Professional Documents
Culture Documents
ii
Acknowledgements
Doing a Phd is a strenuous and cumbersome work, which I was not able
to finish without the help and support of many people. Therefore I want to
thank everyone who contributed in any way to the making of my thesis.
I wish to thank the head of the research group Fluid Mechanics and
Thermodynamics, Prof. Dr. Ir. Chris Lacor for giving me the opportunity
to work on this very interesting field of research. Im particular grateful
that he gave me the chance to develop my own ideas, which helped me to
grow as a researcher.
Secondly, I would like to thank my co-promotor, Prof. Dr. Sylvia Verbanck, who always helped me to focus not only on the computational part
of the research but also the physiological side of the research. For not being
a CFD-specialist, she posed many questions, which helped me to look in a
critical way to the obtained results.
Thirdly, I would like to thank my colleagues and former colleagues at
the Fluid Mechanics research group: Kris Van den Abeele, Sergey Smirnov,
Patryk Widera, Santosh Jayaraju, Ghader Ghorbaniasl, Matteo Parsani,
Mahdi Zakyani Roudsari, Dean Vucinic and former colleagues Jan Ramboer and Tim Broeckhoven. First, Tim and Santhosh thanks a lot for the
proofreading of this dissertation. I know you both had a lot work and reading someones Phd can be quite strenuous. Tim, also thanks for sharing an
office during 4 years, we had a lot of fun together. Jan, you always helped
me to put things into perspective. Things are quite different without our
office-ninja. Kris, you are now almost 2 years in the department but it looks
a lot longer... We had and hopefully will have a lot of fun together. Sergey,
Patryk, Ghader, thanks for the nice discussions during the coffee breaks. I
would express my gratitude to Alain Wery for his unlimited help with all
the computer problems and lab problems, I encountered over the past years
and also our secretary Jenny Dhaes for her administrative support.
I also want to thank my family and friends, who always reminded me
that there was more in life than upper human airways.
Last but certainly not least I want to thank my girlfriend Annemie. You
always put up with my bad mood after an unsuccessful day and gave me
iii
the courage to finish my Phd. Like you supported me during my Phd, I will
help you to go through the upcoming difficult time.
iv
Contents
Abstract . . . . .
Acknowledgments
List of Figures . .
List of Tables . .
List of Symbols .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
i
. iii
. xiv
. xiv
. xiv
1 Introduction
1.1 Asthma . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 History of asthma . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Asthma in the world . . . . . . . . . . . . . . . . . . . . . .
1
1
1
2
4
4
4
6
Fluid Dynamics
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
32
32
32
33
34
34
35
35
36
36
37
37
41
41
Fluid Dynamics
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
43
43
44
45
46
48
49
53
56
7.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8 CFD of the Flow in a Model
8.1 Method . . . . . . . . . .
8.1.1 Grid . . . . . . . .
8.1.2 Numerical Method
8.2 Results and Discussion . .
8.3 Conclusions . . . . . . . .
of the
. . . .
. . . .
. . . .
. . . .
. . . .
92
the
115
. . 115
. . 116
. . 118
. . 118
. . 122
. . 124
. . 126
. . 128
. . 131
. . 135
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
138
138
140
143
148
153
Bibliography
170
A List of publications
171
vii
List of Figures
2.1
2.2
2.3
2.4
2.5
2.6
.
.
.
.
.
.
.
.
.
.
.
.
I recorded at time t + t . . . . . . . . . . . . . . . . . . .
3.9 Schematic representation of the cross-correlation of the in
tensity fields I and I . . . . . . . . . . . . . . . . . . . . . .
3.10 Idealized linear digital signal processing describing the functional relationship between two successively recorded particle
image frames . . . . . . . . . . . . . . . . . . . . . . . . . .
3.11 Measurement uncertainty in digital cross-correlation PIV evaluation with respect to varying particle image diameter . . .
3.12 Arbitrary example of a PIV measurement result containing
spurious displacement vectors . . . . . . . . . . . . . . . .
viii
5
6
7
8
9
10
13
14
16
17
18
19
20
21
22
24
27
28
31
40
6.1
6.2
6.3
6.4
.
.
.
.
59
60
63
65
70
geometry
geometry
geometry
geometry
ix
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
33
35
38
71
75
76
76
79
80
81
82
83
86
87
7.14 Cartesian plots of the normalized velocity at different locations in the trachea( 0.5, 1, 2 and 3 tracheal diameters downstream the glottis) for 15 l/min and 30 l/min air flow rate .
88
91
94
95
98
8.4 contour plots of normalized velocity (left) and normalized kinetic energy (right) of k--sst (a), k--realizable(b), reynolds
stress model (c) and experiments (d) at 15 l/min . . . . . . 100
8.5 contour plots of normalized velocity (left) and normalized kinetic energy (right) of k--sst (a), k--realizable(b), reynolds
stress model (c) and experiments (d) at 30 l/min . . . . . . 101
8.6 zoom of the streaklines in the pharynx at 30 L/min: panel
a: k--sst, panel b: k--realizable, panel c: reynolds stress
model and panel b: experiments at 15 l/min . . . . . . . . . 102
x
103
103
107
108
111
112
113
119
120
122
124
125
125
10.5 CFD simulated pressure drops between model inlet and outlet for different flows up to 60 L/min, in the case of no stenosis
(open triangles), of 60% constriction (solid squares) and of
85% constriction (solid circles). The line plots are the corresponding best-fit power laws, leading to power values of 1.77
(no stenosis), 1.92 (60% stenosis), and 2.00 (85% stenosis) . 149
xiii
List of Tables
7.1 The measured air flow rates and corresponding Reynolds
numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1 Influence on total injected particles on the total deposition
percentage . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2 Comparison of total deposition between mean flow tracking
and Eddy Interaction Model . . . . . . . . . . . . . . . . . .
9.3 Comparison of total deposition for Heliox and air as carrier
gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.4 Comparison of total deposition for steady and unsteady flow
with a FIR of 2 L/s2 ) . . . . . . . . . . . . . . . . . . . . .
9.5 Comparison of total deposition for steady and unsteady flow,
where the particles are released at the moment the flow rate
reaches the maximum value . . . . . . . . . . . . . . . . . .
xiv
78
118
129
130
133
134
List of Symbols
Latin symbols
Xi
magnification factor
V0
R1
ai
interrogation area i
RC
RF
RC
RP
self-correlation peak
d
RD
d (t)
U, u
dp
Up , up
particle velocity
residual
Dp
Gb
mass of a particle
FD
CD
s
S
gravity vector
rp
particle position
TL
Le
tcross
tji
sij
pressure
xvi
time
fk
Gk , G
Sk
Yk
YM
Gij
DT,ij
DL,ij
Cij
Pij
Fij
Mt
y+
friction velocity
DH
hydraulic diameter
damping function
u, v, w
p, P
total energy
Cp
temperature
Greek symbols
tot
resid
sys
systematic errors
relaxation time
particle density
ij
turbulent viscosity
xviii
ij
deposition efficiency
preconditioning matrix
Subscripts
air
mixture
of air
of the water/glycerine mixture
inlet
at inlet
norm
normalized
rms
ref
reference
loc
local
gauge
Other symbols
.
averaged
e.
favre averaged
xix
fluctuating part
Abbreviations
LDV
LDA
PDI
PIV
PTV
LSV
FFT
CFD
DNS
LES
DES
RANS
EIM
RNG
Renormalization group
SST
RSM
MRI
CT
Computer Tomography
STL
stereolithography
DPI
pMDI
FIR
Dimensionless numbers
Re
Reynolds number
Stk
Stokes number
De
Dean number
xxi
xxii
Chapter 1
Introduction
1.1
Asthma
1.2
History of asthma
The word asthma is derived from the Greek aazein, meaning to exhale
with open mouth, to pant. The expression asthma appeared for the first
time in the Iliad (written by Homerus), with the meaning of a short-drawn
breath, but the earliest text where the word is found as a medical term is
the Corpus Hippocraticum. However it is difficult to determine whether
in referring to asthma, Hippocrates and his school (460-360 B.C.) meant
an autonomous clinical entity or simply a symptom. He thought that the
1
spasms associated with asthma were more likely to occur in tailors, anglers,
and metalworkers.The best clinical description of asthma in later antiquity is
offered by the master clinician, Aretaeus of Cappadocia (1st century A.D.).
The numerous mentions of asthma in the extensive writings of Galen (130200 A.D.) appear to be in general agreement with the Hippocratic texts and
to some extent with the statements of Aretaeus [80].
Moses Maimonides, a renowned 12th century rabbi, philosopher, and
physician practiced in the court of Saladin (1137-1193), sultan of Egypt
and Syria. He wrote a treatise on asthma for his royal patient, Prince AlAfdal. He noted that his patients symptoms often began with a common
cold, especially in the rainy season, forcing him to gasp for air until phlegm
was expelled [3].
Jean Baptiste Van Helmont, a Belgium physician during the 16th century, wrote that asthma originated in the pipes of the lungs. In the 17th
century, Bernardino Ramazzini, an Italian physician, noted a connection between asthma and organic dust. During the early 1800s asthma was rarely
mentioned in medical literature. At that time 5 patients with asthma constituted a case report. Asthma was first described in the medical literature
in the mid-1800s and still considered rare at that time [3].
The use of bronchodilators started in 1901. Early 20th century studies
focused on the premise that asthma was a psychosomatic disease, and this
side-tracked the major advances which loomed on the horizon. Eventually
researchers would refute these erroneous psychiatric theories, and prove
that asthma was a physical condition. It was not until the 1960s that the
inflammatory component of asthma was recognized, and anti-inflammatory
medications were added to the regimen [3].
1.3
Chapter 2
Anatomy of the Human
Respiratory Tract
This chapter describes the terminology of the respiratory tract
2.1
Each human being breaths about 20000 times every day, this results into
600 million breaths at the age of seventy. At rest an average adult breaths
about 15 kg of air (10000 to 20000 liters) each day. This section provides
a simplified explanation on how breathing works and a description of the
lung physiology.
2.1.1
The main function of the respiratory system is to supply the blood (and
cells) with oxygen and to remove the carbon-dioxide from the blood (and
cells).
The inhalation process is driven by the diaphragm. When it contracts,
the contents of the abdomen is pushed downwards and the thorax or ribcage
expands. This creates a larger thoracic volume and thus an under pressure
in the lungs with respect to the atmospheric pressure at the level of the
mouth and nose. This makes the air, which contains about 21% oxygen
to travel down into the deeper lung. During forced inhalation, the external
4
intercostal muscles and accessory muscles come into play and further expand
the thoracic volume.
The exhalation process is passive. The lungs are by nature elastic and
because of the recoil from the stretch of inhalation, the air is pushed outwards until the pressure in the thorax reaches equilibrium with the atmospheric pressure. During forced exhalation the expiratory muscles, the
abdominal muscles and internal intercostal muscles force the air to flow out
of the lungs. Figure 2.1 shows a representation of the respiration cycle.
2.1.2
In order to reach the alveolar zone of the lungs, where the gas exchange
takes place, the air has to pass several complex structures. In the following
section a detailed description of the most important structures, which are
dealt with in this dissertation, is given.
usually, when normal breathing, air enters the respiratory tract through the
nostrils or nares and flows through the nose. The open spaces in the nose
are celled nasal passageways or nasal cavities, which act as a filter of dust
and other foreign material. The nasal cavities are covered with tiny hairs
6
Figure 2.3: picture from the cilia in the respiratory tract [5]
The cilia move back and forth pushing the particles and mucus either
toward the the pharynx or the nostrils. The nasal cavities also warm up
and moisten the incoming air. Going down the nasal cavity the air passes
through the nasopharynx, which extends to the level of the uvula.
The other opening of the respiratory tract, the mouth has the same
function also warms up and moistens the air but to a lesser degree, because
air travels much faster through the mouth compared to the nose. At the
roof of the mouth,the hard palate, a thin, bony plate of the skull is situated.
This is followed by the soft palate or the palatine velum. The uvula is in
fact a soft process that extends from the posterior edge of this soft palate.
At the end of the mouth, the air travels through the fauces (Latin plural
for throat) to the oropharynx, which is also connected to the nose by the
nasopharynx. The fauces is the hinder part of the mouth and are regarded
as the two pillars of mucous membrane. One being anterior, known as the
palatoglossal arch and the second is posterior, the palatopharyngeal arch.
Between these two arches is the palatine tonsil, which protects the body
from infection as shown on figure 2.4.
The oropharynx extends from the uvula to the epiglottis and is lined
with stratified squamous epithelium that protects against abrasion due to
7
Figure 2.4: frontal view of the mouth with the different structures [2]
the high volume of food intake. Coming from the oropharynx, air moves
into the laryngopharynx, which extends to the opening of the larynx and the
esophagus, which leads to the digestive system, as can be seen on figure 2.5.
The laryngopharynx is like the oropharynx lined with squamous epithelium.
The nasa-, oro- and laryngopharynx form together the pharynx.
Now the air flows into the complex structure of the larynx. It consist
of an outer casing of nine cartilages connected to each other by muscles
and ligaments. The most well known and also being the largest and most
superior cartilage is the thyroid cartilage or the Adams Apple. The most
inferior is the cricoid cartilage, which forms the base of the larynx. The
already mentioned epiglottis is also one of the nine cartilages and it prevents
material (e.g.food) from entering the larynx by covering its opening. The
six remaining cartilages are stacked in two pillars between the cricoid and
thyroid cartilage. Two pairs of ligaments, known as vestibular or false vocal
folds (the superior pair) and the (true) vocal cords (the inferior pair) are
8
cartilage, which form the anterior and lateral side of the trachea. It has a
protective function and maintains an open passageway for air. The posterior
wall contains no cartilage and consists of a ligamentous membrane and
smooth muscle, which can alter the diameter of the trachea. The esophagus
lies immediately posterior to the cartilage-free wall of the trachea. The
trachea leads down the thoracic cavity where it divides into the right and
left bronchus. The right bronchus is shorter, wider and more vertical than
the left bronchus. This difference is caused by the heart which is situated
more to the left than the right side of the chest. The subdivision of the
bronchi are primary, secondary and tertiary divisions. In all, they divide 16
times into even smaller bronchioles. These lead to the respiratory zone of
the lungs, which consists of respiratory bronchioles, alveolar ducts and the
alveoli where finally, the gas exchange finds place. The surface available for
gas exchange in an average adult is between 100 to 140 m2 .
Figure 2.6: detailed front(left) and top (right) view of the larynx [8]
The respiratory tract from nasal cavities to the smallest bronchi is covered by a layer of sticky mucus, secreted by the epithelium and small ducted
glands. Foreign particles (e.g. dust) which hit the walls of the tract are
trapped in this mucus. Once the foreign material is stuck in the mucus, it
has to be removed. This is carried out by the cilia on the epithelial cells
which move continually up and down the tract. The cilia in the trachea
10
and in the bronchi push the mucus, with the particles towards the pharynx
where it is swallowed.
11
Chapter 3
Theoretical Background of
Particle Image Velocimetry
(PIV)
In this chapter the principles and theoretical background of particle image
velocimetry is described.
3.1
Figure 3.1: Leonardo da Vinci sketched various flow fields over objects in
a flowing stream
Prandtl gained insight to the basic features of fluid flow. However this was a
very interesting experiment which described for the first time methodologically fluid flow, no quantitative measurements were possible at that time.
However these efforts only gave a qualitative view of the flow.
The oldest well-know technique to quantatively measure fluid flow is the
pitot tube, named after French Engineer Henry de Pitot(1695-1771). He was
the first person to measure velocity with an upstream pointed tube, while
the French engineer Henry Darcy (1803-1858) developed most of the features
of the instrument we use today. Pitot tubes are used in wind tunnels,
airplanes, etc. The major disadvantages of this measurement method are:
the tube has to be placed into the fluid flow, and thus disturbs the
flow
only one point can be simultaneously measured and thus it can take
a lot of time to measure the complete profile of a flow
In the late 1950s hot-wire anemometers were introduced . As the name
implies, hot-wire (-film) anemometers uses a very thin wire, which is placed
into the flow and through convective cooling by the flow of a wire which
is heated by an electric current, the flow velocity can be measured. Most
hot-wires have a diameter of 5 m and a length of approximately 1 mm and
are made of tungsten and can take thousands of velocity measurements per
second, allowing to study the details of fluctuations in turbulent flow. The
major drawbacks of this method are:
13
in frequency of the reflected radiation due to the Doppler effect, and relating this frequency shift to the flow velocity of the fluid in the target area.
