C

[
2
is the energy flow of the opposite-directed
wave.
The traveling-waves method has two advantages to the Fusco method. The position
x
Q
, at which we want to measure the acoustic energy flow, is not limited to the position
in-between the microphones. When multiple microphones are present in a tube, this
method can be used as a multi-microphone method, instead of a Fusco method, result-
ing in a higher accuracy. The disadvantage of this method is that it is sensitive to errors
in ξ, as the gas temperature is often not exactly known.
4.4.4 Acoustic-energy-flow measurements
At both sides of the stack, three microphones are positioned. The microphone positions
and the stack position are shown in table 4.3. For the Fusco method, at both sides of
the stack three different p
1A
− p
1B
combinations can be used. We use the following con-
vention:
˙
W
i j
is the acoustic energy flow at position (x
i
+ x
j
)/2 determined by using Eq.
4.24 with p
1A
the pressure measured with microphone i and p
1B
the pressure measured
with microphone j.
To compare the different methods for determining the acoustic energy flow, we mea-
sure the pressures at six different positions in the empty resonator tube as functions of
the frequency. For practical reasons the measurement is performed at constant speaker
voltage. In figure 4.7 the dashed plots represent the Fusco method at positions x
12
. . . x
56
and the solid plots represent the traveling-waves method. All the plots are normalized
by dividing by
˙
W
12
determined by the traveling-waves method. The complex wave
amplitudes C
+
and C

are here determined by using all six microphones.
For f < 120 Hz the plots show a lot of peaks. We are not sure what is causing
those peaks, but they are also present in a measurement of the reflection coefficient.
4.4 Acoustic energy losses 63
Table 4.3: Microphone positions x
i
and acoustic-energy-flow positions x
i j
.
x
i
[cm] x
i j
[cm]
x
1
-210.0 x
12
-195.0
x
2
-180.0 x
23
-168.0
x
3
-156.0 x
34
-125.0
x
4
-94.0 x
45
-77.0
x
5
-60.0 x
56
-45.0
x
6
-30.0
The reason of the peak in
˙
W
34
at 278 Hz is that at this frequency λ/2 ≈ x
4
− x
3
. This
effect has been discussed in section 4.3. Apart from these peaks the two methods are in
good agreement with each other. The results from the traveling-waves method are more
smooth, since they are determined by the pressures at six positions. What the traveling-
waves method in this case is effectively doing is determining the total dissipation in the
resonator and then using the complex wave number to determine the energy flow at
positions x
i j
.
Now we install a stack (sample 1) in the resonator at position x
S
= -100.3 cm. At
the left side of the stack, microphones 1 and 2 give us
˙
W
12
using the Fusco method and
microphones 2 and 3 give us
˙
W
23
. Using microphones 1, 2, and 3 the traveling wave
amplitudes C
+
and C

at the left side of the stack are determined. With the traveling
wave method from C
+
and C

we determine
˙
W
12
and
˙
W
23
to compare both methods.
At the right side of the stack we do the same for
˙
W
45
and
˙
W
56
. In figure 4.8 the
˙
W
i j
for
four different microphone combinations are plotted as function of the frequency. The
solid plots represent the Fusco method and the dashed ones the traveling wave method.
The black dashed plots are the acoustic energy flows determined by using the traveling-
wave method at the positions of the two stack ends.
The acoustic energy dissipated in the stack ∆
˙
W
S
will be determined as the difference
between the two black dashed plots in figure 4.8.
4.4.5 Acoustic energy losses in a stack
Using the definition of acoustic work flow (Eq. 4.18) we find the acoustic energy
˙
W that
is dissipated per unit length
d
˙
W
dx
=
1
2
Re
_
p
1
dU

1
dx
+ U

1
dp
1
dx
_
. (4.29)
For dU

1
/dx and dp
1
/dx we use Eqs. 2.35(a-b)
d
˙
W
dx
=
1
2
Re
_
p
1
(−R

1
+ R

3
U
1
) −U

1
R
2
U
1
_
= −
1
2
Re¦R
1
¦[U
1
[
2

1
2
Re¦R
2
¦[p
1
[
2
+
1
2
Re¦R

3
p
1
U

1
¦, (4.30)
64 Acoustic measurements
100 200 300 400
0
0.2
0.4
0.6
0.8
1
˙
W
i
j
/
˙
W
1
2
f [Hz]


12
23
34
45
56
12
23
34
45
56
Figure 4.7: The normalized acoustic energy flows
˙
W
i j
as functions of the frequency at different
positions x
i j
(as shown in the legend) in the empty resonator. The dashed plots represent the
Fusco method and the solid ones the traveling-waves method. All the plots are normalized by
dividing by
˙
W
12
determined by the traveling-waves method. The complex wave amplitudes C
+
and C

for the traveling-waves method are here determined using all six microphones.
with
Re¦R
1
¦ = −
ωρ
0
A
Im( f
ν
)
[1 − f
ν
[
2
, (4.31a)
Re¦R
2
¦ = −
(γ −1)ωA
γp
0
Im( f
κ
). (4.31b)
The total acoustic energy loss ∆
˙
W in the stack is found by integrating d
˙
W/dx over the
whole stack length.
The pressure p
1
and volume flow rate U
1
as function of the position in the stack are
determined by solving the thermoacoustic equations. As the boundary conditions for
the these equations we use the pressure and volume flow rate at the left end of the stack,
that are measured using the traveling-wave method.
4.4.6 Minor losses
The major losses in a stack are caused by viscous dissipation and thermal relaxation.
These effects are included in the linear theory. At high amplitudes minor losses will
become a significant factor. In case of a stack these minor losses occur at the stack ends.
The sudden change of the cross-sectional area of the gas causes losses. For instance,
4.4 Acoustic energy losses 65
50 100 150 200 250 300 350 400
0
0.2
0.4
0.6
0.8
1
˙
W
i
j
/
˙
W
1
2
f [Hz]
12
23
45
56
12
23
45
56
stack end
stack end
Figure 4.8: The acoustic energy flows as functions of the frequency at different positions in a
resonator with a stack installed inside. The solid plots represent the Fusco method and the dashed
ones the traveling-waves method. All the plots are normalized by dividing by
˙
W
12
determined
by the Fusco method. The black dashed plots are the acoustic energy flows determined by using
the traveling-wave method at the positions of the two stack ends.
when the flow is directed out of the stack vortices are formed.
We will derive an equation to describe these minor losses quantitatively. First we
will do this for a steady flow, since the minor losses in steady flows are well-known in
literature. Unfortunately the equations are very empirical. Then we will extend this
steady theory to oscillatory flows, determine the energy loss due to minor losses, and
verify it with measurements.
Steady flow
First we study a sudden increase of cross section (also called expansion), from A
1
to A
2
,
in a steady flow (figure 4.9). Before the expansion, the flow is developed with velocity
u
1
(y, z), pressure p
1
, and density ρ
1
. The dashed rectangle represents a control volume.
Its surface at the left, S
1
, is directly behind the expansion and its surface at the right, S
2
,
is chosen at the position where the flow is developed again. At surface S
2
the velocity
is u
2
(y, z), the pressure is p
2
, and the density is ρ
2
. We apply the integral law of mass
conservation to the control volume,
¸u
1
¸ρ
1
A
1
= ¸u
2
¸ρ
2
A
2
, (4.32)
66 Acoustic measurements
S2
S1
A
2
u2
p2 p1
A
1
u1
Figure 4.9: A schematic drawing of sudden increase in cross section in a steady flow.
where ¸¸ denotes the cross-sectional average. We apply the integral momentum equa-
tion to the control volume
__
S
1
(ρu
2
+ p)dA =
__
S
2
(ρu
2
+ p)dA, (4.33)
and assume that the pressures and densities are uniform over the cross section
β
1
ρ
1
¸u
1
¸
2
A
1
+ p
1
A
2
= β
2
ρ
2
¸u
2
¸
2
A
2
+ p
2
A
2
, (4.34)
with β
i
the momentum correction factor defined as
β
i
=
¸u
2
i
¸
¸u
i
¸
2
. (4.35)
We consider the integral energy equation

˙
E
ml
=
__
A
(h +
1
2
[v[
2
)(ρv) dA, (4.36)
with ∆
˙
E
ml
the energy loss due to minor losses and h the enthalpy per unit mass, which
is given by
h =
p
ρ
+ c
v
T =
c
p
R
s
p
ρ
, (4.37)
for ideal gases. We apply the energy equation to the control volume

˙
E
ml
=
__
S
1
h
1
ρ
1
u
1
dA −
__
S
2
h
2
ρ
2
u
2
dA. (4.38)
In the enthalpy per unit mass (Eq. 4.37) we write the c
p
/R
s
term as 1/(1 −γ
−1
). If the
4.4 Acoustic energy losses 67
temperature increases during the expansion due to losses, we consider this as a loss,
which is included in ∆
˙
E
ml
. After substituting Eq. 4.37 into Eq. 4.38 and integrating over
the cross section

˙
E
ml
=
1
2
α
1
ρ
1
¸u
1
¸
3
A
1
+
p
1
¸u
1
¸A
1
1 −γ
−1

1
2
α
2
ρ
2
¸u
2
¸
3
A
2
+
p
2
¸u
2
¸A
2
1 −γ
−1
, (4.39)
with α the kinetic-energy correction factor defined as
α =
¸u
3
¸
¸u¸
3
. (4.40)
If we assume that ρ
1
= ρ
2
= ρ and use Eq. 4.32 it can be shown that

˙
E
ml
=
1
2
α
1
ρ¸u
1
¸
3
A
1

1
2
α
2
ρ¸u
1
¸
3
A
1
A
2
1
A
2
2
+
(p
1
− p
2
)¸u
1
¸A
1
1 −γ
−1
. (4.41)
Substituting the p
1
− p
2
of Eq. 4.34 into Eq. 4.41 results in

˙
E
ml
= K
e

1
2
A
1
ρ¸u
1
¸
3
, (4.42)
with minor-loss coefficient K for an expansion given by
K
e
= α
1


1
ψ
1 −γ
−1
+ (

2
1 −γ
−1
−α
2

2
, (4.43)
with the porosity ψ = A
1
/A
2
. The pressure drop due to minor losses is given by
∆p =

˙
E
ml
A
1
¸u
1
¸
= K
e

1
2
ρ¸u
1
¸
2
. (4.44)
Note that ∆p differs from p
1
− p
2
. The pressure difference p
1
− p
2
is from S
1
to S
2
,
whereas ∆p is from the location just before the contraction to S
2
.
If the velocity is uniform over the cross section (i.e. α
1
= β
1
= 1 and α
2
= β
2
= 1)
K
e
= (1 −ψ)
2
. (4.45)
For a developed laminar flow in-between two parallel plates
u =
3
2
¸u¸
_
1 −
y
2
y
2
0
_
, for − y
0
≤ y ≤ y
0
, (4.46)
it follows that α = 54/35 and β = 6/5. In case of a developed laminar flow through a
cylinder
u = 2¸u¸
_
1 −
r
2
R
2
_
, for 0 ≤ r ≤ R, (4.47)
68 Acoustic measurements
it is found that α = 2 and β = 4/3.
In case of a contraction only empirical data is available. According to Streeter et
al. [22] the energy loss and pressure drop for a contraction can also by approximated by
Eqs. 4.42 and 4.44, with the minor-loss coefficient for a contraction
K
c
=
_
1
C
c
−1
_
2
. (4.48)
The contraction coefficient C
c
for water was determined by Weisbach [22] for different
A
2
/A
1
ratios, which we fitted according to a third order polynomial
C
c
= C
c3
ψ
3
+ C
c2
ψ
2
+ C
c1
ψ+ C
c0
, (4.49)
with C
c3
= 0.55, C
c2
= −0.34, C
c1
= 0.17, and C
c0
= 0.61.
Now that we have discussed the theory of minor losses through a sudden expansion
and contraction in the cross section for a steady flow, we will extent the analysis to the
more complicated oscillatory flow.
Oscillatory flow
We use Eq. 4.44 for a time-dependent flow
∆p(t) = −
K(t)ρ(t)
2A
2
[U(t)[U(t), (4.50)
with the pressure p and U real and time dependent. The pressure difference due to
minor losses is 180 degrees out of phase with the velocity and its amplitude proportional
to the velocity squared. For the first order term of the pressure variation we can ignore
the density variations: ρ(t) = ρ
0
. The minor loss coefficient K(t) is time dependent
since its value is different for contraction phase than for the expansion phase. As an
approximation we use a time-independent average value K(t) = (K
e
+ K
c
)/2. We write
the volume flow rate as
U(t) = [U
1
[ cos(ωt), (4.51)
with [U
1
[ the amplitude. The acoustic energy loss due to minor losses we find with

˙
W =
ω

_
2π/ω
0
∆p(t)U(t)dt =
2


0
A
2
[U
1
[
3
. (4.52)
The drop in pressure ∆p
1
due to minor losses, can be approached by the first harmonic
[ cos(ωt)[ cos(ωt) ≈ a
1
cos(ωt), (4.53)
where a
1
follows from taking the first-order amplitude of the Fourier series
a
1
=
ω
π
_
π/ω
−π/ω
[ cos(ωt)[ cos
2
(ωt)dt =
8

. (4.54)
4.4 Acoustic energy losses 69
In figure 4.10 [ cos(ωt)[ cos(ωt) and a
1
cos(ωt) are shown as function of t. By substi-
tuting Eq. 4.53 into Eq. 4.50 the minor-loss pressure-drop amplitude yields
∆p
1
= −
4


0
A
2
[U
1
[U
1
. (4.55)
Eq. 4.55 is consistent with 4.52, since
1
2
Re¦∆p
1
U

1
¦ =
2


0
A
2
[U
1
[
3
. (4.56)
−1
−0.5
0
0.5
1
ωt
0
π 2π 3π 4π
cos(ωt)| cos(ωt)|
8/3π cos(ωt)
Figure 4.10: The [ cos(ωt)[ cos(ωt) dependency of the minor losses is approximated by
8/3π cos(ωt).
Nowlet us discuss the value of K
e
for a transition of a stack into a resonator tube. The
diameter of the resonator tube is much larger than the viscous penetration depth, thus
α
2
≈ 1 and β
2
≈ 1. In figure 4.11a the values of α and β are shown for a parallel plate
stack (solid curve), circular pores (dash-dotted curve), and rectangular pores (dotted
curve) as a function of the hydraulic radius. For ψ = 0 the parallel plates have α =
54/35 and β = 6/5, which are the same as for steady flow. This resemblance with the
steady flow also holds for circular pores, where α = 54/35 and β = 6/5, for ψ = 0. In
the limit of an infinitely large resonator diameter both α and β approach unity.
In figure 4.11b the minor-loss coefficient K
e
is plotted as function of the porosity
for a stack with squared pores with three different sizes: D
h

ν
→ 0, D
h

ν
= 1,
and D
h

ν
→ ∞. The minor-loss coefficient K
c
for a contraction is also shown, using
the empirical data of Weisbach. For an expansion the α and β correction factors for the
velocity profile have a huge effect on the minor loss coefficient. The K coefficient is twice
70 Acoustic measurements
0 1 2 3 4 5
1
1.2
1.4
1.6
1.8
2
2.2
D
H

ν
α
,
β


circular pores
squared pores
parallel plates
α
β
(a)
0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
2
2.5
ψ
K


K
e
, D/δ
ν
→0
K
e
, D/δ
ν
= 1
K
e
, D/δ
ν
→∞
K
c
(b)
Figure 4.11:
(a) The kinetic-energy correction factor α and momentum correction factor β as function of the
D
H

ν
ratio for a developed oscillatory flow through three different stack geometries: circular
pores, squared pores, and parallel plates.
(b) The minor-loss coefficient K as function of the porosity ψ. For an expansion of squared pores
into a resonator tube, K
e
is plotted for three different pore sizes and for a contraction K
c
is plotted
according to the empirical data of Weisbach (Eq. 4.48).
as big as for a uniform velocity profile (D
h

ν
→ ∞), even for zero porosity. For large
porosities the effect of the velocity profile is relatively much larger. For a porosity of 1,
there is no expansion, but still the difference in velocity distribution, leads to K
e
= 0.40
for D
h

ν
→ 0 and K
e
= 0.32 for D
h

ν
= 1. The difference between D
h

ν
→ 0 and
D
h

ν
= 1 is very small for all porosities.
4.5 Minor-loss correction
In figure 4.12 the acoustic power losses of sample 4 are shown as a function of the vol-
ume flow rate. In figure 4.12a the frequency is 36.5 Hz and in figure 4.12b 105.5 Hz. The
dash-dotted curve represents measured data. The dashed curve is a numerical calcula-
tion without taking into account minor losses, the solid curve is with taking into account
minor losses and the black dotted curve is with also including a K
p
[p
1
[
4
term. Both fre-
quencies (36.5 Hz and 105.5 Hz) are resonance frequencies. The pore size D = 1.1 mm,
the porosity ψ = 0.72, and K = 0.41. This value for K = 0.41 is determined by a fit
in figures 4.12. The minor loss coefficient K agrees very well with the theoretical value:
0.39.
In figure 4.13 we show similar plots, but this time for sample 5. The pore size D =
0.83 mm, the porosity ψ = 0.85, and K = 0.35. This value for K is determined by a
fit in figures 4.13. The minor loss coefficient K = 0.35 is higher than the theoretical
value: 0.28. This can be explained by inaccuracies in the porosity and pore size, which
4.5 Minor-loss correction 71
0 0.02 0.04 0.06
0
20
40
60
80

˙
W
S
[
W
]
|U
1
| [m
3
/s]


K = 0
K = 0.41
K = 0.41, K
p
|p
1
|
4
measurement
(a) f = 36.5 Hz, D
r,max
= 19.3%
0 0.01 0.02 0.03 0.04
0
5
10
15
20
25
30

˙
W
S
[
W
]
|U
1
| [m
3
/s]


K = 0
K = 0.41
K = 0.41, K
p
|p
1
|
4
measurement
(b) f = 105.5 Hz, D
r,max
= 5.9%
Figure 4.12: The acoustic-power losses ∆
˙
W
S
as functions of the volume flowrate at two different
frequency and drive ratio combinations. The dash-dotted curve represents the measurement and
the other three curves are analytical calculations. The dashed curves are the theoretical value
without correcting for minor losses (K = 0). The theoretical functions corrected for minor losses
(K = 0.41) are shown by the solid curves. An additional correction term K
p
[p
1
[
4
term is added
for the dotted curves.
are known with a 5% accuracy, and also influence the energy dissipation. Additionally
the pores are not perfectly squared, with exactly the same pore size along the whole
stack length, which can also increase the dissipation. And also in the derivation of the
minor loss theory some assumptions are made which leave room for betterment. The
flow at the entrance of the pore is assumed to be developed. Other nonlinear effects like
turbulence and higher harmonics are not taken into account. Also, for determining the
acoustic energy flows by the traveling-waves method, it is required that no temperature
gradients are present in the resonator tube. Although the tube wall is very thick and is
heat conducting, at high drive ratios, up to 19.5%, temperature gradients can occur. So
it is not surprising that the K-value determined from the fit is higher than the theoretical
value.
72 Acoustic measurements
0 0.02 0.04 0.06
0
20
40
60
80

˙
W
S
[
W
]
|U
1
| [m
3
/s]


K = 0
K = 0.35
K = 0.35, K
p
|p
1
|
4
measurement
(a) f = 36.5 Hz, D
r,max
= 19.5%
0 0.01 0.02 0.03 0.04
0
5
10
15
20
25

˙
W
S
[
W
]
|U
1
| [m
3
/s]


K = 0
K = 0.35
K = 0.35, K
p
|p
1
|
4
measurement
(b) f = 105.5 Hz, D
r,max
= 5.7%
Figure 4.13: The acoustic-power losses ∆
˙
W
S
as functions of the volume flowrate at two different
frequency and drive ratio combinations. The dash-dotted curve represents the measurement and
the other three curves are analytical calculations. The dashed curves are the theoretical value
without correcting for minor losses (K = 0). The theoretical functions corrected for minor losses
(K = 0.35) are shown by the solid curves. An additional correction term K
p
[p
1
[
4
term is added
for the dotted curves.
4.6 Single-stack-position method 73
4.6 Single-stack-position method
0
xL xR xend
x
x1
x2
x4
x3
uL
pL
uR
pR
Rend
C+
C-
D+
D-
Figure 4.14: A schematic drawing of the experimental set-up for an acoustic characterization
of a sample using the two-microphone method. A loudspeaker connected to a tube which is
terminated by a reflection coefficient R
end
. Between x
L
and x
R
the tube is filled with the acoustic
sample and at both sides of the sample two microphones are located.
To study the acoustical properties of a thermoacoustic sample, e.g. a stack or a re-
generator, a multi-microphone set-up is built, which is shown in figure 4.14. At the left
end of the tube a loudspeaker is used as a sound source and the right end of the tube
is closed by a plate with a reflection coefficient R
end
, which will be close to unity. The
advantage of a closed tube compared to an anechoic termination, which is commonly
used, is that it allows higher pressure amplitudes, especially in resonance. The complex
pressure in the tube is the sum of two traveling waves:
p(x, t) = (C
+
e
−iξ(x−x
L
)
+ C

e
iξ(x−x
L
)
)e
iωt
, (4.57)
for (0 ≤ x ≤ x
L
) and
p(x, t) = (D
+
e
−iξ(x−x
R
)
+ D

e
iξ(x−x
R
)
)e
iωt
, (4.58)
for (x
R
≤ x ≤ x
end
).
The complex amplitudes of the traveling waves C
+
, C

, D
+
and D

are found by
using the multi-microphone method (Eq. 4.12). At each side of the stack three micro-
phones are installed in the resonator, but in figure 4.14 only two of them are shown.
The p
1
-u
1
vectors at the left and the right side of the sample can be expressed in
terms of the traveling wave amplitudes:
_
p
L
u
L
_
=
_
C
+
+ C

(C
+
−C

)/(ρ
0
c)
_
, (4.59a)
_
p
R
u
R
_
=
_
D
+
+ D

(D
+
−C

)/(ρ
0
c)
_
. (4.59b)
74 Acoustic measurements
Using Eq. 4.11 the traveling wave amplitudes can be determined, after which Eq. 4.59
provides us with two equations for determining the transfer matrix components. Two
additional equations are required. We make use of the reciprocity of the linear acoustic
equations. Reciprocity refers to situations for which a magnitude associated with an
”effect” at a point is unchanged when the locations of ”cause” and ”point of observa-
tion” are interchanged. The transfer matrix has to obey the reciprocity condition for all
p-u vectors. From this it follows that the determinant of the transfer matrix is unity [23].
In case of a symmetrical sample it can be shown that also the condition T
11
= T
22
ap-
plies [23]. Now we have a total of four equations for the transfer matrix:
p
L
= T
11
p
R
+ T
12
u
R
, (4.60a)
u
L
= T
21
p
R
+ T
22
u
R
, (4.60b)
T
11
T
22
−T
12
T
21
= 1, (4.60c)
T
11
= T
22
. (4.60d)
Combining these four equations the following expressions for the transfer matrix ele-
ments can be found:
T
11
= T
22
=
p
L
u
L
+ p
R
u
R
p
R
u
L
+ p
L
u
R
, (4.61a)
T
12
=
p
2
L
− p
2
R
p
R
u
L
+ p
L
u
R
, (4.61b)
T
21
=
u
2
L
−u
2
R
p
R
u
L
+ p
L
u
R
. (4.61c)
The transfer matrix elements as functions of the frequency are difficult to interpret.
As Eq. 2.42 relates the Rott functions to the transfer matrix elements, we can calculate
the Rott functions from the measured transfer matrix elements. In figure 4.15 the real
and imaginary components of the viscous Rott function f
ν
are shown as a function of
the frequency. The reference frequency f
0
is 174.9 Hz for this stack. The data points are
the measurement data and the solid curve is a fit according to [24]:
f
ν
= 1 −
64
π
4


m=0


n=0
_
(2m + 1)
2
(2n + 1)
2
(1 + i

2
δ
2
ν
D
2
((2m + 1)
2
+ (2n + 1)
2
)
_
−1
(4.62)
for rectangular pores with the pore size D. Pore size D and the porosity ψ are fitting
parameters.
The peaks at f /f
0
= 0.97 and f /f
0
= 1.95 are caused by velocity nodes located at the
stack position at these frequencies. Since viscous effects are a consequence of velocity,
the viscous effects are small when the stack is in a velocity node, which makes them
difficult to measure and more sensible to errors.
To study the performance of the single-stack-position method at high driver ratios
we have increased the maximum D
r
in the resonator from 1% in figure 4.15 to 8% in
figure 4.16. Both figures are almost identical. Since the f
ν
functions have not deformed
at D
r
= 8%, the flow in the stack is probably still laminar. Also we can conclude that
the single-stack-position method still works well at this D
r
.
4.7 Multiple-stack-positions method 75
0 0.5 1 1.5 2
0
0.05
0.1
0.15
0.2
f/f
0
ψ

R
e
[
1

f
ν
]
0 0.5 1 1.5 2
0
0.1
0.2
0.3
0.4
f/f
0
ψ

I
m
[
1

f
ν
]
Figure 4.15: Plot of the real and imaginary part of f
ν
as functions of the normalized frequency.
The data points are measurements and the solid curve is a fit. The maximum D
r
reached in the
tube is 1 %.
0 0.5 1 1.5 2
0
0.05
0.1
0.15
0.2
f/f
0
ψ

R
e
[
1

f
ν
]
0 0.5 1 1.5 2
0
0.1
0.2
0.3
0.4
f/f
0
ψ

I
m
[
1

f
ν
]
Figure 4.16: Plot of the real and imaginary part of f
ν
as functions of the normalized frequency.
The data points are measurements and the solid curve is a fit. The maximum D
r
reached in the
tube is 8 %.
4.7 Multiple-stack-positions method
To determine the different elements of the transfer matrix, two extra equations were
used, which came from the reciprocity of the linear acoustic equations and from the
76 Acoustic measurements
symmetry of the stack (Eqs. 4.60c and 4.60d).
To determine if the use of these equations is valid, the overdetermined transfer ma-
trix method is used. In this method frequency sweep measurements are performed at
two or more stack positions. Between these measurements the sample is moved inside
the tube. This way extra equations for the transfer-matrix are derived. For example, The
first measurement is performed at x = −1.00 m and the second at x = −1.05 m. For ev-
ery measurement the pressure and velocity left and right from the stack are determined.
From these pressures and velocities the transfer matrix can be determined:
_
p
L1
p
L2
p
Ln
u
L1
u
L2
u
Ln
_
= T
_
p
R1
p
R2
p
Rn
u
R1
u
R2
u
Rn
_
. (4.63)
For two stack positions, the transfer-matrix can be determined exactly. In case of more
than two positions, the system of equations is overdetermined. In that case, the best
values for the transfer matrix elements, T
11
, T
12
, T
21
, and T
22
, are determined using a
least-squares method.
The measurement using the single-stack-position method, from which the results
were shown in figure 4.15, is repeated using the multiple-stack-position method. In fig-
ure 4.17 the real and imaginary components of the viscous Rott function, f
ν
, are shown
as a function of the frequency. The reference frequency f
0
is 174.9 Hz for this stack.
0 0.5 1 1.5 2
0
0.05
0.1
0.15
f/f
0
ψ

R
e
[
1

f
ν
]
0 0.5 1 1.5 2
0
0.05
0.1
0.15
0.2
f/f
0
ψ

I
m
[
1

f
ν
]
Figure 4.17: Plot of the real and imaginary part of f
ν
as a function of the normalized frequency.
The data points are measurements and the solid curve is a fit. The maximum D
r
reached in the
tube is 1 %.
The multiple-stack-position results are in excellent agreement with the single-stack-
position results. This confirms the validity of both methods. Since we do the mea-
surements for five different stack positions there is no frequency at which the stack is
always in a velocity node. For this reason we do not measure the peaks at f /f
0
= 0.97
and f /f
0
= 1.95 as we saw in the single-stack-position measurements.
4.8 Discussion and conclusion 77
4.8 Discussion and conclusion
The multi-microphone method proved to be a valuable technique for determining
acoustic-energy flows in a resonator tube as well as the transfer matrix of a stack.
The two-microphone and multi-microphone methods are well-known techniques
for determining the transfer matrix of porous media. This is the first time the multi-
microphone method is used in case of a stack as porous medium. Also, as far as we
know, the traveling-wave method has not been used before in determining the acoustic-
energy dissipation of a stack.
The multi-microphone method is more accurate than the two-microphone method.
Especially for frequencies at which the distances between the two microphones are
multiples of a half wavelength, the two-microphone method fails, whereas the multi-
microphone method works well for a whole range of frequencies (100 Hz to 2000 Hz).
For determining the acoustic-energy flow two methods have been tested: the Fusco
method and the traveling-wave method. The two methods are in good agreement with
the exception of the frequencies at which the microphone are a multitude of λ/2 sepa-
rated from each other. At these frequencies the Fusco method gives unreliable results.
The traveling-wave method does not show this problem.
The acoustic-energy dissipation at low amplitudes is in agreement with the linear
theory. At higher amplitudes the linear theory starts to deviate from the measurements.
These deviations can partially be explained by minor losses at the stack ends, especially
for high velocity amplitudes. For high pressure amplitudes the deviations are probably
caused by higher harmonics, but this has not been verified.
For the first time the multi-microphone method has been used to determine the
transfer matrix elements of a stack. From the transfer matrix elements the f
ν
functions
of a stack can be determined as functions of the frequency. Two methods have been
tested: the single-stack-position method and the multiple-stack-positions method. Both
methods are in excellent agreement with each other. The single-stack-position method
only shows a few small deviations from the analytical f
ν
function. The multiple-stack-
positions method is even in better agreement with the analytical plot of f
ν
as function
of the frequency.
78 Acoustic measurements
Bibliography
[1] A. F. Seybert and D. F. Ross, “Experimental-determination of acoustic properties
using a 2-microphone random-excitation technique,” Journal of the Acoustical Society
of America, vol. 61, no. 5, pp. 1362–1370, 1977.
[2] J. Y. Chung and D. A. Blaser, “Transfer-function method of measuring in-duct
acoustic properties .1. theory,” Journal of the Acoustical Society of America, vol. 68,
no. 3, pp. 907–913, 1980.
[3] J. Y. Chung and D. A. Blaser, “Transfer-function method of measuring in-duct
acoustic properties .2. experiment,” Journal of the Acoustical Society of America,
vol. 68, no. 3, pp. 914–921, 1980.
[4] J. Y. Chung and D. A. Blaser, “Transfer-function method of measuring acoustic
intensity in a duct system with flow,” Journal of the Acoustical Society of America,
vol. 68, no. 6, pp. 1570–1577, 1980.
[5] W. T. Chu, “Extension of the 2-microphone transfer-function method for
impedance tube measurements,” Journal of the Acoustical Society of America, vol. 80,
pp. 347–348, July 1986.
[6] W. T. Chu, “Transfer-function technique for impedance and absorption-
measurements in an impedance tube using a single microphone,” Journal of the
Acoustical Society of America, vol. 80, pp. 555–560, Aug. 1986.
[7] H. Bod´ en and M.
˚
Abom, “Influence of errors on the 2-microphone method for
measuring acoustic properties in ducts,” Journal of the Acoustical Society of Amer-
ica, vol. 79, pp. 541–549, Feb. 1986.
[8] W. T. Chu, “Further experimental studies on the transfer-function technique for
impedance tube measurements,” Journal of the Acoustical Society of America, vol. 83,
pp. 2255–2260, June 1988.
[9] W. T. Chu, “Impedance tube measurements - a comparative-study of current prac-
tices,” Noise Control Engineering Journal, vol. 37, pp. 37–44, July 1991.
[10] M. G. Jones and T. L. Parrott, “Evaluation of a multi-point method for determining
acoustic-impedance,” Mechanical Systems and Signal Processing, vol. 3, pp. 15–35,
Jan. 1989.
80 BIBLIOGRAPHY
[11] S. H. Jang and J. G. Ih, “On the multiple microphone method for measuring in-duct
acoustic properties in the presence of mean flow,” Journal of the Acoustical Society of
America, vol. 103, pp. 1520–1526, Mar. 1998.
[12] G. Penelet, S. Job, V. Gusev, P. Lotton, and M. Bruneau, “Dependence of sound am-
plification on temperature distribution in annular thermoacoustic engines,” Acta
Acustica united with Acustica, vol. 91, no. 3, pp. 567–577, 2005.
[13] W. P. Arnott, H. E. Bass, and R. Raspet, “General formulation of thermoacoustics
for stacks having arbitrarily shaped pore cross sections,” Journal of the Acoustical
Society of America, vol. 90, no. 6, pp. 3228–3237, 1991.
[14] L. A. Wilen, “Measurements of scaling properties for acoustic propagation in a
single pore,” Journal of the Acoustical Society of America, vol. 101, no. 3, pp. 1388–
1397, 1997.
[15] L. A. Wilen, “Measurements of thermoacoustic functions for single pores,” Journal
of the Acoustical Society of America, vol. 103, no. 3, pp. 1406–1412, 1998.
[16] G. Petculescu and L. A. Wilen, “Thermoacoustics in a single pore with an applied
temperature gradient,” Journal of the Acoustical Society of America, vol. 106, no. 2,
pp. 688–694, 1999.
[17] G. Petculescu and L. A. Wilen, “High-amplitude thermoacoustic effects in a single
pore,” Journal of the Acoustical Society of America, vol. 109, no. 3, pp. 942–948, 2001.
[18] A. F. Seybert and B. Soenarko, “Error analysis of spectral estimates with application
to the measurement of acoustic parameters using random sound fields in ducts,”
Journal of the Acoustical Society of America, vol. 69, no. 4, pp. 1190–1199, 1981.
[19] H. Bailliet, P. Lotton, M. Bruneau, V. Gusev, J. Vali` ere, and B. Gazengel, “Acoustic
power flow measurement in a thermoacoustic resonator by means of laser doppler
anemometry (l.d.a.) and microphonic measurement,” Applied Acoustics, vol. 60,
no. 1, pp. 1–11, 2000.
[20] A. Fusco, W. Ward, and G. Swift, “Two-sensor power measurements in lossy
ducts,” Journal of the Acoustical Society of America, vol. 91, no. 4 I, pp. 2229–2235,
1992.
[21] T. Biwa, Y. Tashiro, H. Nomura, Y. Ueda, and T. Yazaki, “Experimental verifica-
tion of a two-sensor acoustic intensity measurement in lossy ducts,” Journal of the
Acoustical Society of America, vol. 124, no. 3, pp. 1584–1590, 2008.
[22] V. L. Streeter, E. B. Wylie, and K. W. Bedford, Fluid Mechanics, vol. 9th edition.
McGraw-Hill Companies, 1998.
[23] B. H. Song and J. S. Bolton, “A transfer-matrix approach for estimating the charac-
teristic impedance and wave numbers of limp and rigid porous materials,” Journal
of the Acoustical Society of America, vol. 107, pp. 1131–1152, Mar. 2000.
[24] G. Swift, A Unifying Perspective for Some Engines and Refrigerators. Melville: Acous-
tical Society of America, 2002.
Chapter 5
Temperature profile in a stack
5.1 Introduction
In this chapter we study the temperature profile in a parallel-plate stack, experimentally
as well as numerically. Both the steady state and the time dependency are studied. The
experimental set-up consists of a stack without heat exchangers in a resonator tube,
which is driven by a loudspeaker. Such a stack without heat exchangers is also called
a thermoacoustic couple [1]. Were the thermoacoustic couple devoid of thermal contact
with the environment, it would keep heating up. In practice, a steady state will be
reached, due to thermal contact with the environment.
In the steady state, the total heat flow to the environment is equal to the absorbed
acoustic power. For a good understanding of a thermoacoustic stack it is essential to
know how the heat flows to the environment. To study the heat losses we studied the
passive cooling down of a stack. First we used the loudspeaker to create a temperature
difference in the stack. Once a steady state is reached, the loudspeaker is turned off
and the temperature profile over the stack is measured as function of time. The pas-
sive cooling down of the stack gives us valuable information of the heat losses to the
environment. It will be shown that the conduction through the plates in radial direction
(from the the middle of the plates to the tube wall) is dominating the heat losses.
Most thermoacoustic models [1–4] assume that the total power in a stack
˙
E is con-
stant in axial direction. This works well in a thermally isolated stack with heat exchang-
ers at both sides. In case of a thermoacoustic couple this assumption does not hold, since
the energy flow at the front of the stack is almost equal to the acoustic loudspeaker out-
put, whereas the energy flow at the end of the stack is almost zero. The difference is
caused by heat losses. One way to solve the energy equation for the temperature pro-
file, is to assume a linear energy profile
˙
E(x). This is a rough approximation, since the
heat losses are not constant over the length of the stack, but depend on the temperature
difference with the environment, T(x) −T
a
. As a consequence the results of the steady
state model are not satisfying.
For this reason we developed a new, time-dependent, model. This model includes a
more realistic approximation of the heat losses.
Detailed measurement on the temperature fields in the gas are done by Wetzel and
Herman [5–8]. They used a combination of holographic interferometry and high-speed
82 Temperature profile in a stack
cinematography to measure the local temperature and the heat transfer qualitatively.
Zoontjens et al. [9, 10] have done detailed numerical calculations of the flow and energy
fields in thermoacoustic couples and have also studied the influence of different plate
edges on the energy flow.
5.2 Experimental set-up
A parallel-plate stack is installed in the resonator tube. A cross section of the stack is
shown in figure 5.1. The plate distance D = 1.0 mm and the plate thickness d = 1.0 mm.
The inner tube diameter 2R
in
= 59.1 mm and the outer diameter 2R
out
= 70.0 mm. The
stack length is 202.0 mm and the porosity is 0.52. The tube wall and the plates are made
of transparent perspex for flow visualization (chapter 6).
59.1 mm
70.0 mm
Figure 5.1: Schematic drawing of a cross section of the stack. The plate distance is 1.0 mm as
is the plate thickness. The center plate, depicted in lighter gray, has a printed circuit board with
temperature sensors on top of it.
To measure the plate temperature T
0
at the different positions in the stack 32 tem-
perature sensors are installed on top of the center plate. The temperature sensors are
platinum resistance thermometers
1
. The temperature sensors are connected to a data-
acquisition system by a thin flexible printed circuit board, with copper lanes towards a
connector outside of the tube, as is shown in figure 5.2. To keep the top of the plate flat
and its thickness at 1.0 mm a thin layer is milled out of the plate, such that the printed
circuit board fits in it. At the positions of the temperature sensors, small cavities are
made in the plate to fit the sensors in. The printed circuit board is glued to the plate and
the cavities are filled with thermally-conducting glue around the temperature sensors.
1
PT1000
5.2 Experimental set-up 83
1 32
32
1
∆x
∆x0
∆xend
8.9 mm 61.0 mm
205.6 mm
64 pin connector
Figure 5.2: Schematic drawing of the center plate, viewed from above, with a printed circuit
board with temperature sensors glued on top of it. The sensor are separated by ∆x = 6.5 mm.
The distances to the boundaries, ∆x
0
= 1.8 mm and ∆x
end
= 2.3 mm. Only two measuring leads
are shown, schematically.
84 Temperature profile in a stack
Having the printed circuit board installed on the stack, the temperature sensors are
calibrated again. The temperature plate has one branch with a connector on it. The
printed circuit board is connected to a data-acquisition system, which monitors the re-
sistance of the temperature sensors. The resistance of all thermometers is measured at a
sample frequency of 1 kHz, averaged over 1 sec and stored as data.
The resistance of a thermometer depends linearly on the temperature
R(T) = R
0
+β(T −T
r
), (5.1)
with R
0
and β empirical constants and T
r
a reference temperature of 300 K. The ther-
mometers have been calibrated by measuring the resistance at room temperature and
after being in a oven at 80

Cfor 6 hours . This way the R
0
and β of all 32 thermometers
are determined. This accuracy of the measured temperature is 0.1

C. The inaccuracy is
mainly caused by small errors in the offset resistance R
0
.
5.3 Model
In section 2.1 three 1-D differential equations were derived (Eq. 2.16, 2.30, and 2.34),
dp
1
dx
= −
iωρ
0
A(1 − f
ν
)
U
1
, (5.2a)
dU
1
dx
= −
iωA[1 + (γ −1) f
κ
]
γp
0
p
1
+
f
κ
− f
ν
(1 − f
ν
)(1 −σ)
dT
0
dx
1
T
0
U
1
, (5.2b)
˙
E =
1
2
Re
_
p
1
U

1
_
1 −
f
κ
− f

ν
(1 + P
r
)(1 − f

ν
)
__
+
ρ
0
c
p
[U
1
[
2
2Aω(1 −P
2
r
)[1 − f
ν
[
2
Im( f
κ
+ P
r
f

ν
)
dT
0
dx
−(Ak + A
so
k
so
)
dT
0
dx
, (5.2c)
with three variables (U
1
(x), p
1
(x), and T
0
(x)) and the total energy (
˙
E(x)).
In the Steady State model (SS model) we assume that
˙
E(x) decreases linearly with x
in the stack and solve the system of equations 5.2 with a Runge–Kutta method. Later an
improved model is developed, which includes the time dependency of the temperature
and also heat losses to the environment. This is the Time Dependent model (TD model).
5.3.1 Steady State model
To solve the set of equations 5.2 we need three boundary conditions and we need to
know
˙
E(x). As boundary conditions we use
p
1
(x
L
) = p
L
, (5.3a)
U
1
(x
L
) = U
L
, (5.3b)
T
0
(x
L
) = T
L
, (5.3c)
5.3 Model 85
as is also indicated in figure 5.3. At position x = x
L
the acoustic energy flow into the
stack is
˙
W
L
=
1
2
Re[p
L
U

L
], (5.4)
and at position x = x
R
the acoustic energy flow out of the stack is
˙
W
R
=
1
2
Re[p
1
(x
R
)U

1
(x
R
)]. (5.5)
TL
pL
UL
WL WR
xL=0 xR=Lp
p1(x) T0(x) E(x) U1(x)
Figure 5.3: Schematic drawing of the SS model. A parallel-plate stack is enclosed in a resonator.
x is the position in the stack, with x
L
denoting the left side and x
R
the right side of the stack.
The boundary conditions on the left side of the stack are the pressure p
L
, the volume flow rate
U
L
, and the temperature T
L
. The total energy flow in the stack,
˙
E(x), is assumed to decrease
linearly with x, with
˙
W
L
the incoming energy flow at x = x
L
and
˙
W
R
the outgoing energy flow
at x = x
R
. By solving the set of three differential equations the pressure p
1
(x), the volume flow
rate U
1
(x), and temperature T
0
(x) can be found as functions of the position in the stack.
Unfortunately, we do not know the x dependency of
˙
E(x). But we do know the total
energy at the stack boundaries:
˙
E(x
L
) =
˙
W
L
and
˙
E(x
R
) =
˙
W
R
, since no heat exchangers
are present. Since
˙
W
L
>
˙
W
R
the total power
˙
E(x) must decrease along the stack. We
have to ask ourselves, where this power is going to. Initially, the stack temperature is
close to the environmental temperature and the dissipated acoustic power is completely
used to heat up the stack. Later, when the stack is heated, part of the dissipated acoustic
power flows to the environment. Once the system is in steady state, the acoustic energy
that is put into the stack is equal to the heat losses to the environment. In the SS model
both the heat losses and power required to heat up the stack are not included. The total
power
˙
E(x) is assumed to decrease linearly with the position in the stack x:
˙
E(x) =
˙
W
L

˙
W
L

˙
W
R
L
p
x. (5.6)
This is a rough assumption, since the heat losses are expected to be proportional with
the temperature difference with the environment, which is not constant over the stack.
The set of three ordinary differential equations 5.2 is solved with a three-stage Lo-
batto IIIa method, which is an implicit Runge–Kutta method and is a standard function
included in Matlab.
As we have discussed how the equations are to be solved, let us have a look at
the parameters in Eqs. 5.2 and their dependencies on the temperature. Since the stack
86 Temperature profile in a stack
consists of parallel plates, we can use Eq. 2.18b for f
ν
and f
κ
:
f
ν
=
tanh[(1 + i)D
p
/2δ
ν
]
(1 + i)D
p
/2δ
ν
, (5.7)
f
κ
=
tanh[(1 + i)D
p
/2δ
κ
]
(1 + i)D
p
/2δ
κ
, (5.8)
with D
p
= 2y
0
the plate separation and δ
ν
and δ
κ
the boundary layer thicknesses, given
by
δ
ν
=
¸