The major advantage of this method over hot-wire anemometry is that is
non-intrusive. Nowadays systems which can measure the three components
of velocity at once become more and more available. However this method
also has its disadvantages:
the desired target area has to be reachable by the laser beams
the major cost of a system
difficult to measure close to a surface
only 1 point can be measured at once
To overcome this last drawback, other measuring methods where developed, called particle-imaging techniques( Planar Laser-Induced Fluorescence, Laser-Speckle Velocimetry, Particle Tracking Velocimetry, Molecular
Tracking velocimetry and Particle Image Velocimetry). An overview of these
methods are described in literature ([76], [13], [55], [88] [37], [35]). Since
Particle Image Velocimetry (PIV) is the only applied method in this work,
it will be discussed in detail.
3.2
The principle of PIV is based on the measurement of the instantaneous velocity of tracer-particles which are carried by the fluid flow through the
detection of the particle displacement with a sophisticated stroboscopic
method. These particles have to be illuminated in a plane of the flow at
least twice within a short time interval (figure 3.3). The light scattered by
the tracer particles is recorded on a photographic negative or on two separate frames on a special cross correlation CCD (Charge Coupled Device)
camera positioned at right angles to the light sheet.
For evaluation the digital recording is divided in small interrogation
areas. The local displacement vector for the image of the tracer particles
reflection is determined for each interrogation area by means of statistical methods. It is assumed that particles moved homogeneously within an
15
Figure 3.4: The three modes of particle image density: (a) low (PTV),
(b) medium (PIV) and high image density (LSV)
no longer possible to identify image pairs by visual inspection. However it
is possible detect the individual particle images on each recording. Medium
image density needs statistical methods to evaluate the PIV recordings. In
case of high image density ((figure 3.4c), it is not even possible to detect the
individual particle images as they overlap in most cases and form speckles.
Therefore this situation is called laser speckle velocimetry (LSV).
3.3
In the previous section, the principle of PIV was explained. The next section
will go more into detail of the PIV evaluation. As mentioned before, the
obtained images are divided into interrogation areas and those areas of
sequential images of PIV recordings are statistically evaluated. A detailed
mathematical description of statistical PIV evaluation has been given by
Adrian [12], Keane [71] and Westerweel [129]. These interrogation areas are
also called interrogation windows The geometric backprojection of these
areas into the light sheet are referred to as interrogation volumes 3.5.
A single exposure recording consists of a random distribution of N tracer
particles:
X1
Xi
X
= 2 with Xi = Yi
(3.1)
Zi
XN
17
xi
yi
x
x =
, Xi =
, Yi =
(3.2)
y
M
M
x refers to the coordinates in the image plane. The particle position and
the image position are related by a constant magnification factor M. The
image intensity field of a single exposure can be expressed by:
N
I = I(
x , ) = ( x)
V0 (Xi )(
x
xi )
(3.3)
i=1
where V0 is the transfer function giving the light energy of the image of an
individual particle within the interrogation volume, the function represents the point spread function of the imaging lens, it is assumed identical
for each tracer particle. This can be rewritten as [96]:
I(
x , ) =
N
X
i=1
V0 (Xi )(
x
xi )
18
(3.4)
3.3.1
Auto-correlation
R1 (
s , ) = hI(
x , )I(
x +
s , )i
(3.5)
R1 (
s , ) =
Z
N
1 X
N
1 X 2
V 0 ( Xi )
a1 i=j
a1
a1
(
x
xi )(
x
xj +
s )d
x
(3.6)
where
s is the separation vector in the correlation plane. The terms i 6= j
represent the correlation of different particle images and therefore randomly
distributed noise in the correlation plane. The terms i = j represent the
correlation of each particle with itself.
Adrian [96] proposed the following decomposition:
19
R1 (
s , ) = RC (
s , ) + RF (
s , ) + RP (
s , )
(3.7)
where RC (
s , ) is the convolution of the mean intensities of I and RF (
s , )
is the fluctuating noise component both resulting from the i 6= j terms.
RP (
s , ) is the self correlation peak located at position (0,0) in the correlation plane, resulting from the components that correspond to the correlation
of each particle with itself.
3.3.2
Figure 3.8: The intensity field I recorded at time t and the intensity field
I recorded at time t + t
R2 = (
s , , D ) =
V0 (Xi )V0 (Xj + D )R (
xi
xj +
s d)
i6=j
+ R (
s d)
where
R (
xi
xj +
s d)=
a1
and
a1
N
X
(3.8)
i=1
(
x
xi )(
x
xj +
s d )d
x (3.9)
Xi + DX
MD
X
Xi = Xi + D = Yi + DY , d =
MDY
Z i + DZ
(3.10)
R2 (
s , , D ) = RC (
s , ) + RF (
s , , d ) + RD (
s , , d )
21
(3.11)
where RD (
s , , d ) represents the component of the cross-correlation function that corresponds to the correlation of images of particles obtained from
the first exposure with images of identical particles obtained from the second exposure ( i = j terms). For a given distribution of particles inside the
and I are given. Correlations of x2 dont appear on figure 3.9 because this
particle image is located outside the interrogation window.
3.3.3
d (t) resid
|U | =
+
Mt
Mt
(3.12)
resid
=
t0 Mt
(3.13)
d (t) is the distance traveled by the particle images within the pulse separation time t and resid are the residual errors of the measured image
displacements. These errors are not affected by the alteration of the pulse
separation time. Therefore, the second term on the right hand side of equation 3.12 will increase if the pulse separation time decreases:
lim
On the other hand, the particle image displacement decreases if the pulse
separation time decreases. This leads to:
d (t)
= |U |
lim
t0 Mt
(3.14)
3.4
As already explained, some kind of interrogation scheme is required to extract displacement information from a PIV recording. Initially, this interrogation was manually performed on selected images with low density seeding
which allowed the individual tracking of particles ([135], [32]). With computers and image processing becoming more and more commonplace, it
became possible to automate the interrogation process ([34], [46], [44]).On
images with medium seeding density it is almost impossible to detect matching image pairs of particle images, by visual inspection. Hence, statistical
methods had to be developed.
Looking from an image perspective view to two successively recorded
particle image frames, the first image can be considered as the input to
system whose output produces the second image of the pair (3.10). The
23
Figure 3.10: Idealized linear digital signal processing describing the functional relationship between two successively recorded particle image frames
input image I is converted, through the displacement function d and an
additive noise process N to the output image I . With both images I and
extending over the edges of I . The template I is smaller than the sample
I . Each choice of the sample shift, one correlation value is computed. For
shift values at which the template matches the sample, the highest value
of the correlation will be found and from that the displacement d can be
found. This method has some drawbacks:
the number of multiplications per correlation value increases in proportion to the interrogation window area.
no rotations or deformations can be recovered by this method.
no efficient method of computing (millions of multiplications)
Alternative is to take advantage of the correlation theorem which states that
the cross-correlation of two functions is equivalent to a complex conjugate
multiplication of their Fourier transforms:
R2 I I
(3.15)
24
using the Fast Fourier Transform (FFT) which reduces computation from
O[N 2 ] operations to O[NlnN] operations ([18], [94]).
The Fourier transform is an integral over a domain extending from negative infinity to positive infinity. In practise, the integrals are computed
over finite domains which is justified by assuming the data to be periodic,
that is, the image sample continually repeats itself in all directions. So, if
the data of length N contains displacements exceeding half the sample size
N/2, then the correlation peak is folded back into the correlation plane to
appear on the opposite side. In this case the Nyquist theorem is violated.
The solution is to reduce the laser pulse delay.
The highest value in the correlation plane can be computed and this
peak permits the displacement to be determined with an uncertainty of 0.5
pixel. But the structure of the correlation peak also contains information.
Therefore peak fitting functions for sub-pixel displacement estimates were
introduced. The three most frequently reported peak interpolation or fitting
schemes are the centroid, parabolic and Gaussian (based on a three-by-three
pixels kernel). The accuracy associated with those schemes has been widely
studied and detailed results from numerical and theoretical investigations
were given by several authors ([129], [91], [96]). The Gaussian scheme shows
the best performance, confirmed by several authors. If no peak fitting function is used or the estimators are misused, the so-called peak-locking effect
can occur. The displacement values are locked at integer values. There are
several reason for this phenomena but the most important one is the particle image being smaller than 1.5 pixel. Also a reduced fill factor of the
image can lead to a peak locking. The interrogation method discussed in
this section is the classical scheme suggested by Willert and Gharib ([136]).
More recently advanced algorithms, like multi-pass and multi-grid were developed. In the multi-pass interrogation technique ,the interrogation of the
image is repeated at least once more. In the following passes the image
sample positions are offset by the integer shift determined in the preceding
pass. Once the residual shifts are less than one pixel, a re-evaluation of the
respective point is no longer necessary (e.g. convergence).
In combination with this multi-pass interrogation scheme a multi-grid
scheme can be applied. With this a pyramid approach is used by starting off
with larger interrogation windows on a coarse grid and refining the windows
and grid with each pass. This is especially useful in PIV recordings with
both a high image density and a high dynamic range in the displacements.
25
3.4.1
Error estimation
3.4.2
After automatic evaluation of the PIV recordings, often a number of spuriousvectors are found back on processed images (figure 3.12).
These vectors deviate unphysically in magnitude and direction from
nearby valid vectors and often appear at the edges of the data field.
(near the surface of the model, at edges of drop-out areas, at the edges of
26
(3.17)
with |U0 | the magnitude of the considered vector, U being the average
value of the nearest neighbors. Problems will arise at the edges of the data
field when there are less than eight neighbors available for comparison.
27
(3.18)
(3.19)
This residual is defined for each vector {Ui |i = 1, , 8}. The median of
the residuals r(median) is used to normalize the residual of U0 :
r0 =
U0 U(median)
r(median) +
(3.20)
3.5
Tracer particles
As mentioned before, tracer particles must be added to the flow and these
particles must rigorously follow the fluid flow. The quantity which describes
the time required for a particle to adjust or relax its velocity to a new
condition of forces is called the relaxation time of the particle and is given
by:
p d2p
=
(3.21)
18
where p is the density of the particle, dp is the particle diameter and
is the dynamic viscosity of the fluid. The particle relaxation time, as defined in 3.21 is restricted to particle motion in the Stokes region, Red < 1.
Nevertheless, the relaxation time remains a convenient measure for the tendency of particles to attain velocity equilibrium with the fluid. The particle
Reynolds number Red is defined by:
Red =
dp Up
(3.22)
with Up , the particle velocity and the kinematic viscosity of the fluid. The
particle relaxation time for a hollow glass sphere of 10m in water glycer2
kg
ine mixture with kinematic viscosity of 5.5e6 ms and density of 1150 m
3 is
6
approximately 1e . which is even more than three times smaller than the
relaxation time of a 1m oil particle in air flow. These particles are often
used in LDA and PIV measurements in air flow. Particles seeded in the flow
29
must scatter the light sufficient enough, to obtain large enough particle images. It is often more effective and economical to use particles with better
scattering behavior than to increase laser power, to increase image intensity. In general, light scattered by small particles is a function of the ratio
of the refractive index of the particles to that of the surrounding medium,
the shape, size and orientation of the particle. Also polarization and observation angle have an influence on the scattered light. The Mie scattering
theory can be applied to spherical particles with larger diameters than the
wavelength of the incident light. The Mie scattering can be characterized
by the normalized diameter, q, defined by:
q=
dp
(3.23)
30
31
Chapter 4
Theoretical Background of
Computational Fluid Dynamics
(CFD): The Particle Phase
In this chapter the principles and theoretical background of the computational method of simulating particle trajectories in combination with RANS,
is described
4.1
Introduction
Particle Laden flows forms a major class of two-phase flows in which the continuous fluid (gas or liquid) and discrete particles (solid, liquid or gaseous)
are treated as two different phases. Laden flows find numerous biological
and industrial applications such as hemodynamics, flow dynamics in human
airways, biological and chemical reactors, sedimentation, filtration etc. The
present chapter describes the governing equation of the particle phase.
4.2
Figure 4.1: Particle size distribution in gas-solid flows (after Soo, 1990)
4.2.1
Particle size
Particle distribution
The volume fractions of dispersed phase is relatively low and for gas-solid
flows it can be roughly considered to be in the range of 102 103 and the
spacing between the particles to be about 10 times more than the particle
size. The specific length scales of particle laden flows should satisfy the
following relationship,
33
4.3
A dilute dispersed phase flow is one in which the particle motion is controlled
by fluid forces such as drag, lift etc. On contrary, the dense phase flow is
dominated by inter-particle collisions. A quantitative assessment of these
flows is established by the ratio of momentum response time (relaxation
time) r to the time between particle collision c . The momentum response
time r is defined as the time required by the particle to loose 63% of its
initial velocity. If r < c , the particle has enough time to respond to
the local fluid dynamic forces before collision. If r > c , collision occurs
before particle responds to the flow field and hence the motion is collision
dominated and the flow is considered dense.
4.4
Phase coupling
4.5
The two most widely used approaches for mathematical modeling of twophase flows are Eulerian continuum approach and Lagrangian trajectory
approach.
4.5.1
In the Eulerian approach, the particles are treated as a second fluid which
behaves like a continuum and the equations are developed for average properties of the particles. For example, the particle velocity is the average
velocity over an averaging volume. This approach is most suitable when
one requires a macroscopic field description of dispersed phase properties
such as pressure, mass flux, concentration, velocity and temperature.
35
4.5.2
The Lagrangian approach is useful when the particle phase is so dilute that
the description of particle behavior by continuum models is not feasible. The
motion of a particle is expressed by ordinary differential equations in Lagrangian co-ordinates and are directly integrated to obtain individual tracks
of particles. To solve the Lagrangian-equation for a particular moving particle, the dynamic behavior of the gas phase (generally obtained by Eulerian
approach) and other particles surrounding this moving particle should be
pre-determined. Since the particle velocity and the corresponding particle
trajectory are calculated for each particle, this approach is more suitable to
obtain discrete nature of motion of particles. However, to obtain statistical
averages with reasonable accuracy, huge number of particles will have to
be tracked. Ideally, one would like to track each and every particle which
is not computationally feasible. Hence a smaller number of computational
particles are chosen which represent the actual particles. Each computational particle is regarded as a group of particles which move in the fluid
phase with the same physical properties.
The Lagrangian approach is classified into two types namely, Deterministic trajectory method and Stochastic trajectory method based on the effect
of turbulence. In the deterministic method, all the turbulent transport processes of the particle phase are neglected, whereas the stochastic method
takes into account the effect of gas turbulence on the particle motion by
considering instantaneous gas velocity in the formulation of the equation of
particle motion.
The flow patterns of both continuous and dispersed phase depend on
the mechanisms of mass, momentum and energy coupling. Below are the
governing equations for a single particle with no short range interactions
such as van der Waals force and collision forces. The momentum balance is
considered in detail as it is of primary importance in the present study.
4.6
Mass Balance
=m
dt
36
(4.1)
Change of mass with respect to time is equal to the mass gained/lost by the
particle due to mass flux over the surface. In the present work, the particles
are not considered to gain or loose mass. So equation 4.1 becomes:
dm
=0
dt
4.7
(4.2)
Momentum Balance
d(m
up )
= F + mGb + (m
up )
dt
(4.3)
4.7.1
Interphase Force
velocity (
up ), termed as the slip velocity (
u
up ), leads to unbalanced
pressure distribution as well as the viscous stresses on the particle surface
which yields a resulting force called drag force. For a rigid sphere the drag
force is given by,
u
u p | (
u
up )
FD = CD A |
2
(4.4)
where A is the exposed frontal area of the particle to the direction of the
37
(4.5)
|
u
up |
(4.6)
18.5
Re0.6
p
CD = 0.44
(4.8)
(4.9)
p
24
1 + 0.179 Rep + 0.013 Rep
Rep
Substituting 4.5, 4.6 and 4.10 in 4.4 results in the following final form
of drag force,
p
a2
a3 2
+
Rep Rep
(4.12)
24
b3 Rep
(1 + b1 Rebp2 +
)
Rep
b4 + Rep
where
39
(4.13)
Figure 4.4: Drag coefficient computed with the different formulations for
spheres as a function of particle Reynolds number; right panel shows a
zoom
b1
b2
b3
b4
=
=
=
=
(4.14)
which is taken from Haider and Levenspiel [52]. The shape factor, , is
defined as
=
s
S
(4.15)
where s is the surface area of a sphere having the same volume as the
particle, and S is the actual surface area of the particle. The Reynolds
number Rep is computed with the diameter of a sphere having the same
volume.
Figure 4.4 show the results for the drag coefficient computed with the
different formulations. As can be seen, the results of the different formulations differ not much.
40
4.7.2
Body Force
The body force does not depend on carrier phase. It is usually due to gravity
and/or the reference system rotation and is given by,
Gb =
g (
rp ) 2
up
(4.16)
where rp is the particle position. In the present work only the gravitational force acts on the particle. So, that equation 4.16 can be simplified
to:
Gb =
g
(4.17)
In conclusion, its worth writing down the final equations used to determine the particle position and the velocity,
d
rp
=
up
(4.18)
dt
d
up
m
= (FD + Gb )
(4.19)
dt
4.8
p
d(m
up )
(4.21)
u =
kN
(4.22)
3
41
where N is a random number drawn from a normal probability distribution with zero mean and unit standard deviation. k is the turbulent kinetic
energy of the flow. The chosen fluctuation is referred to a turbulent eddy
whose size (length scale) and life-time(time scale) is known. The Lagrangian
integral time, TL can be approximated by
TL = CL
(4.23)
(4.25)
k
C3/4
3/2
(4.26)
Le
]
|u up |
(4.27)
(4.28)
As soon as the eddy life-time or the crossing time expires, a new fluctuation
is chosen by considering another random number.
42
Chapter 5
Theoretical background of
Computational Fluid Dynamics
(CFD): The Fluid Phase
In this chapter the principles and theoretical background of Computational
Fluid Dynamics is described.
5.1
Introduction
boundaries and where the turbulent length scale is less than the maximum
grid dimension are handled by the RANS turbulence model. As the turbulent length scale exceeds the grid dimension, the regions are solved using
the LES mode. Therefore the grid resolution is not as demanding as pure
LES but is still considerably high.
RANS is the oldest approach to turbulence modeling. The required grid
resolution is relatively small and is therefore the most widely used approach
for modeling the turbulence.
In this dissertation all fluid simulations are performed by using RANS.
This chapter first gives a short histarical overview of CFD and then describes
the equation governing the fluid phase followed by the description of the
RANS method. To conclude the applied turbulence model equations are
given.
5.2
History of CFD
In the late 17th Century Isaac Newton tried to quantify and predict fluid
flow phenomena through his elementary Newtonian physical equations. His
contributions to fluid mechanics included his second law: F=m.a, the concept of Newtonian viscosity in which stress and the rate of strain vary
linearly, the reciprocity principle: the force applied upon a stationary object by a moving fluid is equal to the change in momentum of the fluid as
it deflects around the front of the object, and the relationship between the
speed of waves at a liquid surface and their wavelength.
In the 18th and 19th centuries, significant work was done trying to
mathematically describe the motion of fluids. Daniel Bernoulli (1700-1782)
derived Bernoullis equation, and Leonhard Euler (1707-1783) proposed the
Euler equations, which describe the conservation of momentum for an inviscid fluid, and conservation of mass. He also proposed the velocity potential
theory. Two other well-known contributors to the field of fluid flow, the
Frenchman, Claude Louis Marie Henry Navier (1785-1836) and the Irishman, George Gabriel Stokes (1819-1903) introduced viscous transport into
the Euler equations. This resulted in the now famous Navier-Stokes equation. These forms of the differential mathematical equations that they proposed nearly 200 years ago are the basis of the modern day computational
fluid dynamics (CFD), and they include expressions for the conservation
44
5.3
For simplicity assume an incompressible, constant property flow. The equations for conservation of mass and momentum are
45
ui
=0
xi
ui
p
tji
ui
+ uj
=
+
t
xj
xi xj
(5.1)
(5.2)
(5.3)
(5.4)
For simple viscous fluids sij = sji, so that tji = tij . To simplify the timeaveraging process, the convecting term has to be written in conservation
form, i.e.,
uj
(uj ui )
uj
ui
=
ui
xj
xj
xj
(5.5)
ui
(uj ui )
=
xj
xj
(5.6)
5.4
(uj ui )
p
(2sij )
ui
+
=
+
t
xj
xi
xj
(5.7)
flow an instantaneous flow variable is expressed by f (x, t). Its time average,
FT (x), is defined by
Z
1 t+T
FT (x) = lim
f (x, t)dt
(5.8)
T T t
Time averaging is the most commonly used form of Reynolds averaging
because most turbulent flows of interest in engineering are stationary. Spatial averaging can be used for homogeneous turbulence, which is a turbulent
flow that, on the average, is uniform in all directions. Ensemble averaging
is the most general type of Reynolds averaging suitable for, e.g., flows that
decay in time.
For a stationary turbulent flow, the instantaneous velocity, ui(x, t), can
be expressed as the sum of the mean, ui (x), and a fluctuating part, ui (x, t),
so that
p
ui
+
(uj ui + uj ui) =
+
(2sji )
t
xj
xi xj
(5.10)
(5.11)
ui
ui
p
+ uj
=
+
(2sji uj ui )
t
xj
xi xj
(5.12)
ij = uj ui
(5.13)
47
The specific Reynolds-stress tensor is symmetric and has six independent components. Hence, by Reynolds averaging six unknown quantities
are produced but no extra equations are created. Now, for general threedimension flows, four unknown mean-flow properties (pressure and the three
velocity components) and the six Reynolds-stress components, resulting in
ten unknowns have to be computed. The mass conservation (equation 5.10)
combined with the three components of equation 5.12 makes four. This
means that the system is not yet closed
When the medium is a compressible fluid one has to account for density
and temperature fluctuations next to velocity and pressure fluctuations.
If standard time averaging is used the complexity of the equations rise.
Therefore a mass-averaging has to be introduced. For details the reader is
referred to the literature. The compressible RANS equations are given by
e
ui
+
= 0
t xi
(5.14)
(e
ui )
e
ui e
uj
2 uk
+
(e
uj u
ei + uj ui ) =
+
[(
+
ij
)]
t
xj
xi xj
xj
xi
3 xk
+
(uj ui )
(5.15)
xj
5.5
Turbulence modeling
uj ui = t (
ui
ui
2
uk
+
) (k + t
)ij
xj
xj
3
xk
(5.16)
But for certain types this approximation fails. The most known applications
where it fails are:
flows with sudden change in mean strain rate
flow over curved surfaces
flow in ducts with secondary motions
flow in rotating fluids
three-dimensional flows
flows with boundary-layer separation
The so-called Reynolds Stress models use an alternative to the employment of the Boussinesq hypothesis. A transport equation is solved for each
term of the Reynolds stress tensor. An additional scale-determining equation is also required. The major drawback is that in 3D seven additional
transport equations have to solved.
5.5.1
k turbulence models
the transport equation for is more an empirically based equation. In developing this model it was assumed that the flow is fully turbulent.
The turbulent kinetic energy and its dissipation rate are obtained from
the following transport equations:
t k
(k) +
(kui ) =
[( + ) ] + Gk + Gb Ym + Sk (5.17)
t
xi
xi
k xi
and
() +
(ui ) =
[( + ) ] + C1 (Gk + C3 Gb ) C2 + S
t
xi
xi
xi
k
k
(5.18)
The turbulent viscosity, t is computed by :
k2
t = C
(5.19)
(5.20)
k
(k) +
(kui ) =
[k ef f ] + Gk + Gb Ym + Sk
t
xi
xi
xi
50
(5.21)
and
() +
(ui ) =
[ ef f ] + C1 (Gk + C3 Gb ) C2 R + S
t
xi
xi
xi
k
k
(5.22)
The quantities k and are the inverse Prandtl numbers for k and ,
respectively. As can be seen the major difference between the RNG and the
standard model lies in the additional term R in the equation, given by:
C 3 (1 /0 ) 2
R =
1 + 3
k
(5.23)
where
2 k
b
d( ) = 1.72 3
db
b 1 + C
(5.24)
ef f
100
b =
t k
(k) +
(kui ) =
[( + ) ] + Gk + Gb Ym + Sk (5.25)
t
xi
xi
k xi
and
t
2
+C1 C3 Gb +S
()+
(ui ) =
[(+ ) ]+C1 SC2
t
xi
xi
xi
k +
k
(5.26)
where
p
k
], = S , S = 2Sij 2Sij
C1 = max[0.43,
eta + 5
and
C1 = 1.44, C2 = 1.9, k = 1.0, = 1.2
The turbulent viscosity is computed from
t = C
k2
(5.27)
5.5.2
k turbulence models
and
k
(k) +
(kui ) =
(k ) + Gk Yk + Sk
t
xi
xi
xi
(5.28)
() +
(ui ) =
( ) + G Y + S
t
xi
xi
xi
(5.29)
k = +
53
(5.30)
(5.31)
where k and are the turbulent Prandtl number for k and respectively.
The turbulent viscosity, t , is computed by
t =
(5.32)
The coefficient damps the turbulent viscosity causing a low-Reynoldsnumber correction. It is given by
=
(
0 + Ret /Rk
)
1 + Ret /Rk
(5.33)
=
=
=
=
=6
i
3
0.072
1
2.0
2.0
k
e k Y k + Sk
(k) +
(kui ) =
(k ) + G
t
xi
xi
xi
54
(5.34)
and
() +
(ui ) =
( ) + G Y + D + S
t
xi
xi
xi
(5.35)
where k and are the turbulent Prandtl number for k and respectively.
The turbulent viscosity, t , is computed as follows
t =
k
1
2
max[ 1 , SF
]
(5.38)
1
F1 /k,1 + (1 F1 )/k,2
1
=
F1 /,1 + (1 F1 )/,2
k =
(5.39)
(5.40)
(5.41)
where
0 + Ret /Rk
)
1 + Ret /Rk
The blending functions, F1 and F2 , are given by
=
(
F1 = tanh(41 )
k 500
4k
, 2 ),
]
0.09y y ,2 D+ y 2
1 1 k
, 1010 ]
= max[2
,2 xj xj
= tanh(22 )
k 500
= max[2
,
]
0.09y y 2
1 = min[max(
D+
F2
2
55
(5.42)
(5.43)
(5.44)
(5.45)
(5.46)
(5.47)
where y is the distance to the next surface and D+ is the positive portion
of the cross-diffusion term. The model constants are given by
k,1 = 1.176, 1 = 2.0, k,2 = 1.0, 2 = 1.168, a1 = 0.31
All other model constants have the same value as for the standard k
model.
5.5.3
The exact transport equations for the transport of the Reynolds stresses
can be written as follows:
(ui uj )
+
(uk uiuj ) = DT,ij +
x
|t {z }
| k {z
}
Local Time Derivative Cij Convection
DL,ij
uj ui
(ui uk
uu
) +Gij + ij ij
xk j k xk
{z
}
|
Pij Stress Production
[
(ui uj )]
x
xk
| k
{z
}
Molecular Diffusion
(5.48)
(5.49)
t
1
2
() +
(ui ) =
[( + ) ] + C1 (Pii + C3 Gii ) C2 + S
t
xi
xj
xj
2
k
(5.50)
where = 1.0, C1 = 1.44, C2 = 1.92, C3 is evaluated as a function of
the local flow direction relative to the gravitational vector. The turbulent
viscosity, t is computed similarly to the k turbulence models:
t = C
where C = 0.09.
57
k2
(5.51)
Chapter 6
State-of-the-Art of the
Research in Upper Airway
Geometries
In this chapter the important publications in the field of flow and deposition
of particles in the upper human airways are summarized and discussed.
Results of different research groups were set together in order to have a
clear view of the progress of the different groups.
Katz and Martonen [68] performed preliminary flow studies using a finite
element analysis on a three-dimensional model of the larynx (Figure 6.1) for
three different flow rates representing sedentary, light and heavy breathing
activities (15 l/min, 30 and 60 l/min). The model of the larynx was based
on morphometric measurements [81] of replica human casts and Weibels
[128] morphology of the tracheal dimensions. The larynx was simulated as
a six cm long cylinder with a circular entrance and exit cross-sections. The
ventricular and vocal cords were modeled ass ellipsis. For obtaining the
simulated flow field the RANS-equations were solved, combined with the
standard k- turbulence model. A central jet, caused by the restrictions of
the ventricular and vocal cords, a major recirculation zone downstream the
glottis and a circumferential secondary flow was observed.
Katz et al [69] studied the influence of the glottal aperture an the flow
velocity and pressure distribution using the same method and model as [68].
It is found that the complex geometries produce jets, recirculation zones
and circumferential flows which may have profound influence on particle
58
diameter.
Gemci et al [43] numerically investigated the influence of 70/30 Helium/Oxygen (Heliox) compared to air as carrier fluid on the deposition of
a series of mono-disperse aerosol injections in a cadaver based throat model.
Rans combined with a standard k - turbulence model was used to simulate the fluid flow. In the computation of the particle deposition, two-way
coupling was included. Simulations of particles with size ranging from 0.25
to 20 micro-meter were carried out. The authors see higher than expected
particle deposition of smaller droplets in the trachea. The influence of using
Heliox as carrier gas is significant on the deposition of smaller particles (up
to 7% less deposition).
Allen et al. [15] used an in vitro computational model of the upper
human airways of a five year old male subject. The model was prepared
from axial Magnetic Resonance Imaging (MRI)- slices taken every 3 mm,
as described in Corcoran et al. [29] The method used to compute the air
flow in the model was RANS with a low Reynolds number shear-stress k-
turbulence model. The applied method was validated with measurements
performed by Corcoran and Chigier [27] on an adult cadaver model. According to the Allen et al. the numerical results obtained are similar to
the experimental results and differences in magnitude of velocity can be
addressed to wrong determination of the flow rate in the experiments. The
laryngeal jet in this pediatric model was compared with the jet in an adult
model and it was been found that axial velocities are comparable. The
turbulent kinetic energy levels obtained in the pediatric model were higher
than could be expected from measurements in adults.
Cheng et al. [24] performed an experimental deposition study on a cast
of the human oral airways. The oral part of the cast was molded from a
dental impression of the oral cavity in a living human subject and the rest of
the airway (up to three generations of bronchi) were made from a cadaver.
Regional deposition of nine different sizes of fluorescent particles was measured at three different inspiratory flow rates (15, 30 and 60 l/min) in the
geometry. The deposition increased with increasing flow rate or particle
diameter. The deposition efficiency was found to be a unique function of
the Stokes number.
Su and Cheng [117] studied the deposition of fiber in a cast of the human
oral airways developed by Cheng et al. [24]. The results of the study showed
that impaction is the dominant deposition mechanism and a data-set of
62
fibers with different aerodynamic diameter at two different flow rates was
provided.
obtained numerical particle deposition results set out to the Stokes number
are close to the experimental correlation of Cheng et al. [24]. It also was
found that turbulence affected the motion of smaller particles in contrary
to larger particles where it had a minor effect. The particle deposition in
the mouth, pharynx and larynx increased with increasing Stokes number.
This was due to the inertial impaction on the particles. Almost no influence
of the inspiratory flow rate on local deposition fractions for a given Stokes
number was observed. Kleinstreuer and Zhang [74] presented basically the
same results, only representation differed.
Zhang et al[139] investigated nano-particle deposition in the simplified
upper airway model combined with a symmetric triple bifurcation representing generation G0-G3, similar to the airway model given by Weibel [128].