ωρ
0
, (5.9)
δ
κ
=
¸
2k
ωρ
0
c
p
. (5.10)
For the dynamic viscosity µ and the thermal conduction k we use the following empiri-
cal temperature dependencies:
µ = µ
0
_
T
0
T
r
_
b
µ
, (5.11)
k = k
0
_
T
0
T
r
_
b
k
, (5.12)
where T
r
is a reference temperature at which µ = µ
0
and k = k
0
and b
µ
and b
k
are two
empirical constants. The dimensionless Prandtl number, defined as P
r
= c
p
µ/k, is also
temperature dependant. The ratio of specific heats is calculated by γ = c
p
/(c
p
− R
s
).
The density is determined by the ideal gas law ρ
0
= p
0
/R
s
T
0
. In our experimental
set-up we use atmospheric air. In table 5.1 the constants that are used in the model are
presented.
5.3.2 Time Dependent model
In the Time Dependent (TD) model both the heat losses to the environment and the
heating up of the stack as function of time are included. Another difference with the SS
model is that the temperature over the stack cross section is not assumed to be uniform.
We focus on the middle plate, as that is the plate where the thermometers are located.
The middle plate is shown in figure 5.4. The plate has thickness d
p
, width W
p
, and
length L
p
. Note that the stack consists of 17 parallel plates, but only the middle one is
shown here, as that is the one we use in the modeling of the energy equation. The plates
are attached to a cylindrical stack holder (dashed circles) with inner radius R
i
and outer
radius R
o
. The stack plates and the stack holder are made of perspex.
The TD model is based on a number of assumptions. The time dependence is as-
sumed to be so small that it does not influence the wave equations. Only the temper-
ature is modeled as function of time. But since the pressure and volume flow rate are
5.3 Model 87
Table 5.1: List of constants used in our model.
Gas area A 14.3 cm
2
Solid area A
so
13.1 cm
2
Stack length L
p
205.6 mm
Pressure p
0
1.0110
5
Pa
Viscosity µ
0
1.85 10
−5
Pas
b
µ
0.76
Gas thermal conduction k
0
2.610
−2
W/Km
b
k
0.89
Reference temperature T
r
300 K
Solid conduction k
so
0.155 W/Km
Specific gas constant R
s
286.7 J/(kgK)
Gas heat capacity c
p
1.003 kJ/(kgK)
Solid heat capacity c
so
1.17 kJ/(kgK)
Plate thickness d
p
1 mm
Plate separation D
p
1 mm
Width of center plate W
p
59.1 mm
R
i
y
z
x
d
p
R
o
L
p
W
p
Figure 5.4: Schematic 3D drawing of the center plate in the stack. The plate has thickness d
p
,
width W
P
, and length L
p
. The plate is located in a cylindrical plate holder, with inner diameter
2R
i
and outer diameter 2R
o
. Note that the other plates in the stack are not shown here and that
the plate dimensions are not on scale.
88 Temperature profile in a stack
x
i
x1
x2
x
i-1 x
i+1 x
n
Ti
Ti-1 Ti+1
Tn T2
T1
y
x
S1
Si-1
Si
Te
Te
WL
WR
Figure 5.5: Schematic drawing of the numerical grid on the center plate of the stack. The plate
is divided in n −1 segments numbered S
1
to S
n−1
. The grid points are located at the borders of
the segments. In total n grid points are located on the dashed center line, indicated with dots.
The x positions of the grid points are x
1
. . . x
n
and the temperatures are T
1
. . . T
n
. The edges of
the plate that connect with the stack holder are assumed to be at environmental temperature T
e
.
The acoustic energy flow on the left side of the plate is
˙
W
L
and on the right side
˙
W
R
.
temperature dependent, they also slowly change in time. This time dependency of p
1
and U
1
is assumed to be much smaller than the acoustic variations. We assume that the
heat losses to the environment can be approximated by
˙ q
e
(x) = α(T(x) −T
e
), (5.13)
where ˙ q
e
(x) is the heat flow, α a constant heat transfer coefficient, T(x) the temperature,
and T
e
the environmental temperature (T
e
= 24

C).
To solve the energy equation for T(x, t) we use a discretization of the plate in the
x direction, which is shown in figure 5.7. The middle stack plate is in the x-direction
(along the tube axis) divided into segments S
1
. . . S
n−1
, which all have a heat capacity.
At every edge of a segment a grid point for the temperature is located. The x positions of
the grid points are x
1
. . . x
n
and the discreticized temperatures are T
1
. . . T
n
. Each plate
segment has its own energy balance. The total net power of the plate segment divided
by the heat capacity determines the temperature increase dT
i
/dt of the segment. The
edges of the plate that are attached to the stack holder are assumed to be at environ-
mental temperature T
e
.
Using the energy equation (Eq. 5.2c) the total power at the two borders of each
segment can be calculated. The first term of the energy equation we define as Y(x),
Y(x) =
1
2
Re
_
p
1
U

1
_
1 −
f
κ
− f

ν
(1 + P
r
)(1 − f

ν
)
__
, (5.14)
is an energy flow and the second term,
Z(x) =
ρ
0
c
p
[U
1
[
2
2Aω(1 −P
2
r
)[1 − f
ν
[
2
Im( f
κ
+ P
r
f

ν
)
dT
0
dx
, (5.15)
is the thermoacoustic pumping effect. The pumping effect can either enhance the thermal-
conduction term or compensate for it, depending on its sign. Now let us look at the total
5.3 Model 89
energy balance of segment i. The energy flow into the segment at the left side is Y(x
i
)
and the energy flow out of the segment is Y(x
i+1
). The contribution of the pumping
term on the left side is Z(x
i
)(T
i
−T
i−1
)/∆x and analogous on the right side. The ther-
mal conduction on the left side is (Ak + A
so
k
so
)(T
i
− T
i−1
)/∆x and analogous on the
right side. And the last contribution is of the heat losses. The change of temperature as
function of time ∂T/∂t time the heat capacity is equal to the net power:
ρc
p
V
∂T
∂t
= Y(x
i
) −Y(x
i+1
) + (Z(x
i
) + Ak + A
so
k
so
)(T
i
−T
i−1
)/∆x −
(Z(x
i+1
) + Ak + A
so
k
so
)(T
i+1
−T
i
)/∆x −
˙
Q
e
, (5.16)
with V the volume of the segment.
The model of the plate is shown as an electrical network in figure 5.6.
Wi
Ci
1/(α∆xWp)
Wi+1
Ti
Wi+1
Ci+1
1/(α∆xWp)
Wi+2
Ti+1
Zi
Figure 5.6: Electrical network of the TD model. The voltages V
i
correspond to temperatures
T
i
and the currents correspond to energy flows. The capacitance to earth is the equivalent of
heat capacity. The heat losses are a resistance to earth. The conduction and pumping effect are
integrated in the resistance Z
i
. The energy flows Y
i
are included by current sources.
The heat flows are approximated by Eq. 5.13. We need to determine α. The heat
losses are dominated by conduction in the y direction of the plate towards the stack
holder, which is at temperature T
e
. The temperature profile in the y direction of the
plate is shown in figure 5.7. The heat flow due to losses is
˙ q = k
so
_
_
∂T
∂y
¸
¸
¸
¸
y=−W
p
/2

∂T
∂y
¸
¸
¸
¸
y=W
p
/2
_
_
. (5.17)
For a quadratic profile it follows that α = 8k
so
/W
p
.
The initial temperature profile that is used for the model is uniformly room temper-
ature. At this temperature profile T(x) we calculated p(x) and U(x) using Rott’s wave
equations (Eqs. 2.16 and 2.30). Consequently
∂T
∂t
is calculated, from which T(x) one
timestep later can be determined. The new determined T(x) can be used for calculating
the new p(x) and U(x) and the whole process is repeated again.
This whole process is repeated timestep after timestep until a steady state is reached.
90 Temperature profile in a stack
-W
p
/2 W
p
/2
0
T
T
e
T
c
y
Figure 5.7: Model of the plate temperature as function the y position, with y = 0 the center of
the plate, W
p
the plate width, T
e
the temperature at the stack holder, and T
c
the temperature of
the center of the plate.
5.4 Heat losses
When performing a long-term measurement, after a while the system reaches a steady
state. The amount of acoustic power dissipated in the stack is equal to the heat loss
of the stack to the environment. We will study how this energy is transferred to the
environment. We ask ourselves, how much is lost by radiation, conduction, forced con-
vection, and free convection? We consider the resonator tube wall as a heat sink at
environmental temperature. Due to its thickness and good heat conductivity, the tube
wall remains a at environmental temperature.
To study which part of the heat loss can be accounted to each heat loss effect, we
monitor the temperature in the stack when the system is in steady state and the speaker
is suddenly turned off. Two things can be noticed. The temperature gradient in the
stack will slowly decay, due to the absence the thermoacoustic effect. Additionally, the
internal energy in the stack as a whole, will decay due to heat flow to the surrounding.
The heat losses in the system with the speaker turned off, are due to radiation, free
convection and conduction.
With static heat losses in a stack we mean the heat losses of the stack plates to the
environment with absence of an acoustic flowthrough the stack. The x direction is along
the tube axis, the z direction is perpendicular to the stack plates, and the y direction is
parallel to the plates and perpendicular to the tube axis.
The temperature in the plate as a function of time is described by the Fourier equa-
tion:
ρc
∂T
∂t
= ∇ (k∇T), (5.18)
As the plate thickness (1 mm) is much smaller than the length, the temperature is as-
sumed to be uniform in the z-direction and ∇ (k∇T) = k(∂
2
T/∂x
2
+ ∂
2
T/∂y
2
). The
k∂
2
T/∂y
2
is the main cause of the heat losses to the environment. The tube wall is close
to environmental temperature and is effectively a heat sink.
In figure 5.8 two cooling down measurements are presented. The temperature pro-
file, i.e. the temperature as function of the position, in the stack is shown at six different
moments in time. At time t = 0 the loudspeaker is turned off and the temperature
measurement is initiated. The difference between figure 5.8a and 5.8b is the initial tem-
5.5 Temperature profile measurement 91
perature profile. The initial temperature difference is 30

C in figure 5.8a and 60

C in
5.8b.
0 0.05 0.1 0.15 0.2
10
20
30
40
50
x [m]
T
[
o
C
]


t =0 s
t =365 s
t =731 s
t =1463 s
t =2927 s
t =7818 s
t =0 s
365 s
731 s
1463 s
2927 s
7818 s
(a)
0 0.05 0.1 0.15 0.2
0
20
40
60
80
x [m]
T
[
o
C
]


t =0 s
t =365 s
t =731 s
t =1463 s
t =2927 s
t =7818 s
t =0 s
365 s
731 s
1463 s
2927 s
7818 s
(b)
Figure 5.8: The temperature as function of the position, measured at 32 positions, at six different
moments in time after the loudspeaker is turned off. The solid curves represent measured data
and the dashed curves are calculated numerically. The initial temperature difference is 30

C in
figure (a) and 60

C in (b).
5.5 Temperature profile measurement
5.5.1 Time dependent measurements
We recorded the temperature profiles as functions of time during a period of 1600 s. The
measured temperature profiles are shown at five moments in time by the solid curves
in figure 5.9a. The initial temperature profile is almost constant around 23

C. The
stack is located in a region with a positive temperature gradient. When the speaker is
turned on, the temperature first starts to deviate from the initial temperature profile at
the edges of the stack, as can be seen after 25 s. At the edges the difference with the
critical temperature gradients is the largest. After 100 s the temperature profile is still
flat in the middle of the plate (50 mm< x <180 mm). At t = 400 s a nonzero temperature
gradient starts to develop in the middle of the plate and at t = 1400 s the profile is almost
linear.
The measured initial temperature is used as the initial temperature for the TDmodel.
The calculated temperature profile is shown by the dashed curves. The model corre-
sponds quite well with the experiments. This justifies the assumptions made in the
model, regarding heat losses. Especially at t = 25 and t = 100 the model and experiments
are in good agreement with each other. At t = 400 and t = 1400 the calculated tempera-
ture gradients at the edges are significantly higher than the measured ones. This can be
explained by extra heat losses at the stack edges, mainly due to convection, which are
not included in the TD model.
92 Temperature profile in a stack
0 0.05 0.1 0.15 0.2
0
10
20
30
40
50
x [m]
T
[
o
C
]


t =0 s
t =25 s
t =100 s
t =400 s
t =1400 s
0 s
25 s
100 s
400 s
1400 s
(a)
0 0.05 0.1 0.15 0.2
0
20
40
60
80
100
x [m]
1
T
[
o
C
]
1


SS model
1
TD model
1
initial
1
final
1
(b)
Figure 5.9: Measurements and calculations of the temperature profile (T(x)) in a stack at a
frequency of 50 Hz.
a) The temperature profile determined by measurement (solid curves) and for the TD model
(dashed curves) at different moments in time.
b) The final temperature profile determined by measurement and for both models.
The measured and calculated profile share a certain point (or small region) where
the temperature stays at (or close to) the environmental temperature during the whole
measurement. This point is located at x = 80 mm. When the thermacoustic effect would
only involve the pumping of heat, this point is expected to be in the middle of the plate
(x = 103 mm). Since the thermoacoustic effect also leads to the heating of the whole
plate, this point is shifted to the left.
In figure 5.9b the final (t=1660 s) temperature profiles are shown. The solid curve
is calculated by the SS model, the dashed curve by the TD model (after 1660 s), and
the dash-dotted curve is measured. For the SS model we use the measured final tem-
perature at x = 0 as the T
L
boundary condition. The SS model predicts a much higher
temperature difference over the stack than the TD model and measurement. The system
is in steady state after 1660 s, which was verified by continuing the TD simulation until
t = 5000 and noticing no significant changes in the temperature profile. As the system is
in steady state, the difference between the SS model and the measurements can only be
explained by heat losses to the environment which are not included in the SS model.
5.5.2 Amplitude sweep
Here we study the final temperature difference ∆T = T
R
− T
L
over the stack. The
measurement is continued until ∆T converged and this measurement is repeated at
different drive ratios. In figure 5.11a the temperature difference is shown as function of
the pressure amplitude at the left side of the stack, at a frequency of 55 Hz. The circular
dots are measurement points, the solid curve represents the SS model and the dashed
curve the TD model. The dash-dotted line is the critical temperature difference. In
5.5 Temperature profile measurement 93
0 0.05 0.1 0.15 0.2
18
20
22
24
26
28
x [m]
T
[
o
C
]


t =0 s
t =25 s
t =100 s
t =400 s
t =1400 s
0 s
25 s
100 s
400 s
1400 s
(a)
0 0.05 0.1 0.15 0.2
10
20
30
40
50
60
70
x [m]
1
T
[
o
C
]
1


SS model
1
TD model
1
initial
1
final
1
(b)
Figure 5.10: Measurements and calculations of the temperature profile (T(x)) in a stack at a
frequency of 80 Hz.
a) The temperature profile determined by measurement (solid curves) and for the TD model
(dashed curves) at different moments in time.
b) The final temperature profile determined by measurement and for both models.
figure 5.12a the acoustic energy flow into the stack W
in
(at the left axis) and the velocity
amplitude [u
L
[ (on the right axis) are shown as a function of the pressure amplitude [p
L
[,
at 55 Hz. The acoustic energy flow is proportional with [p
L
[
2
and [u
L
[ is proportional
with [p
L
[.
A similar measurement is performed at 158 Hz. The temperature differences at 158
Hz are shown in figure 5.11b and the acoustic energy flow and velocity amplitude are
shown in figure 5.12b.
The critical temperature is determined by integrating the critical temperature gradi-
ent over the stack
(∆T)
c
=
_
x
R
x
L
_
dT
dx
_
c
dx, (5.19)
with (
dT
dx
)
c
given by Eq. 2.27. The [p
1
[ and [U
1
[ that are used in Eq. 2.27 are calculated
with the SS model, at the highest [p
L
[. It has been verified that when [p
L
[ is increased
further the temperature difference predicted with the SS model approaches (∆T)
c
.
5.5.3 Frequency sweep
In figure 5.13a the temperature difference over the stack is plotted versus the frequency.
The stack is located at 69 mm from the closed end of the resonator. In figure 5.13b the
drive ratio and velocity are shown. During this measurement the loudspeaker voltage is
constant. The resonance frequencies of the resonator are at 55 Hz and at 165 Hz, which
can be clearly seen from the peaks in the pressure and the velocity. The temperature
difference also shows a peak at these frequencies, as the acoustic power input to the
94 Temperature profile in a stack
0 5 10
0
20
40
60
80
100
120
p
L
[kPa]

T
[
o
C
]


SS model
TD model
measurement
(∆T)
c
(a) f =55 Hz
0 0.5 1 1.5 2
0
20
40
60
80
100
p
L
[kPa]

T
[
o
C
]


SS model
TD model
measurement
(∆T)
c
(b) f =158 Hz
Figure 5.11: The temperature difference as function of the pressure amplitude [p
L
[ at the left
side of the stack. The measurement is performed at 55 Hz and at 158 Hz. The temperature
difference is determined from a measurement as well as from both models.
0 5 10 15
0
0.2
0.4
W
i
n
[
W
/
c
m
2
]
|p
L
| [kPa]
0 5 10 15
0
5
10
|
u
L
|
[
m
/
s
]
(a) f =55 Hz
0 1 2 3
0
0.05
0.1
W
i
n
[
W
/
c
m
2
]
|p
L
| [kPa]
0 1 2 3
0
2
4
|
u
L
|
[
m
/
s
]
(b) f =158 Hz
Figure 5.12: The energy flow
˙
E
0
that is applied to the stack and the velocity amplitude [u
L
[
as functions of the pressure amplitude [p
L
[ at the left side of the stack. The measurement is
performed at 55 Hz and at 158 Hz.
stack is higher and the stack has not reached the critical temperature yet. The TD model
is in very good agreement with the measurements. The SS model overestimates the
temperature difference, as it does not include heat losses. The shape of the temperature
plots is determined heavily by the pressure and velocity dependence of the frequency.
In figure 5.14a the temperature difference over the stack is plotted versus the fre-
5.6 Discussion and conclusion 95
50 100 150
0
50
100
150
f [Hz]

T
[
o
C
]


SS model
TD model
measurement
(∆T)
c
(a)
50 100 150
0
0.05
0.1
D
R
f [Hz]


50 100 150
0
0.02
0.04
|
u
L
|
/
c
0
DR
|u
L
|/c
0
(b)
Figure 5.13: a) The temperature difference T
R
−T
L
in steady state as function of the frequency,
determined by the SS model, by the TD model and from measurements. The calculated critical
temperature difference (∆T)
c
is also plotted. b) The drive ratio D
r
= [p
L
[/p
0
and the Mach
number [u
L
[/c
0
at the left side of the stack as function of the frequency. The stack is located at
69 mm from the end of the resonator.
quency. In figure 5.14b the normalized pressure and velocity are shown. The stack is
located at 469 mm from the closed end of the resonator. During this measurement the
loudspeaker voltage is constant. The resonance frequencies of the resonator is at 43 Hz.
This resonance frequency is lower than in the previous measurement, because in this
case the tube is 40 cm longer. The temperature difference shows a peak at the resonance
frequency, but this peak is flatter than in the previous measurement. The reason for
this is that the temperature is close to the critical temperature and therefore can not go
higher. The critical temperature difference is much lower this time, because the stack is
much farther away from the resonator end. The temperature difference goes towards
zero at a frequency of 120 Hz, as a knot in the pressure amplitude reaches the stack.
At one part of the stack the temperature gradient is positive and at the other part it
is negative. The critical temperature gradient does not go to zero because the critical
temperature gradient is always positive, by definition. At even higher frequencies the
temperature difference becomes negative.
The TD model is in very good agreement with the measurements, except at the res-
onance frequency, where it approaches the SS model.
5.6 Discussion and conclusion
Literature shows many to have measured the temperature difference between the two
stack ends, or between two heat exchangers, but, as far as we know, this is the first time
that the temperature profile in a stack is measured. We have measured the temperature
at 32 positions in the stack solid. Wetzel and Herman have visualized and quantified
96 Temperature profile in a stack
0 50 100 150 200
−20
0
20
40
60
f [Hz]

T
[
o
C
]


SS model
TD model
measurement
(a)
50 100 150 200
0
0.01
0.02
0.03
0.04
0.05
0.06
f [Hz]
D
R
,
|
u
L
|
/
c
0


DR
|u
L
|/c
0
(b)
Figure 5.14: a) The temperature difference T
R
−T
L
in steady state as function of the frequency,
determined by the SS model, by the TD model and from measurements. The calculated critical
temperature difference (∆T)
c
is also plotted. b) The drive ratio D
r
= [p
L
[/p
0
and the Mach
number [u
L
[/c
0
at the left side of the stack as function of the frequency. The stack is located at
469 mm from the end of the resonator.
the temperature fields in the neighborhood of a single plate with a combination of holo-
graphic interferometry and high-speed cinematography.
Not only did we measure the steady-state temperature profile, but also the dynamic
effects. We made a model of the energy balance in the center plate of the stack. We
have included the heat capacity of the plate, to model the time-dependency. For the
temperature in the radial direction a parabolic profile in the plates is assumed.
The modeled temperature profiles as functions of time are consistent with the mea-
surements. So the model works well for a stack that is not thermally isolated.
For further research it would be interesting to use thermometers at different radial
positions to verify the parabolic temperature profile in this direction. The heat losses
to the environment could be reduced significantly by reducing the thermal contact be-
tween the stack and the resonator tube.
Bibliography
[1] J. Wheatley, T. Hofler, G. Swift, and A. Migliori, “Intrinsically irreversible ther-
moacoustic heat engine.,” Journal of the Acoustical Society of America, vol. 74, no. 1,
pp. 153–170, 1983.
[2] A. Atchley, T. Hofler, M. Muzzerall, M. Kite, and C. Ao, “Acoustically generated
temperature gradients in short plates,” Journal of the Acoustical Society of America,
vol. 88, no. 1, pp. 251–263, 1990.
[3] A. Piccolo and G. Cannistraro, “Convective heat transport along a thermoacoustic
couple in the transient regime,” International Journal of Thermal Sciences, vol. 41,
no. 11, pp. 1067–1075, 2002.
[4] P. in ’t panhuis. PhD thesis.
[5] M. Wetzel and C. Herman, “Experimental study of thermoacoustic effects on a
single plate. part i: Temperature fields,” Warme- und Stoffubertragung Zeitschrift,
vol. 36, no. 1, pp. 7–20, 2000.
[6] M. Wetzel and C. Herman, “Experimental study of thermoacoustic effects on a
single plate. part ii: Heat transfer,” Warme- und Stoffubertragung Zeitschrift, vol. 35,
no. 6, pp. 433–441, 1999.
[7] M. Wetzel and C. Herman, “Limitations of temperature measurements with holo-
graphic interferometry in the presence of pressure variations,” Experimental Ther-
mal and Fluid Science, vol. 17, no. 4, pp. 294–308, 1998.
[8] M. Wetzel and C. Herman, “Accurate measurement of high-speed, unsteady tem-
perature fields by holographic interferometry in the presence of periodic pressure
variations,” Measurement Science and Technology, vol. 9, no. 6, pp. 939–951, 1998.
[9] L. Zoontjens, C. Howard, A. Zander, and B. Cazzolato, “Numerical study of flow
and energy fields in thermoacoustic couples of non-zero thickness,” International
Journal of Thermal Sciences, vol. 48, no. 4, pp. 733–746, 2009.
[10] L. Zoontjens, C. Howard, A. Zander, and B. Cazzolato, “Numerical comparison of
thermoacoustic couples with modified stack plate edges,” International Journal of
Heat and Mass Transfer, vol. 51, no. 19-20, pp. 4829–4840, 2008.
98 BIBLIOGRAPHY
Chapter 6
Flow visualization in and around
a stack
6.1 Introduction
The last decades a vast amount of research in the field of thermoacoustics has been
conducted [1]. Rott [2–7] and Swift [8, 9] have derived analytical equations describing
thermoacoustics. This linear theory, however, is only valid at a relatively small drive
ratio, which is defined as the pressure amplitude to average pressure ratio. The linear
theory is in good agreement with measurement for drive ratios smaller than 10% [10].
The theory is used as basis for designing thermoacoustic applications. In many appli-
cations, however, large drive ratios (more than 10%) are used. At larger drive ratios,
unfortunately, the coefficient of performance (COP), which is defined as the cooling
power to work ratio, decreases significantly. This might be attributed to nonlinear ef-
fects occurring at these high drive ratios, like turbulence, vortex formation, streaming,
and higher harmonics. Therefore the study of nonlinear effects has become an impor-
tant issue in thermoacoustics. This chapter focusses on the vortex formation at the end
of a parallel-plate stack and the velocity profile in the middle of the stack.
Steady flow around obstacles, especially circular cylinders, is well-known [11]. In
case of R
e
< 1 (the Reynolds number is defined here as R
e
=
V

D
ν
, with V

the velocity
far away from the cylinder, D the cylinder diameter, and ν the kinematic viscosity) the
Stokes approximation can be used. The vorticity, which is created close to the cylinder
surface due to the no-slip boundary condition, is not advected, resulting in a symmetri-
cal flow. At increasing Reynolds numbers twin vortices appear (4 < R
e
< 40), a vortex
street develops (40 < R
e
< 200), which finally becomes unstable (R
e
> 200), and even
turbulent (R
e
> 5000).
Oscillatory flows around single circular cylinders have been studied recently by Ne-
hari et al. [12], Anagnostopoulo and Minear [13], Wybrowet al. [14], and others. Parallel
plates have a rectangular shape instead of circular. Ozgoren [15] studied the flow struc-
ture behind both square and circular cylinders using DPIV. A stack of parallel plates
is much different from a single circular cylinder. Not much is known about oscillatory
flow around rectangular cylinders and certainly not about oscillatory flow around a
100 Flow visualization in and around a stack
stack of parallel plates. Because the length of the plates (130 mm) is much larger than
the acoustic displacement (< 7 mm), the two stack edges can be studied separately. A
stack edge can be modeled as an oscillating flowthrough an array of channels into a vol-
ume. This geometry is very similar to a channel with a sudden change in cross section.
Oscillating flow in channels with a sudden change in cross-section has been studied by
Ibrahim and Hashim [16] and Morris et al. [17] numerically and by Smith [18, 19] and
Skulina [20, 21] experimentally, using PIV and microphone measurements. Oscillatory
flow through a rapid expansion was also studied by Smith [22, 23]. This geometry is
similar geometry to a jet pump, which was studied by Petculescu and Wilen [24]. Al-
ready in 1979 Disselhorst and Van Wijngaarden [25, 26] studied oscillatory flow near
tube entrances both experimentally and theoretically. They used schlieren to study the
vortex shedding near both sharp and round edges and showed that different vortex-
shedding modes exist, depending on the initial state. We will show both numerically
and experimentally that a similar behavior occurs for the flow at the end of a stack plate.
Not much research has been conducted on the flow around a stack of parallel plates
specifically. Stoltenkamp studied steady flow around stacked plates at an angle con-
cerning a rotor [27, 28]. Oscillating flow around stacked plates has been studied exper-
imentally using PIV and also numerically by Blanc-Benon et al. [29] at relatively low
amplitudes, where a vortex pair originates behind the stack plates and no vortex streets
appear. Berson et al. [30–32] and Mao et al. [33] recently have measured a vortex street
behind parallel plates at higher amplitudes using PIV. The interaction of the flow with
a stack and heat exchangers is studied numerically by Besnoin and Knio [34]. The heat
exchange, velocity field, vorticity and temperature distribution at the end of a stack
are calculated. However, until now, no systematic experimental research has been un-
dertaken on the flow patterns behind stack plates, in which all the relevant parameters
(plate thickness and separation, velocity amplitude, frequency, and plate-end shape) are
varied and their influence on the flow structure is studied.
This chapter focusses on oscillatory flow through a stack. In section 6.2 the experi-
mental set-ups are described. PIV measurements of the vortex shedding behind a stack
plate are discussed in section 6.3, the influence of different dimensionless numbers in
section 6.4, different plate-end shapes in section 6.5, numerical simulations of the vor-
tex shedding in section 6.6, and the evolution of a vortex street in section 6.7. In section
6.8 the velocity profile in-between two parallel plates is discussed, including a laminar-
turbulent transition. The natural convection at the hot end of a stack is discussed in
section 6.9 and the streaming velocity at the end of a stack is discussed in section 6.10.
We conclude with section 6.11.
6.2 Experimental Set-up
For the flow visualization we use a method called Particle Image Velocimetry (PIV). For
this method we inject small oil droplets (with a mean particle diameter of 0.2 µm) into
the set-up, which follow the acoustic flow and are used as tracer particles to visualize
the flow. The movement of the particles is recorded by a combination of a camera and a
laser. The details of the PIV method and hardware are discussed in section 6.2.3.
Two different set-ups were built for PIV measurements. The first set-up consisted of
a perspex resonator tube and is discussed in detail in section 6.2.1. This set-up was a
6.2 Experimental Set-up 101
relatively simple one, in the sense that it was easy and cheap to build. The idea of this
set-up was mainly to test and familiarize ourselves with the PIVhardware and software,
and also the particle generator and particle behavior in an acoustic flow. Although this
set-up was relatively simple, it enabled us to obtain many results of the vortex shedding
behind plates.
With the experience gained with the perspex set-up, we designed a second, more
advanced, PIV set-up. The Aluminium set-up was designed to have a number of im-
provements in comparison to the perspex set-up:
• Thick aluminium resonator walls, to allow for accurate acoustic measurements,
which were not possible in the perspex set-up.
• Obtain higher drive ratios and velocities, to study nonlinear effects.
• Longer resonator tube, to use lower frequencies and therefore larger boundary
layers.
• Larger inner diameter of resonator tube, to increase the number of plates in a stack.
• More flexibly in stack position, tube length, plate thickness, and separation.
The geometry, loudspeaker, and microphone details of this set-up were already dis-
cussed in section 3.3 and the flow visualization part of this set-up is described in section
6.2.2.
6.2.1 Perspex set-up
The perspex set-up consists of a loudspeaker-driven standing-wave resonator. The res-
onator is a cylindrical perspex tube, with an inner diameter of 25 mm and a total length
of 1400 mm (figure 6.1). The tube is closed with a transparent end plate on the left end,
connected to a Dynaudio D54AF loudspeaker on the right end, and is filled with atmo-
spheric air at room temperature. The first resonance frequency in the setup occurs at
125 Hz. At different locations microphones are inserted in the tube wall to measure the
pressure. A stack of parallel plates is placed 350 mm in front of the endplate. The flow
field measurements occur at the left end of the stack. A close-up of this measurement
area is shown at the bottom of the figure. At the measurement window (field of view)
the tube has a smaller wall thickness in order to increase its optical transparency. The
observation window is still thick enough not to act as a membrane and start vibrating.
The measurement window (field of view) is in the center of the tube and perpendicular
to the stack plates.
In most thermoacoustic devices, in general, the first resonance frequency is of the
order 0.1 - 1 kHz. In our experiments we have chosen a relatively low resonance fre-
quency: 125 Hz. A low frequency has two advantages for visualization:
1. it lowers the required time resolution of the laser and camera;
2. we can increase the plate separation of the stack as a consequence of larger pene-
tration depths.
3. due to larger scales, visualization is easier.
102 Flow visualization in and around a stack
Figure 6.1: Schematic drawing of the experimental set-up. At the top a resonance tube connected
to a loudspeaker. (a) Transparent endplate, (b) injection nozzles for seeding, (c) stack of parallel
plates, (d) loudspeaker. At the bottom a close-up of the measurement area. The dotted area is the
measurement window, which is 20 mm 15 mm in size.
6.2 Experimental Set-up 103
The 2nd advantage follows from the penetration depths being proportional to ω
−1/2
(Eq. 2.12). The stacks that are used in this set-up consist of five parallel perspex plates.
The plate thickness d = 1 mm and the plate distance D = 4 mm. To study the influence of
the plate thickness in some measurements stacks with a plate thickness of 0.5 mm and a
plate distance of 4.5 mm are used, as well as single plates with a thickness of 3 mm. The
stack length is 130 mm. To reduce laser light reflections all stack plates are coated black.
6.2.2 Aluminium set-up
The geometry of the Aluminium set-up, the positions of the microphones, and the
speaker details are discussed in section 3.3. Here we will focus on the PIV Lavision
part of the set-up. In figure 6.2 a part of the resonator tube and the laser optics are
shown. Two different configurations are used to create a light sheet at the measurement
window. In configuration (a) the light sheet enters the resonator through the transpar-
ent end plate, whereas in configuration (b) the light sheet enters the resonator tube from
above. For most measurements both configurations can be used, but for measurements
in the middle of the stack configuration (a) is required and for measurements in the
regenerator-based set-up (chapter 7) configuration (b) is required. The laser and the
first 45
o
mirror are standing on an horizontal rail, beneath the table. The positions of
the mirrors as well as their angle can be modified for a good alignment of the laser beam
and laser sheet. The laser beam passes through a hole in the table and is then reflected
by a second mirror. In configuration (a) this mirror is located at the tube axis and in
configuration (b) above the tube axis. The laser beam is turned into a light sheet by the
light sheet optics. The focal point and the angle of the light sheet can be manually set.
In configuration (a) the light sheet passes the transparent end plate. In configuration
(b) the light sheet is reflected by a third mirror and enters the resonator tube, which
is transparent at this location. The focal point of the light sheet is at the measurement
window and the angle is such that at least the whole measurement window is covered.
The CCD camera is located perpendicular to the plane of the figure and is directed at
the measurement window. In front of the camera zoom lens a filter is placed, to reduce
the influence of background light. Only light in a wavelength range around that of the
laser can pass the filter. To reduce laser reflections we use fluorescent paint. Fluorescent
paint converts the wavelength of the original laser light into a different wavelength that
cannot pass through the filter in front of the camera.
The camera is located on a horizontal rail, so we can easily change the position of
the measurement window along the tube. The light sheet optics and third mirror of
configuration (a) are attached to a frame that moves along with the camera over the
horizontal rail. A photo of this frame, the camera, the horizontal rail, and the stack
is shown in figure 6.3. A second horizontal rail is used to change the distance of the
camera to the resonator tube, in order to vary the size of the measurement window.
104 Flow visualization in and around a stack
45
o
mirror
45
o
mirror
light-sheet optics
measurement window
double-pulsed laser
stack
transparent window
(a) a light sheet through the transparent end plate
45
o
mirror
45
o
mirror
light-sheet optics
measurement window
double-pulsed laser
stack
transparent window
45
o
mirror
(b) a light sheet from above
Figure 6.2: Schematic drawing of the Aluminium set-up.
(a) The laser beam from a double-pulsed laser is reflected by two 45
o
mirrors. Here the beam
goes through the light-sheet optics and a light sheet is formed that enters the tube through a
transparent end plate and illuminates the particles in the measuring window.
(b) The beam goes through the light-sheet optics above the tube, a light sheet is formed that is
reflected by a 45
o
mirror and enters the tube from above.
In both configurations the CCD camera is perpendicular to the plane of picture aimed at the
measurement window.
6.2 Experimental Set-up 105
6.2.3 Particle Image Velocimetry
Particle Image Velocimetry (PIV) is an optical method of flow visualization, that can
provide an accurate quantitative measure of the instantaneous flow field across a planar
area of a flow field. It is used to obtain an instantaneous 2D velocity vector field in and
around a stack. The fluid is seeded with tiny tracer particles, oil droplets smaller than
1 µm, which are assumed to faithfully follow the flow dynamics. This method is to a
large degree nonintrusive, as the added oil droplets cause negligible distortion of the
fluid flow. It is the motion of these seeding particles that is used to calculate velocity
information of the flow being studied. During PIV, the particle concentration is such
that it is possible to identify individual particles in an image, but not with certainty to
track it between images. Therefore, instead of following individual particles, particle
clusters are followed.
Using a light sheet, formed by passing a double pulsed laser beam through light
sheet optics, the particles in the flow are illuminated twice with a small time separation
∆t. One picture is taken at t
1
= t − ∆t/2 and a second at t
2
= t + ∆t/2. The time
between the pictures, ∆t, is chosen such that the maximum particle displacement is
roughly 5 pixels, which is 1/6 of the size of an interrogation window [35]. With the
illumination of these two short duration laser flashes, two pictures of the illuminated
particles are captured by a charge-coupled device (CCD) camera and are stored in two
separate frames. The frames are split into a large number of interrogation windows,
as is shown in figure 6.4. It is then possible to calculate a displacement vector for each
window with the help of signal processing and cross-correlation techniques. The spatial
displacement that produces the maximum cross-correlation statistically approximates
the average displacement of the particles in the interrogation cell. Velocity associated
with each interrogation spot is the displacement divided by the time between the laser
pulses ∆t. The size of the interrogation window is chosen to have at least 6 particles per
window on average.
With PIV a subpixel accuracy can be reached. But since the correlation function is
determined on the discrete pixels, the determination of its maximum is biased toward
integer values. This is called the peak-locking effect, which can be problematic, espe-
cially if the particles are smaller than one pixel and the particle concentration is low. In
our measurements the probability density function of the displacement did not show
a bias towards integer values, so we concluded that the peak-locking effect was not an
issue for us.
In our set-ups we use a 15 Hz PIV System from LAVISION, including a double laser,
CCD, optics and software. For illumination a New Wave ND-YAG laser is used, which
produces 2 x 50 mJ of energy with a pulse length of 5 ns. Laser sheet optics in front of the
laser create a laser sheet that coincides with the measurement area and has a thickness
of 0.5 mm. The optics consist of a spherical lens and cylindrical lens combination. The
cylindrical lens expands the laser into a plane while the spherical lens compresses the
plane into a thin sheet. Images are recorded with a 1600x1200 pixels
2
Imager Pro CCD
camera.
The PIV system is calibrated by placing a calibration grid at the field of view in the
plane of the light sheet. This calibration is used to convert the displacement in pixels to
physical length scales and also to correct for spherical abberations due to the cylindrical
shape of the resonator tube.
106 Flow visualization in and around a stack
Figure 6.3: Photo of the PIV part of the aluminium set-up. (a) resonator tube, (b) 45
o
mirror,
(c) 2 mm parallel-plate set-up, (d) camera lens, (e) horizontal rail, (f) bellows, (g) CCD camera.
Figure 6.4: Schematic view of the cross-correlation method that is used to determine the dis-
placement vectors within an interrogation window. Two frames time ∆t apart are shown. Both
frames are divided in a number of interrogation windows (16 in this picture, but much more
in the actual experiment). A close-up of one interrogation window of each frame is shown. We
determined the cross-correlation function of the two frames, by varying the displacement ∆x and
∆y. By taking the highest peak, we find the best displacement vector (∆x, ∆y).
6.3 A typical measurement 107
Diethyl hexyl sebacate (DEHS) seeding particles are created by a Palas AGF2.0 seed-
ing generator. The mean particle size, distributed by the seeding generater, is 200 nm,
with a maximum particle size of 1 µm. An airflow through the resonance tube is gen-
erated during the preparation of each measurement, to create a homogeneous aerosol
distribution over our measuring volume. The particles are injected through a nozzle in
front of the measurement section and are released through a nozzle behind the mea-
surement section. After closing the nozzles, the seeding density in the measurement
section is sufficiently high. When the injection nozzles are closed, the injection flow has
to be decayed sufficiently, before measurements can be performed. This is important for
streaming measurements in particular. From test measurements it was concluded that
1 minute after closing the injection nozzles, no significant flow (smaller than 0.2 mm/s)
was present in the measurement area.
Before starting the PIV algorithm we use an image preprocessing method, called
”subtract a sliding background”. With this method large intensity fluctuations in the
background due to laser reflections are filtered out while small intensity fluctuations
of the particle signal will pass through. This way we receive an image with a con-
stant background level without affecting the particle signal we correlate on. The cross-
correlation algorithm is used iteratively, with decreasing the interrogation-window size
each time. In the final cross-correlation step the interrogation windows are 32 by 32
pixels
2
.
6.3 A typical measurement
The vortex formation is studied using PIV at the left end side of a stack. In this typical
measurement the stack consists of 5 rectangular plates with the following geometry
parameters: d = 1.0 mm, D = 4.0 mm, and L = 130 mm. The Reynolds and Strouhal
numbers are here defined as:
R
e
=
[u
1
[d
ν
, (6.1)
S
t
=
f d
[u
1
[
, (6.2)
with [u
1
[ the peak velocity of the main flow just outside the stack. The peak velocity
inside the stack is higher than V due to a smaller gas area inside the stack. The flow
can be characterized by f = 125 Hz, [p
1
[ = 3.7 kPa, [u
1
[ = 5 m/s. This corresponds to
R
e
= 330 and S
t
= 0.025. The viscous penetration depth δ
ν
= 0.2 mm. The phase angle
at which the measurement has been performed ωt
0
= π. This is the moment at which
the main flow is at rest and the particles are at their leftmost excursion. During the mea-
surement one picture is taken at t
1
= t
0
− ∆t/2 and a second at t
2
= t
0
+ ∆t/2. The
time between the pictures, ∆t, is chosen such that the maximum particle displacement
is roughly 5 pixels, which is 1/6 of the size of an interrogation window [35]. By a cross-
correlation algorithm the two pictures are compared and the particle displacement at
different positions is determined. The particle displacement is divided by the time dif-
ference ∆t, resulting in a velocity field that is shown in the upper plot of figure 6.5. The
vector lengths and colors are a measure of the magnitude of the velocity vectors. The
color bar to the right of the picture indicates which vector color corresponds to which
108 Flow visualization in and around a stack
velocity. The black rectangles represent a part of the stack plates. In total 10075 vec-
tors are determined, which corresponds to a resolution of 14 vectors/mm
2
. However, to
keep the individual vectors visible, the resolution of the images in this paper are lower.
The relative errors of the measured velocities are typically 1% of the maximum velocity.
When studying vortices originating at the end of stack plates the vorticity
ω
z
=
∂v
∂x

∂u
∂y
, (6.3)
is a useful parameter. A vorticity plot of the same measurement is shown in the bottom
plot of figure 6.5. The color bar at the right side of the plot shows the correspondence
between color and vorticity. At the end of each plate a street of alternating vortices is
present. Note that there is a slight phase shift between the forming of vortices behind
the different plates. Sufficiently far away from the stack (more than five times the dis-
placement length) the flow in the stack is axisymmetrical. The asymmetry of the flow
patterns can be caused by small asymmetries in the geometry. In section 6.6 we will
show that the same asymmetries are also present in numerical simulations with a sym-
metrical geometry.
6.4 Different categories of vortex patterns and dimension-
less numbers
Figure 6.7 is a simplified drawing of the different vortex patterns that occur at the end
of a plate when the amplitude is increased. We can divide the vorticity plots at the stack
end into four categories:
A two vortices: Two vortices of opposite sign (a dipole) are formed behind each
stack plate, as is shown in figures 6.6(a) and 6.7(a).
B four vortices: Behind each stack plate four vortices are formed, as is shown in
figures 6.6(b) and 6.7(b).
C transition area: A category in-between four vortices and a vortex street, as is
shown in figure (6.6(c)). The distinction between this category and categories B
and D is somewhat arbitrary. A vortex pattern is placed in category B when the
number of vortices is four and the vortices are located directly behind the stack
in a symmetrical way. In the transition area the number of vortices is larger than
four or the vortices are not symmetrically located in the extension of the plate.
D vortex street: A row of at least seven vortices is formed behind each plate (figures
6.6(d) and 6.7(c)). A vortex pattern is called a vortex street when at least four
alternating vortices can be distinguished.
The objective is to find a correlation between the category of vortex formation (A
to D) and relevant dimensionless parameters (R
e
and S
t
). At higher Reynolds numbers
the instationary term of the Navier-Stokes equation becomes more important. Therefore
we expect that at higher Reynolds numbers the flow is more inclined towards a vortex
street than to two or four vortices. The Strouhal number is a ratio of the plate thickness
6.4 Different categories of vortex patterns and dimensionless numbers 109
−10 −5 0 5
−6
−4
−2
0
2
4
6
8
x/d
y
/
d


u
[
m
/
s
]
0
1
2
3
4
5
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
Figure 6.5: A typical flow visualization measurement at the end of a stack. The top figure shows
the velocity vector field and the bottom picture its corresponding vorticity plot. The three black
rectangles represent (a part of) the stack plates, which are 1 mm thick and at a distance of 4 mm
from each other. The frequency is 125 Hz and the velocity amplitude is 5.0 m/s.
110 Flow visualization in and around a stack
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
ω
z
[
1
0
4
s