The triple bifurcation model is already separately investigated by Zhang
and Kleinstreuer [140], [141]. To capture the airflow structures, the same
method as used in Zhang et al. [145] was applied. The particle transport
and depositions employed , were described by the convection-diffusion equation. This approach is called the eularian approach. The main conclusions
of the study were: (1) the total deposition of nano-size particles for cyclic
inspiratory flow differed not significantly of those obtained by using steady
inlet flow boundary conditions; (2) turbulent fluctuations didnt influence
the deposition of nano-size particles in the upper human airways; (3) transient effects appeared most prominently in the decelerating phase of the
inspiratory cycle.
Zhang et al [144] compared micro- and nano-size particle depositions
in the model described by Zhang et al [145]. As in previous publications
the Rans-equations with an adapted k- turbulence model were solved. In
the treatment of the influence of the effect of turbulence on the particle
movement was different. A near-wall correction proposed by Matida et
al. [85] was applied. Only minor differences in the particle deposition
fractions could be observed when compared with results obtained without
the near-wall correction [145] could be observed, this in contrary to the
results obtained by Matida et al. [85]. In the same geometry , the heat and
mass transfer of ultra-fine particles was numerically investigated by Zhang
and Kleinstreuer [143]. It was found that at an inhalation flow rate of 15
l/min big ambient temperature variations influenced local velocity fields
but total deposition fractions remained unaffected. At higher flow rates no
influence could be observed.
64
al. the LES results reasonably agree with experiments but according to the
author no clear similarity can be discovered. A particle deposition study
in combination with LES was performed. The obtained numerical results
reasonably agree with the experimental results obtained by Grgric et al.
[48] and Zhang and Finlay [138].
Jayaraju et al. [64] developed a lagrangian particle tracking module for
unstructured grids to analyze deposition patterns in a CT based realistic
extra-thoracic airway model. The fluid flow was computed using RANS in
combination with a low-Reynolds-number shear-stress-transport k turbulence model. Inhalation flow rates of 15, 30 and 60 l/min were considered
with particle diameters ranging between 2 and 20 micrometer. Good agreement with the best fit curve of Grgic et al. [47] was attained. The deposition
in the mouth region was considerably high, which was not the case in more
simplified models, which indicated that there was a need for more realistic representation of the mouth cavity to reliably predict regional mouth
deposition in simplified models.
Renotte et al. [97] simulated time-varying three-dimensional flow during quiet breathing in an anatomically representative model of the human
larynx with a pseudo-time-varying glottic aperture. The RANS-equations
were solved in combination with a RNG k - turbulence model. Only minor
differences between inspiration and expiration profiles have bee outlined. A
double pair of counter-rotating vortices develop shortly after the glottis and
merge to a single pair at 25mm after the glottis.
68
Chapter 7
PIV of the Flow in a Model of
the Upper Human Airways
In this chapter the development of a computer generated geometry of the
upper human airways and the step by step creation of a physical transparent
phantom is described. This is followed by a description of the experimental
setup and method. Detailed results of the experiments are presented and
physical phenomena are explained.
7.1
7.1.1
From the scans obtained on the five subjects, one representative image of
upper and tracheal airway anatomy was selected by the pulmonary radiologist and pneumologist. The multi-slice CT images were imported and a raw
upper and tracheal airway 3D geometry was constructed by triangulation
(Amira, Mercury Computer Systems, Chelmsford, Massachusetts). Figure
7.1 shows the different structures (skin, bone, lungs and the realistic upper
airway model)obtained from the CT-scans.
geometry was derived from this raw geometry. In this way the critical features of the flow field, like shape and location of tracheal jet, recirculation
zones and maximum velocity were preserved. The main reason for smoothing the geometry was to facilitate the creation of the physical model. On
Figure 7.2 the magnitude of velocity and streaklines of the velocity in a central sagittal plane of both realistic (left) and smoothed (right) upper airway
geometry are presented. As can be seen most physical features, like the
shape of mouth cavity, position of the trachea and the epiglottis valve were
maintained in the simplification of the model. The uvula was not withheld
in the smoothed geometry because of the extra complication in the physical
model creation. From the preliminary CFD-simulation, the influence of the
uvula was negligible.
Figure 7.2: Comparison of flow field in realistic (left) and simplified geometry (right)
7.1.2
The use of PIV as experimental method implies the model of being optical
accessible. From the non-scaled computer designed model a physical model,
using Stereo-lithography was created by Materialise NV (Leuven, Belgium).
Stereo-lithography is a common rapid manufacturing and rapid prototyping
technology for producing parts with high accuracy and good surface finish.
With a UV-laser a liquid UV-curable photopolymer resin was solidified
layer by layer until the physical model was created with a standard accuracy
of 0.2%.
A 2-block non-transparent silicone mold was fabricated around the stereolithographic model, which was, as already mentioned created by Materialize.
The reasons for choosing this type of mould material was two-fold:
other commonly used mould-materials like plaster are brittle. In the
epiglottic region a major undercut is located and removing the stereolithographic model would damage the model or the mould. Both situations are not desirable. The silicone mold was deformable, hence
the model could be easily removed without damaging the mould or
model.
Silicone molds are very accurate
First an L-shaped mold container was constructed and on the bottom of
this container plasticine (clay) was put. The stereo-lithographic model was
then suspended into the stereo-lithographic model in such a way that the
level of the plasticine reached till central sagittal plane of the STL-model
as can be seen on Figure 7.4. Liquid silicone was poured into the container
and after twelve hours it was completely cured (hard). Before the other half
of the mold was poured, the plasticine had to be removed. On the cured
part a very thin layer of grease was put. This way, the two mold halves
could be easily separated after they were cured. After the application of
this thin layer the liquid silicone was poured on the already cured silicone
part. After 24 hours of curing, the second half was hardened and STL-model
was removed without damaging any parts of the mold.
After re-assembly of both mold halves, the void where the original stereolithography part was originally located was filled with a low-melting point
alloy MCP 70 (Mining & Chemical product, Wellingbourough). In the
72
plasticine, shown on Figure 7.4 black wires are placed. This is done in
such a way that when the complete silicone model was finished, several air
ducts would be present. These ducts make the excessive air to escape, when
the metal alloy is poured into the mold. The holes in this plasticine will
form the so-called keys, when the first half of the silicone is poured on
the plasticine. In the second half, dents will occur at similar locations. This
makes it possible to correctly place the two mold halves together.
The surface quality of the new cast replica was not yet suited for making the transparent silicone negative and had to be polished. The metal
positive prototype was suspended in a transparent Perspex box (figure
7.5. The box had flat sides to minimize optical distortion during the PIVrecordings stage. The distance from the model to the edges of the box had
to be thin enough that the scattered light was not overly attenuated, but
not to thin that pressure in the flow passage would damage or deform it.
Typical spacing was about 1.5 cm in this application.
Care had to be taken when mixing the curing agent and base of the
transparent silicone (Silicone Elastomer Sylgard 184, Dow Corning, USA)
to insure homogeneous properties while minimizing air entrainment. The
thoroughly mixed silicone was exposed to a vacuum (under-pressure) to remove air bubbles and dissolved air. Excessive exposure to vacuum may also
remove the curing agent thereby slowing the curing process. The silicone
was then carefully poured into the Perspex box to minimize air entrainment. After the silicone had cured, the low-melting point alloy prototype
was removed by submerging it in hot (boiling) water. The remaining metal
residuals had to be manually removed.
In Hopkins et al. [60] an alternative method of replicating transparent
model of arbitrary geometries, suitable for flow diagnostics with PIV is described. A water-soluble cornstarch model of a human nose was generated,
using rapid prototyping. The cornstarch model is very porous and had to
be protected from the silicone penetrating the model by five thin layers of
water-soluble glue. The following steps were similar to the ones described
in the previous paragraphs, except for removing the model from the silicone
block, which could be removed with cold water. This method is less time
consuming than the one described and applied in this dissertation. However
this option was not chosen because no company in Belgium could produce
such a cornstarch model and letting it deliver from outside Belgium was
impossible due to the difficulty of shipping such a brittle cornstarch model.
73
74
76
7.2
7.2.1
Flow measurements
The experimental-set-up
U1 D1
U2 D2
=
= Re2
1
2
(7.1)
air
Qmixture
(7.2)
mixture
where Q is the volumetric flow rate.
By the use of this Reynolds similitude method, four different inhalation
flow rates (10, 15, 30 and 45 l/min air flow rate) were measured in a central
sagittal plane. The corresponding Reynolds numbers are represented in
table 7.2.1
Flow rate
Reynolds number
10 l/min
676
15 l/min
1014
30 l/min
2028
45 l/min
3042
Table 7.1: The measured air flow rates and corresponding Reynolds
numbers
PIV set-up
Figure 7.7 shows a schematic diagram of the experimental set-up. A New
Wave MinilaseII Nd-Yag laser (532 nm wavelength, 100mJ/pulse) was synchronized with a pulse separation, depending on the flow rate: the pulse
separation was chosen in such a way that the reflection of the tracer particles (10 micro-meter hollow glass spheres) shift 5 pixels between an image
pair. The laser beams were combined and formed into a sheet with cylindrical optics. This pulsed sheet was passed through the model, parallel
to the flow, and the light scattered from the particles was recorded with a
PCO sensicam QE 5Hz camera.
The fluid was pumped in a reservoir placed approximately 1.5 meter
above the model in order to create a developed velocity profile. The reservoir had an inlet, which was connected to the pump, an outlet connecting
the model and an overflow exit, which guaranteed a constant level in the
78
Figure 7.6: Phantom of upper human airway model with the glycerine/water mixture in the pharynx
reservoir. The outlet of the reservoir was separated from the inlet and overflow exit by a fine maze to stabilize the level and to remove any fluctuations
caused by the pump. The flow rate was regulated by a valve between 10 and
45 l/min air flow rate, placed behind a flow meter (type VKM710800R20 of
KOBOLD Instrumentation NV/SA with accuracy of 4 % f.s.) placed well
behind the model.
Method
Approximately 4000 image pairs at the higher flow rates (30 and 45 l/min)
and approximately 2000 pairs for the lower flow rates (10 and 15 l/min)
were recorded. The images were analyzed using PIVview 2C (PIVTEC
GmbH, Gttingen Germany). The vector fields were generated using crosscorrelation fast Fourier transform (FFT) with a multi-grid procedure combined with a sub-pixel based image shifting or image deformation with a
third order interpolation scheme [101]. The final interrogation region was
79
Figure 7.8: an image pair (a. image 1, b. image 2) with the obtained
correlation coefficient (c)
error in the displacement estimation for a given interrogation region using a
cross-correlation FFT. The multi-grid algorithm combined with image deformation reduces the uncertainty and the influence of the measurement
noise significantly [103], [101]. The typical interrogation region parameters
were:
Average particle image diameter: 2 to 4 pixels
An average pixel displacement: 5
Number of particles: 10-15
According to Raffel et al. [96] and Scarano [101], this results in a RMS
random error less than 0.02 pixels (1 % of the average displacement). On
Figure 7.8 an image pair (7.8a: image 1, 7.8b: image 2) is shown together
81
Figure 7.9: comparison of results obtained with evaluating 3000 and 4000
image pairs; left: normalized magnitude of velocity at 1 tracheal diameter
downstream the glottis for an air flow rate of 45 l/min; right normalized
turbulent kinetic energy at 1.5 tracheal downstream the glottis for the same
flow rate
7.3
Results
In Figure 7.10 the contour plots and streaklines of the normalized (with
the average inlet velocity based on the flow rate) velocity magnitude are
represented for 15 l/min and 30 l/min air flow rates. At the left the results
for 15 l/min and on the right for 30 l/min are shown. A streak line represents
the path a massless particle would follow in the 2-D velocity field described
by the PIV measurements. At first sight the flows are very similar in both
cases. The flow is characterized by three major zones of recirculation: in
the mouth, the pharynx and the trachea. The two recirculation regions in
the mouth are caused by the sudden change in flow direction from the inlet
tube to the mouth and high curvature of the mouth. The flow impinges on
82
occurs under the epiglottis valve. At the posterior side of the end of the
laryngopharynx the flow stagnates and forms a small recirculation zone.
The so-called laryngeal jet impinges on the anterior side of the trachea
and follows this wall to the end of the trachea. At the posterior side of the
trachea a huge separation bubble is formed, which extends downstream beyond the limit of measurements. The shape of the streaklines violate the
two dimensional continuity which could indicate the presence of truly threedimensional flow. Visualization with air bubbles showed very irregular behavior of the bubbles. The tracheal flow/jet has been numerically [68], [97],
[74], [19] and experimentally [27], [53]investigated by several researchers. In
the numerical study of Katz and Martonen [68] and Kleinstreuer and Zhang
[74], the formation of this jet is described to be in the center of the trachea.
In the experimental study of Corcoran and Chigier [27] and the numerical
study Renotte et al. [97], the jet was found to be at the posterior side of
the trachea while Heenan et al. [53] claimed that the jet is formed at the
anterior side of the trachea. In Brouns et al. [19] a numerical investigation
showed that the location of the jet depends on the geometry of the glottis
and the overall mouth-throat geometry.
In Figure 7.11 a detailed view of the streaklines of velocity in the mouth
at 10 l/min (a), 15 l/min (b), 30 l/min (c), 45 l/min (d) air flow rate is
shown. Looking at the recirculation zone, a difference can be observed between quiet breathing (10 and 15 l/min) and normal breathing (30 and 45
l/min). At the higher flow rates a double vortex structure occurs where
at the lower flow rates only a single vortex appears. Also more streaklines are broken, which could be pointing to a non-negligible third velocity
component leading to a non conservation of the two dimensional continuity
equation. Thus, at higher flow rates the flow in this region could be more
three dimensional.
A more detailed view of the streamlines in the pharynx for all flow
rates can be seen on Figure 7.12. Again a clear difference can be observed
between quiet breathing (10 and 15 l/min) and normal breathing (30 and 45
l/min). The streaklines at the posterior side of the oropharynx for the lower
flow rates seem to violate the 2-D continuity (broken streaklines) while
at the higher flow rates the vortex appears to be two-dimensional. The
separation bubble is a little bit smaller in case of 45 l/min compared to 30
l/min. Heenan et al. [54] reported a small growth of the separation bubble
with increasing flow rates. The mean reattachment length of the pharynx
84
Figure 7.13: Cartesian plot of the normalized axial velocity in the pharynx
for 10, 15, 30 and 45 l/min air flow rate
Figure 7.14 shows the velocity magnitudes normalized by the inlet plug
velocity for 15 and 30 l/min at half, one, two and three tracheal diameters
downstream the glottis. As can be seen the normalized peak velocity at 15
l/min air flow rate is higher than the one at 30 l/min. The shape of the
tracheal jet is also different; the jet at 15 l/min shows two peaks while the
jet at 30 l/min air flow rate consists of one broad peak.
The laryngeal jet expands more slowly at the lower flow rates compared
to the higher two flow rates. Similar observations were made by Corcoran et
al. [27] in the phase Doppler study of the laryngeal jet. As can be seen the
normalized magnitude of velocity at the beginning of the trachea is higher
for 15 l/min compared to the magnitude of velocity at 30 l/min. This was
also observed in the hot wire measurements of Johnstone et al. [66].
2
Turbulent Kinetic Energy k( ms2 ) is the mean kinetic energy per unit mass
87
Figure 7.14: Cartesian plots of the normalized velocity at different locations in the trachea( 0.5, 1, 2 and 3 tracheal diameters downstream the
glottis) for 15 l/min and 30 l/min air flow rate
associated with eddies in turbulent flow. The normalized turbulent kinetic
energy, knorm based on the two measured velocity components is defined by:
knorm
2
u2rms + vrms
= 0.5
2
Uinlet
(7.3)
2
where u2rms and vrms
represents the fluctuating components of velocity
2
and Uinlet is the inlet plug flow velocity . As only two of the three components are measured the turbulent kinetic energy is underestimated. However
some interesting conclusions can be drawn. In Figure 7.15 color plots of the
normalized kinetic energy are shown for the four measured flow rates. The
regions of high turbulent kinetic energy, which also has been reported by
Heenan et al. [54], can be found for all flow rates in the separated shear
layer of the tracheal jet. At the lowest flow rate almost no signs of turbulence are visible only at the end of the trachea a zone of higher kinetic
88
energy is visible. At 15 l/min air flow rate the zone of higher normalized
kinetic energy is situated a little higher compared to the lowest flow rate (10
l/min). At even higher flow rates (30 and 45 l/min air flow rate) this zone
is located at the beginning of the glottis. At these flow rates some increase
in kinetic energy is visible in the shear layers in pharynx and even in the
mouth. For the lower flow rates the flow is more or less laminar from the
mouth to the larynx. This could be a possible explanation of the differences
in velocity streaklines and profiles of 10 and 15 l/min on one side and 30
and 45 l/min on the other side. Corcoran and Chigier [27]found in their
Phase Doppler study of the human larynx and trachea that lower Reynolds
number cases have larger regions of higher turbulence intensity, which is
contrary to what is found in the present experimental study.