1
]
−0.06
−0.04
−0.02
0
0.02
0.04
0.06
(a) v=0.50 m/s
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
ω
z
[
1
0
4
s

1
]
−0.1
−0.05
0
0.05
0.1
(b) v=0.83 m/s
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
ω
z
[
1
0
4
s

1
]
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
(c) v=2.5 m/s
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
(d) v=5.0 m/s
Figure 6.6: The vorticity behind plates for four different velocity amplitudes. The phase angle
θ = π. The velocity amplitude in-between the stacks [u
1
[ = 5 m/s and the frequency is 125 Hz.
The plates, represented by black rectangles, have a thickness of 1 mm and are 4 mm apart from
each other.
+
-
(a) cat A: two vortices
+
- +
-
(b) cat B: four vortices
+
-
+
-
-
-
+
+
(c) cat D: vortex street
Figure 6.7: Different categories of vortex formation at the end of a stack plate.
6.4 Different categories of vortex patterns and dimensionless numbers 111
and the displacement amplitude. The plate thickness is a measure for the vortex size in
the direction perpendicular to the plate and the displacement amplitude determines the
number of vortices in the direction parallel to the plate. The vortex size is based on the
optical appearance in the vorticity plots. At low Strouhal numbers we expect the vortex
to be more elongated and farther away from the plate and is therefore expected to be
more inclined towards a vortex street.
In figure 6.8 the Strouhal number is plotted versus the Reynolds number for a set of
25 measurements with different frequencies (5, 25, 45, 114, 125, 240, and 372 Hz) and
different velocity amplitudes (0.1 to 10.0 m/s). For each measurement the vorticity pat-
tern is studied and is placed in one of the four categories. Depending on the category
the data points in figure 6.8 have different symbols. There is a clear distinct correlation
between the two relevant dimensionless parameters and the different categories. Each
category has its own area in the Reynolds-Strouhal space. The two black lines indi-
cate the boundaries between these areas. The lines can be described by the following
function:
S
t
/R
e
= f ν/V
2
= α, (6.4)
in which α is a constant. For the border line between category B and C it follows that
α = (1.9 ±0.4) 10
−4
and for the border line between C and D it follows that α = (8 ±
1) 10
−4
. For the range of R
e
and S
t
it seems that only the ratio S
t
/R
e
is determinative
for the category of vortex patterns occurring at the end of a plate.
1/S
t
is a measure for the number of vortices that can be formed. For 1/S
t
∼ 2π, i.e.
V/2π f ∼ d, only two vortices are expected, since the displacement length is approx-
imately of the same size as the vortices. For a vortex street to be formed, two criteria
need to be fulfilled:
A 1/S
t
> 2π, otherwise the displacement length is not long enough for more than
two vortices to be formed,
B the R
e
number needs to be high enough for vortices to be formed.
Another dimensionless parameter that is often used in oscillating flows is the
Keulegan-Carpenter number, KC, also called the period number. It is the ratio of a
displacement amplitude (V/ω) to a characteristic length scale.
When this characteristic length scale is the length of the stack, the Keulegan-
Carpenter number is defined as
KC
L
=
V
ωL
. (6.5)
In our set-up KC
L
¸1. This implies that the stack is much longer than the displacement
amplitude. The stack can be considered as semi-infinite. Sufficiently far away (more
than 2V/ω) from the plate edges the flow is fully developed in the stack. There is no
interaction between the flow effects at the two ends of the stack (apart from streaming
effects). The situation is different when KC
L
≈ 1, in which case the stack plates are of the
same size as the displacement amplitude. This can occur with very small stacks or very
high amplitudes. Gas parcels can move from one stack end to the other. In the middle
of the stack the flow is not fully developed and there is a strong interaction between the
flow effects at the two stack ends.
112 Flow visualization in and around a stack
0 50 100 150 200 250 300 350 400 450 500 550 600 650
0.0
0.1
0.2
0.3
0.4
0.5
S
t

Re
two vortices
four vortices
transition
vortex street
Figure 6.8: The Strouhal number plotted versus the Reynolds number for different data points.
For each data point it is determined in which category it belongs by studying the vorticity plots.
The different categories are indicated by different colors and shapes, as is indicated in the legend.
The groups of data points can be separated by straight lines trough the origin.
6.5 The influence of the plate-end shape 113
When the characteristic length scale of the KC number is the plate separation D, we
get the following definition
KC
D
=
V
ωD
. (6.6)
This number is related to the Strouhal number as
KC
D
=
1
2πS
t
. (6.7)
Another important parameter is the Reynolds number based on the viscous pene-
tration depth
R

=

ν
ν
. (6.8)
This number is important in the transition of laminar to turbulent for oscillatory flows.
This number is related to α
R

=
_
2
α
. (6.9)
The two border lines in figure 6.8 correspond to R

= 50 and R

= 102.
6.5 The influence of the plate-end shape
So far all measurements were performed on rectangular plates with a flat end (figure
6.9(a)). In this subsection we discuss the influence of the shape of the plate ends on the
vortex shedding at the end of the plates. We measured the flow around single plates
with a thickness of 3 mm. In figure 6.9(b) the flow around a circular end is shown,
in figure 6.9(c) around a 90
o
-triangle and in figure 6.9(d) around a 25
o
-triangle. Due
to the absence of sharp corners for the circular end the vorticity plot is more smooth
and the two vortices are almost circles. The 90
o
triangle does have sharp corners and
is very similar to a rectangular end. At the end of a 25
o
triangle however, two thin
layers of vorticity are present behind the sharp corner. These two layers do not roll
up. The vortex on top of the triangle is caused by the transition of a straight plate
towards a triangle. A sharp triangular shape is clearly giving different results as the
other geometries. The choice of a shape can influence the dissipation due to the vortices
that are shed off and the heat transfer in the heat exchanger. At triangular ends with
even sharper edges (12
o
or smaller) vortices completely disappear and the shear layer
dissolves smoothly in the main stream.
6.6 Numerical simulations
6.6.1 CFD model
We have used commercial CFD packages to run a numerical simulation of a oscillating
flow at the end of a stack. We have used the same geometry as in the experiment which
is discussed in section 6.3, with the same frequency f =125 Hz, and the same velocity
amplitude [u
1
[ = 5 m/s. We are only interested in flow effects and therefore do not
account for the compressibility of the gas.
114 Flow visualization in and around a stack
x/d
y
/
d


−2 0 2
−2
−1
0
1
2
ω
z
[
1
0
3
s

1
]
−5
0
5
(a) circle
x/d
y
/
d


−2 0 2
−2
−1
0
1
2
ω
z
[
1
0
3
s

1
]
−5
0
5
(b) rectangle
x/d
y
/
d


−2 0 2
−2
−1
0
1
2
ω
z
[
1
0
3
s

1
]
−5
0
5
(c) triangle
x/d
y
/
d


−2 0 2
−2
−1
0
1
2
ω
z
[
1
0
3
s

1
]
−5
0
5
(d) knife
Figure 6.9: The vorticity behind plates with different end shapes. The plate thickness d = 3 mm,
the velocity amplitude of the main flow is 5 m/s, f = 125 Hz. The angle of the triangle (c) is 90
o
and the angle of the knife (d) is 25
o
.
6.6 Numerical simulations 115
The geometry is two dimensional and is shown in figure 6.10. The modeled region
is indicated with a gray color and the black regions represent the wall (stack plates and
resonator wall). In total we have four stack plates in our model, with thickness d = 1
mm and with 4 mm in-between them. At the interface with the walls a no-slip condition
is applied. We have used two additional boundary conditions:
u(t) = [u
1
[ sin(ωt), for x/d = −30, (6.10)
∂u
∂x
= 0, for x/d = 30. (6.11)
We have created a multi block structured mesh, with 113,120 rectangular elements.
The mesh is refined at the boundary layers, where the velocity gradient is highest. The
differential equations are solved using the so called SIMPLE (Semi-Implicit Method
for Pressure-Linked Equations) algorithm. A second-order implicit time discretization
scheme together with a second-order upwind space discretization for convective terms
was chosen. No turbulence modeling was applied. We use 200 time steps for each pe-
riod and run the simulation for 10 periods. The fluid is atmospheric air at 300 K and
incompressible.
To verify the algorithm the simulation is performed two times, with different codes.
First the simulation was ran in CFX and subsequently in FLUENT, with a different mesh
and a different time step. Since the results were very similar, we conclude that the
algorithm works well. But this does not imply that the model is correct. Therefore we
will compare the simulation results with the PIV measurements.
x/d
0 -30 30
y/d
0
-5
d
-9.5
14.5
u(t)
0 = x / u
Figure 6.10: A schematic drawing of the two-dimensional geometry that used for the numerical
simulations. The dimensions are scaled by the plate thickness d = 1 mm. The black rectangles
represent walls, at which a no-slip condition is applied.
In figure 6.11 a typical result of the numerical simulation at θ = 3π/2 + 2πn is
shown. This result can be compared to the PIV measurement in figure 6.5.
When comparing the numerical simulation with the measurement we see that qual-
itatively both results are very similar. The size of the vortices (in-between 1 and 2 mm
in diameter) and distance in-between the vortices (approximately 2 mm) are in good
agreement. Differences can be caused by averaging over 50 samples, which is executed
for the measurement, whereas the numerical results are instantaneous measurements.
116 Flow visualization in and around a stack
−10 −5 0 5
−6
−4
−2
0
2
4
6
8
x/d
y
/
d


u
[
m
/
s
]
0.5
1
1.5
2
2.5
3
3.5
4
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
Figure 6.11: Numerical simulation. A plot of the velocity field (top) and of the vorticity (bot-
tom).
6.6 Numerical simulations 117
6.6.2 Subsequent cycles
From the PIV measurements we observed that the measurements are not perfectly re-
producible. When we repeat the measurement at the same phase angle the vorticity plot
differs. We expect this to be caused by statistical fluctuations. This can be observed most
clearly at the phase angle θ = 3π/2. Based on symmetry one would expect the vorticity
pattern to be completely symmetrical with the centers of the plates the symmetry axes.
The measurements show a pattern in which the vortices are created alternately, which is
similar to a Von K´ arm´ an vortex street. This seems to be energetically favorable. The vor-
tex street can start either with a positive or with a negative vortex. This is determined
by statistical fluctuations or by the previous cycle. In this subsection we will study the
reproducibility of the numerical simulations. We will look into the vorticity pattern at
θ = 3π/2 at five subsequent cycles, which are shown in figure 6.12. In figure 6.12a
behind the top plate, a negative vortex is attached to the plate end, followed by a vortex
with a positive sign and then negative and positive vortices alternatingly. The second
vortex resembles much more a circular shape then the third and fourth vortices, which
are elongated ellipses. To the left of the fifth vortex the vorticity is not uniform, but the
vortices can not be distinguished clearly. The vortex street behind the middle plate is
very similar to the vortex street behind the top plate. The bottom plate has a different
vortex pattern behind it. Here the vortex attached to the plate is positive in sign. This
vortex street is almost a mirror image of the other two.
In figure 6.12a the vortex pattern is similar to that of 6.12b, with the major difference
being that at the bottom plate the first and third vortex (both positive in sign) start to
attach to each other. In figure 6.12c we see that the attachment of the first and third
vortex continues and that they have unified into one vortex. The unified vortex and the
second vortex are next to each other. And also the third and fourth vortex are next to
each other instead of being behind each other.
In figure 6.12d an interesting vortex pattern is created behind the top plate. One
could even argue that this is not a vortex street, but instead two vortices next to each
other, both attached to the plate and both of them with a filament of vorticity behind
them.
In figure 6.12e all three plates show a very similar vortex pattern.
We see that different modes are possible for the vortex shedding, depending on his-
tory. We see the same behavior in our measurements. Also small asymmetries in the
geometry can have a significant influence on the vortex shedding mode.
6.6.3 Minor losses
In chapter 3 we discussed the minor losses occurring at the ends of a stack, due to
a sudden change in cross section. In this section we will determine the minor losses
by a numerical simulation. Near the edge of the stack, the flow is heavily distorted
due to vortices. Sufficiently far away from the stack edge (more than two times the
displacement amplitude) the velocity profile is laminar. The numerical code does not
include the energy equation, only the Navier-Stokes and the continuity equation. The
gas is considered by the code as an incompressible fluid with a temperature that is
118 Flow visualization in and around a stack
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
Figure 6.12: The vorticity plots at θ = 3π/2 for five subsequent cycles, determined by a nu-
merical simulation. The three black rectangles represent (a part of) the stack plates which have
thickness d=1 mm and are at a distance of 4 mm from each other. The frequency is 125 Hz and
the velocity amplitude in the stack is 5.0 m/s.
6.6 Numerical simulations 119
constant in time and position. In the laminar area the energy flow can be determined by
˙
W =
__
1
τ
_
τ
0
ρ(h +[v[
2
/2)udtdA, (6.12)
with the enthalpy per unit mass given by h = + p/ρ. The internal energy per unit
mass is not relevant here, as the numerical simulation is at constant temperature. The
energy flow per unit of area can be written as
˙
W(x)/A =
1
τ
_
τ
0
_
(p(t, x, y) +ρ[v(t, x, y)[
2
/2)u(t, x, y)
_
dt, (6.13)
with τ = 2π/ω the time of one period, and ¸¸ denoting the cross-sectional average. In
the laminar area the viscous losses per unit of length are given by
d
˙
W
dx
= −
1
2
Re¦R
1
¦[U
1
[
2
, (6.14)
with R
1
given by Eq. 4.31:
Re¦R
1
¦ =
ωρ
0
A
Im¦−f
ν
¦
[1 − f
ν
[
2
. (6.15)
Note that thermal relaxation is not included in this simulation, because the gas is at
constant temperature in the simulation.
In figure 6.13
˙
W(x)/A is plotted as function of x. The left ends of the stack plates are
located at x = 30 mm. In this plot the slope, d
˙
W/dx, is always negative, as is expected as
a result of a decrease in the energy flow due to viscous dissipation. The kinetic energy
has a significant contribution to the energy flow. At x/d = 60 the energy flow is zero.
For 10 < x/d < 20 (out of the stack) the energy flow
˙
W(x)/A is expected to decrease
linearly by 0.07 Wm
−2
mm
−1
according to Eq. 6.14. The numerical data are fitted lin-
early for 10 < x/d < 20 to a linear curve with a slope of 0.07 Wm
−2
mm
−1
. The fit is
shown by the dashed line. For 25 < x/d < 35 the flow is distorted due to vortices. For
x/d > 35 (in the stack) the flow is laminar again and according to Eq. 6.14 the energy
flow is expected to decrease by 0.50 Wm
−2
mm
−1
. The numerical data are fitted linearly
for 40 < x/d < 55 to a linear curve with a slope of 0.50 Wm
−2
mm
−1
. The fit is shown
by the dash-dotted line. At the x/d = 30 the two fit lines would intersect if there would
be no minor losses. The difference between the two fit lines is the minor loss ∆
˙
W/A =
(2.3 ± 0.5) W/m
2
, indicated by the arrow. Using the relation between minor losses and
K, which was derived in section 4.4,

˙
W =
2


0
A
2
[U
1
[
3
(6.16)
we find that the corresponding minor loss coefficient K = 0.07 ± 0.02. The analytical
value of minor loss coefficient for a sudden expansion K
e
= 0.04. The minor losses
120 Flow visualization in and around a stack
during the contraction are determined by the vena contracta
K
c
=
_
1
C
c
−1
_
2
, (6.17)
with C
c
the ratio between the area of the jet at the vena contracta to the area between
the plates. According to Weisbach, for ψ = 0.8 this C
c
ratio is 0.81, which results in
K
c
= 0.05. But this value of C
c
is measured in a circular geometry instead of parallel
plates. In section 6.7 we will show that it is difficult to determine the vena contracta
from our numerical simulations.
15 20 25 30 35 40 45 50 55
5
10
15
20
25
x/d
˙
W
/
A
[
W
/
m
2
]

˙
W/A
Figure 6.13: The energy flow
˙
W/A including kinetic energy as function of the position. Four
parallel plates are present in the geometry for x ≥ 30 mm. The plates are 1 mm thick and at
a distance of 4 mm from each other. The frequency is 125 Hz and the velocity amplitude is 5.0
m/s. The dashed line represents the energy flow outside of the stack if no change in cross section
would occur and the dash-dotted line is the energy flow between the plates sufficiently far away
from the cross-section change. The difference between the two fits, at x = 30 mm, is the minor
loss ∆
˙
W/A (indicated by the double arrow).
6.7 Vortex street evolution
The velocity as function of the phase angle at the x/d = −30 is shown in figure 6.14.
The velocity u(t) is normalized by its peak value u
L
= 5 m/s and the phase angle θ is
normalized by 2π. The squares indicated by A to F are the phase angles at which we
will show plots of the vorticity. The plots corresponding to A, B, and C are shown in
figure 6.15 and the plots corresponding to D, E, and F are shown in figure 6.16. In the
6.7 Vortex street evolution 121
left columns the PIV measurements are shown and in the right columns the numerical
simulations.
0 0.25 0.5 0.75 1
−1
−0.5
0
0.5
1
θ/2π
u
(
t
)
/
u
L
A
B
C
D
E
F
Figure 6.14: The velocity at x/d = −30 as function of the phase angle. The squares A to F
indicate at which phase angle a vorticity plot is shown in the next two figures.
As a starting point (A) we choose θ = 0, the moment at which the gas is at its
rightmost position. The velocity is zero at this moment. In the plot of the numerical
simulation the boundary layer near the plates is not developed. The disturbances in the
boundary layer are caused by vortices, that were formed during the previous period,
interfering with the boundary layer. In the measurement plot these disturbances in the
boundary layer are less clear. As the spatial resolution (the number of velocity vector
per area) is lower, the disturbances are smoothed out. The vorticity pattern behind the
top plate is a mirror of the one behind the middle plate. This is a consequence of a
symmetry axis in the geometry at y/d = 2.5. The disturbances in the boundary at the
start of the period will have an effect on the vortex shedding during the period. In the
measurements plots, at the top and bottom of the plate a small white region is present,
in which no measurements could be done due to laser reflections.
At θ = π/2 (B) the velocity is at its peak, so the vortices formed at this phase are
the strongest. When we focus on the center plate (y/d = 0) the positive vortex is elon-
gated and is almost split by the negative vortex. Once this positive vortex is split into
two parts, a vortex is shed off. Not only in the numerical simulation, but also in the
measurements, we see some vortices that have already been shed off and located now
around x/d = 5. The displacement amplitude, given by

1
[ = [u
1
[/ω, (6.18)
is in this case equal to figure 6.4d. This is the distance the gas has traveled from moment
A to moment B. For x/d <6.4 the vorticity is close to zero ([ω
z
[ < 0.2 10
4
s
−1
) in both
the measurement and the numerical simulation.
At moment (C), θ = 3π/4, the vortex streets have grown in length and in number
122 Flow visualization in and around a stack
measurement numerical simulation
x/d
y
/
d


−10 −5 0 5
−5
0
5
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
(A) θ = 0
x/d
y
/
d


−10 −5 0 5
−5
0
5
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
(B) θ = π/2
x/d
y
/
d


−10 −5 0 5
−5
0
5
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
(C) θ = 3π/4
Figure 6.15: The vorticity behind plates as function of time during the first half of a period.
In the left column the PIV measurements are shown and in the right column the numerical
simulations. The velocity amplitude in-between the stacks [u
1
[ = 5 m/s and the frequency is 125
Hz. The plates, represented by black rectangles, have a thickness of 1 mm and are 4 mm apart
from each other.
6.7 Vortex street evolution 123
measurement numerical simulation
x/d
y
/
d


−10 −5 0 5
−5
0
5
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
(D) θ = π
x/d
y
/
d


−10 −5 0 5
−5
0
5
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
(E) θ = 5π/4
x/d
y
/
d


−10 −5 0 5
−5
0
5
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
x/d
y
/
d


−10 −5 0 5
−6
−4
−2
0
2
4
6
8
ω
z
[
1
0
4
s

1
]
−1
−0.5
0
0.5
1
(F) θ = 3π/2
Figure 6.16: The vorticity behind plates as function of time during the second half of a period.
In the left column the PIV measurements are shown and in the right column the numerical
simulations. The velocity amplitude in-between the stacks [u
1
[ = 5 m/s and the frequency is 125
Hz. The plates, represented by black rectangles, have a thickness of 1 mm and are 4 mm apart
from each other.
124 Flow visualization in and around a stack
of vortices. Also notice that the vortices in numerical simulation have tilted a little
towards a more vertical orientation. In the measurement plot the vortices are still more
horizontally oriented.
The phase angle θ = π is shown in figure 6.16(D). At this phase the gas is at its
leftmost position. The vortex streets are at its maximum length and can therefore be
identified most clearly at this moment of the period. The length of the vortex streets
divided by d is (12 ± 1) in the numerical simulation and (11 ± 1) in the measurement.
The displacement in-between the plates 2[ζ
1
[/d equals 12.8. Since the jet out of the stack
spread out, the displacement value is smaller here.
At θ = 5π/8, in figure 6.16(E), the flow is into the stack. The vortices move with the
flow into the stack. The vortices that are away from the plates get a circular shape. The
strongest vortices are the ones that are going into the stack but are still connected with
the corners of the plates. We can recognize the vortices also in the measurement plot.
At θ = 3π/4, in figure 6.16(F), the vortices disappear into the boundary layers, cre-
ating disturbances in these boundary layers. These disturbances will still be present a
the start of the next period and will influence the vortex pattern of the next period.
6.8 Velocity profile in-between two plates 125
6.8 Velocity profile in-between two plates
6.8.1 Introduction
The velocity parallel to the plates is given by [9]
u(y, t) = Re
_
i
ωρ
m
(1 −
cosh[(1 + i)y/δ
ν
]
cosh[(1 + i)y
0

ν
]
)
dp
dx
e
iωt
_
, (6.19)
with y = 0 in the center of the channel, ρ
m
the mean density, 2y
0
the plate distance and
Re¦x¦ representing the real part of x. Eq. 6.19 can also be written as
u(y, t) = Re
_
C(1 −
cosh[(1 + i)y/δ
ν
]
cosh[(1 + i)y
0

ν
]
)e
iωt
_
, (6.20)
with C a complex constant.
At higher R
e
or higher Wo the flow becomes turbulent [9]. The transition occurs at
a critical Reynolds number based on the viscous penetration depth:
Re
δ

2
=
V

νω
≈ 400
[36].
6.8.2 A single measurement of the velocity profiles
A velocity profile is the velocity parallel to the plates, u, as a function of the distance to
the bottom plate y. The velocity profile between two plates, at a distance of 2 mm from
each other, is measured using PIV. First we will consider measurements at a frequency
of 55 Hz and at a drive ratio of 1.3%. The velocity is measured at 21 different phase
angles, with a time difference of 1 ms between them, covering a whole period (figure
6.17).
In the next section we will show a good method to compare the measurements with
the analytical solution.
6.8.3 Fitting procedure
Since the velocity is a function of two variables (position and time), it is difficult to find
a good way to plot it and compare it with the analytical solution (Eq. 6.19). When we
look at the velocity function for a fixed position, y = y
A
, it is a harmonic function of t
u(y
A
, t) = C
1
sin(ωt +θ
1
), (6.21)
with C
1
and θ
1
, the amplitude and phase angle respectively, as fitting parameters. Eq.
6.21 is only valid in an ideal case in which
A no streaming is present,
B no higher harmonics are present,
C the flow is laminar and
D no entrance effects are present.
126 Flow visualization in and around a stack
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
−10
−8
−6
−4
−2
0
2
4
6
8
10
y [mm]
v
e
l
o
c
i
t
y

u

[
m
/
s
]
Figure 6.17: The velocity profile at 21 different phase angles at a frequency of 55 Hz and a drive
ratio of 1.3%.
We can correct for the effects of streaming and the harmonic by fitting according to
u(y
A
, t) = u
0
+[u
1
[ sin(ωt +θ
1
) +[u
2
[ sin(2ωt +θ
2
), (6.22)
with u
0
, [u
1
[, θ
1
, [u
2
[ and θ
2
as fitting parameters. For every position y
A
we can fit the
measured data according to Eq. 6.22. This way we find the five fitting parameters as a
function of y.
In figure 6.18 the time dependence of the velocity at position y = y
0
is shown. The
red curve is a fit using Eq. 6.22.
Edge effects are avoided by measuring sufficiently far away from the plate ends
(more than 2 times the displacement amplitude away). Fluctuations in time due to tur-
bulence are reduced averaging over 50 measurements. If the presence of turbulence
effects the average velocity profile, we can observe this by comparing the fitting param-
eter C
1
with the amplitude of Eq. 6.20.
6.8.4 First harmonic velocity and the transition to turbulence
The measured fitting parameters [u
1
[(y) and θ
1
(y) are the amplitude and phase angle
acoustic velocity profile. We can compare the fitting parameters with the amplitude
and phase angle of the analytical solution (Eq. 6.20). In figure 6.19 the amplitudes
and in figure 6.20 the phase angle of the acoustic velocities are shown at three different
drive ratios. We can add an arbitrary offset constant to the phase angle. The phase
angle θ
1
is therefore defined as such that the θ
1
=0 for y = 0. The velocity amplitude
6.8 Velocity profile in-between two plates 127
0 5 10 15 20 25
−10
−8
−6
−4
−2
0
2
4
6
8
10
t [ms]
v
e
l
o
c
i
t
y

[
m
/
s
]
Figure 6.18: The velocity as a function of time at position y = y
0
. The solid curve is a fit.
[u
1
[(y) is normalized by its value at y = 0. The viscous penetration depth is 0.29 mm.
The analytical solution is very consistent with the measurements. The linear theory
−1 −0.5 0 0.5 1
0
0.2
0.4
0.6
0.8
1
y/y
0
|
u
1
(
y
)
|
/
|
u
1
(
0
)
|


experimental
analytical
(a) DR = 0.1%, [u
1
(0)[=0.66 m/s
−1 −0.5 0 0.5 1
0
0.2
0.4
0.6
0.8
1
y/y
0
|
u
1
(
y
)
|
/
|
u
1
(
0
)
|


experimental
analytical
(b) DR = 0.4%, [u
1
(0)[=3.45 m/s
−1 −0.5 0 0.5 1
0
0.2
0.4
0.6
0.8
1
y/y
0
|
u
1
(
y
)
|
/
|
u
1
(
0
)
|


experimental
analytical
(c) DR = 1.3%, [u
1
(0)[=12.4 m/s
Figure 6.19: The velocity amplitude [u
1
[ as function of the y-position in-between the plates.
DR is the drive ratio at the end of the resonator tube.
produces good results for these amplitudes. We can conclude that the flow is linear.
We are interested in what happens at higher amplitudes. Therefore we have performed
measurements at the resonance frequency (44 Hz). These measurements are shown in
figures 6.21 and 6.22. The viscous penetration depth is 0.33 mm.
At DR = 1.7% the flow is still laminar, as no significant deviations from the analytical
solution can be seen. At DR = 3.4% the measured phase angle starts to deviate signifi-
cantly from the analytical solution. At DR = 6.8% the measured phase angle is constant
over the cross section. In a laminar flow pattern the flow at the boundary layers is lag-
ging in correspondence to the main flow in the center of the plates. This is no longer
true at DR = 6.8%. The flow has become turbulent and has turned into a plug flow, with
128 Flow visualization in and around a stack
−1 −0.5 0 0.5 1
0
0.05
0.1
0.15
0.2
y/y0
|
θ
1
(
y
)
|
/
2
π


experimental
analytical
(a) DR = 0.1%
−1 −0.5 0 0.5 1
−0.05
0
0.05
0.1
0.15
y/y0
|
θ
1
(
y
)
|
/
2
π


experimental
analytical
(b) DR = 0.4%
−1 −0.5 0 0.5 1
−0.05
0
0.05
0.1
0.15
y/y0
|
θ
1
(
y
)
|
/
2
π


experimental
analytical
(c) DR = 1.3%
Figure 6.20: The phase angle θ
1
of the velocity as function of the y-position in-between the
plates. DR is the drive ratio at the end of the resonator tube.
−1 −0.5 0 0.5 1
0
0.5
1
1.5
y/y
0
|
u
1
(
y
)
|
/
|
u
1
(
0
)
|


experimental
analytical
(a) DR = 1.7%, [u
1
(0)[=10.7 m/s
−1 −0.5 0 0.5 1
0
0.2
0.4
0.6
0.8
1
y/y
0
|
u
1
(
y
)
|
/
|
u
1
(
0
)
|


experimental
analytical
(b) DR = 3.4%, [u
1
(0)[=21.3 m/s
−1 −0.5 0 0.5 1
0
0.5
1
1.5
y/y
0
|
u
1
(
y
)
|
/
|
u
1
(
0
)
|


experimental
analytical
(c) DR = 6.8%, [u
1
(0)[=42.7 m/s
Figure 6.21: The velocity amplitude [u
1
[ as function of the y-position in-between the plates.
DR is the drive ratio at the end of the resonator tube.
−1 −0.5 0 0.5 1
0
0.02
0.04
0.06
0.08
0.1
0.12
y/y0
|
θ
1
(
y
)
|
/
2
π


experimental
analytical
(a) DR = 1.7%
−1 −0.5 0 0.5 1
−0.05
0
0.05
0.1
0.15
y/y0
|
θ
1
(
y
)
|
/
2
π


experimental
analytical
(b) DR = 3.4%
−1 −0.5 0 0.5 1
0
0.05
0.1
0.15
0.2
y/y0
|
θ
1
(
y
)
|
/
2
π


experimental
analytical
(c) DR = 6.8%
Figure 6.22: The phase angle θ
1
of the velocity as function of the y-position in-between the
plates. DR is the drive ratio at the end of the resonator tube.
6.8 Velocity profile in-between two plates 129
a constant phase angle over the cross section. The amplitude dependency of the y posi-
tion only shows small deviations from the analytical solution. At DR = 3.4% the phase
angle profile is in-between that of a laminar and a turbulent flow. This is a transition
point. The Reynolds number based on the viscous boundary thickness is defined as
R

=
[u
1
(0)[δ
ν
ν
. (6.23)
For a drive ratio of 3.4% the Reynolds number R

= 468.
6.8.5 Velocity fluctuations
The velocity fluctuations are shown in figure 6.23. To determine the fluctuations in
the boundary layer the root-mean-square values (rms) are averaged over the region
0.88 < y < 0.97, which are indicated with squares. The fluctuations in the bulk flow,
indicated by diamonds, are averaged over the region 0 < y < 0.88. The rms values are
determined by:
[v
/
[ =
¸
1
n −1
n

k=1
[v
k
−v[
2
. (6.24)
0 5 10 15 20
0
0.02
0.04
0.06
0.08
0.1
t [ms]
|
R
M
S
(
v
)
|
/
|
u
1
|
(a) DR = 0.2%, [u
1
(0)[=2.06 m/s
0 5 10 15 20
0
0.02
0.04
0.06
0.08
0.1
t [ms]
|
R
M
S
(
v
)
|
/
|
u
1
|
(b) DR = 0.4%, [u
1
(0)[=3.45 m/s
0 5 10 15 20
0
0.02
0.04
0.06
0.08
0.1
t [ms]
|
R
M
S
(
v
)
|
/
|
u
1
|
(c) DR = 1.3%, [u
1
(0)[=12.4 m/s
0 5 10 15 20
0
0.02
0.04
0.06
0.08
0.1
t [ms]
|
R
M
S
(
v
)
|
/
|
u
1
|
(d) DR = 1.7%, [u
1
(0)[=10.7 m/s
0 5 10 15 20
0
0.02
0.04
0.06
0.08
0.1
t [ms]
|
R
M
S
(
v
)
|
/
|
u
1
|
(e) DR = 3.4%, [u
1
(0)[=21.3 m/s
0 5 10 15 20
0
0.02
0.04
0.06
0.08
0.1
t [ms]
|
R
M
S
(
v
)
|
/
|
u
1
|
(f) DR = 6.8%, [u
1
(0)[=42.7 m/s
Figure 6.23: The velocity fluctuations [v’[ as function of time. The squares represent the fluctu-
ations in the boundary layer and the diamonds in the bulk flow. The fluctuations are normalized
by the peak value of the velocity [u
1
[. DR is the drive ratio at the end of the resonator tube.
130 Flow visualization in and around a stack
6.8.6 Streaming velocity
From the time dependent fit we can also determine the DC-component of the velocity,
also called the streaming velocity. We define the streaming velocity as the time average
velocity,
u
0
=
1
τ
_
τ
0
u(t)dt, (6.25)
not to be confused with the mass streaming, which is given by
˙ m =
1
τ
_
τ
0
ρ(t)u(t)dt, (6.26)
The ratio of the streaming velocity u
0
(y) to the acoustic velocity in the center [u
1
(0)[ is
shown in figure 6.24 as function of the position in-between the plates.
0 0.5 1
−5
0
5
10
x 10
−3
y/y
0
u
0
(
y
)
/
|
u
1
(
0
)
|
(a) DR = 0.4%, [u
1
(0)[=3.45 m/s
0 0.5 1
−0.04
−0.02
0
0.02
0.04
y/y
0
u
0
(
y
)
/
|
u
1
(
0
)
|
(b) DR = 1.3%, [u
1
(0)[=12.4 m/s
Figure 6.24: The streaming velocity [u
1
[ as function of the y-position in-between the plates.
DR is the drive ratio at the end of the resonator tube. The streaming velocity is normalized by
the acoustic velocity in middle of the plates, [u
1
(0)[.
The fluctuations of temperature and density in the boundary layers cause a net
streaming velocity in the boundary layers. The mass streaming integrated over the tube
cross section should always be zero, since the system is closed. To compensate for the
net streaming in the boundary layers, a streaming velocity in the opposite direction is
generated in the bulk of the gas. The streaming in the bulk of the gas is expected to have
a parabolic shape.
6.9 Natural Convection
6.9.1 Scale analysis on a vertical wall
We want to do a scale analysis of the natural convection in order to compare the order
of magnitude of the convection velocity with the measurements and to approximate the
heat losses due to convection. One simplification we make is that we replace the stack
6.9 Natural Convection 131
x
Ta
Tw
y
0
0
δκ
δν
ρ
u=0
g
0
v
x
δν
0
T
x
δκ
Figure 6.25: A schematic view of natural convection near a solid vertical wall at temperature
T
w
, which is higher than the ambient temperature T
a
. Due to buoyancy the gas moves up and
both a viscous and a thermal boundary layer are formed. The boundary layer thicknesses, δ
ν
and
δ
κ
respectively, grow with the y-position along the wall. Schematic plots of the two boundary-
layer thicknesses as functions of x are also shown.
end, which consists of 1 mm thick plates, separated 1 mm from each other, with a solid
vertical wall. This wall is at temperature T
w
and has height H.
First we want to consider the situation of a finite vertical wall in an infinitely large
space at temperature T
w
higher than the ambient temperature T
a
, as is shown in fig-
ure 6.25. Both a viscous and a thermal boundary layer develop at the wall and their
thicknesses, δ
ν
and δ
κ
, grow with the vertical position y.
In steady state (∂ρ/∂t = 0) the mass conservation law yields
∂u
∂x
+
∂v
∂y
= 0, (6.27)
and the momentum conservation law in y direction yields
ρ
_
u
∂v
∂x
+ v
∂v
∂y
_
= −
∂p
∂y
−ρg +µ
_

2
v
∂x
2
+

2
v
∂y
2
_
, (6.28)
with g the acceleration due to gravity. The energy conservation law yields
u
∂T
∂x
+ v
∂T
∂y
= κ
_

2
T
∂x
2
+

2
T
∂y
2
_
. (6.29)
According to Bejan [37], the density variation due to temperature changes can be ap-
proximated by
ρ ≈ ρ

−βρ

(T −T
a
), (6.30)
where β is the volume expansivity (β = 1/T in ideal gases) and T
a
is the ambient
132 Flow visualization in and around a stack
temperature. Since the pressure distribution is hydrostatic
∂p
∂y
= −ρ

g. (6.31)
Substituting Eqs. 6.31 and 6.30 into Eq. 6.28 and dividing by ρ

results in
u
∂v
∂x
+ v
∂v
∂y
ρ

[1 −β(T −T
a
)] = βg(T −T
a



_

2
v
∂x
2
+

2
v
∂y
2
_
ρ

[1 −β(T −T
a
)].
(6.32)
Since T −T
a
is 40 K at maximum, β(T −T
a
) ∼ 10
−1
. We only want to make a rough
model of the natural convection and therefore neglect β(T −T
a
) in comparison to 1. The

2
v/∂x
2
term is of the order v/δ
2
κ
and ∂
2
v/∂y
2
∼ v/y
2
. Since y ¸ δ
κ
the ∂
2
v/∂y
2
term
can be neglected in comparison to ∂
2
v/∂x
2
. Eq. 6.32 simplifies to
u
∂v
∂x
+ v
∂v
∂y
= βg(T −T
a
) +ν

2
v
∂x
2
. (6.33)
The two terms at the left hand side are the inertia terms, and the first term at the right
hand side is the friction and the second term is the Buoyancy. The buoyancy is the
driving force of the natural convection and is caused by the decrease in density due to
the heating up of the gas by the hot wall.
To determine whether the inertia or the friction is balancing the buoyancy force, we
will now make an order approximation of the three different terms.
v
2
y
∼ βg(T −T
a
) +ν
v
δ
2
κ
. (6.34)
Applying a scale analysis to the mass and energy conservation:
u
δ
κ

v
y
, (6.35a)
v
∆T
y
∼κ
∆T
δ
2
κ
. (6.35b)
By using Eq. 6.35b we find for the friction term in Eq. 6.34 that νv/δ
2
κ
∼ P
r
v
2
/y.
Thus for P
r
¸ 1 the friction is dominating the inertia, whereas for P
r
¸ 1 the inertia is
dominating the friction.
When the buoyancy is balanced by friction (P
r
¸ 1) the momentum equation in y
direction gives
ν
v
δ
2
κ
∼ gβ∆T. (6.36)
We have three scale equations (Eqs. 6.36a,b and Eq. 6.34) and three unknowns (v, δ
κ
, u).
6.9 Natural Convection 133
This set of equations can be solved easily resulting in
v ∼
κ
y
R
1/2
a
, (6.37a)
δ
κ
∼ yR
−1/4
a
, (6.37b)
u ∼
κ
y
R
1/4
a
, (6.37c)
with the Rayleigh number
R
a
=
gβ∆Ty
3
κν
. (6.38)
When the buoyancy is balanced by inertia (P
r
¸ 1) by solving the scale equations
we find very similar Eqs. as in Eqs. 6.37, but now the R
a
number is replaced by the
Boussinesq number, B
o
:
v ∼
κ
y
B
1/2
o
, (6.39a)
δ
κ
∼ yB
−1/4
o
, (6.39b)
u ∼
κ
y
B
1/4
o
, (6.39c)
in which the Boussinesq number is given by
B
o
= R
a
P
r
. (6.40)
For P
r
≈ 1, as is the case in our set-up (P
r
= 0.72), both the inertia and the friction
need to be taken into account. Eqs. 6.39 and 6.37 converge and either of them can be
used.
The Nusselt number is defined as
N
u
=
αy
k
, (6.41)
where α is the heat transfer coefficient, defined as
α =
q
w
T
w
−T
a
=
−k(∂T/∂x)
x=0
T
w
−T
a
. (6.42)
By using the approximation −(∂T/∂x)
x=0
∼ (T
w
−T
a
)/δ
κ
, we find for the Nusselt num-
ber
N
u
∼ R
1/4
a
, for P
r
¸1, (6.43a)
N
u
∼ B
1/4
o
, for P
r
¸1. (6.43b)
Although these equations are based only on scale analysis they are in good agreement
134 Flow visualization in and around a stack
with detailed numerical calculations, which resulted in:
N
u
= 0.503R
1/4
a
, for P
r
¸1, (6.44a)
N
u
= 0.600B
1/4
o
, for P
r
¸1, (6.44b)
and for the whole range of P
r
numbers:
N
u
= 0.503
_
P
r
Pr + 0.986P
1/2
r
+ 0.492
_
1/4
R
1/4
a
. (6.45)
Using Eq. 6.39 for our set-up, with atmospheric air, ∆T = 40

C, and H = 60 mm, we
find that v ≈ 0.3 m/s, δ
κ
≈ 2 mm, and δ
ν
≈ 1.7 mm, at position y = H.
6.9.2 Measurements
When the hot end of the stack heats up, this induces natural convection. The flow
around the hot end of the stack is measured using PIV. We use thermoacoustics to heat
up the hot end of the stack. The loudspeaker is used as the source of acoustic power.
Once the hot end reaches 60

C, the loudspeaker is turned off. The reason for turning
off the loudspeaker is that the absence of acoustics makes it much easier to measure the
convective flow. An additional advantage is that we exclude other forms of streaming,
e.g. acoustic streaming. The temperature of the stack is registered by thermometers,
as is described in chapter 5. Due to heat losses to the environment and due to conduc-
tion, the hot-end temperature decreases over time. Every time the temperature drops
approximately 5

C we do a PIV measurement of the flow. In figure 6.26 the flow field
behind the stack is shown, for two different hot-end temperatures. The colors used in
the two plots are on the same scale, whereas the vector sizes in the plot are normalized
by the maximum velocity. The measurement plane is vertical and intersects with the
tube axis. The hot end of the stack is located 63 mm from the resonator end. The tube
diameter is 59.1 mm.
It can be seen in figure 6.26(a) that near the stack end (x/d = 0) a viscous boundary
develops, that grows with the y position. The velocity v
y
increases with the y position,
as expected, but for y/d ≥ 10 it decreases again. The reason for this is that the gas is
confined in a closed volume, so the gas has to move to the right, due to mass conser-
vation. A circulation within the stack due to streaming is also a possibility (to the left
at the upper side of the stack and to the right at the bottom side), but from the results
it seems that this effect is negligible. The viscous resistance within the stack is much
higher than outside of the stack.
The flow structure in figure 6.26(b) has changed. The boundary layers are much
thicker than in figure 6.26(a). This cannot be explained by the lower temperature dif-
ference T
w
− T
a
. A possible cause for this difference is that the flow in figure 6.26(a)
is not fully developed yet. As the measurements are done during one cooling down
process, the flow in figure 6.26(b) had more time to develop and is therefore close to
the stationary state. In a stationary state the buoyancy flow can be modeled by using
a closed volume, as is shown in figure 6.28. Figure 6.26(a) resembles the vertical wall
model more than figure 6.26(b).
6.10 Streaming 135
0 10 20 30 40
−30
−20
−10
0
10
20
30
x/d
y
/
d


u
[
m
/
s
]
0
0.01
0.02
0.03
0.04
0.05
(a) T = 60

C
0 10 20 30 40
−30
−20
−10
0
10
20
30
x/d
y
/
d


u
[
m
/
s
]
0
0.01
0.02
0.03
0.04
0.05
(b) T = 30

C
Figure 6.26: A velocity vector field of the convection flow behind the hot end of the stack at two
different stack-end temperatures T. The black rectangles represent the stack plates, which are 1
mm in thickness and 1 mm apart from each other. The stack end is located at 63 mm from the
end of the resonator.
In figure 6.27 a velocity profile of v
y
as function of x at height y/d = 10. The solid
plot is at temperature T
w
= 60