In Figure 7.16 the Cartesian plots of the normalized turbulent kinetic
energy at four different locations in the trachea are shown for all four measured flow rates. At a half tracheal diameter downstream the glottis, the
normalized turbulent kinetic energy is the highest for 45 l/min air flow rate
but further down the trachea the normalized kinetic energy at the airflow
rate of 45 l/min becomes equal and even lower than the one of 30 l/min
air flow rate. For the higher flow rates (30l/min and 45 l/min) the peak
in turbulent kinetic energy can be observed starting from the posterior side
(0.5D) and moving towards the center (1D), and continues moving towards
the anterior side while the amplitude reduces. For the lower flow rates (10
l/min 15 l/min), almost no amplification in turbulent kinetic energy is seen
till two tracheal diameters downstream the glottis. However a clear transition in flow is seen due to amplification in turbulent kinetic energy levels
for both flow rates at three diameters downstream.
89
91
7.4
Conclusion
92
Chapter 8
CFD of the Flow in a Model of
the Upper Human Airways
In this chapter the results of RANS simulations in a model of the upper
human respiratory tract are discussed and compared with the experimental
results of chapter 7. Different turbulence models of an available commercial
code are applied in the simulations of the fluid flow for four different flow
rates (10, 15, 30 and 45 l/min). Qualitative and quantitative comparison
with experiments is performed.
8.1
8.1.1
Method
Grid
An unstructured hexahedral computational mesh was generated with Hexpress (Numeca international, Brussels, Belgium). Due to the complex structure of the airway model, an unstructured mesh generation was chosen in
order to create a high quality mesh. The advantage of unstructured grids
is the fast generation of the grid. Structured grids take much more time
to generate but require less computational time. In the near-wall region a
structured very dense mesh was generated. The thickness of these near-wall
cells were chosen to contain the viscous sub-layers and to fully resolve all
geometric features. Figure 8.1 shows a three dimensional view of the grid
in the central sagittal plane. In the insert the dense boundary layer cells
are shown. The cells displayed in Figure 8.1 appear to be triangular but
93
Figure 8.1: Three dimensional view of the grid in the central sagittal plane
with a detailed view of the boundary layer cells
this is merely a visualization artefact. The k- turbulence model required
a value of y + 1 in the first grid point near the wall. The dimensionless
wall distance y + is defined by:
y+ =
u y
(8.1)
with the local kinematic viscosity, y the distance to the nearest wall and
u being the friction velocity which can be defined by
r
w
u =
(8.2)
where w is the wall shear stress and is the density. This criterion is
strictly maintained for all computations.
94
8.1.2
Numerical Method
The segregated solver of Fluent uses a finite volume method, where the governing partial differential equations are integrated over each control volume
to convert these equations into algebraic equations. This can be illustrated
by considering the steady-state conservation equation for the transport of
a scalar quantity , which is integrated over a finite volume V :
I
I
Z
v .d A = .d A +
S dV
(8.3)
V
where
= density
= velocity vector
=
=
=
=
X
f
Nf aces
X
f v f f . A f =
()n . A f + S V
(8.4)
where
Nf aces
f
Af
f
v f.Af
()n
V
In Fluent the discrete values for the scalar are stored in the cell centers,
but for the convective terms face values of the scalar are needed. To obtain
96
these face values the cell center values have to be interpolated. In this
dissertation a second-order upwind scheme was applied for the convective
terms of the momentum equation and a third-order MUSCL (Monotone
Upstream-Centered Schemes for Conservation Laws) scheme [78], which is
a combination of a central differencing scheme and a second-order upwind
scheme. It is more accurate in space than the second-order upwind scheme
for all types of meshes.
After discretization of the momentum equation the pressure at the faces
has to be interpolated. The standard or Rhie-Chow pressure interpolation scheme [99], used in this dissertation uses the momentum equation
coefficients to interpolate the face values for pressure.
In the sequential procedure, the continuity equation is used as an equation for pressure. But pressure does not appear explicitly in the discretized
continuity equation for incompressible flows. In this work the SIMPLE
(Semi-Implicit Method for Method for Pressure-Linked Equations) is used
to introduce pressure into the continuity equation.
The linearized form of the non-linear, with respect to the unknown variables, discretized scalar transport equations can be written for each cell in
the grid. This results in a linear system of algebraic equations, which is
solved with a point implicit (Gauss-Seidel) linear equation solver in combination with an algebraic multigrid method.
Because of the nonlinearity of the set of equation being solved, it is
necessary to control the change of the transported scalar. This is achieved by
under-relaxation, which reduces the change of the scalar. Under relaxation
factors used in the computations were 0.3 for pressure, 1 for density, 0.7 for
momentum, 0.8 for turbulent kinetic energy and dissipation rate and 1 for
turbulent viscosity.
When solving the RANS-equations combined with a two-equation turbulence model for a sub-sonic flow, an estimate for the turbulent kinetic energy and turbulent (specific) dissipation, besides the necessary inlet velocity
components, have to be given for inlet boundary conditions. A well-known
rule of thumb for estimation of the turbulent kinetic energy is given by:
3
kinlet = (Iinlet uinlet)2
2
(8.5)
where Iinlet is the inlet turbulent intensity usually taken as 5% for internal flows and uinlet is the mean inlet velocity. An estimation for the inlet
97
inlet =
c k 3/2
l
(8.6)
(8.7)
All simulations are performed using air as carrier gas with a density
kg
5 m2
of 1.2 m
.
3 and a kinematic viscosity of of 1.57e
s
Computations took approximately 5-8 hours on a transtec AMD Quad
Opteron Server with four AMD dual core Opteron 880 (2.4 GHz) processors
and 32 GB ECC DDR400-RAM.
8.2
Simulations were performed with inlet flow rates of 10, 15, 30 and 45 l/min
for all available turbulence models in Fluent. First qualitative comparison
between three turbulence models and the experiment is presented in figures
8.4,8.5. Contour plots of the normalized magnitude of velocity (left), based
on two velocity components, and the normalized turbulent kinetic energy
(right) of the results obtained with the k- sst (a), k- realizable (b) and
the Reynolds Stress Model (c) are shown for inlet flow rates of 15 l/min
and 30 l/min respectively. Panel d of both figures represent the contour
plots obtained with experiments. It has to be noted that the maximum
values of normalized turbulent kinetic energy differed too much between
experiments and simulations Therefore each contour plot of turbulent kinetic energy comes with the respective legend in both figures. The legend
for the normalized velocity shown in panel a of both figures is valid for all
panels.
As can be seen all simulations show a similar velocity field, which more
or less resembles the one obtained with experiments but when looking in
detail differences can be found. The streaklines in the mouth at the location
of the epiglottis are similar for all turbulence models and experiments. The
closed streaklines, at 30 l/min inlet flow rate in the posterior side of the
pharynx, as shown on figure 8.6 are more or less predicted by the Reynolds
Stress Model and the realizable k- turbulence model , while the k--sst
turbulence model predicts a more three-dimensional flow, as explained in
chapter 7. At 15 l/min the three-dimensional streaklines at this location,
as shown on figure 8.7 are well captured by the Reynolds Stress Model and
the k--sst turbulence model. The streaklines of the k- turbulence model
show a more two-dimensional structure.
Figure 8.8 shows a zoom of the streaklines in the trachea at 15 L/min.
At the posterior side of the trachea the streaklines for this flow rate show a
99
100
101
102
small vortex structure embedded in the big structure for the Reynolds Stress
Model, while the streaklines of the experiments and the results obtained
with the other turbulence models do not show this double vortex structure.
Similar structures are visible at a flow rate of 30 L/min. The streaklines of
the k--sst turbulence model resemble the most the streaklines obtained in
the experiment.
As already mentioned the turbulent kinetic energy obtained with the
experiments is based on only two velocity components, while in the RANSsimulations the three-dimensional turbulent kinetic energy field is modeled. Therefore, differences in magnitudes of the turbulent kinetic energy
field can appear. At an inlet flow rate of 15 l/min the location of the area
with maximum turbulent kinetic energy in the trachea is not predicted by
any turbulence model. All models predict this area to start at the posterior
side of the glottis, while in experiments this area is more shifted towards
the middle part of the trachea. At an inlet flow rate of 30 l/min the region
of maximum turbulent kinetic energy is located in the shear layer of the
tracheal jet for all turbulence models and starts at the posterior side of the
trachea. The shape of this region is predicted best by the Reynolds Stress
turbulence model and the k--sst model, while the k- turbulence model
appears to smear out this region. The turbulence levels in the mouth and
pharynx are overpredicted by all turbulence models for a flow rate of 15
l/min. In the case of 30 l/min the turbulent kinetic energy obtained with
the k--sst model in the mouth and pharynx seems to be very similar to
the one found in experiments. It has to be remarked that this is only a
qualitative comparison. The Reynolds Stress turbulence model and the k-
realizable model, in a lesser degree over predicts the turbulence levels in
mouth and pharynx relative to the maximum level in the shear layer of the
tracheal jet.
Figures 8.9 and 8.10 show the velocity profiles obtained with the PIV
measurements and with all turbulence models for 15 (8.9) and 30 l/min
(8.10) at 5mm above the epiglottis (a), one (b) and three (c) tracheal diameters downstream the glottic aperture. All turbulence models have difficulties to predict the velocity profile in the pharynx (Figures 8.9a and 8.10a)
for both flow rates. At the anterior side of the pharynx the models predict
a recirculation zone, which can be seen in the dip of the velocity profiles
around 0.8, where no recirculation is observed in the measurements. The
magnitude of the velocity in this recirculation given by the k- is much
104
smaller than other models. The Reynolds Stress model even predicts a
width of the recirculation zone of 25% and 20% of the posterior to anterior
distance of the pharynx.
At approximately 15 % of the posterior to anterior distance of the pharynx (Figures 8.9a and 8.9b) the k- turbulence models give a small velocity
peak at an inhalation flow rate of 15 l/min, while in the experiment no peak
can be observed. Other turbulence models do not predict such a raise in
velocity for this flow rate. At 30 l/min, however a peak is visible in the
experimental results and both k- models are able to predict this peak.
The maximum velocity is reasonably well predicted by all models. At an
inhalation flow rate of 15 l/min the k- models give the correct maximum
velocity, while at 30 l/min the other models score better. The profile of the
pharyngeal jet given by the k- turbulence models at a flow rate of 15 l/min,
is flatter than the measured profile. This can indicate an the overestimation
of the turbulence levels by these models. From both k- turbulence models,
the sst gives the best prediction of the velocity profile in this region.
The velocity profile in the trachea is best predicted by the k- sst turbulence model. Certainly for an inhalation flow rate of 15 l/min the differences
between the turbulence models are clearly visible. At one tracheal diameter
downstream the glottis the magnitude of the tracheal jet is well predicted
but the two peaks at 15 l/min do not appear in the results of the simulations. Further downstream, at three tracheal diameters downstream the
glottis the difference between the k- turbulence models and the other models is more pronounced. The shape of the velocity profile obtained with the
experiment at a flow rate of 15 l/min is almost similar to the one from simulations with the k- turbulence models. However the tracheal jet is located
more towards the center in the PIV-measurements than in the simulations.
At a flow rate of 30 l/min the k- sst turbulence model over predicts the
maximum of the tracheal jet, while the standard k- turbulence model gives
more or less the correct maximum. The shape of the velocity profile at one
and three tracheal diameters downstream the glottis is best predicted by
the k- sst turbulence model for both flow rates.
To the authors knowledge no in depth comparison between experimental
measurements and RANS simulations in a human upper respiratory tract
is available in the literature. Heenan et al. [54] showed mainly a qualitative
comparison between simulations using the standard k- turbulence model
with a low-Reynolds number modification (CFX) and PIV experiments in
105
a simplified model of the upper human airways. This was probably due to
the use of endoscopic PIV, which resulted in a dotted flow field. Only for
the velocity field at the entrance of the pharynx a more detailed comparison
was given. The main conclusion was that this turbulence model was able
to capture the main features of the fluid flow but did not capture well the
increased viscous effects at lower Reynolds numbers.
Allen et al [15] validated the use of the low Reynolds number version
of the SST k- turbulence model (Fluent) with measurements at the flow
rate of 18 l/min in a model of the trachea derived from the cadaver of a
female, aged 84. This model was able to capture sufficiently the core of
the laryngeal jet and recirculation zone in the trachea. This is similar to
the findings in this dissertation but no measurements were available in the
pharynx, where the complexity of the fluid flow was even higher.
Zhang and Kleinstreuer [142] proposed a Low Reynolds number modification of the turbulent viscosity of the standard k- model. The turbulent
viscosity is expressed as:
t = 0.09exp(3.4/(1 +
k 2
) )
50
(8.8)
Figure 8.9: comparison of velocity profiles for all tested turbulence models
with experiment at 5 mm above the epiglottis (a), one (b) and three (c)
tracheal diameter downstream the glottis for 15 l/min
with all turbulence models and experiments at 15 l/min (left) and 30 l/min
(right) inhalation flow rate. It has to be remarked that the turbulent kinetic
energy obtained with the experiments is based on two velocity components,
while as already mentioned the turbulent kinetic energy modeled with the
turbulence models is based on the three components. At 15 l/min all turbu107
Figure 8.10: comparison of velocity profiles for all tested turbulence models with experiment at 5 mm above the epiglottis (a), one (b) and three
(c) tracheal diameter downstream the glottis for 30 l/min; same legend as
figure 8.9
lence models, when compared to the levels obtained with the experiment,
clearly overpredict the turbulence levels in the trachea. At two tracheal
diameters downstream the glottis. At 30 l/min the profiles of the simula108
tion are much closer to the profiles of the experiments. The shape of the
velocity profile obtained with the k- turbulence models resembles the best
the experimental profile. However at one tracheal diameter downstream
the glottis the peak is located more toward the center than the peak of the
measured turbulent kinetic energy. The measured velocity profile of the
tracheal jet is smaller than the simulated one, making the shear layer of the
tracheal jet occur more towards the anterior side of the trachea. The region
of high turbulence is located in the shear layer of the jet.
Heenan et al. [54] only showed a contour plot of the normalized RMS velocity obtained with PIV, but did not compare this result with the obtained
turbulent kinetic energy obtained with the numerical simulations. Allen et
al. [15] found that the turbulent intensity is generally predicted well by the
low Reynolds Number sst k--model at three diameters downstream but
overpredicted at half a diameter downstream the glottis. The centerline
turbulence intensity in a locally constricted conduit at the relatively high
Reynolds number of 2000 computed with the LRN k- turbulence model by
Zhang et al. [142] showed quite good agreement downstream the constriction but it also predicted an elevated level in the constriction itself, which
was not reported by Ahmed and Giddens [14].
Figure 8.12 shows a cross-sectional view of the streaklines in the mouth
(a), pharynx (b) and at two tracheal diameters downstream the glottis (c).
On the left panel of the figure the exact location of the cross-sections are
shown. A double counter-rotating vortex pair is located in the middle part
of the mouth. The secondary motion caused by strong curvature in bend
channels is the so-called Dean flow. The Dean number, used in momentum
transfer in general and curved channels simulations in particular is normally
defined in the following form:
Re.centrif ugalf orce
Dv L 1/2
De =
=
inertialf orce
2R
(8.9)
human airway model, first described by Stapleton et al. [115] has a very
big mouth section with the inlet tube aligned with the tongue, while in the
present geometry the mouth is much smaller and an inlet tube which is more
directed towards the tongue. Kleinstreuer and Zhang [74] also reported a
single vortex pair in the mouth region.
Further downstream, in the pharynx a complex secondary motion is
present. At the posterior side the counter-rotating vortices occur, which is
caused by the backward facing step when entering the pharynx. The two
small vortices at the sides are probably caused by the upstream effect of the
epiglottis.
In the trachea a single vortex pair appears at the posterior side. The
fluid, coming from the laryngeal jet flows along the walls from the anterior
to posterior side and returns to the anterior side around the central symmetry plane. Heenan et al. [54] also reported such secondary motion, while
Renotte et al.[97] describe a double vortex pair. Brouns et al. [19] found
that the shape of the triangular shape glottis cause a double vortex pair to
occur, while the upstream geometry influences the location of the secondary
motion and tracheal jet.
Figure 8.13 gives a qualitative three-dimensional view of the streamlines
in the upper airway geometry. Streamlines represent the path, a mass-less
particle follows when released at the inlet. As can be seen part of the flow
impinges on the epiglottis and is diverted towards the side of the pharynx
and leaves the pharynx through the pharyngeal jet.
110
Figure 8.11: comparison of turbulent kinetic energy profiles for all tested
turbulence models with experiment at one (a and c) and two (b and d)
tracheal diameters downstream the glottis for 15 (a and b) and 30 (c and
d) l/min
111
112
113
8.3
Conclusions
To the authors knowledge, this is the first elaborate comparison in literature of different commercially implemented RANS turbulence models with
experimental results at different sections in an upper human airway model.
Velocity profiles in the trachea are best predicted by the low Reynolds
number sst k- turbulence models. In literature, low Reynolds number k turbulence models seem to predict quite well the velocity profiles in the
trachea [15] or in a trachea shaped geometry [142]. However, in the pharynx
the velocity profile of the pharyngeal jet differs from the measured profile.
For the higher flow rates the turbulence kinetic energy is quite well
predicted but for lower flow rates the turbulence models overpredict the
turbulent kinetic energy. It is expected that the deposition of particles is
mainly dominated by the inertial effects. The overprediction of turbulence
kinetic energy will probably only have influence on the smallest particles.
Therefore, using RANS in combination with the low-Reynolds Number SST
k- turbulence model is a good compromise for the calculation of the fluid
phase in the upper human airway model, because LES or DNS, certainly in
combination with the particle phase is still too time consuming.
114
Chapter 9
Numerical Particles Deposition
Study in a Model of the Upper
Human Airways
In this chapter the results of particle simulations in a model of the upper
human respiratory tract are presented and compared with the experimental
results available in the literature. Different influences, like gravity, carrier
gas, ... on the particle deposition are discussed.
9.1
Introduction
Inhaled medications have been available for many years for the treatment
of lung diseases and are widely accepted as the optimal route of administration for the first-line therapy for asthma and chronic obstructive pulmonary
diseases. The advantage of pulmonary drug delivery through inhalation
has recently led to the development of a series of new aerosol medication.
For some medications, this kind of drug administration is chosen because
it offers topical treatment of a specific lung condition while limiting the
whole-body effects.
To be effective, the alveolar zone of the respiratory tract has to be
reached but the extra-thoracic airways with its complex structures act as
a filter. In order to improve the delivery of aerosolized medication to the
alveolar zone, where the drugs are most effective, it is necessary to first
115
understand the mechanisms of aerosol transport and deposition in the pulmonary airways and more specific the upper airways.
9.2
Method
(9.1)
this near-wall correction [145]. Zhang et al. [139] suggested using the nearwall correction up to a dimensionless distance from the wall of y + = 100
for a flow rate of 60 and 90 l/min and y + = 20 for flow rate of 30 l/min.
This suggestion of using different parameters in the simulation for different circumstances (e.g. flow rate) makes this method cumbersome and not
straightforward to use. As already explained above, this method is based on
data of a channel flow, but the flow in the upper human airway geometry is
dominated by drastic geometrical changes (e.g. 90o bend, backward facing
steps, ...). For all these reasons, the method of near-wall correction was not
used in this dissertation.
A particle inlet plane with uniform distributed cells was created. From
each cell, 5 particles were injected from the cell center. As mentioned in
chapter 8. Particles injected at the proximity of the walls, neighboring the
inlet plane have more chance to deposit on this wall. There were little
differences in percentage when injected from the inlet plane, where the
cells are clustered towards the wall. Since uniform distribution is closer
to what happens in practice, a uniform distribution of particles in space
was chosen. The diameter of standard pharmaceutical aerosols range from
1 up to 20 micrometer [56], [90]. Therefore mono-dispersed inert particles
with a diameter of 1, 2, 4, 6, 8, 10, 12, 16 and 20 micrometer and a density of
1000 kg/m3 were injected from this uniform distributed plane and tracked
throughout the upper airway geometry for four different flow rates (10, 15,
30 and 45 l/min).
The adequacy of grid resolution for particle deposition is tested by Jayaraju et al. [64] on a realistic upper airway geometry, which was the basis
for the studied upper airway geometry. The grids, used for that study were
also created using Hexpress (Numeca International, Brussels). Particle deposition in the geometry for two different grid sizes (550 000 cells and 950
000 cells) were compared and it only marginal differences on total deposition was found between the two different mesh sizes. This proves that
a resolution of 800 000 cells in the present geometry is adequate for the
simulation of the particle deposition.
The effect of total number of injected particles on the deposition percentage is also tested for nine different particle diameters in case of 30 l/min
inlet flow rate. The maximum difference is less than 1.5% (Table 9.2) and
hence 4500 particles suffice for accurate prediction of deposition percentage.
This observation is consistent with Matida et al. [85] who stated that their
117
9.3
9.3.1
Results
Validation and total deposition analysis
Lippmann and Albert [79], Stahlhofen et al. [112], [113] and Bowes and
Swift [16] measured regional deposition in the oral airway, using monodispersed radiolabeled particles. The measured data were deposition fractions
in the oral airway and consisted of inspiratory and expiratory deposition.
The inspiratory deposition can be estimated from the oral deposition, lung,
and total deposition fraction from the reported data assuming that the inspiratory and expiratory oral deposition was the same Cheng et al. [22], [23].
Figure 9.1 summarizes inspiratory oral deposition efficiency data in living
human subjects as a function of the impaction parameter d2p Q, where dp is
the aerodynamic diameter (m) and Q is the flow rate (L/min). Stahlhofen
et al. [114] proposed following equation for upper airway particle deposition,
118
1
4.2e6 (d2p Q)1.7
+1
(9.3)
(9.4)
the human upper airways. As can be seen, the numerically simulated total deposition in the upper human respiratory tract are very close to the
experimental fit 9.4, suggesting the adequacy of the used method. However, for particle impaction parameter smaller than 100, the numerically
simulated deposition is around 5%, where as the experimental fit gives a
total deposition around 1%. This difference can probably be attributed to
the difference in airway geometry, the relative position of inlet tube and
the overprediction of the turbulent kinetic energy by the applied turbulence
model.
Figure 9.2: Simulated total deposition and experimental best fit as a function of Stokes number and Reynolds number as defined in Grgic et al. [47]
As can be seen on Figure 9.4, the experimental data available in the
literature show a lot of scatter when plotted with respect of the impaction
parameter d2p Q and this scatter is generally attributed to intersubject vari120
(9.7)
Grgic et al. [48] found also a Reynolds number effect on the deposition
of particles in a simplified upper human airway model. Therefore Grgric
et al. [47] incorporated this effect by plotting the extra-thoracic deposition
against Stk.Re0.37 , where Re is defined by:
Re =
Umean Dmean
(9.8)
Expression 9.9, proposed by Grgic et al. [47], was found by applying a least
squares best fit on deposition data, obtained by using Gamma scintigraphy
and gravimetry on a selection of seven out of 80 realistic upper airway
geometries.
100
= 100
(9.9)
11.5(StkRe0.37 )1.912 + 1
Using this best representation, simulated total deposition data for 10, 15,
30 and 45 L/min are plotted in Figure 9.2, along with this best fit curve 9.9
of Grgic et al. [47].
The general good agreement with three reported experimental best fit
curves (Stahlhofen et al. [114], Cheng et al. [24] and Grgic et al. [47]) represents the validation of the applied method on the upper human respiratory
tract.
121
9.3.2
The different sites of deposition are shown on Figure 9.3. The mouth is
colored green, the pharynx red, the larynx and trachea are both yellow.
The inlet tube is colored blue, while deposition in the inlet tube is not
represented in the total deposition.
Figure 9.3: sites of deposition: inlet tube (blue), mouth (green), pharynx
(red) and larynx + trachea (yellow)
On Figure 9.4 the simulated deposition values for all nine particle diameters in the three model subparts for 10, 15, 30 and 45 L/min inlet flow
rate are shown. The deposition in the oral cavity increases with increasing
particle diameter, while deposition in the pharynx first increases and then
decreases. For a flow rate of 10 L/min the decrease starts for a particle
diameter larger than 16 m, while in the case of 45 L/min, the decrease
in pharyngeal deposition already starts for particle diameters larger than 8
m. The same behavior occurs for the deposition in the larynx and trachea.
In fact, there is almost no tracheal deposition noticeable. This behavior can
be attributed the effective filtering function of the mouth. The number of
particles, which deposit in the oral cavity becomes so high, that only a small
part of the injected particles reaches the lower localized regions. This filtering function is positive thing for the inhalation of toxic particles, which
122
will not reach the alveolar zone of lungs, where they can harm the human
subject but will leave the body through digestive system by transport of
saliva. However, this also has a drawback for the administration of medical
aerosol, which are supposed to reach the alveolar zone of the lungs in order
to be effective.
This observation agrees with the experimental study of Grgic et al. [47]
where the oral cavity accounted for the most intense deposition in six out
of the seven studied realistic upper airway casts. The numerical study
of deposition of micro-sized particles in a realistic upper human airway
geometry of jayaraju et al. [64], also has shown that the major part of
the deposition of particles occurs in the oral cavity deposition. In contrast,
numerical and experimental studies on other idealized upper airway models,
e.g., Zhang et al. [145] and Grgic et al. [48] show considerably smaller
mouth depositions. Jayaraju et al. [64] suggested the need for realistic
models of the oral cavity to reliably estimate the oral deposition. However,
the deposition in the present idealized upper human airway model shows
the same trends in deposition as the realistic model, making it also adequate
for the estimation of the mouth deposition.
Figure 9.5 shows, in addition to the deposition values in the model subparts, a more detailed view of the deposition patterns of individual particles
(1, 10 and 20 m) for three different flow rates (10, 15, 30 L/min). The
particle coordinates are projected into a two dimensional plane because a
three dimensional view of the rather complex particle deposition patterns
in the upper human airway is difficult to represent in a compact and clear
way. Along with the plotted particle coordinates, the central sagittal plane
is represented. The small particles show very low and scattered deposition
for all flow rates. The mid-sized particles (10 m) deposit, in the case of
10 L/min on the surface of the tongue, where as for an inlet flow rate of
30 L/min, particles tend to deposit on the roof of the mouth and the tip
of the tongue. This can probably be attributed to the inertial effects of the
heavier particles. With increasing flow rate more particles (10 m) deposit
at the end of the oral cavity, also called soft palate. In the pharynx, the
10 micron particles mainly deposit on the top of the epiglottis. In the case
of 30 L/min another important deposition site appears on the upper site of
the pharynx. The main deposition sites of the 20 m are similar to those of
the 10 micron particles in the case of 15 L/min. However in the case of 30
L/min, less particles tend to deposit on the tip of the tongue but deposit
123
9.3.3
Influence of gravity
Figure 9.6: Top view of the deposited 10 m particles for a flow rate of
15 L/min
In all preceding simulations of the particle deposition, the model was
positioned with the trachea in the vertical direction. This corresponds to a
125
9.3.4
Figure 9.8 and table 9.2 represent the influence of the Eddy Interaction
on the total deposition in the upper airway model for all flow rates. As
can be seen, the total deposition without taking into account the effect of
turbulence (mean flow tracking) greatly differs from the total deposition
values obtained with the eddy interaction model. For the smallest particles
(1 and 2 m) the total deposition was, in the case of mean flow tracking
almost zero, which is comparable with the experimental best fit of Grgic
et al. [47]. As already mentioned the total deposition for these particles
obtained with the EIM gave a small overprediction. In case of 10 L/min
the over prediction of the total deposition by the EIM ranges up to 10
m particles, where in case of 15 L/min the deposition of particles with a
diameter up to 4 m are over predicted by the EIM. In both cases mean
126
Figure 9.7: Total deposition in zero gravity, gravity vector under an angle
of 45 ,gravity vector under an angle of 90 , normal gravity vector
flow tracking gave better results. This can probably attributed to the over
prediction of turbulent kinetic energy by the turbulence models in case
of 10 and 15 /min. In case of 30 and 45 L/min, the deposition of the
smallest particles (1 and 2 m) is higher compared to the experimental
best fit of Grgic et al. [47]. However this overprediction of total deposition
is negligible compared to the deposition obtained by Matida et al. [85] in
case of using the standard EIM. Even in case of near-wall correction,
the overprediction in deposition is still significant compared to the present
results. For all other combination of particle diameters and flow rates, the
use of the Eddy Interaction Model gave significantly better results. The
difference in total deposition between mean flow tracking and EIM is up to
almost 40 %, in case of 45 L/min and 8 m particles.
127
9.3.5
particle
diameter
(m)
1
2
4
6
8
10
12
16
20
10 l/min
deposition(%)
mean EIM
0.64
3.12
0.97
2.77
1.38
4.54
1.97
6.17
4.02
9.93
7.04 16.87
13.83 26.43
31.88 48.79
49
70.28
15 l/min
deposition(%)
mean EIM
0.65
5.03
0.98
5.53
1.89
6.51
2.05
8.94
2.82 14.02
6.99 25.95
12.77 43.3
37.98 75.12
76.08 92.31
30 l/min
deposition(%)
mean EIM
1.47
4.61
1.58
5.11
2.2
9.05
4.27 18.09
14.42 41.89
39.83 68.61
70.44 84.72
88.96 97.31
98.68 99.58
45 l/min
deposition(%)
mean EIM
2.68
8.1
2.51
9.69
4.28 20.72
9.39 45.82
36.19 75.79
72.05 90.23
87.8 96.37
97.98 99.43
99.91 99.94
Table 9.2: Comparison of total deposition between mean flow tracking and
Eddy Interaction Model
identified a similar improvement for stable asthmatics in lung deposition of
inhaled particles delivered with Heliox. Deposition of radiolabeled 3.6 m
particles for flow rates of 500 and 1200 mL/s in the mouth, throat and tracheobronchial region was decreased with corresponding increase in alveolar
deposition. Svartengren et al. [119] also found an improved alveolar deposition when using Heliox as carrier gas as compared to air for radiolabeled
3.6-3.8 m Teflon particles for a flow rate of 500 mL/s. Habib et al. [50]
carried out an in vitro study of Heliox in a mechanically ventilated pediatric
model, using albuterol delivered by a pressurized Metered Dose Inhaler. An
increase in aerosol delivery by 8% (12% for air and 20 % for Heliox) was
seen.
Gemci et al. [43] performed numerical deposition study, using the standard k- turbulence model, which was proven to overpredict the deposition
of aerosols in the upper human airways [115], in a cadaver-based throat,
without oral cavity. Grgic et al. [47] and Jayaju et al. [64] showed that
most deposition of aerosol occur in the oral cavity and pharynx. Gemci et
al. [43] compared total deposition for monodispersed particles with diameter ranging from 0.25 to 20 m for a flow rate of 18 L/min , with Heliox
70/30 as carrier gas with results obtained in the same geometry, using air
[41]. At the smallest droplet radius of 0.25 m, for the Heliox case, 31.8 %
of the injected particle mass deposited compared to the 38.7 % deposition
129
when air delivery was used. Similar difference were found for 1 and 2 m
particles. For bigger particles the difference in deposition was negligible.