C and the dashed plot is at T
w
= 30

C. The maximum
velocities are v
m
= 0.046 m/s and v
m
= 0.014 m/s, respectively. The boundary layer
thickness is here defined as the position at which the velocity is at its maximum value.
We find that δ
ν
/d = 3 and δ
ν
/d = 5, respectively. For T
w
= 60

C the found δ
ν
/d of 3
is higher than the expected 1.7. As the scale analysis can only determine the order of
magnitude, this result is still quite good.
We can use the Nusselt-number approximation to approximate the heat losses due
to buoyancy. For T
w
= 60

Cand H = 0.06 m we find that Nu ≈ 11 and that q ≈17 W/m
2
.
As a comparison the heat losses to the environment due to radial conduction in the stack
are 10 times a high. Thus, in this set-up the effects of buoyancy on the energy balance of
the stack are relatively small. The heat flow due to buoyancy can increase significantly
when the flow becomes turbulent, which takes place at R
a
/P
r
≈ 10
9
, according to Bejan
and Lage [38]. Since the R
a
number grows with H
3
, this can be an important effect for
large devices.
6.10 Streaming
Streaming is a net velocity field that is obtained from averaging the velocity over a
whole period. When the oscillation is a sine wave the streaming velocity is zero. In
the presence of objects of wall the sine wave is disturbed locally, resulting in a local
streaming. The local streaming behind a sudden expansion is called jet streaming. It
is cause be a difference in flow pattern during the expansion phase, when the flow
behaves like a jet and vortices are formed, and the suction phase, when the flow goes
into the stack. The measured streaming velocity around a single plate with d = 1 mm,
136 Flow visualization in and around a stack
0 5 10 15
−0.01
0
0.01
0.02
0.03
0.04
0.05
x/d
v
y
δ
ν
v
m
Figure 6.27: The vertical velocity v
y
, due to buoyancy, as function of the distance from the stack
x measured at height y/d = 10. The solid plot is at temperature T
w
= 60

C and the dashed plot
is at T
w
= 30

C. The maximum velocities are v
m
= 0.046 m/s and v
m
= 0.014 m/s, respectively.
x
T
a
T
w
y
0
0
g
W
H
insulated
Figure 6.28: A schematic view of natural convection in a rectangular volume with height H and
width W. The left wall is at temperature T
w
, the right wall is at temperature T
a
(< T
w
), and the
top and bottom walls are adiabatic. Due to buoyancy the gas moves up at the hot wall on the left
and down at the cold wall on the right.
6.11 Discussion and conclusion 137
[u
1
[ outside the stack is 0.5 m/s, and f =100 Hz is shown in figure 6.29. Behind the
plate one vortex pair can be distinguished. Two other vortices are more difficult to
distinguish, one is located above and the other beneath the plate, with the two vortex
centers near x/d ≈ 1.
−3 −2 −1 0 1 2
−2.5
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
2.5
x/d
y
/
d


u
[
m
/
s
]
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
Figure 6.29: The measured velocity vector field of the streaming at the the end of a stack plate.
The black rectangle represent the stack plate, which is 1 mm in thickness. [u
1
[ outside the stack
is 0.5 m/s, f =100 Hz.
In figure 6.30 the measured streaming velocity of a parallel-plate stack is shown, in
which the plate ends are not perfectly parallel. Since the plate ends are not tight up, the
plates can bend a little at their ends. This result in asymmetries of the stack at its ends,
resulting in asymmetrical flow patterns. The flow channels that have a smaller outlet
show a stronger jet flow. As a consequence the channels with a small outlet have a
higher dissipation and are less preferred when the flow is directed out of the stack. This
difference in preference of the different channels leads to an internal convection, which
is found to be approximately 1% of the acoustic flow, for [u
1
[ is 5 m/s. This shows the
importance of a careful stack design. In all other measurements these asymmetries in
the geometry are avoided by keeping the plate ends at the same distance with the use
of a comb.
6.11 Discussion and conclusion
PIV has proven to be a useful measurement technique in the study of oscillating flow
around stacks of parallel plates.
It is shown that different vortex patterns (one vortex pair, two vortex pairs, vortex
street) can originate behind a parallel plate. The Reynolds and Strouhal numbers, and
their ratio in particular, are the relevant dimensionless numbers for determining which
vortex pattern occurs. The influence of porosity and plate-end shape are also studied.
The vorticity pattern behind sharp edged plate ends (25

or smaller) differs fromsquared
or circular plate ends: instead of a vortex pair or a vortex street, two thin layers of
vorticity are formed.
Next we have shown that the numerical simulations, which are in good agreement
138 Flow visualization in and around a stack
−2 −1 0 1 2 3 4
−12
−11
−10
−9
−8
−7
−6
−5
x/d
y
/
d


u
[
m
/
s
]
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Figure 6.30: The measure streaming velocity at the end of a stack for which the plate ends are
not perfectly parallel. The black rectangles represent the ends of the stack plates, which are 1 mm
in thickness. [u
1
[ outside the stack is 1 m/s, f = 50 Hz.
with each other, are in good agreement with the PIV measurement. Not only does this
verify the numerical simulations but also the PIV measurement principle. We have
shown that the vortex pattern behind the plates show huge differences when compar-
ing the vorticity plot at the same phase angle for different periods. These fluctuations
are also be found in the numerical calculations, even without including any turbulence
modeling. These fluctuations are not a start-up effect, as after calculating for ten periods,
the fluctuations persist. Apparently the vorticity pattern is sensitive to small changes
in the velocity field at the start of the cycle, which is typical for nonlinear effects. This
shows that the vorticity pattern is strongly determined by its prehistory.
Furthermore we have measured the velocity in-between two plates in the middle of
the stack, where no entrance effects takes place, as functions of the position in-between
the plates y and the phase angle. By fitting the velocity as function of the phase angle
at a single position to a sine wave, we can determine the amplitude and phase shift. We
have studied this amplitude and phase angle as function of the position y and compared
them with an analytical solution. For sufficiently low Reynolds numbers R

200 the
measurements were very consistent with the analytical solution. For R

≈ 500 the
phase shift differs from the analytical solution. And at R

≈ 1000 the measured phase
shift is constant over the cross section. This is caused by turbulence. To study the
turbulence in more detail, we have studied the instantaneous velocity profiles instead
of averages.
The types of streaming that can be present in a standing-wave device are jet stream-
ing, internal streaming, natural convection, and Rayleigh streaming. Of these four
streaming types, we have shown the first three types in PIV measurements.
Bibliography
[1] S. L. Garrett, “Resource letter: Ta-1: Thermoacoustic engines and refrigerators,”
American Journal Of Physics, vol. 72, pp. 11–17, Jan. 2004.
[2] N. Rott, “Damped and thermally driven acoustic oscillations in wide and narrow
tubes,” Journal of Applied Mathematics and Physics, vol. 20, pp. 230–243, 1969.
[3] N. Rott, “Thermally driven acoustic oscillations. part II: Stability limit for helium,”
Journal of Applied Mathematics and Physics, vol. 24, pp. 54–72, 1973.
[4] N. Rott, “Thermally driven acoustic oscillations. part III: Second-order heat flux,”
Journal of Applied Mathematics and Physics, vol. 26, pp. 43–49, 1975.
[5] N. Rott and G. Zouzoulas, “Thermally driven acoustic oscillations. part IV: Tubes
with variable cross-section,” Journal of Applied Mathematics and Physics, vol. 27,
pp. 197–224, 1976.
[6] N. Rott, “Thermoacoustics,” Advances in Applied Mechanics, vol. 20, pp. 135–175,
1980.
[7] N. Rott, “The influence of heat conduction on acoustic streaming,” Journal of Ap-
plied Mathematics and Physics, vol. 25, pp. 417–421, 1974.
[8] G. Swift, “Thermoacoustic engines,” Journal of Acoustical Society of America, vol. 84,
pp. 1146–1180, 1988.
[9] G. Swift, Thermoacoustics: A Unifying Perspective for Some Engines and Refrigerators.
Acoustical Society of America, 2002.
[10] G. Swift, “Thermoacoustic engines and refrigerators,” Physics Today, pp. 22–28,
1995.
[11] P. Kundu and I. Cohen, Fluid Mechanics. Academic Press, San Diego, CA., 3th
edition ed., 2004.
[12] D. Nehari, V. Armenio, and F. Ballio, “Three-dimensional analysis of the unidirec-
tional oscillatory flow around a circular cylinder at low keulegan-carpenter and
beta numbers,” Journal of Fluid Mechanics, vol. 520, pp. 157–186, Dec. 2004.
[13] P. Anagnostopoulos and R. Minear, “Blockage effect of oscillatory flow past a fixed
cylinder,” Applied Ocean Research, vol. 26, pp. 147–153, May 2004.
140 BIBLIOGRAPHY
[14] M. F. Wybrow, B. Yan, and N. Riley, “Oscillatory flow over a circular cylinder close
to a plane boundary,” Fluid Dynamics Research, vol. 18, pp. 269–288, Oct. 1996.
[15] M. Ozgoren, “Flow structure in the downstream of square and circular cylinders,”
Flow Measurement And Instrumentation, vol. 17, pp. 225–235, Aug. 2006.
[16] M. Ibrahim and W. Hashim, “Oscillating flow in channels with a sudden change in
cross-section,” Computers & Fluids, vol. 23, pp. 211–224, Jan. 1994.
[17] P. Morris, S. Boluriaan, and C. Shieh, “Numerical simulation of minor losses due
to a sudden contraction and expansion in high amplitude acoustic resonators,”
Journal of the Acoustical Society of America, vol. 90, pp. 393–409, 2004.
[18] B. L. Smith, “Pressure recovery in a radiused sudden expansion,” Experiments in
Fluids, vol. 36(6), pp. 901–7, 2004.
[19] B. L. Smith and G. W. Swift, “Power dissipation and time-averaged pressure in
oscillating flow through a sudden area change,” Journal of the Acoustical Society of
America, vol. 113(5), pp. 2455–63, 2003.
[20] D. Skulina, A study of non-linear flows at the open end of a tube using particle imaging
velocimetry. PhD thesis, University of Edinburgh, 2005.
[21] D. Skulina, “A study of non-linear acoustic flows at the open end of a tube using
particle image velocimetry,” Journal of the Acoustical Society of America, vol. 120(5),
p. 3363, 2006.
[22] B. L. Smith and C. V. King, “Time-resolved piv and pressure measurements of os-
cillating and pulsating flow in a rapid expansion,” in 2007 5th Joint AMSE/JSME
Fluids Engineering Conference, 2007.
[23] B. L. Smith, K. V. Mortensen, and S. Wendel, “Oscillating flow in adverse pressure
gradients,” in 2005 AMSE Fluids Engineering Summer Conference, 2005.
[24] A. Petculescu and L. A. Wilen, “Oscillatory flowin jet pumps: nonlinear effects and
minor losses,” Journal of the Acoustical Society of America, vol. 113(3), pp. 1282–92,
2003.
[25] J. Disselhorst and L. Van Wijngaarden, “Flow in the exit of open pipes during
acoustic resonance,” Journal of Fluid Mechanics, vol. 99, no. pt 2, pp. 293–319, 1980.
[26] J. Disselhorst and L. v. Wijngaarden, “Oscillatory flow around the edge of a flat
plate,” Journal of Engineering Mathematics, vol. 13, no. 3, pp. 271–287, 1979.
[27] P. Stoltenkamp, S. Araujo, H. Riezebos, J. Mulder, and A. Hirschberg, “Spurious
counts in gas volume flow measurements by means of turbine meters,” Journal of
Fluids and Structures, vol. 18, no. 6, pp. 771–781, 2003.
[28] P. Stoltenkamp, Dynamics of turbine flow meters. PhD thesis, Eindhoven University
of Technology, 2007.
BIBLIOGRAPHY 141
[29] Ph. Blanc-Benon, E. Besnoin, and O. Knio, “Experimental and computational vi-
sualization of the flow field in a thermoacoustic stack,” Comptes Rendus M´ ecanique,
vol. 331, pp. 17–24, Jan. 2003.
[30] A. Berson, Ph. Blanc-Benon, and M. Michard, “Ecoulements secondaires aux
extr´ emit´ es du stack d’un r´ efrig´ erateur thermoacoustique : mesure des champs de
vitesse oscillante ` a l’aide de la piv,” in 8` eme Congr` es Fran¸ cais d’Acoustique, pp. 749–
752, 2006.
[31] A. Berson, M. Michard, and Ph. Blanc-Benon, “Measurement of acoustic velocity in
the stack of a thermoacoustic refrigerator using particle image velocimetry,” Heat
and Mass Transfer, vol. 44, pp. 1015–1023, 2008.
[32] A. Berson and Ph. Blanc-Benon, “Nonperiodicity of the flow within the gap of a
thermoacoustic couple at high amplitudes,” Journal of the Acoustical Society of Amer-
ica, vol. 122(4), pp. EL122–127, 2007.
[33] X. Mao, D. Marx, and A. J. Jaworski, “Piv measurement of coherent structures and
turbulence created by an oscillating flow at the end of a thermoacoustic stack,” in
Progress in Turbulence II Proceedings iTi Conference in Turbulence (M. Oberlack, ed.),
vol. 109, pp. 99–102, 2007.
[34] E. Besnoin and O. M. Knio, “Numerical study of thermoacoustic heat exchangers,”
Acta Acustica united with Acustica, vol. 90, pp. 432–444, May 2004.
[35] M. Raffel, C. Willert, and J. Kompenhans, Particle Image Velocimetry. A Practical
Guide. Springer, 1998.
[36] P. Merkli and H. Thomann, “Transition to turbulence in oscillating pipe-flow,” Jour-
nal of Fluid Mechanics, vol. 68, pp. 567–575, 1975.
[37] A. Bejan, Heat Transfer. John Wiley & Sons, 1993.
[38] A. Bejan and J. Lage, “Prandtl number effect on the transition in natural convection
along a vertical surface,” Journal of Heat Transfer, vol. 112, no. 3, pp. 787–790, 1990.
142 BIBLIOGRAPHY
Chapter 7
Flow measurements in co-axial
regenerator-based devices
7.1 Introduction
An important type of regenerator-based engines is depicted in Fig. 7.1. It consists of a
coaxial loop, that is designed by Aster Thermoacoustics and ECN. The advantage of the
coaxial design in comparison to the loop design (figure 2.11) is that it is compact and
easy to manufacture. This coaxial loop is created by inserting an open cylinder into the
resonator tube. Inside of the cylinder a regenerator and thermal buffer zone (TBZ) are
located. In-between the cylinder and the resonator wall the inertance is located and to
the right of the cylinder the compliance.
It is valuable to know how an oscillatory flow behaves in such a coaxial loop. Espe-
cially the flow at the sharp corners, at both sides of the inertance, are of interest. In front
of the TBZ one or more flow straighteners can be placed. It is known that the presence
of flow straighteners improves the performance. We expect that the flow straighteners
reduce the turbulence in the TBZ.
7.2 Experimental set-up
The experimental set-up is shown in figure 7.1. The concentric coaxial regenerator-
based loop is enclosed in a transparent cylindrical tube (c), which is closed on the right
side and is connected to the subwoofer set-up (which is described in section 3.3) on
the left side. A transparent cylindrical open tube (d) separates the intertance (b), also
called bypass, from the thermal buffer zone (TBZ) (e). At the right end of the TBZ a
stacked-screen regenerator (f), is located. The volume on the right of the regenerator
is the compliance volume (g). In front of the TBZ one or more wire-gauze screens are
installed which act as straighteners. The set-up is designed to work as an regenerator-
based thermoacoustic couple at a frequency of 50 Hz. The system is filled with air at
atmospheric pressure and drive ratios up to 15 percent can be reached. The dimensions
of the various components are shown in table 7.1.
From the top of the set-up a laser sheet is projected into the set-up, in the plane of
144 Flow measurements in co-axial regenerator-based devices
0
D
i
D
o
Dr Dt
x1 x2 x3x4
x5
b
e f g
d
a
c
Figure 7.1: A schematic drawing of the experimental set-up of a regenerator-based loop that can
be used for flow visualization. (a) one or more flow straighteners, (b) intertance, (c) outer tube
wall, (d) cylindrical loop wall, (e) thermal buffer zone (TBZ), (f) regenerator, (g) compliance.
Table 7.1: The dimensions of the geometry in mm.
x
1
8
x
2
104
x
3
112
x
4
128
x
5
220
D
r
40
D
t
50
D
i
60
D
o
70
the picture, as described in section 4.2. The camera, which is directed perpendicular
to the plane of the picture, is focussed on the light sheet. Three different interrogation
windows are used (figure 7.2). To reduce laser reflections, large parts of the outside and
inside cylindrical tubes are covered with fluorescent paint.
(2) (1)
(3)
Figure 7.2: A schematic drawing of the three interrogation windows (1)-(3) in the experimental
set-up.
As the tube is cylindrical it is difficult to visualize the flowusing PIV. The tube acts as
7.3 Results 145
a lens in vertical direction and therefore changes the focal point, whereas in the horizon-
tal direction the focal point remains undisturbed. As a consequence, it is not possible to
have both the horizontal and the vertical direction in focus. A small point in the tube
is projected as a vertical line or a horizontal line on the CCD of the camera (depending
on which direction is in focus). When performing measurements in the TBZ this effect
is even stronger, as both the tube wall and the loop wall act as lenses. Measurements in
the intertance are also difficult because they are located very close to the wall.
It is very difficult to measure the Gedeon streaming in this geometry using PIV. This
is due to the fact that the Gedeon streaming velocity is small in relation to the veloc-
ity amplitude and that it is not possible to do measurements close to the wall due to
reflections.
7.3 Results
First we will show instantaneous PIV results through window 1. One cycle is divided
into 20 equidistant phase angles, at which the measurements are done. In section 7.3.1
the results at four different phase angles are shown. The streaming velocity is deter-
mined by averaging the velocity over a complete cycle. We approximate the streaming
velocity by averaging over the 20 phase angles. In sections 7.3.2 and 7.3.3 the flow
around the sharp edges is studied. These are indicated as interrogation windows 2 and
3. In these sections we will restrict ourselves to show only the streaming results.
7.3.1 Instantaneous measurements
In figures 7.3 the instantaneous velocity fields through window 1 are shown. One cycle
is divided into 20 equidistant phase angles, at which the velocity fields are measured,
but here only four of them are shown.
At ωt = 0 the flowin the resonator is directed towards the loop. Close to the loop the
velocity at the top and bottom is higher than in the middle. The gas has a preference to
flow into the inertance, due to a lower flow resistance than in the TBZ. The regenerator
in the TBZ has a high flow resistance.
At ωt = π/2 the flow velocity in the resonator is approximately zero. Because the
acoustic velocities are small, disturbances have a larger effect. At the bottom of the
figure the flow is directed to the right. This is a consequence of a background flow, as
will be shown in the streaming results. At this moment there is a small flow out of the
TBZ and into the inertance. The velocity in the TBZ is ahead in phase to the flow in the
resonator and the flow in the inertance is behind in phase.
At ωt = π the flow is directed out of the loop. Analogous to ωt = 0 the flow out of
the inertance is higher than out of the TBZ.
At ωt = 3π/2 the flow velocity in the resonator is approximately zero again. At the
top right and the bottom right of the figure, two vortices originate. The phase difference
between the velocity in the inertance and TBZ can not explain these vortices. The phase
difference has only a small effect, as was shown at ωt = π/2. The two vortices are
caused by a jet flow out of the inertance.
By averaging all 20 phase angles, we determine the time-average velocity, called
streaming velocity. In figures 7.4 the streaming velocity through window 1 is shown at
146 Flow measurements in co-axial regenerator-based devices
−60 −40 −20 0
−20
−10
0
10
20
0.2
0.4
0.6
0.8
1
1.2
1.4
(a) ωt = 0
−60 −40 −20 0
−20
−10
0
10
20
0.1
0.2
0.3
0.4
0.5
(b) ωt = π/2
−60 −40 −20 0
−20
−10
0
10
20
0.2
0.4
0.6
0.8
1
1.2
1.4
(c) ωt = π
−60 −40 −20 0
−20
−10
0
10
20
0.2
0.4
0.6
0.8
1
1.2
(d) ωt = 3π/2
Figure 7.3: The instantaneous velocity field in window 1 at different moments during the cycle,
at D
r
= 5.23%. The positions at the axes are in mm. The color and the vectors lengths are
measures of the velocity. The velocities to the right of the color bars are in m/s.
7.3 Results 147
two different D
r
.
−60 −40 −20 0
−20
−10
0
10
20
0.1
0.2
0.3
0.4
0.5
0.6
(a) D
r
= 5.23%
−60 −40 −20 0
−20
−10
0
10
20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
(b) D
r
= 10.5%
Figure 7.4: Vector plots of the streaming velocity field in window 1. The positions at the axes
are in mm. The color and the vectors lengths are measures of the velocity. The velocities to the
right of the color bars are in m/s.
At D
r
= 5.23% the streaming velocity is a combination of jet streaming (the two vor-
tices) and a background flow. The background flow is probably a remaining flow from
the oil particles that were injected before the measurement. This could be tested by
repeating the experiment with a longer pause between the particle injection and the
measurement. At D
r
= 10.5% the jet streaming dominates the background flow.
7.3.2 Streaming in window 2
Window 2 is located at the T-junction of the device. Here the inertance, TBZ and res-
onator tube are all connected. At a D
r
of 2.60% (figure 7.5a) a vortex originates in the
streaming velocity field at the sharp angle at the start of the loop (indicated with the
dashed ellipse). This vortex is caused by jet streaming. The velocity in the bypass is
much higher than in the tube, due to the high resistance of the regenerator. When the
flow is leftwards, a jet flow is going from the bypass into the tube, which has a much
larger cross section, causing a vortex.
When the D
r
is increased (figures 7.5 b and c), the vortex increases in strength and
dimensions. Also the center of the vortex shifts. In the D
r
range from 2.60% to 5.23% it
mainly shifts leftwards. In the range from 5.23% to 10.5% it mainly shifts upwards.
From the streaming plots in figure 7.5 it seems that the measured velocity vectors do
not obey the law of mass conservation. This can be explained by 3-D effects. As we only
measure the velocity in one plane, we do not know the velocity in axial direction.
Another effect that can be relevant is that the measurements in this particular
traveling-wave set-up have a higher uncertainty, due to the cylindrical wall of the res-
onator, as was discussed in section 7.2. When summing over 20 measurements that
compensate each other, resulting in a small quantity, the uncertainty in the individual
measurements become more clearly expressed.
148 Flow measurements in co-axial regenerator-based devices
Now that we have shown the presence of the jet effect, we want to discuss its conse-
quences:
A Due to the jet effect the resistance at the inertance end is asymmetric, resulting in
Gedeon streaming.
B Jet streaming, which can enhance the heat losses to the environment.
C The extra resistance at the edges of the inertance increases the dissipation.
Gedeon streaming is a net time-average mass flow through the regenerator [1] that
can be present in thermoacoustic devices containing a loop, as is the case in traveling-
wave devices, including the set-up we are studying. The time-average mass flow
˙
M
causes an unwanted enthalpy flow,
˙
H =
˙
Mc
p
∆T trough the regenerator from hot to
cold. This effect can have a severe negative effect on the efficiency of a device. This
phenomenon is well-known in the literature about double-inlet pulse-tube refrigerators
[2]. Gedeon streaming is caused by second order effects in the regenerator and can be
enhanced or reduced by an asymmetric resistance at another location within the loop.
Swift et al. [3] used a jet pump to reduce streaming. This is similar to the jet-pumping
effect that is present at the inertance entrance in this set-up. The resistance of the flow
into the inertance differs from resistance of the flow out of the inertance. According to
Swift [4] the time-average pressure drop is
∆p = −
ρ
0
[U
1
[
2
(K
out
−K
in
)
8A
2
In
, (7.1)
where U
1
is the volume-flow rate in the inertance, A
In
= π/4(D
2
i
− D
2
r
) is the cross-
sectional area, and K
in
and K
out
are the minor-loss coefficients of the flow into and out
of the inertance respectively. The minor loss coefficient is defined by Eq. 4.42 for steady
flows. This can reduce the Gedeon streaming. To completely eliminate the Gedeon
streaming, in many regenerator-based devices an elastic membrane is installed in the
loop, that is transparent for acoustic waves, but not for time-average mass flows.
The vortices we have seen in figures 7.4 are a consequence of jet streaming. Because
this streaming is local, it does not contribute to Gedeon streaming. This vortex can
enhance the heat flow from the TBZ wall to the resonator wall, due to forced convection.
Because the location of the streaming vortex is relatively far away from the regenerator,
this effect is relatively small.
The dissipation due to a sudden change in cross-section has been studied in section
4.4. The energy loss is given by Eq. 4.52