Heliox is an inert, non-toxic gas which has a similar dynamic viscosity ()
as air, but a much lower density (three times less). This lead to an almost
threefold lower kinematic viscosity (), making the Reynolds number three
times lower. The more viscous Heliox will probably cause more laminar
regions to appear in the upper human airway compared to air. Simulations
were performed for flow rates of 15 and 30 L/min using Heliox 80/20, with
2
s
a dynamic viscosity of 1.98e5 N
and kinematic viscosity of 4.95e5 ms .
m2
particle
diameter
(m)
1
2
4
6
8
10
12
16
20
15 l/min
deposition(%)
Heliox
air
0.22
5.03
0.24
5.53
0.46
6.51
0.77
8.94
2.29 14.02
5.11 25.95
8.06
43.3
30.61 75.12
63.99 92.31
30 l/min
deposition(%)
Heliox
air
0.78
4.61
0.66
5.11
2.51
9.05
7.21 18.09
22.88 41.89
40.94 68.61
53.97 84.72
82.95 97.31
98.20 99.58
Table 9.3: Comparison of total deposition for Heliox and air as carrier gas
In table 9.3 numerically computed total deposition for particles with
a diameter ranging from 1 to 20 m entrained in Heliox and air for flow
rates of 15 and 30 L/min are shown. Particles suspended into Heliox are
less susceptible to deposit in the extra-thoracic airways. For a flow rate
of 15 L/min only particles larger than or equal to 16 m, entrained in
Heliox have significantly total deposition values (< 30%). In the case of 30
L/min particles larger than 8 m, have deposition values higher than 20%.
The differences in total deposition values between particles carried by
air and Heliox can go up to 35 % (16 m at 15 L/min). These difference
are significantly higher than the ones reported by Gemci et al [43]. The
difference in geometry and turbulence model can explain this difference.
On Figure 9.9 the total deposition, listed in table 9.3 is plotted against
130
9.3.6
Figure 9.10: Scheme of the inhalation profile and particle injection for
unsteady flow accelerating through 30 L/min for FIR of 2 L/s2
taken into account. For transient simulations, the governing equations must
be discretized in both space and time. The spatial discretization for the
time-dependent equations is identical to the steady-state case. Temporal
discretization involves the integration of every term in the differential equations over a time step t. For time integration a second order fully implicit
scheme was applied. Simulation were performed with a time step of 0.1 ms.
To avoid numerical instabilities, a primary flow solution was performed for
1 L/min and thus the transient calculation started at 1L/min with the converged steady state solution. For comparison, [49] took a time step of 10 ms
and started the unsteady simulation at a flow rate of 10 L/min. The Basset
force term or a term describing the force applied on particles as a result of
pressure gradients arising from fluid acceleration were not included, because
they are expected to by negligible in case of high particle fluid density ratio
[57]. Grgic et al. [49] decided to take into account both terms.
Grgic et al. [49] found no significant difference in deposition between an
accelerating flow of 2 and 4 sL2 . Therefore is was chosen to only perform a
simulation for particles ranging from 1 to 20 m for an accelerating flow of
2 sL2 . A scheme for the inhalation profile with a maximum of 46 L/min and
particle injection profile are presented in Figure 9.10.
In table 9.4 total deposition for steady and unsteady flow with a FIR of
132
particle
diameter
(m)
1
2
4
6
8
10
12
16
20
30 l/min
steady flow
4.61
5.11
9.05
18.09
41.89
68.61
84.72
97.31
99.58
30 l/min
unsteady flow
FIR 2 L/s2
4.67
5.8
11.05
26.77
51.42
70.58
83.87
95.76
99.58
Table 9.4: Comparison of total deposition for steady and unsteady flow
with a FIR of 2 L/s2 )
2 L/s2 are shown. As can be seen the accelerating flow only has a significant
influence on particles ranging from 4 up to 8 m, with the biggest difference for the particles with a diameter of 8 m, which has a total deposition
difference of approximately 30 %. Grgic et al. experimentally found the
mouth-throat deposition of 5 m particles to be 275 % (meanstandard
deviation) and 375 % for unsteady flow accelerating through 30 L/min for
FIR of 2 L/s2 . The present numerical simulation has a difference of 2 %
for 4 m particles and almost 9 % for 6 m particles. The difference can
be probably attributed to the difference in upper airway geometry. The deposition for steady inhalation in the present upper human airway geometry
situates in the oral cavity, where the main deposition site in the simplified
geometry used in the study of Grgic et al [49] is located in the larynx. The
numerical study of Grgic et al [49] showed no deposition difference between
steady and unsteady inhalation profile for 5 m particles, where a difference
of 5 % was found for 10 m particles. This supports the adequacy of the
applied method for unsteady flow rate.
133
30 l/min
steady flow
30 l/min
unsteady flow
4.61
5.11
9.05
18.09
41.89
68.61
84.72
97.31
99.58
4.13
4.42
7.29
16.05
33.31
56.54
75.55
91.8
98.76
Table 9.5: Comparison of total deposition for steady and unsteady flow,
where the particles are released at the moment the flow rate reaches the
maximum value
Influence of particle injection time
A very popular device for administration of aerosolized drugs is the pressure
metered dose inhaler (pMDI). Today approximately 500 million pMDIs are
134
produced annually [90]. The pMDI is a small, pressurized can that contains
aerosol medicine and the propellant (CFC, HFA)
There are five parts to an pMDI: the medication, the propellant, the
canister, the metering valve and the mouthpiece. Each time, the pMDI
is used, a precise measured, or metered, amount of medicine is released
and carried by the propellant. The spray is released with a typical initial
velocity of 30 m/s. Often a spacer device, which is a large plastic container
attached to the inhaler that act as a reservoir or holding chamber, is used
in combination with a pMDI. They serve to hold the medication that is
sprayed by the pMDI and is then inhaled into the mouth. It is not clear
when the particles start entering the mouth. Until now it was accepted
that at the moment the aerosols enter the mouth, the flow is constant and
fully developed (steady flow rate). However, it is interesting to study the
influence of particle release time on the total particle deposition in the
human upper airways.
A simulation was carried out with the same parameters as mentioned
for the unsteady simulation, described in the previous paragraph. Figure
9.11 displays a schematic view of the inhalation profile with a maximum of
30 L/min and FIR of 2 L/s2 and particle injection profile. The particles
are injected at the moment when the flow rate reaches the maximum value
(30 L/min).
Comparison between total deposition values, obtained with steady flow
rate and unsteady inhalation flow are presented in table 9.5. Deposition of
particles, released immediately after reaching maximum flow rate is considerably lower (more than 6% difference) for particles with diameter ranging
from 8 to 16 m than particles injected at the moment of fully developed
flow. Particles with larger and smaller diameter also have higher deposition
in assumption of steady flow rate, but differences are smaller (approximately
2 %). Releasing particles at the moment maximum flow rate is reached has
a positive influence, when compared to releasing particles at fully developed
flow on the alveolar deposition of medical aerosols.
9.4
Conclusion
erably lower (more than 6% difference) for particles with diameter ranging
from 8 to 16 m than particles injected at the moment of fully developed
flow.
137
Chapter 10
Clinical Application: Tracheal
Stenoses
In this chapter the effect of stenosis on the pressure drop is studied. CFD
simulations were carried out in the simplified upper human airway model.
The focus is less directed towards local flow patterns (which are crucial
for local particle deposition) than towards overall pressure drops over the
airway structure (which is a determinant of the work of breathing in stenosis
patients).
10.1
Introduction
Patients with trachea airway stenosis often report a relatively sudden appearance of breathing impairment, which at the stage of admission to the
clinic, is observed when a loss of 75% or more of the airway lumen has
occurred [109]. Interventional bronchoscopic techniques such as laser or
electrocautery resection with mechanical debulking of obstructive tissue,
usually followed by airway stenting, then often have to be performed in a
relatively urgent setting. There obviously is time for a progressive increase
in tracheal stenosis before clinically significant breathing impairment is experienced by a patient, yet, when the constriction reaches a certain value,
the pressure drop suddenly becomes critical and symptoms rapidly occur.
The aim of CFD simulations of upper airway flow dynamics in a smoothed
realistic upper and tracheal airway structure, was to provide a better understanding of this clinical observation and to suggest a means of monitoring
138
patients who are at risk for development of tracheal stenosis (e.g., patients
with a history of longstanding or complicated intubation, post-tracheostomy
patients, lung cancer patients, or patients with a history of prior tracheal
stenosis treated with laser or stenting).
10.2
Figure 10.1 shows the two types of stenoses inserted in the upper airway geometry. The unstructured hexahedral mesh (Hexpress, Numeca, Brussels,
Belgium) contains approximately 650000 cells with a local mesh refinement
around the area of stenosis. A preliminary grid resolution study (using
meshes with 433000, 650000 and 1580000 cells) had indicated that the velocity profile just downstream of the stenosis was not altered by a grid
refinement above 650000 cells, and that the pressure drop between model
in- and outlet varied by less than 1% between simulations on the 650000
and 1580000 meshes.
Hexstream (Numeca international, Brussels, Belgium) solves the compressible RANS equations. In contrary to the segregated solver of Fluent
the RANS equations are not longer segregated but coupled. The governing equations of continuity, momentum, and (where appropriate) energy are
solved simultaneously as a set, or vector, of equations. When the magnitude
of the flow velocity becomes small in comparison with the acoustic speeds
(which is the case in our application) time marching algorithms designed
for compressible flows have difficulties to converge. The problems faced by
compressible codes at low Mach number flows are:
1. High disparity between the convective and acoustic eigenvalues, leading to a much too restrictive time step for the convecting waves causing
poor convergence characteristics.
2. Round off errors due to the use of absolute pressure in the momentum
equation.
3. Impossibility to treat strictly incompressible flow.
In order to overcome these problems in compressible flow solvers a technique called preconditioning is introduced. For steady state applications,
solved by time marching algorithms, the time derivatives of the unknowns
in the flow equations have no physical meaning and can therefor be altered
without altering the final steady solution. The idea of preconditioning is
multiplying the time derivatives of the dependant variables with a preconditioning matrix. This matrix removes the stiffness of the eigenvalues,
introduces reduced flow variables such as dynamic pressure and enthalpy
140
Figure 10.1: Side view of the realistic (smoothed) 3D upper and tracheal
airway model including stenosis with 3D grid refinements in the stenotic
area. Inserts are a zoom of a weblike stenosis (length 2mm) and an elongated
stenosis (length 30mm). Cross-sections A-H refer to different locations along
the model
which reduce the round-off errors and the acoustic speed c is replaced by a
1 Q
d +
ZZ
S
141
F dS =
ZZ
S
V dS
(10.1)
with
Q = (pg , u, v, w, Eg , k, )
(10.2)
velocity components, F and V are the inviscid and viscous fluxes respectively and Eg is the total gauge energy. For an incompressible fluid with
constant Cp , Eg is given by:
v2
Eg = Cp (T Tref +
)
(10.3)
2
The general form both for compressible and incompressible fluids of the
preconditioning matrix employed here is a combination of those suggested
by Choi and Merkle [25] and Turkel [123] and is given by:
1
0 0 0 0 0 0 0 0
2
(1+)u 0 0 0 0 0 0 0
2
(1+)v
0
0
0
0
0
1
(1+)w
=
(10.4)
0
0
0
0
0
0
0
2
v +Eg
2
0 0 0 0 0 0 0
0
0 0 0 0 1 0 0 0
0
0 0 0 0 0 1 0 0
The two parameters and are chosen so that the stiffness of the eigenvalues is minimized at low speed. The optimal value for was found to
be -1. The preconditioning parameter is imposed by the user through a
coefficient and a characteristic velocity Uref :
2
2
2 = max(Uref
, Uloc
)
(10.5)
with Uloc representing the local velocity. If the value of is too large,
it will introduce excessive artificial dissipation into the solution. Therefor
the value of should be chosen as small as possible. A standard central
scheme with Jameson type dissipation of second and fourth order is used
for spatial discretisation [63], and a fourth order Runge-Kutta scheme for
time integration. A full multigrid V-cycle strategy [58] with four grid levels
was used for convergence acceleration.
In the present application a low-Reynolds number Yang-shih k- turbulence model [137] is used, the accuracy of which has been previously tested
for the purpose of a similar application [19].
142
10.3
Results
Figure 10.2: Velocity streaklines in the model for 0%, 50% and 90% stenosis (panel A, B and C, respectively). Dark grey areas represent regions where
the velocity is equal or higher than 80% of the peak velocity anywhere in
the model. Inserts represent 3D streamlines in the stenotic area
Figure 10.2 illustrates the flow fields for 30 L/min in the model geometries without stenosis (panel A) and with two degrees of a weblike stenosis
(panels B and C). The degree of constriction is quantified here as the obstructed cross sectional area as a percentage of the nominal (unobstructed)
airway cross sectional area. The model insets in Figure 10.2 show selected
3D streamlines to illustrate the increasing complexity of flow field with increasing constriction. The main models in panels A-C show streaklines
(i.e.,the intersection of 3D streamlines with a central plane) and the dark
grey areas delimit areas where flows exceed 80% of peak velocity generated
anywhere in the model. For instance, in the case of 0% and 50% stenosis,
the highest velocities (5.5m/s) are essentially in the vicinity of the jet generated by the glottic constriction, whereas in the case of 90% constriction,
the peak velocities (22.5m/s) are generated at the level of the stenosis and
143
Figure 10.3: CFD simulated pressures along the model with a stenosis
of 50%, 75%, 85% and 90% (weblike stenosis; solid circles) and with no
stenosis (indistinguishable from 50% stenosis). For the 90% constriction,
an elongated stenosis was also considerd (open circles, triangles and squares
refer to 10mm, 20mm and 30mm stenosis length); inlet flow is 30 L/min
In Figure 10.3, pressure averaged over the cross sections along the upper
airway axis are plotted as a function of distance from the model inlet, for
144
Figure 10.4: Panel A: CFD simulated pressure drops over the stenosis as a
function of degree of stenosis constriction. Open and closed symbols refer to
CFD simulations with air breathing at 15 and 30 L/min flow rate; crosses
refer to Heliox breathing at 30 L/min. The line plots are corresponding
pressure drop estimates obtained by use of Equation 3 with K=1.2 for 15
L/min (solid line) and 30 L/min (dotted lines). Panel B: K values for use
in Eq.(10.7), obtained for all simulation conditions of panel A (see text for
details).
P13
X
V32
S32
Piloss
= (1 2 ) +
2
S1
i
(10.6)
146
Q2
2S22
(10.7)
By fitting the simple equation 10.7 to the CFD simulated P over the
stenosis represented in Figure 10.4A, K was found to vary between 0.4-1.5
for 15 and 30 L/min flows and for constrictions between 50 and 90% (Figure
10.4B). For the sake of simplicity, with the aim to turn equation 10.7 into
one rule of thumb which is valid for all conditions within the above range,
one single value for K which provided a good fit for the 15 and 30 L/min
curves in Figure 10.4A, was chosen. The solid and dotted line in Figure
10.4A are pressure drops, for respectively 15 and 30 L/min, obtained from
equation 10.7 with K = 1.2, plotted alongside the 15 and 30 L/min CFD
simulated data points for air and heliox.
The general form of equation 10.7, typically applied in engineering problems, considers a dependence on bulk flow with a power law and exponent
2, which is a good approximation for a rule of thumb designed to roughly
estimate a local pressure drop generated by a specific feature (stenosis) as a
function of bulk flow. In the present context, it was useful to also perform
a more detailed inspection of pressure drop dependence on flow, considering the entire upper and tracheal airway model of Figure 10.1. Figure 10.5
shows the pressure drops obtained between model inlet and outlet for different flows up to 60 L/min, in the case of no stenosis (open symbols) and of
60% and 85% constriction (solid symbols). On the represented data points
a power law ( P = aQb ) was fitted to obtain an exponent b to characterize
pressure dependence on flow. In the presence of 60% and 85% constriction,
the best power fit exponent b was 1.92 and 2.00 (R2 =1.00 for both). In the
absence of stenosis, the best fit b was 1.77 (R2 =0.999). When using only
data points below 30 L/min, b was similar (1.72; R2 =0.999) but a linear
regression forced through zero in this flow range up to 30 L/min led to a
worse fit with R2 =0.89.