˙
W =
2


0
A
2
In
[U
1
[
3
. (7.2)
Because the inertance size D
i
− R
r
is much smaller than the viscous penetration depth,
the minor loss coefficient can be approximated by Eq. 4.45
K = (1 −ψ)
2
, (7.3)
7.4 Discussion and conclusion 149
with ψ the ratio of cross sections:
ψ =
A
In
A
Re
= 1 −
D
2
r
D
2
i
, (7.4)
where A
Re
= πD
2
i
/4 is the cross-sectional area of the resonator. For this approximation
of ψ the flow to the TBZ is neglected, thus this approximation of ψ is a little too low, but
it is good enough to make a rough approximation of the minor losses.
At D
r
= 5.23% and f = 50 Hz in the inertance [u
1
[ = 4 m/s. From Eq. 7.2 it
follows that ∆
˙
W = 8 mW and for D
r
= 10.5 this is 66 mW. As a comparison, the viscous
dissipation at the inertance wall is 10 mW for D
r
= 5.23% and 40 mW for D
r
= 10.5%.
7.3.3 Streaming in window 3
In figure 7.6 the streaming in window 3 is shown. In figure 7.6a the vortex, due to jet
streaming, is difficult to distinguish. In figures 7.6b and 7.6c the jet streaming vortex is
clearly visible. The vortices at this location are much stronger than in front of the loop,
whereas the velocity amplitude is smaller. The reason for this is that at the end of the
bypass the change in cross section is very abrupt. In front of the loop the change in cross
section is much more gradually. Similar vortices were calculated by a CFD simulation
by Lycklama ` a Nijeholt et. al. [5].
7.4 Discussion and conclusion
We have measured the flow at both ends of the inertance. Due to the jet effect vortices
originate at these locations. The major consequences of the jet effect on the performance
of a device are the dissipation due to minor losses, a locally enhanced heat transfer
due to forced convection caused by jet streaming, and Gedeon streaming due to an
asymmetry in flow resistance for the in and outgoing flow at the inertance entrance.
Unfortunately, in this set-up, it was not possible to measure the Gedeon stream-
ing. First of all the Gedeon-streaming velocity is small in comparison to the oscillatory
flow. Secondly, the thick cylindrical walls make it very difficult to performPIVmeasure-
ments. Since the wall acts as a lens in vertical direction and not in horizontal direction,
the vertical and horizontal directions have different focus points. A recommendation
for future measurements is to use thinner walls or even better is to install plane mea-
surement windows into the wall. It is also interesting to study the influence of different
shapes of the inertance entrances on the flow.
150 Flow measurements in co-axial regenerator-based devices
−10 −8 −6 −4 −2 0
0
2
4
6
8
10
0.05
0.1
0.15
0.2
0.25
0.3
0.35
(a) D
r
= 2.60%
−10 −8 −6 −4 −2 0
0
2
4
6
8
10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
(b) D
r
= 5.23%
−10 −8 −6 −4 −2 0
0
2
4
6
8
10
0.5
1
1.5
2
2.5
(c) D
r
= 10.5%
Figure 7.5: Vector plot of the streaming velocity field in window 2. The positions at the axes are
in mm. The color and the vectors lengths are measures of the velocity. The velocities to the right
of the color bars are in m/s.
−4 −2 0 2 4 6
0
2
4
6
8
10
0.1
0.2
0.3
0.4
0.5
0.6
(a) D
r
= 2.60%
−4 −2 0 2 4 6
0
2
4
6
8
10
0.5
1
1.5
(b) D
r
= 5.23%
−4 −2 0 2 4 6
0
2
4
6
8
10
0.5
1
1.5
2
2.5
3
3.5
(c) D
r
= 10.5%
Figure 7.6: Vector plot of the streaming velocity field behind the bypass.
Bibliography
[1] D. Gedeon, “Dc gas flows in stirling and pulse-tube cryocoolers,” in Cryocoolers 9
(R. G. Ross, ed.), pp. 385–392, 1997.
[2] S. Zhu, P. Wu, and Z. Chen, “Double inlet pulse tube refrigerators: An important
improvement,” Cryogenics, vol. 30, pp. 514–520, 1990.
[3] S. Backhaus and G. W. Swift, “A thermoacoustic-stirling heat engine: Detailed
study,” Journal of the Acoustical Society of America, vol. 107, pp. 3148–3166, 2000.
[4] G. Swift, A Unifying Perspective for Some Engines and Refrigerators. Melville: Acousti-
cal Society of America, 2002.
[5] J. A. Lycklama ` a Nijeholt, M. E. H. Tijani, and S. Spoelstra, “Simulation of a
traveling-wave thermoacoustic engine using computational fluid dynamics,” Jour-
nal of the Acoustical Society of America, vol. 118, pp. 2265–2270, 2005.
Summary
Thermoacoustics concerns phenomena in which an interaction of acoustics with ther-
modynamics takes place. In an acoustic wave the gas parcels always undergo temper-
ature variations, which is a consequence of their compression and expansion. In adia-
batic sound waves these temperature variations go unnoticed. However, when a solid is
present near the acoustic wave, the wave interacts with the solid and can cause a trans-
fer of heat from one location in the solid to another. This is called the thermoacoustic
heat-pumping effect. This effect is the driving mechanism in stack-based coolers and
heat pumps. Vice versa, when a sound wave interacts with a solid with a temperature
gradient above a certain critical value, the temperature gradient enhances the sound
wave.
One of the major disadvantages of thermoacoustics is that the power density of
acoustic waves is relatively low. One way to increase the power density is to use higher
drive ratios. Unfortunately the linear theory of thermoacoustics is only valid at low am-
plitudes (drive ratios up to 3%). Nonlinear effects that are not taken into account by the
linear theory include vortex shedding at the end of a stack, dissipation at the ends of
a stack due to the sudden change in cross section, transition to turbulence in-between
plates, and streaming. The objective of this PhD work is to gain a better understanding
of different phenomena that are occurring in thermoacoustic devices with an emphasize
on nonlinear effect, that occur at high amplitudes.
Thermoacoustics is a complicated but also very interesting phenomena, since three
fields of research, i.e. acoustics, flow dynamics, and thermodynamics all come together
and all of them bring different key quantities, i.e. pressure, velocity, and temperature.
To gain a better understanding of different phenomena that are occurring in thermoa-
coustic devices we have measured, calculated, and studied all three quantities.
In order to measure the pressure, three microphones were installed in the resonator
on each side of the stack. Following the multi-microphone method, we determined
the acoustic-energy flows at both sides of the stack. The found difference between the
acoustic-energy flows at both sides of the stack equals the energy absorbed by the stack.
We have been able to show that at low amplitudes the absorbed energy is in good agree-
ment with the linear theory of thermoacoustics. Unfortunately at high drive ratios the
linear theory underestimates the energy losses in the stack, which is mainly caused by
minor losses at the two stack ends. Furthermore, the multi-microphone method is used
to determine the transfer matrix of a stack. These transfer-matrix elements are employed
to determine the Rott functions as functions of the frequency, which we found to be in
good agreement with analytical fits.
A flow visualization technique called PIV is used to measure the velocity. By de-
termining the displacement of very small oil droplets (typically 1 µm in diameter) that
move along with the gas during a short time interval (typically 1 µs), a velocity vector
field is determined. Even using a measurement window as small as 3 mm 2 mm, a
velocity vector field of 100 75 vectors can be obtained, during a time interval as small
as 0.1 µs. Moreover, the powerful PIV technique is used to study the vortex shedding
behind the plates of a parallel-plate stack, the velocity profile in-between two parallel
plates, and a time-average velocity, called streaming velocity. The velocity profiles in-
between two plates showa change in phase dependency when above a critical Reynolds
number, indicating a transition from laminar to turbulent flow.
In order to register the stack temperature profile as a function of time, we installed
32 thermometers in a stack plate. Furthermore we made a model of the energy balance
of the stack. The calculated temperature profiles as functions of time were found to be
in good agreement with the temperature measurements.
We have used different measurement techniques, microphones, thermometers, and
PIV, to get a more complete view of different thermoacoustic phenomena. We have
studied various nonlinear effects: minor losses at the stack ends, streaming, and a tran-
sition to turbulence in a parallel-plate stack. Also we have made a complete model
of linear thermoacoustics, based on the established linear theory, that can predict the
temperature profile in a stack, even dynamically. We have gained more insight in ther-
moacoustics and nonlinear effects in particular and we hope that other researchers and
the industry will benefit from this.
Samenvatting
Thermoakoestiek betreft fenomenen waarin een interactie van akoestiek met thermo-
dynamica plaatsvindt. In een akoestische golf ondergaan de gaspakketten altijd tem-
peratuurvariaties, ten gevolge van compressie en expansie. In adiabatische geluids-
golven blijven deze temperatuurvariaties in het algemeen onopgemerkt. Wanneer de
geluidsgolf echter in thermisch contact komt met een vast lichaam, vindt er een inter-
actie plaats die ertoe kan leiden dat warmte wordt verplaatst van de ene locatie naar
een andere. Dit wordt het thermoakoestisch warmtepompeffect genoemd. Op dit ef-
fect is het mechanisme in staandegolfkoelers en -warmtepompen gebaseerd. Vice versa,
wanneer een geluidsgolf in wisselwerking staat met een vast lichaam, waarvan de tem-
peratuursgradi¨ ent een kritieke waarde overschreidt, versterkt de temperatuurgradi¨ ent
de geluidsgolf.
E´ en van de belangrijkste nadelen van thermoakoestiek is dat de energiedichtheid
van de akoestische golven vrij laag is. De energiedichtheid kan vergroot worden door
hogere drukamplituden (drive ratio’s) te gebruiken. Helaas is de lineaire theorie van
thermoakoestiek slechts geldig bij relatief lage drukamplituden (drive ratio’s tot 3%).
Niet-lineaire effecten die niet worden meegenomen in deze lineaire theorie zijn o.a.
wervelafschudding aan het uiteinde van stackplaten en de hierdoor veroorzaakte dissi-
patie aan het uiteinde van de stack, een overgang naar turbulentie tussen de stackplaten,
en een DC-component in de snelheid, beter bekend als streaming. De doelstelling van
dit promotieonderzoek is het verkrijgen van beter inzicht in de verschillende fenome-
nen - in thermoakoestische machines, waarbij de nadruk ligt op niet-lineaire effecten -
die optreden bij relatief hoge amplituden.
Thermoakoestiek omvat gecompliceerde maar ook zeer interessante fenomenen, om-
dat het zich afspeelt daar waar de drie vakgebieden akoestiek, stromingsleer en thermo-
dynamica elkaar overlappen. Deze drie vakgebieden brengen elk hun eigen grootheden
met zich mee, te weten druk, snelheid en temperatuur. Om beter inzicht te krijgen in
de verschillende fenomenen die voorkomen in thermokoestische apparaten, zijn al deze
drie grootheden gemeten, berekend en bestudeerd.
Voor de drukmetingen zijn zes microfoons in de resonator geplaatst, aan ieder stack-
uiteinde drie. De akoestische energiestroom aan beide stackuiteinden is bepaald met
de multimicrofoonmethode. Het gemeten verschil tussen de twee energiestromen aan
beide uiteinden is gelijk aan het vermogen geabsorbeerd door de stack. We hebben
laten zien dat voor lage drukamplituden het geabsorbeerde vermogen consistent is met
de lineaire theorie van thermoakoestiek. Voor relatief hoge drukamplituden onderschat
deze theorie het geabsorbeerde vermogen, hetgeen voornamelijk te wijten is aan minor
losses aan de stackuiteinden. Daarnaast is deze multimicrofoonmethode gebruikt voor
het bepalen van de overdrachtsmatrix van een stack. De overdrachtsmatrixelementen
zijn gebruikt voor het bepalen van de Rott-functies als functie van de frequentie, welke
consistent met analytische vergelijkingen bleken te zijn.
Voor het bepalen van snelheid is de stromingsvisualisatietechniek PIV gebruikt. Het
bepalen van de verplaatsing van kleine oliedruppeltjes (typisch 1 µm in diameter), die
meebewegen met het gas, tijdens een kort tijdinterval (typisch 1 µs), resulteert in een
snelheidsvectorveld. Zelfs gebruikmakend van een meetvenster zo klein als 3 mm bij
2 mm, kan een snelheidsvectorveld van 100 bij 75 vectoren worden gemeten tijdens
een tijdsinterval van 0.1 µs klein. Voorts wordt de krachtige PIV techniek gebruikt om
wervelafschudding te bestuderen aan de uiteinden van een parallelle-plaatstack, het
snelheidsprofiel tussen twee parallele platen en een tijdsgemiddelde snelheid, ook wel
streaming genoemd. De snelheidsprofielen tussen twee platen vertonen een verander-
ing in faseafhankelijkheid wanneer het Reynoldsgetal een kritische waarde overschrijdt,
hetgeen op een overgang van laminair naar turbulent duidt.
Om het profiel van de stacktemperatuur als functie van tijd te registreren, zijn 32
thermometers in een stackplaat ingebouwd. Om de metingen te kunnen verklaren is
de energiebalans van de stack gemodelleerd. De berekende temperatuurprofielen als
functies van tijd bleken consistent te zijn met de temperatuurmetingen.
We hebben verschillende meettechnieken, microfoons, thermometers, en PIV ge-
bruikt, om een vollediger beeld te krijgen van verschillende thermoakoestische fenome-
nen. We hebben diverse niet-lineaire effecten bekeken: minor losses aan de stackuitein-
den, streaming en een overgang van laminair naar turbulent tussen de stackplaten.
Ook hebben we een volledig model gemaakt van lineaire thermoakoestiek, gebaseerd
op de gevestigde lineaire theorie, dat het temperatuurprofiel in een stack kan voor-
spellen, zelfs tijdsafhankelijk. We hebben op deze manier meer inzicht gekregen in
thermoakoestiek en niet-lineaire effecten in het bijzonder en we hopen dat andere on-
derzoekers en de industrie hiervan kan profiteren.
Dankwoord
Een groot project, zoals een promotieonderzoek, doe je niet alleen. Er zijn talloze mensen
aan wie ik dank verschuldigd ben, omdat ze elk op hun eigen wijze hebben bijgedragen
aan mijn promotieonderzoek en aan het tot standkomen van dit proefschrift. Enkele
mensen wil ik in het bijzonder vermelden.
Allereerst natuurlijk mijn promotor, Fons de Waele, vanwege zijn grote kennis van
thermodynamica en pulsbuiskoelers, zijn didactische en theoretische kwaliteiten en
omdat hij altijd een originele en verfrissende kijk had op mijn resultaten. Ook voor
de adviseur in dit project, Mico Hirschberg, was een belangrijke rol weggelegd. Als
akoestisch specialist had hij altijd goede adviezen, over het bouwen van de opstelling,
meetmethodes, analyses, numerieke simulaties, interpretatie van meetresultaten en het
aanbevelen van relevante literatuur.
Met de technici Leo van Hout, Paul Bloemen, Paul Ni¨ el en Peter Helfferich heb
ik altijd zeer plezierig samengewerkt, veel dank voor al jullie technische ondersteun-
ing. Daarnaast ben ik de medewerkers van de werkplaats, Marius Bogers, Henk van
Helvoirt, Frank van Hoof, Han den Dekker en Ginny Fransen-Ter Plegt, veel dank ver-
schuldigd voor de vervaardiging van onderdelen en stacks. Zij stonden altijd voor
me klaar voor het doen van technische klussen. In het tot stand komen van de grote
akoestische opstelling gaat in het bijzonder mijn dank uit naar Jaap Hoevenaar, die alle
hoornsegmenten heeft gedraaid op de grote freesbanken bij de GTD.
Met het wiskundeteam binnen dit project heb ik altijd heel prettig samengewerkt.
Allereerst wil ik Peter in ’t panhuis bedanken voor de goede samenwerking en vriend-
schap en verder ook Sjoerd Rienstra, Jaap Molenaar en Han Slot.
Ik wil gaarne de stichting STW bedanken voor hun financi¨ ele bijdrage, en in het bij-
zonder Corine Meuleman voor de begeleiding. Voorts wil ik de medewerkers van ECN,
Aster Thermoacoustics en Shell bedanken voor de interessante discussies. In het bijzon-
der wil ik noemen Okke Ouweltjes van Philips voor zijn bijdrage aan de electroakoestis-
che berekeningen en Cees de Blok voor het maken van het co-axiale koppelstuk voor de
lopende golf metingen.
Gunes Nakiboglu en Wenqing Liang ben ik dank verschuldigd voor hun assistentie
bij de numerieke simulaties en mijn stagiaires Jan van Kemenade en Wim Weltjens voor
alle metingen, analyses en discussies.
Met Yan Li heb ik als kamergenote en collega een goede band opgebouwd. Tijdens
de lunchpauzes zorgden Paul Ni¨ el, Yan Li en Wenqing Liang voor gezelschap en ver-
maak, mede door de vele spelletjes die gespeeld zijn in de koffieruimte.
Gezelligheid en steun buiten het werk is ook erg belangrijk om een dergelijk groot
project af te ronden. Allereerst wil ik mijn ouders bedanken, omdat ze me altijd zijn blij-
ven steunen. Mijn vriendin, Nicole, die ik pas in de laatste fase van mijn promotie heb
leren kennen, heeft mij ook altijd gesteund en bovendien geholpen met het corrigeren
van spelling en grammatica. Heel belangrijk waren ook mijn makkers Coen Oosse, Joost
Heltzel, Prahlada Belle en Raoul Lemmen. Verder wil ik mijn vrienden van schaakclub
’t Pionneke en mijn oud-medestudenten, Niels, Jeroen, Leon, Rob, Bart en Joost hier
graag vermelden.
In de periode van het schrijven van mijn proefschrift heeft Sonja Feiner-Valkier me
geholpen om door een moeilijke fase heen te komen.
Degene die ik verreweg het meeste dankbaar ben is Jos Zeegers. Hij stond altijd voor
me klaar, was altijd erg betrokken en wist me steeds te motiveren. Vanaf het aanvragen
van project, het ontwerpen van de opstelling, het leggen van contacten met univer-
siteiten, de bezoeken aan universiteiten en conferenties in Frankrijk tot het schrijven van
papers en dit proefschrift. Hij heeft op vele manieren geholpen en een betere begeleider
had ik me niet kunnen wensen.
Curriculum Vitae
Paul Aben was born the first of September 1981 in Sittard, the Netherlands. After finish-
ing his pre-university education (Atheneum) in 1999 at Bisschoppelijk College Broekhin
in Roermond, he started his studies in Technical Physics at Eindhoven University of
Technology in Eindhoven that same year. Within in one year he finished his first-year
diploma (propedeuse) and won the CIVI encouragement award for being the student
with the highest grades in the Netherlands. During his studies he did an internship at
Oc´ e, Venlo, entitled “Warmtehuishouding in digitale kopieerapparaten”. In 2005 he gradu-
ated within the Elementary Processes in Gas discharges (EPG) group in Eindhoven on
“Optical Study of Breakdown Phenomenology in Metal Halide Lamps”, under the supervision
of dr. Freddy Manders and prof.dr.ir. Marco Haverlag.
In 2005 he started a PhD project at Eindhoven University of Technology, within the
Low-Temperature Physics (LTE) group. This project, under the supervision of dr.ir. Jos
Zeegers and prof.dr. Fons de Waele, was sponsored by the Dutch Technology Foun-
dation (STW). He defended his thesis, entitled “High-Amplitude Thermoacoustic Flow In-
teracting with Solid Boundaries” on the eight of December 2010. As of July 2010 he is
employed at ASML in Veldhoven.

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->