147
10.4
Discussion
In the present chapter, CFD simulated pressure drops over tracheal weblike
stenoses of different constriction in a realistic upper and tracheal airway
geometry roughly follow a dependence on the squared ratio of bulk flow
to stenosis lumen cross section. This relationship predicts a dramatic and
relatively sudden increase in breathing impairment experienced by patients
once a relatively severe degree of stenosis is reached, in agreement with
what is observed in clinical practice. From this characteristic dependence
of pressure drop on increasing degrees of weblike stenosis a simple equation
to actually quantify pressure drops generated by different degrees of stenosis (equation 10.7), provided that breathing flow and stenosis cross section
can be reliably estimated, was derived. The coefficient K in equation 10.7
is expected to partly depend on the development of the velocity profile upstream from the stenosis and on the exact geometry of the constriction,
explaining slightly varying K-values between 15 and 30 L/min, and for different degrees of constriction (Fig.10.4B). However, the agreement between
the CFD simulations and equation 10.7 by use of a single K value (K=1.2;
Fig.10.4A) indicates that this rule of thumb can be adequately used for
most clinically relevant situations (slow to quiet breathing).
For the purpose of decision making on weblike stenosis treatment, pressure drops owing to stenosis can be roughly estimated by means of equation
10.7 with K=1.2, given that a non-invasive measure of the stenosis cross
section is available, for instance by an acoustic reflection technique [59].
The resulting resistance at a given flow can then be computed, and related
in terms of abnormality with respect to a reference value of upper airway
resistance. In Wassermann et al. [127], an inspiratory resistance 7-fold
greater than that measured in normal control subjects, was used as a cutoff to discriminate between stenosis patients who needed surgical treatment
or not; for this purpose Wassermann et al. [127] measured tracheal pressure drops in situ. In fact, 15 and 30 L/min simulations show that a 7-fold
increase in pressure drop would already correspond to 85-90% constriction
(Fig.10.4A). Irrespective of the preferred choice for a cut-off, the obtained
results suggests that a similar approach could be used but by estimating
pressure drops from equation 10.7, where only stenosis cross section needs
to be measured, which may be of considerable practical advantage with
respect to actually having to measure the pressure drops in situ [127].
148
Figure 10.5: CFD simulated pressure drops between model inlet and outlet
for different flows up to 60 L/min, in the case of no stenosis (open triangles),
of 60% constriction (solid squares) and of 85% constriction (solid circles).
The line plots are the corresponding best-fit power laws, leading to power
values of 1.77 (no stenosis), 1.92 (60% stenosis), and 2.00 (85% stenosis)
or nasal pathway structures. Yet, even if it is not possible to directly access pressure in situ in the case of tracheal stenosis, one could imagine that
tracheal contribution to a non-invasive measure of upper resistance would
suffice to identify a characteristic flow dependence of resistance. An increase of the best-fit power law exponent between 1.75 and 2 during a given
patients follow up, could then signal a progression towards stenosis which
impairs breathing at rest, especially because the exponent is seen to increase
to 1.92 even in the case of 60% constriction, which in Fig.10.4A corresponds
only to a preliminary stage of P increase. In any case, Figure 10.5 clearly
illustrates how resistance measurements at higher flows could be more useful as an indicator of flow limitation due to stenosis (see for instance, a 50%
increase in resistance between no stenosis and 60% stenosis when measured
at 1 L/s)
Little is known about pulmonary function measurements in stenosis patients that could actually lead to a better management of stenosis treatment.
Because of the rapid onset and progression of symptoms in most patients
with tracheal stenosis, pulmonary function testing including spirometry and
flow-volume loop analysis can rarely be obtained in patients referred in
extremis. Diagnosis severity assessment and therapeutic decision in tracheal stenosis are therefore based on history, physical examination, imaging
studies and flexible bronchoscopy. The role of pre-treatment pulmonary
function testing outside of research protocols is unclear [26]. Significant improvement in forced vital capacity, forced expiratory flow in one second and
in peak expiratory flows, forced inspiratory volumes and airway resistance
have nonetheless been demonstrated after airway stenting [125]. There is
no standardized approach in the follow-up of patients treated for, or at
risk for tracheal stenosis. Timing of follow up visits, including history and
clinical examination, imaging, pulmonary function testing, and /or flexible bronchoscopic control, should be individualized; the role of surveillance
bronchoscopy has not been established [86]. The present study suggests
that flow dependence of upper airway resistance measurement (Fig.10.5)
could provide a valuable diagnostic tool. In addition, the heliox simulations
(crosses in Fig.10.4A) also lend support to the experimental observation of
a temporary relief of breathing impairment during heliox administration as
has been suggested for use with post-intubation patients [61].
The glottic constriction of the tracheal model used here was 125 mm2 ,
which corresponds to the average 126mm2 peak value of the glottic area
150
(10.8)
152
Chapter 11
Conclusions and Future
Challenges
The goal of this PhD research work was to develop an simplified but still
realistic upper airway geometry and research the particle deposition in the
upper human airways.
Chapter 7 first describes the development of a simplified computer-model
of the upper human airways, based on the available CT-scans of a male
subject. From this computer model a suitable model for PIV experiments
was created and via Reynolds similitude, four different inhalation flow rates
(10, 15, 30 and 45 l/min air flow rate) were measured in a central sagittal
plane by the use of PIV. The obtained results were analyzed and the different
flow structures, inherent to the geometry of the upper human airways, were
discussed.
Chapter 8 describes a thorough and to the authors knowledge first elaborate comparison in literature of different commercially implemented RANS
turbulence models with experimental results at different sections in an upper
human airway model. Reported comparison of measurements with simulations using low Reynolds number k- turbulence models seem to predict
quite well the velocity profiles in the trachea [15] or in a trachea shaped
geometry [142]. However, in the pharynx of the developed upper airway
model, the simulated velocity profiles of the pharyngeal jet differ quite
strongly from the measured profile. For the higher flow rates the turbulent kinetic energy in the trachea is predicted quite well, but for lower flow
rates the turbulence models overpredict the turbulent kinetic energy.
153
in the power law that relates upper airway resistance to flow can be used
as a diagnostic tool in the non-invasive monitoring of stenosis patients.
For this dissertation 2C-2D (two components of velocity - two dimensional) PIV measurements were conducted in a central sagittal plane in the
upper airway geometry. Because of the three-dimensionality of this flow
a next step could be a 3C-2D (three component - two dimensional) PIV
experiment in the central sagittal plane. Measurements in the trachea and
pharynx using PIV in the coronal plane are much more difficult to perform because the flow of the jet perpendicular to this plane is opposite to a
part of the flow in the recirculation bubbles, making it extremely difficult
to catch the secondary vortices. This could be solved by using a three
component LDA (Laser Doppler Anemometry) system for measurements in
this coronal plane.
In this work numerically simulated particle deposition is compared with
best fits obtained with deposition experiments in several different realistic
upper airway geometries and living subjects. However every geometry leads
to a different particle deposition behavior. Therefore measurements using a
method like gamma scintigraphy could be used for determining the local and
total deposition in the developed upper airway geometry. The numerically
obtained deposition could be compared in detail with these experimentally
obtained deposition values.
Pressurized Metered Dose Inhalers(pMDI) are the most used devices to
generate the medical aerosols. The generation of aerosols by these pMDIs
is a very complex phenomenon and the existing models to predict such generation can be used the simulate the deposition in the developed geometry
and in that way predict and ameliorate the transport of aerosols through
the upper human airways.
The fast growing availability of computational power makes it possible
to use more advanced computational methods, like LES and DNS. These
methods will probably better predict the very complex flow in the upper
human airways (especially in the pharynx) and thus will be able to better
estimate the particle deposition in the upper airways. When using LES
or DNS, particles are tracked in real time together with every fluid time
step. This in contrary to the post-processing strategy when using RANS
for the simulation of the carrier fluid. The use of LES or DNS eliminates
the use of the Eddy Interaction model. Therefore the overprediction in the
total deposition of particles, with a value for Stk.Re0.37 smaller than 0.1
155
156
Bibliography
[1] Fluents User Guide ver. 6.2.
[2] http://education.yahoo.com/reference/gray/subjects/subject/242.
[3] http://www.aafa-ca.org/asthma history.php.
[4] http://www.breath2000.org/medical.html.
[5] http://www.cayuga-cc.edu/people/facultypages/greer/biol204/resp2/resp2.html.
[6] http://www.ems-ceu.com/courses/182/index ems.html.
[7] http://www.fluent.com/about/cfdhistory.htm.
[8] http://www.gbmc.org/voice/anatomyphysiologyofthelarynx.cfm.
[9] http://www.who.int/mediacentre/factsheets/fs307/en/.
[10] www.asthma.org.uk/all about asthma/publications/asthma at work.html.
[11] The European Lung White Book: The First Comprehensive Survey on
Respiratory Health in Europe. 2003.
[12] R. J. Adrian. Image shifting technique to resolve directional ambiguity
in double-pulsed velocimetry. Appl. Optics, 25:38553858, 1986.
[13] R. J. Adrian. Multi-point optical measurements of simultaneous vectors in unsteady flow - a review. Int. J. Heat Fluid Flow, 7:127145,
1986.
[14] S. A. Ahmed and P. D. Giddens. Velocity measurements in steady flow
through axisymmetric stenoses at moderate reynolds number. Journal
of Biomechanics, 16-7:505516, 1983.
157
[26] H. G. Colt. Functional evaluation before and after interventional bronchoscopy. in: Interventionalbronchoscopy edited by c. t. bolliger and
p. v. mathur. Progr Respir Res., 30.
[27] T. E. Corcoran and N. Chigier. Characterization of the laryngeal jet
using phase doppler interferometery. Journal of Aerosol Medicine,
13:125137, 2000.
[28] T. E. Corcoran and N. Chigier. A numerical and experimental study
of spray dynamics in a simple throat model. Journal of Biomechanics,
13:125137, 2000.
[29] T. E. Corcoran, B. P. Shortall, I. K. Sin, and M. P. Meza N. Chigier.
Aerosol drug delivery using heliox and nebulizer reservoirs: Results
from a mri based pediatric model. Journal of Aerosol Medicine,
6(3):263272, 2003.
[30] E. R. Corino and R. S. Brodkey. A visual investigation of the wall
region in turbulent flow. Journal of Fluid Mechanics, 37:130, 1969.
[31] C. T. Crowe, T. R. Troutt, and J. N. Chung. Numerical methods for
two phase turbulent flows. Ann. Rev. Fluid Mech., 28:1143, 1996.
[32] M. Dadi, M. Stanislas, O. Rodriguez, and A. Diment. A study by
holographic velocimetry of the behaviour of free particles in a flow.
Expiments in Fluids, 10:285294, 1991.
[33] M. Dockrell, M. R. Partridge, and E. Valovirta. The limitations of
severe asthma: the results of a european survey. 62:134141, 2007.
[34] Th. Dracos. Three-dimensional velocity and vorticity Measurering and
image analysis tecniques. Kluwer Academic Publishers, Dordrecht,
1996.
[35] Gad el Hak. Advances in Fluid Mechanics Measurements. Springer,
New York, 1989.
[36] S. E. Elghobashi and W. T. Abou-Arab. On predicting particle-laden
turbulent flows. Applied Science Research, 52:309329, 1994.
159
[69] I. M. Katz, T. B .Martonen, and W. Flaa. Three-dimensional computational study of inspiratory aerosol flow through the larynx: the
effect of glottal aperture modulation. journal of aerosol medicine,
28-6:10731083, 1997.
[70] R. D. Keane and R. J. Adrian. Optimization of particle image velocimeters. part i: Double pulsed systems. Meas. Sci Tech., 1:1202
1215, 1990.
[71] R. D. Keane and R. J. Adrian. Optimization of particle image velocimeters. part ii: Multiple pulsed systems. Meas. Sci Tech., 2:963
974, 1991.
[72] R. D. Keane and R. J. Adrian. Theory of cross-correlation analysis of
piv images. Appl. Sci. Res., 49:191215, 1992.
[73] R. D. Keane, R. J. Adrian, and Y. Zhang. Super-resolution particle
image velocimetry. Meas. Sci Tech., 6:754768, 1995.
[74] C. Kleinstreuer and Z. Zhang. Laminar-to-turbulent fluid-particle
flows in a human airway model. International Journal of Multiphase
Flow, 29:271289, 2003.
[75] J. P. Kress. The utility of albuterol nebulized with heliox during acute
asthma exacerbations. American Journal of Respiratory and Critical
Care Medicine, 165:13171321, 2002.
[76] W. Lauterborn and A. Vogel. Modern optical techniques in fluid
mechanics. Annu.Rev. Fluid Mech., 16:233244, 1984.
[77] H. Le, P. Moin, and J. Kim. Direct numerical simulation of turbulent flow over a backward-facing step. Journal of Fluid Mechanics,
330:349374, 1997.
[78] B. Van Leer. Toward the ultimate concervative difference scheme. iv.
a second order sequel to godunovs method. Jounal of Computational
Physics, 32:101136, 1979.
[79] M. Lippmann and R. E. Albert. The effect of particle size on the
regional deposition of inhaled aerosols in the human respiratory tract.
Am. Ind. Hyg. Assoc. J., 30:257275, 1969.
163
[114] W Stahlhofen, G Rudolf, and A C James. Intercomparison of experimental regional aerosol deposition data. Journal of Aerosol Medicine,
2:285308, 1989.
[115] K. W. Stapleton, E. Guentsch, M. K. Hoskinson, and W. H. Finlay.
On the suitability of k turbulence modeling for aerosol deposition
in the mouth and throat: A comparison with experiment. Journal of
Aerosol Science, 31:739749, 2000.
[116] F. Stern, R. V. Wilson, H. W. Coleman, and E. G. Paterson. Comprehensive approach to verification and validation of cfd-simulationspart 1: Methodology and procedures. Journal of Fluids Engineering,
123:793802, 2001.
[117] W. C. Su and Y. S. Cheng. Deposition of fiber in a human airway
model. Journal of Aerosol Science, 37:14291441, 2006.
[118] M. Andrson M. Svartengren, G. Bylin, M. Philipson, and P. Camner. Deposition in asthmatics of particles inhaled in air or in heliumoxygen. American review of respiratory disease, 147(3):524528, 1993.
[119] K. Svertengren, P. A. Lindestad, M. Svartengren, M. Philipson,
G. Bylin, and P. Camner. Deposition of inhaled particles in the
mouth and throat of astmathic subjects. European Respiratory Journal, 7:14671473, 1994.
[120] Z. J. Takamura, A. Yamaoka, Y. Odajima, and Y. Iikura. Altered
eosinophil levels as a result of viral infection in asthma exacerbation
in childhood. J Pediatr Allergy Immunol, 13(1):4750, 2002.
[121] J. B. Tobias and J. S. Garrett. Therapeutic options for severe, refractory status asthmaticus: Inhalation anaesthetic agents, extracorporeal membrane oxygenation and helium/oxygen ventilation. Pediatric
Anaesthesia, 7:4757, 1997.
[122] S. C Tough, P. A. Hessel, M. Ruff, F. H. Green, I. Mitchell, and
J. C. Butt. Features that distinguish those who die from asthma from
community controls with asthma. J Asthma, 35:657665, 1998.
167
170
Appendix A
List of publications
articles in scientific journals with an international referee system(1st author)
Influence of glottic aperture on the tracheal flow, journal of Biomechanics, Volume: 40, pp: 165 - 172, , 2007, Mark Brouns, Sylvia Verbanck,
Christian Lacor
Tracheal stenosis: a flow dynamics study, Journal of Applied Physiology, Volume: 102, pp: 1178 - 1184, , 2007, Mark Brouns, Santhosh Tovinakere Jayaraju, Christian Lacor, Johan De Mey, Marc Noppen, Walter
Vincken, Sylvia Verbanck
Particle Image Velocimetry in a upper human airway model, Medical Engineering and Physics, Mark Brouns, Santhosh Jarayaju, Steve Vanlanduit, Sylvia Verbanck, Bachir Belkassem,Chris Lacor (in review)
172